#include "llvm/ADT/SetVector.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/Analysis/AliasAnalysis.h"
+#include "llvm/Analysis/AssumptionCache.h"
#include "llvm/Analysis/CallGraph.h"
#include "llvm/Analysis/CaptureTracking.h"
#include "llvm/Analysis/InstructionSimplify.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/DebugInfo.h"
#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/DIBuilder.h"
#include "llvm/IR/Dominators.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/Instructions.h"
using namespace llvm;
static cl::opt<bool>
-EnableNoAliasConversion("enable-noalias-to-md-conversion", cl::init(false),
+EnableNoAliasConversion("enable-noalias-to-md-conversion", cl::init(true),
cl::Hidden,
cl::desc("Convert noalias attributes to metadata during inlining."));
+static cl::opt<bool>
+PreserveAlignmentAssumptions("preserve-alignment-assumptions-during-inlining",
+ cl::init(true), cl::Hidden,
+ cl::desc("Convert align attributes to assumptions during inlining."));
+
bool llvm::InlineFunction(CallInst *CI, InlineFunctionInfo &IFI,
- bool InsertLifetime) {
- return InlineFunction(CallSite(CI), IFI, InsertLifetime);
+ AAResults *CalleeAAR, bool InsertLifetime) {
+ return InlineFunction(CallSite(CI), IFI, CalleeAAR, InsertLifetime);
}
bool llvm::InlineFunction(InvokeInst *II, InlineFunctionInfo &IFI,
- bool InsertLifetime) {
- return InlineFunction(CallSite(II), IFI, InsertLifetime);
+ AAResults *CalleeAAR, bool InsertLifetime) {
+ return InlineFunction(CallSite(II), IFI, CalleeAAR, InsertLifetime);
}
namespace {
- /// A class for recording information about inlining through an invoke.
- class InvokeInliningInfo {
+ /// A class for recording information about inlining a landing pad.
+ class LandingPadInliningInfo {
BasicBlock *OuterResumeDest; ///< Destination of the invoke's unwind.
BasicBlock *InnerResumeDest; ///< Destination for the callee's resume.
LandingPadInst *CallerLPad; ///< LandingPadInst associated with the invoke.
SmallVector<Value*, 8> UnwindDestPHIValues;
public:
- InvokeInliningInfo(InvokeInst *II)
+ LandingPadInliningInfo(InvokeInst *II)
: OuterResumeDest(II->getUnwindDest()), InnerResumeDest(nullptr),
CallerLPad(nullptr), InnerEHValuesPHI(nullptr) {
// If there are PHI nodes in the unwind destination block, we need to keep
CallerLPad = cast<LandingPadInst>(I);
}
- /// getOuterResumeDest - The outer unwind destination is the target of
+ /// The outer unwind destination is the target of
/// unwind edges introduced for calls within the inlined function.
BasicBlock *getOuterResumeDest() const {
return OuterResumeDest;
LandingPadInst *getLandingPadInst() const { return CallerLPad; }
- /// forwardResume - Forward the 'resume' instruction to the caller's landing
- /// pad block. When the landing pad block has only one predecessor, this is
+ /// Forward the 'resume' instruction to the caller's landing pad block.
+ /// When the landing pad block has only one predecessor, this is
/// a simple branch. When there is more than one predecessor, we need to
/// split the landing pad block after the landingpad instruction and jump
/// to there.
void forwardResume(ResumeInst *RI,
SmallPtrSetImpl<LandingPadInst*> &InlinedLPads);
- /// addIncomingPHIValuesFor - Add incoming-PHI values to the unwind
- /// destination block for the given basic block, using the values for the
- /// original invoke's source block.
+ /// Add incoming-PHI values to the unwind destination block for the given
+ /// basic block, using the values for the original invoke's source block.
void addIncomingPHIValuesFor(BasicBlock *BB) const {
addIncomingPHIValuesForInto(BB, OuterResumeDest);
}
};
}
-/// getInnerResumeDest - Get or create a target for the branch from ResumeInsts.
-BasicBlock *InvokeInliningInfo::getInnerResumeDest() {
+/// Get or create a target for the branch from ResumeInsts.
+BasicBlock *LandingPadInliningInfo::getInnerResumeDest() {
if (InnerResumeDest) return InnerResumeDest;
// Split the landing pad.
return InnerResumeDest;
}
-/// forwardResume - Forward the 'resume' instruction to the caller's landing pad
-/// block. When the landing pad block has only one predecessor, this is a simple
+/// Forward the 'resume' instruction to the caller's landing pad block.
+/// When the landing pad block has only one predecessor, this is a simple
/// branch. When there is more than one predecessor, we need to split the
/// landing pad block after the landingpad instruction and jump to there.
-void InvokeInliningInfo::forwardResume(ResumeInst *RI,
- SmallPtrSetImpl<LandingPadInst*> &InlinedLPads) {
+void LandingPadInliningInfo::forwardResume(
+ ResumeInst *RI, SmallPtrSetImpl<LandingPadInst *> &InlinedLPads) {
BasicBlock *Dest = getInnerResumeDest();
BasicBlock *Src = RI->getParent();
RI->eraseFromParent();
}
-/// HandleCallsInBlockInlinedThroughInvoke - When we inline a basic block into
-/// an invoke, we have to turn all of the calls that can throw into
-/// invokes. This function analyze BB to see if there are any calls, and if so,
+/// When we inline a basic block into an invoke,
+/// we have to turn all of the calls that can throw into invokes.
+/// This function analyze BB to see if there are any calls, and if so,
/// it rewrites them to be invokes that jump to InvokeDest and fills in the PHI
/// nodes in that block with the values specified in InvokeDestPHIValues.
