bool CallAnalyzer::visitAlloca(AllocaInst &I) {
// Check whether inlining will turn a dynamic alloca into a static
// alloca, and handle that case.
if (I.isArrayAllocation()) {
if (Constant *Size = SimplifiedValues.lookup(I.getArraySize())) {
ConstantInt *AllocSize = dyn_cast<ConstantInt>(Size);
assert(AllocSize && "Allocation size not a constant int?");
Type *Ty = I.getAllocatedType();
AllocatedSize += Ty->getPrimitiveSizeInBits() * AllocSize->getZExtValue();
return Base::visitAlloca(I);
}
}
// Accumulate the allocated size.
if (I.isStaticAlloca()) {
Type *Ty = I.getAllocatedType();
AllocatedSize += (DL ? DL->getTypeAllocSize(Ty) :
Ty->getPrimitiveSizeInBits());
}
// We will happily inline static alloca instructions.
if (I.isStaticAlloca())
return Base::visitAlloca(I);
// FIXME: This is overly conservative. Dynamic allocas are inefficient for
// a variety of reasons, and so we would like to not inline them into
// functions which don't currently have a dynamic alloca. This simply
// disables inlining altogether in the presence of a dynamic alloca.
HasDynamicAlloca = true;
return false;
}
void Lint::visitAllocaInst(AllocaInst &I) {
if (isa<ConstantInt>(I.getArraySize()))
// This isn't undefined behavior, it's just an obvious pessimization.
Assert(&I.getParent()->getParent()->getEntryBlock() == I.getParent(),
"Pessimization: Static alloca outside of entry block", &I);
// TODO: Check for an unusual size (MSB set?)
}
SizeOffsetEvalType ObjectSizeOffsetEvaluator::visitAllocaInst(AllocaInst &I) {
if (!I.getAllocatedType()->isSized())
return unknown();
// must be a VLA
assert(I.isArrayAllocation());
Value *ArraySize = I.getArraySize();
Value *Size = ConstantInt::get(ArraySize->getType(),
TD->getTypeAllocSize(I.getAllocatedType()));
Size = Builder.CreateMul(Size, ArraySize);
return std::make_pair(Size, Zero);
}
SizeOffsetType ObjectSizeOffsetVisitor::visitAllocaInst(AllocaInst &I) {
if (!I.getAllocatedType()->isSized())
return unknown();
APInt Size(IntTyBits, TD->getTypeAllocSize(I.getAllocatedType()));
if (!I.isArrayAllocation())
return std::make_pair(align(Size, I.getAlignment()), Zero);
Value *ArraySize = I.getArraySize();
if (const ConstantInt *C = dyn_cast<ConstantInt>(ArraySize)) {
Size *= C->getValue().zextOrSelf(IntTyBits);
return std::make_pair(align(Size, I.getAlignment()), Zero);
}
return unknown();
}
bool NVPTXAllocaHoisting::runOnFunction(Function &function) {
bool functionModified = false;
Function::iterator I = function.begin();
TerminatorInst *firstTerminatorInst = (I++)->getTerminator();
for (Function::iterator E = function.end(); I != E; ++I) {
for (BasicBlock::iterator BI = I->begin(), BE = I->end(); BI != BE;) {
AllocaInst *allocaInst = dyn_cast<AllocaInst>(BI++);
if (allocaInst && isa<ConstantInt>(allocaInst->getArraySize())) {
allocaInst->moveBefore(firstTerminatorInst);
functionModified = true;
}
}
}
return functionModified;
}
bool IRTranslator::translateStaticAlloca(const AllocaInst &AI) {
assert(AI.isStaticAlloca() && "only handle static allocas now");
MachineFunction &MF = MIRBuilder.getMF();
unsigned ElementSize = DL->getTypeStoreSize(AI.getAllocatedType());
unsigned Size =
ElementSize * cast<ConstantInt>(AI.getArraySize())->getZExtValue();
// Always allocate at least one byte.