-static void HandleCallsInBlockInlinedThroughInvoke(BasicBlock *BB,
- InvokeInliningInfo &Invoke) {
+static BasicBlock *
+HandleCallsInBlockInlinedThroughInvoke(BasicBlock *BB, BasicBlock *UnwindEdge) {
for (BasicBlock::iterator BBI = BB->begin(), E = BB->end(); BBI != E; ) {
Instruction *I = BBI++;
// Create the new invoke instruction.
ImmutableCallSite CS(CI);
SmallVector<Value*, 8> InvokeArgs(CS.arg_begin(), CS.arg_end());
- InvokeInst *II = InvokeInst::Create(CI->getCalledValue(), Split,
- Invoke.getOuterResumeDest(),
+ InvokeInst *II = InvokeInst::Create(CI->getCalledValue(), Split, UnwindEdge,
InvokeArgs, CI->getName(), BB);
II->setDebugLoc(CI->getDebugLoc());
II->setCallingConv(CI->getCallingConv());
// Delete the original call
Split->getInstList().pop_front();
-
- // Update any PHI nodes in the exceptional block to indicate that there is
- // now a new entry in them.
- Invoke.addIncomingPHIValuesFor(BB);
- return;
+ return BB;
}
+ return nullptr;
}
-/// HandleInlinedInvoke - If we inlined an invoke site, we need to convert calls
+/// If we inlined an invoke site, we need to convert calls
/// in the body of the inlined function into invokes.
///
/// II is the invoke instruction being inlined. FirstNewBlock is the first
/// block of the inlined code (the last block is the end of the function),
/// and InlineCodeInfo is information about the code that got inlined.
-static void HandleInlinedInvoke(InvokeInst *II, BasicBlock *FirstNewBlock,
- ClonedCodeInfo &InlinedCodeInfo) {
+static void HandleInlinedLandingPad(InvokeInst *II, BasicBlock *FirstNewBlock,
+ ClonedCodeInfo &InlinedCodeInfo) {
BasicBlock *InvokeDest = II->getUnwindDest();
Function *Caller = FirstNewBlock->getParent();
// The inlined code is currently at the end of the function, scan from the
// start of the inlined code to its end, checking for stuff we need to
// rewrite.
- InvokeInliningInfo Invoke(II);
+ LandingPadInliningInfo Invoke(II);
// Get all of the inlined landing pad instructions.
SmallPtrSet<LandingPadInst*, 16> InlinedLPads;
for (Function::iterator BB = FirstNewBlock, E = Caller->end(); BB != E; ++BB){
if (InlinedCodeInfo.ContainsCalls)
- HandleCallsInBlockInlinedThroughInvoke(BB, Invoke);
+ if (BasicBlock *NewBB = HandleCallsInBlockInlinedThroughInvoke(
+ BB, Invoke.getOuterResumeDest()))
+ // Update any PHI nodes in the exceptional block to indicate that there
+ // is now a new entry in them.
+ Invoke.addIncomingPHIValuesFor(NewBB);
// Forward any resumes that are remaining here.
if (ResumeInst *RI = dyn_cast<ResumeInst>(BB->getTerminator()))
InvokeDest->removePredecessor(II->getParent());
}
-/// CloneAliasScopeMetadata - When inlining a function that contains noalias
-/// scope metadata, this metadata needs to be cloned so that the inlined blocks
+/// If we inlined an invoke site, we need to convert calls
+/// in the body of the inlined function into invokes.
+///
+/// II is the invoke instruction being inlined. FirstNewBlock is the first
+/// block of the inlined code (the last block is the end of the function),
+/// and InlineCodeInfo is information about the code that got inlined.
+static void HandleInlinedEHPad(InvokeInst *II, BasicBlock *FirstNewBlock,
+ ClonedCodeInfo &InlinedCodeInfo) {
+ BasicBlock *UnwindDest = II->getUnwindDest();
+ Function *Caller = FirstNewBlock->getParent();
+
+ assert(UnwindDest->getFirstNonPHI()->isEHPad() && "unexpected BasicBlock!");
+
+ // If there are PHI nodes in the unwind destination block, we need to keep
+ // track of which values came into them from the invoke before removing the
+ // edge from this block.
+ SmallVector<Value *, 8> UnwindDestPHIValues;
+ llvm::BasicBlock *InvokeBB = II->getParent();
+ for (Instruction &I : *UnwindDest) {
+ // Save the value to use for this edge.
+ PHINode *PHI = dyn_cast<PHINode>(&I);
+ if (!PHI)
+ break;
+ UnwindDestPHIValues.push_back(PHI->getIncomingValueForBlock(InvokeBB));
+ }
+
+ // Add incoming-PHI values to the unwind destination block for the given basic
+ // block, using the values for the original invoke's source block.
+ auto UpdatePHINodes = [&](BasicBlock *Src) {
+ BasicBlock::iterator I = UnwindDest->begin();
+ for (Value *V : UnwindDestPHIValues) {
+ PHINode *PHI = cast<PHINode>(I);
+ PHI->addIncoming(V, Src);
+ ++I;
+ }
+ };
+
+ // Forward EH terminator instructions to the caller's invoke destination.
+ // This is as simple as connect all the instructions which 'unwind to caller'
+ // to the invoke destination.
+ for (Function::iterator BB = FirstNewBlock, E = Caller->end(); BB != E;
+ ++BB) {
+ Instruction *I = BB->getFirstNonPHI();
+ if (I->isEHPad()) {
+ if (auto *CEPI = dyn_cast<CatchEndPadInst>(I)) {
+ if (CEPI->unwindsToCaller()) {
+ CatchEndPadInst::Create(CEPI->getContext(), UnwindDest, CEPI);
+ CEPI->eraseFromParent();
+ UpdatePHINodes(BB);
+ }
+ } else if (auto *CEPI = dyn_cast<CleanupEndPadInst>(I)) {
+ if (CEPI->unwindsToCaller()) {
+ CleanupEndPadInst::Create(CEPI->getCleanupPad(), UnwindDest, CEPI);
+ CEPI->eraseFromParent();
+ UpdatePHINodes(BB);
+ }
+ } else if (auto *TPI = dyn_cast<TerminatePadInst>(I)) {
+ if (TPI->unwindsToCaller()) {
+ SmallVector<Value *, 3> TerminatePadArgs;
+ for (Value *Operand : TPI->operands())
+ TerminatePadArgs.push_back(Operand);
+ TerminatePadInst::Create(TPI->getContext(), UnwindDest, TPI);
+ TPI->eraseFromParent();
+ UpdatePHINodes(BB);
+ }
+ } else {
+ assert(isa<CatchPadInst>(I) || isa<CleanupPadInst>(I));
+ }
+ }
+
+ if (auto *CRI = dyn_cast<CleanupReturnInst>(BB->getTerminator())) {
+ if (CRI->unwindsToCaller()) {
+ CleanupReturnInst::Create(CRI->getCleanupPad(), UnwindDest, CRI);
+ CRI->eraseFromParent();
+ UpdatePHINodes(BB);
+ }
+ }
+ }
+
+ if (InlinedCodeInfo.ContainsCalls)
+ for (Function::iterator BB = FirstNewBlock, E = Caller->end(); BB != E;
+ ++BB)
+ if (BasicBlock *NewBB =
+ HandleCallsInBlockInlinedThroughInvoke(BB, UnwindDest))
+ // Update any PHI nodes in the exceptional block to indicate that there
+ // is now a new entry in them.
+ UpdatePHINodes(NewBB);
+
+ // Now that everything is happy, we have one final detail. The PHI nodes in
+ // the exception destination block still have entries due to the original
+ // invoke instruction. Eliminate these entries (which might even delete the
+ // PHI node) now.
+ UnwindDest->removePredecessor(InvokeBB);
+}
+
+/// When inlining a function that contains noalias scope metadata,
+/// this metadata needs to be cloned so that the inlined blocks
/// have different "unqiue scopes" at every call site. Were this not done, then
/// aliasing scopes from a function inlined into a caller multiple times could
/// not be differentiated (and this would lead to miscompiles because the
// Walk the existing metadata, adding the complete (perhaps cyclic) chain to
// the set.
- SmallVector<const Value *, 16> Queue(MD.begin(), MD.end());
+ SmallVector<const Metadata *, 16> Queue(MD.begin(), MD.end());
while (!Queue.empty()) {
const MDNode *M = cast<MDNode>(Queue.pop_back_val());
for (unsigned i = 0, ie = M->getNumOperands(); i != ie; ++i)
// Now we have a complete set of all metadata in the chains used to specify
// the noalias scopes and the lists of those scopes.
- SmallVector<MDNode *, 16> DummyNodes;
- DenseMap<const MDNode *, TrackingVH<MDNode> > MDMap;
+ SmallVector<TempMDTuple, 16> DummyNodes;
+ DenseMap<const MDNode *, TrackingMDNodeRef> MDMap;
for (SetVector<const MDNode *>::iterator I = MD.begin(), IE = MD.end();
I != IE; ++I) {
- MDNode *Dummy = MDNode::getTemporary(CalledFunc->getContext(), None);
- DummyNodes.push_back(Dummy);
- MDMap[*I] = Dummy;
+ DummyNodes.push_back(MDTuple::getTemporary(CalledFunc->getContext(), None));
+ MDMap[*I].reset(DummyNodes.back().get());
}
// Create new metadata nodes to replace the dummy nodes, replacing old
// node.
for (SetVector<const MDNode *>::iterator I = MD.begin(), IE = MD.end();
I != IE; ++I) {
- SmallVector<Value *, 4> NewOps;
+ SmallVector<Metadata *, 4> NewOps;
for (unsigned i = 0, ie = (*I)->getNumOperands(); i != ie; ++i) {
- const Value *V = (*I)->getOperand(i);
+ const Metadata *V = (*I)->getOperand(i);
if (const MDNode *M = dyn_cast<MDNode>(V))
NewOps.push_back(MDMap[M]);
else
- NewOps.push_back(const_cast<Value *>(V));
+ NewOps.push_back(const_cast<Metadata *>(V));
}
- MDNode *NewM = MDNode::get(CalledFunc->getContext(), NewOps),
- *TempM = MDMap[*I];
+ MDNode *NewM = MDNode::get(CalledFunc->getContext(), NewOps);
+ MDTuple *TempM = cast<MDTuple>(MDMap[*I]);
+ assert(TempM->isTemporary() && "Expected temporary node");
TempM->replaceAllUsesWith(NewM);
}
// which instructions inside it might belong), propagate those scopes to
// the inlined instructions.
if (MDNode *CSM =
- CS.getInstruction()->getMetadata(LLVMContext::MD_alias_scope))
+ CS.getInstruction()->getMetadata(LLVMContext::MD_alias_scope))
NewMD = MDNode::concatenate(NewMD, CSM);
NI->setMetadata(LLVMContext::MD_alias_scope, NewMD);
} else if (NI->mayReadOrWriteMemory()) {
if (MDNode *M =
- CS.getInstruction()->getMetadata(LLVMContext::MD_alias_scope))
+ CS.getInstruction()->getMetadata(LLVMContext::MD_alias_scope))
NI->setMetadata(LLVMContext::MD_alias_scope, M);
}
// which instructions inside it don't alias), propagate those scopes to
// the inlined instructions.
if (MDNode *CSM =
- CS.getInstruction()->getMetadata(LLVMContext::MD_noalias))
+ CS.getInstruction()->getMetadata(LLVMContext::MD_noalias))
NewMD = MDNode::concatenate(NewMD, CSM);
NI->setMetadata(LLVMContext::MD_noalias, NewMD);
} else if (NI->mayReadOrWriteMemory()) {
- if (MDNode *M =
- CS.getInstruction()->getMetadata(LLVMContext::MD_noalias))
+ if (MDNode *M = CS.getInstruction()->getMetadata(LLVMContext::MD_noalias))
NI->setMetadata(LLVMContext::MD_noalias, M);
}
}
-
- // Now that everything has been replaced, delete the dummy nodes.