Size = std::max(Size, 1u);
unsigned Alignment = AI.getAlignment();
if (!Alignment)
Alignment = DL->getABITypeAlignment(AI.getAllocatedType());
unsigned Res = getOrCreateVReg(AI);
int FI = MF.getFrameInfo().CreateStackObject(Size, Alignment, false, &AI);
MIRBuilder.buildFrameIndex(LLT::pointer(0), Res, FI);
return true;
}
int IRTranslator::getOrCreateFrameIndex(const AllocaInst &AI) {
if (FrameIndices.find(&AI) != FrameIndices.end())
return FrameIndices[&AI];
MachineFunction &MF = MIRBuilder.getMF();
unsigned ElementSize = DL->getTypeStoreSize(AI.getAllocatedType());
unsigned Size =
ElementSize * cast<ConstantInt>(AI.getArraySize())->getZExtValue();
// Always allocate at least one byte.
Size = std::max(Size, 1u);
unsigned Alignment = AI.getAlignment();
if (!Alignment)
Alignment = DL->getABITypeAlignment(AI.getAllocatedType());
int &FI = FrameIndices[&AI];
FI = MF.getFrameInfo().CreateStackObject(Size, Alignment, false, &AI);
return FI;
}
SizeOffsetType ObjectSizeOffsetVisitor::visitAllocaInst(AllocaInst &I) {
if (!I.getAllocatedType()->isSized())
return unknown();
APInt Size(IntTyBits, DL.getTypeAllocSize(I.getAllocatedType()));
if (!I.isArrayAllocation())
return std::make_pair(align(Size, I.getAlignment()), Zero);
Value *ArraySize = I.getArraySize();
if (const ConstantInt *C = dyn_cast<ConstantInt>(ArraySize)) {
APInt NumElems = C->getValue();
if (!CheckedZextOrTrunc(NumElems))
return unknown();
bool Overflow;
Size = Size.umul_ov(NumElems, Overflow);
return Overflow ? unknown() : std::make_pair(align(Size, I.getAlignment()),
Zero);
}
return unknown();
}
/// performCallSlotOptzn - takes a memcpy and a call that it depends on,
/// and checks for the possibility of a call slot optimization by having
/// the call write its result directly into the destination of the memcpy.
bool MemCpyOpt::performCallSlotOptzn(Instruction *cpy,
Value *cpyDest, Value *cpySrc,
uint64_t cpyLen, unsigned cpyAlign,
CallInst *C) {
// The general transformation to keep in mind is
//
// call @func(..., src, ...)
// memcpy(dest, src, ...)
//
// ->
//
// memcpy(dest, src, ...)
// call @func(..., dest, ...)
//
// Since moving the memcpy is technically awkward, we additionally check that
// src only holds uninitialized values at the moment of the call, meaning that
// the memcpy can be discarded rather than moved.
// Deliberately get the source and destination with bitcasts stripped away,
// because we'll need to do type comparisons based on the underlying type.
CallSite CS(C);
// Require that src be an alloca. This simplifies the reasoning considerably.
AllocaInst *srcAlloca = dyn_cast<AllocaInst>(cpySrc);
if (!srcAlloca)
return false;
// Check that all of src is copied to dest.
if (TD == 0) return false;
ConstantInt *srcArraySize = dyn_cast<ConstantInt>(srcAlloca->getArraySize());
if (!srcArraySize)
return false;
uint64_t srcSize = TD->getTypeAllocSize(srcAlloca->getAllocatedType()) *
srcArraySize->getZExtValue();
if (cpyLen < srcSize)
return false;
// Check that dest points to memory that is at least as aligned as src.
unsigned srcAlign = srcAlloca->getAlignment();
if (!srcAlign)
srcAlign = TD->getABITypeAlignment(srcAlloca->getAllocatedType());
bool isDestSufficientlyAligned = srcAlign <= cpyAlign;
// If dest is not aligned enough and we can't increase its alignment then
// bail out.
if (!isDestSufficientlyAligned && !isa<AllocaInst>(cpyDest))
return false;
// Check that accessing the first srcSize bytes of dest will not cause a
// trap. Otherwise the transform is invalid since it might cause a trap
// to occur earlier than it otherwise would.
if (AllocaInst *A = dyn_cast<AllocaInst>(cpyDest)) {
// The destination is an alloca. Check it is larger than srcSize.