- for (unsigned i = 0, ie = DummyNodes.size(); i != ie; ++i)
- MDNode::deleteTemporary(DummyNodes[i]);
}
-/// AddAliasScopeMetadata - If the inlined function has noalias arguments, then
-/// add new alias scopes for each noalias argument, tag the mapped noalias
+/// If the inlined function has noalias arguments,
+/// then add new alias scopes for each noalias argument, tag the mapped noalias
/// parameters with noalias metadata specifying the new scope, and tag all
/// non-derived loads, stores and memory intrinsics with the new alias scopes.
static void AddAliasScopeMetadata(CallSite CS, ValueToValueMapTy &VMap,
- const DataLayout *DL) {
+ const DataLayout &DL, AAResults *CalleeAAR) {
if (!EnableNoAliasConversion)
return;
if (!NI)
continue;
+ bool IsArgMemOnlyCall = false, IsFuncCall = false;
SmallVector<const Value *, 2> PtrArgs;
if (const LoadInst *LI = dyn_cast<LoadInst>(I))
if (ICS.doesNotAccessMemory())
continue;
+ IsFuncCall = true;
+ if (CalleeAAR) {
+ FunctionModRefBehavior MRB = CalleeAAR->getModRefBehavior(ICS);
+ if (MRB == FMRB_OnlyAccessesArgumentPointees ||
+ MRB == FMRB_OnlyReadsArgumentPointees)
+ IsArgMemOnlyCall = true;
+ }
+
for (ImmutableCallSite::arg_iterator AI = ICS.arg_begin(),
- AE = ICS.arg_end(); AI != AE; ++AI)
+ AE = ICS.arg_end(); AI != AE; ++AI) {
// We need to check the underlying objects of all arguments, not just
// the pointer arguments, because we might be passing pointers as
// integers, etc.
- // FIXME: If we know that the call only accesses pointer arguments,
+ // However, if we know that the call only accesses pointer arguments,
// then we only need to check the pointer arguments.
+ if (IsArgMemOnlyCall && !(*AI)->getType()->isPointerTy())
+ continue;
+
PtrArgs.push_back(*AI);
+ }
}
// If we found no pointers, then this instruction is not suitable for
// pairing with an instruction to receive aliasing metadata.
// However, if this is a call, this we might just alias with none of the
// noalias arguments.
- if (PtrArgs.empty() && !isa<CallInst>(I) && !isa<InvokeInst>(I))
+ if (PtrArgs.empty() && !IsFuncCall)
continue;
// It is possible that there is only one underlying object, but you
// need to go through several PHIs to see it, and thus could be
// repeated in the Objects list.
SmallPtrSet<const Value *, 4> ObjSet;
- SmallVector<Value *, 4> Scopes, NoAliases;
+ SmallVector<Metadata *, 4> Scopes, NoAliases;
SmallSetVector<const Argument *, 4> NAPtrArgs;
for (unsigned i = 0, ie = PtrArgs.size(); i != ie; ++i) {
bool IsNonPtrConst = isa<ConstantInt>(V) || isa<ConstantFP>(V) ||
isa<ConstantPointerNull>(V) ||
isa<ConstantDataVector>(V) || isa<UndefValue>(V);
- if (!IsNonPtrConst &&
- !isIdentifiedFunctionLocal(const_cast<Value*>(V))) {
+ if (IsNonPtrConst)
+ continue;
+
+ // If this is anything other than a noalias argument, then we cannot
+ // completely describe the aliasing properties using alias.scope
+ // metadata (and, thus, won't add any).
+ if (const Argument *A = dyn_cast<Argument>(V)) {
+ if (!A->hasNoAliasAttr())
+ UsesAliasingPtr = true;
+ } else {
UsesAliasingPtr = true;
- if (!isa<Argument>(V)) {
- CanDeriveViaCapture = true;
- break;
- }
}
+
+ // If this is not some identified function-local object (which cannot
+ // directly alias a noalias argument), or some other argument (which,
+ // by definition, also cannot alias a noalias argument), then we could
+ // alias a noalias argument that has been captured).
+ if (!isa<Argument>(V) &&
+ !isIdentifiedFunctionLocal(const_cast<Value*>(V)))
+ CanDeriveViaCapture = true;
}
-
+
+ // A function call can always get captured noalias pointers (via other
+ // parameters, globals, etc.).
+ if (IsFuncCall && !IsArgMemOnlyCall)
+ CanDeriveViaCapture = true;
+
// First, we want to figure out all of the sets with which we definitely
// don't alias. Iterate over all noalias set, and add those for which:
// 1. The noalias argument is not in the set of objects from which we
// definitely derive.
// 2. The noalias argument has not yet been captured.
+ // An arbitrary function that might load pointers could see captured
+ // noalias arguments via other noalias arguments or globals, and so we
+ // must always check for prior capture.
for (const Argument *A : NoAliasArgs) {
if (!ObjSet.count(A) && (!CanDeriveViaCapture ||
// It might be tempting to skip the
}
if (!NoAliases.empty())
- NI->setMetadata(LLVMContext::MD_noalias, MDNode::concatenate(
- NI->getMetadata(LLVMContext::MD_noalias),
- MDNode::get(CalledFunc->getContext(), NoAliases)));
+ NI->setMetadata(LLVMContext::MD_noalias,
+ MDNode::concatenate(
+ NI->getMetadata(LLVMContext::MD_noalias),
+ MDNode::get(CalledFunc->getContext(), NoAliases)));
// Next, we want to figure out all of the sets to which we might belong.
// We might belong to a set if the noalias argument is in the set of
// scopes to arbitrary functions unless we know they don't access any
// non-parameter pointer-values.
bool CanAddScopes = !UsesAliasingPtr;
- if (CanAddScopes && (isa<CallInst>(I) || isa<InvokeInst>(I))) {
- // FIXME: We should have a way to access the
- // IntrReadArgMem/IntrReadWriteArgMem properties of intrinsics, and we
- // should have a way to determine that for regular functions too. For
- // now, just do this for the memory intrinsics we understand.
- CanAddScopes = isa<MemIntrinsic>(I);
- }
+ if (CanAddScopes && IsFuncCall)
+ CanAddScopes = IsArgMemOnlyCall;
if (CanAddScopes)
for (const Argument *A : NoAliasArgs) {
}
if (!Scopes.empty())
- NI->setMetadata(LLVMContext::MD_alias_scope, MDNode::concatenate(
- NI->getMetadata(LLVMContext::MD_alias_scope),
- MDNode::get(CalledFunc->getContext(), Scopes)));
+ NI->setMetadata(
+ LLVMContext::MD_alias_scope,
+ MDNode::concatenate(NI->getMetadata(LLVMContext::MD_alias_scope),
+ MDNode::get(CalledFunc->getContext(), Scopes)));
+ }
+ }
+}
+
+/// If the inlined function has non-byval align arguments, then
+/// add @llvm.assume-based alignment assumptions to preserve this information.
+static void AddAlignmentAssumptions(CallSite CS, InlineFunctionInfo &IFI) {
+ if (!PreserveAlignmentAssumptions)
+ return;
+ auto &DL = CS.getCaller()->getParent()->getDataLayout();
+
+ // To avoid inserting redundant assumptions, we should check for assumptions
+ // already in the caller. To do this, we might need a DT of the caller.
+ DominatorTree DT;
+ bool DTCalculated = false;
+
+ Function *CalledFunc = CS.getCalledFunction();
+ for (Function::arg_iterator I = CalledFunc->arg_begin(),
+ E = CalledFunc->arg_end();
+ I != E; ++I) {
+ unsigned Align = I->getType()->isPointerTy() ? I->getParamAlignment() : 0;
+ if (Align && !I->hasByValOrInAllocaAttr() && !I->hasNUses(0)) {
+ if (!DTCalculated) {
+ DT.recalculate(const_cast<Function&>(*CS.getInstruction()->getParent()
+ ->getParent()));
+ DTCalculated = true;
+ }
+
+ // If we can already prove the asserted alignment in the context of the
+ // caller, then don't bother inserting the assumption.
+ Value *Arg = CS.getArgument(I->getArgNo());
+ if (getKnownAlignment(Arg, DL, CS.getInstruction(),
+ &IFI.ACT->getAssumptionCache(*CS.getCaller()),
+ &DT) >= Align)
+ continue;
+
+ IRBuilder<>(CS.getInstruction())
+ .CreateAlignmentAssumption(DL, Arg, Align);
}
}
}
-/// UpdateCallGraphAfterInlining - Once we have cloned code over from a callee
-/// into the caller, update the specified callgraph to reflect the changes we
-/// made. Note that it's possible that not all code was copied over, so only
+/// Once we have cloned code over from a callee into the caller,
+/// update the specified callgraph to reflect the changes we made.
+/// Note that it's possible that not all code was copied over, so only
/// some edges of the callgraph may remain.
static void UpdateCallGraphAfterInlining(CallSite CS,
Function::iterator FirstNewBlock,
// If the call was inlined, but then constant folded, there is no edge to
// add. Check for this case.
Instruction *NewCall = dyn_cast<Instruction>(VMI->second);
- if (!NewCall) continue;
+ if (!NewCall)
+ continue;
+ // We do not treat intrinsic calls like real function calls because we
+ // expect them to become inline code; do not add an edge for an intrinsic.
+ CallSite CS = CallSite(NewCall);
+ if (CS && CS.getCalledFunction() && CS.getCalledFunction()->isIntrinsic())
+ continue;
+
// Remember that this call site got inlined for the client of
// InlineFunction.
IFI.InlinedCalls.push_back(NewCall);
static void HandleByValArgumentInit(Value *Dst, Value *Src, Module *M,
BasicBlock *InsertBlock,
InlineFunctionInfo &IFI) {
- LLVMContext &Context = Src->getContext();
- Type *VoidPtrTy = Type::getInt8PtrTy(Context);
Type *AggTy = cast<PointerType>(Src->getType())->getElementType();
- Type *Tys[3] = { VoidPtrTy, VoidPtrTy, Type::getInt64Ty(Context) };
- Function *MemCpyFn = Intrinsic::getDeclaration(M, Intrinsic::memcpy, Tys);
- IRBuilder<> builder(InsertBlock->begin());
- Value *DstCast = builder.CreateBitCast(Dst, VoidPtrTy, "tmp");
- Value *SrcCast = builder.CreateBitCast(Src, VoidPtrTy, "tmp");
-
- Value *Size;
- if (IFI.DL == nullptr)
- Size = ConstantExpr::getSizeOf(AggTy);
- else
- Size = ConstantInt::get(Type::getInt64Ty(Context),
- IFI.DL->getTypeStoreSize(AggTy));
+ IRBuilder<> Builder(InsertBlock->begin());
+
+ Value *Size = Builder.getInt64(M->getDataLayout().getTypeStoreSize(AggTy));
// Always generate a memcpy of alignment 1 here because we don't know
// the alignment of the src pointer. Other optimizations can infer
// better alignment.