ConstantInt *destArraySize = dyn_cast<ConstantInt>(A->getArraySize());
if (!destArraySize)
return false;
uint64_t destSize = TD->getTypeAllocSize(A->getAllocatedType()) *
destArraySize->getZExtValue();
if (destSize < srcSize)
return false;
} else if (Argument *A = dyn_cast<Argument>(cpyDest)) {
// If the destination is an sret parameter then only accesses that are
// outside of the returned struct type can trap.
if (!A->hasStructRetAttr())
return false;
Type *StructTy = cast<PointerType>(A->getType())->getElementType();
uint64_t destSize = TD->getTypeAllocSize(StructTy);
if (destSize < srcSize)
return false;
} else {
return false;
}
// Check that src is not accessed except via the call and the memcpy. This
// guarantees that it holds only undefined values when passed in (so the final
// memcpy can be dropped), that it is not read or written between the call and
// the memcpy, and that writing beyond the end of it is undefined.
SmallVector<User*, 8> srcUseList(srcAlloca->use_begin(),
srcAlloca->use_end());
while (!srcUseList.empty()) {
User *UI = srcUseList.pop_back_val();
if (isa<BitCastInst>(UI)) {
for (User::use_iterator I = UI->use_begin(), E = UI->use_end();
I != E; ++I)
srcUseList.push_back(*I);
} else if (GetElementPtrInst *G = dyn_cast<GetElementPtrInst>(UI)) {
if (G->hasAllZeroIndices())
for (User::use_iterator I = UI->use_begin(), E = UI->use_end();
I != E; ++I)
srcUseList.push_back(*I);
else
return false;
} else if (UI != C && UI != cpy) {
return false;
}
}
// Since we're changing the parameter to the callsite, we need to make sure
// that what would be the new parameter dominates the callsite.
DominatorTree &DT = getAnalysis<DominatorTree>();
if (Instruction *cpyDestInst = dyn_cast<Instruction>(cpyDest))
if (!DT.dominates(cpyDestInst, C))
return false;
// In addition to knowing that the call does not access src in some
// unexpected manner, for example via a global, which we deduce from
// the use analysis, we also need to know that it does not sneakily
// access dest. We rely on AA to figure this out for us.
AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
AliasAnalysis::ModRefResult MR = AA.getModRefInfo(C, cpyDest, srcSize);
// If necessary, perform additional analysis.
if (MR != AliasAnalysis::NoModRef)
MR = AA.callCapturesBefore(C, cpyDest, srcSize, &DT);
if (MR != AliasAnalysis::NoModRef)
return false;
// All the checks have passed, so do the transformation.
bool changedArgument = false;
for (unsigned i = 0; i < CS.arg_size(); ++i)
if (CS.getArgument(i)->stripPointerCasts() == cpySrc) {
Value *Dest = cpySrc->getType() == cpyDest->getType() ? cpyDest
: CastInst::CreatePointerCast(cpyDest, cpySrc->getType(),
cpyDest->getName(), C);
changedArgument = true;
if (CS.getArgument(i)->getType() == Dest->getType())
CS.setArgument(i, Dest);
else
CS.setArgument(i, CastInst::CreatePointerCast(Dest,
CS.getArgument(i)->getType(), Dest->getName(), C));
}
if (!changedArgument)
return false;
// If the destination wasn't sufficiently aligned then increase its alignment.
if (!isDestSufficientlyAligned) {
assert(isa<AllocaInst>(cpyDest) && "Can only increase alloca alignment!");
cast<AllocaInst>(cpyDest)->setAlignment(srcAlign);
}
// Drop any cached information about the call, because we may have changed
// its dependence information by changing its parameter.
MD->removeInstruction(C);
// Remove the memcpy.
MD->removeInstruction(cpy);
++NumMemCpyInstr;
return true;
}
/// performCallSlotOptzn - takes a memcpy and a call that it depends on,
/// and checks for the possibility of a call slot optimization by having
/// the call write its result directly into the destination of the memcpy.
bool MemCpyOpt::performCallSlotOptzn(Instruction *cpy,
Value *cpyDest, Value *cpySrc,
uint64_t cpyLen, unsigned cpyAlign,
CallInst *C) {
// The general transformation to keep in mind is
//
// call @func(..., src, ...)
// memcpy(dest, src, ...)
//
// ->
//
// memcpy(dest, src, ...)
// call @func(..., dest, ...)
//
// Since moving the memcpy is technically awkward, we additionally check that
// src only holds uninitialized values at the moment of the call, meaning that
// the memcpy can be discarded rather than moved.
// Deliberately get the source and destination with bitcasts stripped away,
// because we'll need to do type comparisons based on the underlying type.