- Value *CallArgs[] = {
- DstCast, SrcCast, Size,
- ConstantInt::get(Type::getInt32Ty(Context), 1),
- ConstantInt::getFalse(Context) // isVolatile
- };
- builder.CreateCall(MemCpyFn, CallArgs);
+ Builder.CreateMemCpy(Dst, Src, Size, /*Align=*/1);
}
-/// HandleByValArgument - When inlining a call site that has a byval argument,
+/// When inlining a call site that has a byval argument,
/// we have to make the implicit memcpy explicit by adding it.
static Value *HandleByValArgument(Value *Arg, Instruction *TheCall,
const Function *CalledFunc,
PointerType *ArgTy = cast<PointerType>(Arg->getType());
Type *AggTy = ArgTy->getElementType();
+ Function *Caller = TheCall->getParent()->getParent();
+
// If the called function is readonly, then it could not mutate the caller's
// copy of the byval'd memory. In this case, it is safe to elide the copy and
// temporary.
if (ByValAlignment <= 1) // 0 = unspecified, 1 = no particular alignment.
return Arg;
+ const DataLayout &DL = Caller->getParent()->getDataLayout();
+
// If the pointer is already known to be sufficiently aligned, or if we can
// round it up to a larger alignment, then we don't need a temporary.
- if (getOrEnforceKnownAlignment(Arg, ByValAlignment,
- IFI.DL) >= ByValAlignment)
+ if (getOrEnforceKnownAlignment(Arg, ByValAlignment, DL, TheCall,
+ &IFI.ACT->getAssumptionCache(*Caller)) >=
+ ByValAlignment)
return Arg;
// Otherwise, we have to make a memcpy to get a safe alignment. This is bad
}
// Create the alloca. If we have DataLayout, use nice alignment.
- unsigned Align = 1;
- if (IFI.DL)
- Align = IFI.DL->getPrefTypeAlignment(AggTy);
-
+ unsigned Align =
+ Caller->getParent()->getDataLayout().getPrefTypeAlignment(AggTy);
+
// If the byval had an alignment specified, we *must* use at least that
// alignment, as it is required by the byval argument (and uses of the
// pointer inside the callee).
Align = std::max(Align, ByValAlignment);
- Function *Caller = TheCall->getParent()->getParent();
-
Value *NewAlloca = new AllocaInst(AggTy, nullptr, Align, Arg->getName(),
&*Caller->begin()->begin());
IFI.StaticAllocas.push_back(cast<AllocaInst>(NewAlloca));
return NewAlloca;
}
-// isUsedByLifetimeMarker - Check whether this Value is used by a lifetime
-// intrinsic.
+// Check whether this Value is used by a lifetime intrinsic.
static bool isUsedByLifetimeMarker(Value *V) {
for (User *U : V->users()) {
if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(U)) {
return false;
}
-// hasLifetimeMarkers - Check whether the given alloca already has
+// Check whether the given alloca already has
// lifetime.start or lifetime.end intrinsics.
static bool hasLifetimeMarkers(AllocaInst *AI) {
Type *Ty = AI->getType();
return false;
}
-/// updateInlinedAtInfo - Helper function used by fixupLineNumbers to
-/// recursively update InlinedAtEntry of a DebugLoc.
-static DebugLoc updateInlinedAtInfo(const DebugLoc &DL,
- const DebugLoc &InlinedAtDL,
- LLVMContext &Ctx) {
- if (MDNode *IA = DL.getInlinedAt(Ctx)) {
- DebugLoc NewInlinedAtDL
- = updateInlinedAtInfo(DebugLoc::getFromDILocation(IA), InlinedAtDL, Ctx);
- return DebugLoc::get(DL.getLine(), DL.getCol(), DL.getScope(Ctx),
- NewInlinedAtDL.getAsMDNode(Ctx));
+/// Rebuild the entire inlined-at chain for this instruction so that the top of
+/// the chain now is inlined-at the new call site.
+static DebugLoc
+updateInlinedAtInfo(DebugLoc DL, DILocation *InlinedAtNode, LLVMContext &Ctx,
+ DenseMap<const DILocation *, DILocation *> &IANodes) {
+ SmallVector<DILocation *, 3> InlinedAtLocations;
+ DILocation *Last = InlinedAtNode;
+ DILocation *CurInlinedAt = DL;
+
+ // Gather all the inlined-at nodes
+ while (DILocation *IA = CurInlinedAt->getInlinedAt()) {
+ // Skip any we've already built nodes for
+ if (DILocation *Found = IANodes[IA]) {
+ Last = Found;
+ break;
+ }
+
+ InlinedAtLocations.push_back(IA);
+ CurInlinedAt = IA;
+ }
+
+ // Starting from the top, rebuild the nodes to point to the new inlined-at
+ // location (then rebuilding the rest of the chain behind it) and update the
+ // map of already-constructed inlined-at nodes.
+ for (const DILocation *MD : make_range(InlinedAtLocations.rbegin(),
+ InlinedAtLocations.rend())) {
+ Last = IANodes[MD] = DILocation::getDistinct(
+ Ctx, MD->getLine(), MD->getColumn(), MD->getScope(), Last);
}
- return DebugLoc::get(DL.getLine(), DL.getCol(), DL.getScope(Ctx),
- InlinedAtDL.getAsMDNode(Ctx));
+ // And finally create the normal location for this instruction, referring to
+ // the new inlined-at chain.
+ return DebugLoc::get(DL.getLine(), DL.getCol(), DL.getScope(), Last);
}
-/// fixupLineNumbers - Update inlined instructions' line numbers to
+/// Update inlined instructions' line numbers to
/// to encode location where these instructions are inlined.
static void fixupLineNumbers(Function *Fn, Function::iterator FI,
Instruction *TheCall) {
DebugLoc TheCallDL = TheCall->getDebugLoc();
- if (TheCallDL.isUnknown())
+ if (!TheCallDL)
return;
+ auto &Ctx = Fn->getContext();
+ DILocation *InlinedAtNode = TheCallDL;
+
+ // Create a unique call site, not to be confused with any other call from the
+ // same location.
+ InlinedAtNode = DILocation::getDistinct(
+ Ctx, InlinedAtNode->getLine(), InlinedAtNode->getColumn(),
+ InlinedAtNode->getScope(), InlinedAtNode->getInlinedAt());
+
+ // Cache the inlined-at nodes as they're built so they are reused, without
+ // this every instruction's inlined-at chain would become distinct from each
+ // other.
+ DenseMap<const DILocation *, DILocation *> IANodes;
+
for (; FI != Fn->end(); ++FI) {
for (BasicBlock::iterator BI = FI->begin(), BE = FI->end();
BI != BE; ++BI) {
DebugLoc DL = BI->getDebugLoc();
- if (DL.isUnknown()) {
+ if (!DL) {
// If the inlined instruction has no line number, make it look as if it
// originates from the call location. This is important for
// ((__always_inline__, __nodebug__)) functions which must use caller
// location for all instructions in their function body.
+
+ // Don't update static allocas, as they may get moved later.
+ if (auto *AI = dyn_cast<AllocaInst>(BI))
+ if (isa<Constant>(AI->getArraySize()))
+ continue;
+
BI->setDebugLoc(TheCallDL);
} else {
- BI->setDebugLoc(updateInlinedAtInfo(DL, TheCallDL, BI->getContext()));
- if (DbgValueInst *DVI = dyn_cast<DbgValueInst>(BI)) {
- LLVMContext &Ctx = BI->getContext();
- MDNode *InlinedAt = BI->getDebugLoc().getInlinedAt(Ctx);
- DVI->setOperand(2, createInlinedVariable(DVI->getVariable(),
- InlinedAt, Ctx));
- }
+ BI->setDebugLoc(updateInlinedAtInfo(DL, InlinedAtNode, BI->getContext(), IANodes));
}
}
}
}
-/// InlineFunction - This function inlines the called function into the basic
-/// block of the caller. This returns false if it is not possible to inline
-/// this call. The program is still in a well defined state if this occurs
-/// though.
+/// This function inlines the called function into the basic block of the
+/// caller. This returns false if it is not possible to inline this call.
+/// The program is still in a well defined state if this occurs though.
///
/// Note that this only does one level of inlining. For example, if the
/// instruction 'call B' is inlined, and 'B' calls 'C', then the call to 'C' now
/// exists in the instruction stream. Similarly this will inline a recursive
/// function by one level.
bool llvm::InlineFunction(CallSite CS, InlineFunctionInfo &IFI,
- bool InsertLifetime) {
+ AAResults *CalleeAAR, bool InsertLifetime) {
Instruction *TheCall = CS.getInstruction();
assert(TheCall->getParent() && TheCall->getParent()->getParent() &&
"Instruction not in function!");
}
// Get the personality function from the callee if it contains a landing pad.
- Value *CalleePersonality = nullptr;
- for (Function::const_iterator I = CalledFunc->begin(), E = CalledFunc->end();
- I != E; ++I)
- if (const InvokeInst *II = dyn_cast<InvokeInst>(I->getTerminator())) {
- const BasicBlock *BB = II->getUnwindDest();
- const LandingPadInst *LP = BB->getLandingPadInst();
- CalleePersonality = LP->getPersonalityFn();
- break;
- }
+ Constant *CalledPersonality =
+ CalledFunc->hasPersonalityFn() ? CalledFunc->getPersonalityFn() : nullptr;
// Find the personality function used by the landing pads of the caller. If it
// exists, then check to see that it matches the personality function used in
// the callee.
- if (CalleePersonality) {
- for (Function::const_iterator I = Caller->begin(), E = Caller->end();
- I != E; ++I)
- if (const InvokeInst *II = dyn_cast<InvokeInst>(I->getTerminator())) {
- const BasicBlock *BB = II->getUnwindDest();
- const LandingPadInst *LP = BB->getLandingPadInst();
-
- // If the personality functions match, then we can perform the
- // inlining. Otherwise, we can't inline.
- // TODO: This isn't 100% true. Some personality functions are proper
- // supersets of others and can be used in place of the other.
- if (LP->getPersonalityFn() != CalleePersonality)
- return false;
-
- break;
- }
+ Constant *CallerPersonality =
+ Caller->hasPersonalityFn() ? Caller->getPersonalityFn() : nullptr;
+ if (CalledPersonality) {
+ if (!CallerPersonality)
+ Caller->setPersonalityFn(CalledPersonality);
+ // If the personality functions match, then we can perform the
+ // inlining. Otherwise, we can't inline.
+ // TODO: This isn't 100% true. Some personality functions are proper
+ // supersets of others and can be used in place of the other.