CallSite CS(C);
// Require that src be an alloca. This simplifies the reasoning considerably.
AllocaInst *srcAlloca = dyn_cast<AllocaInst>(cpySrc);
if (!srcAlloca)
return false;
ConstantInt *srcArraySize = dyn_cast<ConstantInt>(srcAlloca->getArraySize());
if (!srcArraySize)
return false;
const DataLayout &DL = cpy->getModule()->getDataLayout();
uint64_t srcSize = DL.getTypeAllocSize(srcAlloca->getAllocatedType()) *
srcArraySize->getZExtValue();
if (cpyLen < srcSize)
return false;
// Check that accessing the first srcSize bytes of dest will not cause a
// trap. Otherwise the transform is invalid since it might cause a trap
// to occur earlier than it otherwise would.
if (AllocaInst *A = dyn_cast<AllocaInst>(cpyDest)) {
// The destination is an alloca. Check it is larger than srcSize.
ConstantInt *destArraySize = dyn_cast<ConstantInt>(A->getArraySize());
if (!destArraySize)
return false;
uint64_t destSize = DL.getTypeAllocSize(A->getAllocatedType()) *
destArraySize->getZExtValue();
if (destSize < srcSize)
return false;
} else if (Argument *A = dyn_cast<Argument>(cpyDest)) {
if (A->getDereferenceableBytes() < srcSize) {
// If the destination is an sret parameter then only accesses that are
// outside of the returned struct type can trap.
if (!A->hasStructRetAttr())
return false;
Type *StructTy = cast<PointerType>(A->getType())->getElementType();
if (!StructTy->isSized()) {
// The call may never return and hence the copy-instruction may never
// be executed, and therefore it's not safe to say "the destination
// has at least <cpyLen> bytes, as implied by the copy-instruction",
return false;
}
uint64_t destSize = DL.getTypeAllocSize(StructTy);
if (destSize < srcSize)
return false;
}
} else {
return false;
}
// Check that dest points to memory that is at least as aligned as src.
unsigned srcAlign = srcAlloca->getAlignment();
if (!srcAlign)
srcAlign = DL.getABITypeAlignment(srcAlloca->getAllocatedType());
bool isDestSufficientlyAligned = srcAlign <= cpyAlign;
// If dest is not aligned enough and we can't increase its alignment then
// bail out.
if (!isDestSufficientlyAligned && !isa<AllocaInst>(cpyDest))
return false;
// Check that src is not accessed except via the call and the memcpy. This
// guarantees that it holds only undefined values when passed in (so the final
// memcpy can be dropped), that it is not read or written between the call and
// the memcpy, and that writing beyond the end of it is undefined.
SmallVector<User*, 8> srcUseList(srcAlloca->user_begin(),
srcAlloca->user_end());
while (!srcUseList.empty()) {
User *U = srcUseList.pop_back_val();
if (isa<BitCastInst>(U) || isa<AddrSpaceCastInst>(U)) {
for (User *UU : U->users())
srcUseList.push_back(UU);
continue;
}
if (GetElementPtrInst *G = dyn_cast<GetElementPtrInst>(U)) {
if (!G->hasAllZeroIndices())
return false;
for (User *UU : U->users())
srcUseList.push_back(UU);
continue;
}
if (const IntrinsicInst *IT = dyn_cast<IntrinsicInst>(U))
if (IT->getIntrinsicID() == Intrinsic::lifetime_start ||
IT->getIntrinsicID() == Intrinsic::lifetime_end)
continue;
if (U != C && U != cpy)
return false;
}
// Check that src isn't captured by the called function since the
// transformation can cause aliasing issues in that case.
for (unsigned i = 0, e = CS.arg_size(); i != e; ++i)
if (CS.getArgument(i) == cpySrc && !CS.doesNotCapture(i))
return false;
// Since we're changing the parameter to the callsite, we need to make sure
// that what would be the new parameter dominates the callsite.
DominatorTree &DT = getAnalysis<DominatorTreeWrapperPass>().getDomTree();
if (Instruction *cpyDestInst = dyn_cast<Instruction>(cpyDest))
if (!DT.dominates(cpyDestInst, C))
return false;
// In addition to knowing that the call does not access src in some
// unexpected manner, for example via a global, which we deduce from
// the use analysis, we also need to know that it does not sneakily
// access dest. We rely on AA to figure this out for us.
AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
AliasAnalysis::ModRefResult MR = AA.getModRefInfo(C, cpyDest, srcSize);
// If necessary, perform additional analysis.
if (MR != AliasAnalysis::NoModRef)
MR = AA.callCapturesBefore(C, cpyDest, srcSize, &DT);
if (MR != AliasAnalysis::NoModRef)
return false;
// All the checks have passed, so do the transformation.
bool changedArgument = false;
for (unsigned i = 0; i < CS.arg_size(); ++i)
if (CS.getArgument(i)->stripPointerCasts() == cpySrc) {
Value *Dest = cpySrc->getType() == cpyDest->getType() ? cpyDest
: CastInst::CreatePointerCast(cpyDest, cpySrc->getType(),
cpyDest->getName(), C);
changedArgument = true;
if (CS.getArgument(i)->getType() == Dest->getType())
CS.setArgument(i, Dest);
else
CS.setArgument(i, CastInst::CreatePointerCast(Dest,
CS.getArgument(i)->getType(), Dest->getName(), C));
}
if (!changedArgument)
return false;
// If the destination wasn't sufficiently aligned then increase its alignment.
if (!isDestSufficientlyAligned) {
assert(isa<AllocaInst>(cpyDest) && "Can only increase alloca alignment!");
cast<AllocaInst>(cpyDest)->setAlignment(srcAlign);
}
// Drop any cached information about the call, because we may have changed
// its dependence information by changing its parameter.
MD->removeInstruction(C);
// Update AA metadata
// FIXME: MD_tbaa_struct and MD_mem_parallel_loop_access should also be
// handled here, but combineMetadata doesn't support them yet
unsigned KnownIDs[] = {
LLVMContext::MD_tbaa,
LLVMContext::MD_alias_scope,
LLVMContext::MD_noalias,
};
combineMetadata(C, cpy, KnownIDs);
// Remove the memcpy.
MD->removeInstruction(cpy);
++NumMemCpyInstr;
return true;
}
/// 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.
///
/// 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) {
Instruction *TheCall = CS.getInstruction();
assert(TheCall->getParent() && TheCall->getParent()->getParent() &&
"Instruction not in function!");
// If IFI has any state in it, zap it before we fill it in.
IFI.reset();
const Function *CalledFunc = CS.getCalledFunction();
if (CalledFunc == 0 || // Can't inline external function or indirect
CalledFunc->isDeclaration() || // call, or call to a vararg function!
CalledFunc->getFunctionType()->isVarArg()) return false;
// If the call to the callee is not a tail call, we must clear the 'tail'
// flags on any calls that we inline.
bool MustClearTailCallFlags =
!(isa<CallInst>(TheCall) && cast<CallInst>(TheCall)->isTailCall());
// If the call to the callee cannot throw, set the 'nounwind' flag on any
// calls that we inline.
bool MarkNoUnwind = CS.doesNotThrow();
BasicBlock *OrigBB = TheCall->getParent();
Function *Caller = OrigBB->getParent();
// GC poses two hazards to inlining, which only occur when the callee has GC:
// 1. If the caller has no GC, then the callee's GC must be propagated to the
// caller.
// 2. If the caller has a differing GC, it is invalid to inline.
if (CalledFunc->hasGC()) {
if (!Caller->hasGC())
Caller->setGC(CalledFunc->getGC());
else if (CalledFunc->getGC() != Caller->getGC())
return false;
}
// Get the personality function from the callee if it contains a landing pad.
Value *CalleePersonality = 0;
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;
}
// 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;
}
}
// Get an iterator to the last basic block in the function, which will have
// the new function inlined after it.
Function::iterator LastBlock = &Caller->back();
// Make sure to capture all of the return instructions from the cloned
// function.
SmallVector<ReturnInst*, 8> Returns;
ClonedCodeInfo InlinedFunctionInfo;
Function::iterator FirstNewBlock;
{ // Scope to destroy VMap after cloning.
ValueToValueMapTy VMap;
assert(CalledFunc->arg_size() == CS.arg_size() &&
"No varargs calls can be inlined!");
// Calculate the vector of arguments to pass into the function cloner, which
// matches up the formal to the actual argument values.