+ else if (CalledPersonality != CallerPersonality)
+ return false;
}
// Get an iterator to the last basic block in the function, which will have
// Keep a list of pair (dst, src) to emit byval initializations.
SmallVector<std::pair<Value*, Value*>, 4> ByValInit;
+ auto &DL = Caller->getParent()->getDataLayout();
+
assert(CalledFunc->arg_size() == CS.arg_size() &&
"No varargs calls can be inlined!");
VMap[I] = ActualArg;
}
+ // Add alignment assumptions if necessary. We do this before the inlined
+ // instructions are actually cloned into the caller so that we can easily
+ // check what will be known at the start of the inlined code.
+ AddAlignmentAssumptions(CS, IFI);
+
// We want the inliner to prune the code as it copies. We would LOVE to
// have no dead or constant instructions leftover after inlining occurs
// (which can happen, e.g., because an argument was constant), but we'll be
// happy with whatever the cloner can do.
- CloneAndPruneFunctionInto(Caller, CalledFunc, VMap,
+ CloneAndPruneFunctionInto(Caller, CalledFunc, VMap,
/*ModuleLevelChanges=*/false, Returns, ".i",
- &InlinedFunctionInfo, IFI.DL, TheCall);
+ &InlinedFunctionInfo, TheCall);
// Remember the first block that is newly cloned over.
FirstNewBlock = LastBlock; ++FirstNewBlock;
CloneAliasScopeMetadata(CS, VMap);
// Add noalias metadata if necessary.
- AddAliasScopeMetadata(CS, VMap, IFI.DL);
+ AddAliasScopeMetadata(CS, VMap, DL, CalleeAAR);
+
+ // FIXME: We could register any cloned assumptions instead of clearing the
+ // whole function's cache.
+ if (IFI.ACT)
+ IFI.ACT->getAssumptionCache(*Caller).clear();
}
// If there are any alloca instructions in the block that used to be the entry
FirstNewBlock->getInstList(),
AI, I);
}
+ // Move any dbg.declares describing the allocas into the entry basic block.
+ DIBuilder DIB(*Caller->getParent());
+ for (auto &AI : IFI.StaticAllocas)
+ replaceDbgDeclareForAlloca(AI, AI, DIB, /*Deref=*/false);
}
bool InlinedMustTailCalls = false;
ConstantInt *AllocaSize = nullptr;
if (ConstantInt *AIArraySize =
dyn_cast<ConstantInt>(AI->getArraySize())) {
- if (IFI.DL) {
- Type *AllocaType = AI->getAllocatedType();
- uint64_t AllocaTypeSize = IFI.DL->getTypeAllocSize(AllocaType);
- uint64_t AllocaArraySize = AIArraySize->getLimitedValue();
- assert(AllocaArraySize > 0 && "array size of AllocaInst is zero");
- // Check that array size doesn't saturate uint64_t and doesn't
- // overflow when it's multiplied by type size.
- if (AllocaArraySize != ~0ULL &&
- UINT64_MAX / AllocaArraySize >= AllocaTypeSize) {
- AllocaSize = ConstantInt::get(Type::getInt64Ty(AI->getContext()),
- AllocaArraySize * AllocaTypeSize);
- }
+ auto &DL = Caller->getParent()->getDataLayout();
+ Type *AllocaType = AI->getAllocatedType();
+ uint64_t AllocaTypeSize = DL.getTypeAllocSize(AllocaType);
+ uint64_t AllocaArraySize = AIArraySize->getLimitedValue();
+
+ // Don't add markers for zero-sized allocas.
+ if (AllocaArraySize == 0)
+ continue;
+
+ // Check that array size doesn't saturate uint64_t and doesn't
+ // overflow when it's multiplied by type size.
+ if (AllocaArraySize != ~0ULL &&
+ UINT64_MAX / AllocaArraySize >= AllocaTypeSize) {
+ AllocaSize = ConstantInt::get(Type::getInt64Ty(AI->getContext()),
+ AllocaArraySize * AllocaTypeSize);
}
}
// Insert the llvm.stacksave.
CallInst *SavedPtr = IRBuilder<>(FirstNewBlock, FirstNewBlock->begin())
- .CreateCall(StackSave, "savedstack");
+ .CreateCall(StackSave, {}, "savedstack");
// Insert a call to llvm.stackrestore before any return instructions in the
// inlined function.
// If we are inlining for an invoke instruction, we must make sure to rewrite
// any call instructions into invoke instructions.
- if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall))
- HandleInlinedInvoke(II, FirstNewBlock, InlinedFunctionInfo);
+ if (auto *II = dyn_cast<InvokeInst>(TheCall)) {
+ BasicBlock *UnwindDest = II->getUnwindDest();
+ Instruction *FirstNonPHI = UnwindDest->getFirstNonPHI();
+ if (isa<LandingPadInst>(FirstNonPHI)) {
+ HandleInlinedLandingPad(II, FirstNewBlock, InlinedFunctionInfo);
+ } else {
+ HandleInlinedEHPad(II, FirstNewBlock, InlinedFunctionInfo);
+ }
+ }
// Handle any inlined musttail call sites. In order for a new call site to be
// musttail, the source of the clone and the inlined call site must have been
// the entries are the same or undef). If so, remove the PHI so it doesn't
// block other optimizations.
if (PHI) {
- if (Value *V = SimplifyInstruction(PHI, IFI.DL)) {
+ auto &DL = Caller->getParent()->getDataLayout();
+ if (Value *V = SimplifyInstruction(PHI, DL, nullptr, nullptr,
+ &IFI.ACT->getAssumptionCache(*Caller))) {
PHI->replaceAllUsesWith(V);
PHI->eraseFromParent();
}