CallSite::arg_iterator AI = CS.arg_begin();
unsigned ArgNo = 0;
for (Function::const_arg_iterator I = CalledFunc->arg_begin(),
E = CalledFunc->arg_end(); I != E; ++I, ++AI, ++ArgNo) {
Value *ActualArg = *AI;
const Argument *Arg = I;
// When byval arguments actually inlined, we need to make the copy implied
// by them explicit. However, we don't do this if the callee is readonly
// or readnone, because the copy would be unneeded: the callee doesn't
// modify the struct.
if (CS.isByValArgument(ArgNo)) {
ActualArg = HandleByValArgument(ActualArg, Arg, TheCall, CalledFunc, IFI,
CalledFunc->getParamAlignment(ArgNo+1));
// Calls that we inline may use the new alloca, so we need to clear
// their 'tail' flags if HandleByValArgument introduced a new alloca and
// the callee has calls.
MustClearTailCallFlags |= ActualArg != *AI;
}
VMap[I] = ActualArg;
}
// 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,
/*ModuleLevelChanges=*/false, Returns, ".i",
&InlinedFunctionInfo, IFI.TD, TheCall);
// Remember the first block that is newly cloned over.
FirstNewBlock = LastBlock; ++FirstNewBlock;
// Update the callgraph if requested.
if (IFI.CG)
UpdateCallGraphAfterInlining(CS, FirstNewBlock, VMap, IFI);
// Update inlined instructions' line number information.
fixupLineNumbers(Caller, FirstNewBlock, TheCall);
}
// If there are any alloca instructions in the block that used to be the entry
// block for the callee, move them to the entry block of the caller. First
// calculate which instruction they should be inserted before. We insert the
// instructions at the end of the current alloca list.
{
BasicBlock::iterator InsertPoint = Caller->begin()->begin();
for (BasicBlock::iterator I = FirstNewBlock->begin(),
E = FirstNewBlock->end(); I != E; ) {
AllocaInst *AI = dyn_cast<AllocaInst>(I++);
if (AI == 0) continue;
// If the alloca is now dead, remove it. This often occurs due to code
// specialization.
if (AI->use_empty()) {
AI->eraseFromParent();
continue;
}
if (!isa<Constant>(AI->getArraySize()))
continue;
// Keep track of the static allocas that we inline into the caller.
IFI.StaticAllocas.push_back(AI);
// Scan for the block of allocas that we can move over, and move them
// all at once.
while (isa<AllocaInst>(I) &&
isa<Constant>(cast<AllocaInst>(I)->getArraySize())) {
IFI.StaticAllocas.push_back(cast<AllocaInst>(I));
++I;
}
// Transfer all of the allocas over in a block. Using splice means
// that the instructions aren't removed from the symbol table, then
// reinserted.
Caller->getEntryBlock().getInstList().splice(InsertPoint,
FirstNewBlock->getInstList(),
AI, I);
}
}
// Leave lifetime markers for the static alloca's, scoping them to the
// function we just inlined.
if (InsertLifetime && !IFI.StaticAllocas.empty()) {
IRBuilder<> builder(FirstNewBlock->begin());
for (unsigned ai = 0, ae = IFI.StaticAllocas.size(); ai != ae; ++ai) {
AllocaInst *AI = IFI.StaticAllocas[ai];
// If the alloca is already scoped to something smaller than the whole
// function then there's no need to add redundant, less accurate markers.
if (hasLifetimeMarkers(AI))
continue;
// Try to determine the size of the allocation.
ConstantInt *AllocaSize = 0;
if (ConstantInt *AIArraySize =
dyn_cast<ConstantInt>(AI->getArraySize())) {
if (IFI.TD) {
Type *AllocaType = AI->getAllocatedType();
uint64_t AllocaTypeSize = IFI.TD->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);
}
}
}
builder.CreateLifetimeStart(AI, AllocaSize);
for (unsigned ri = 0, re = Returns.size(); ri != re; ++ri) {
IRBuilder<> builder(Returns[ri]);
builder.CreateLifetimeEnd(AI, AllocaSize);
}
}
}
// If the inlined code contained dynamic alloca instructions, wrap the inlined
// code with llvm.stacksave/llvm.stackrestore intrinsics.
if (InlinedFunctionInfo.ContainsDynamicAllocas) {
Module *M = Caller->getParent();
// Get the two intrinsics we care about.
Function *StackSave = Intrinsic::getDeclaration(M, Intrinsic::stacksave);
Function *StackRestore=Intrinsic::getDeclaration(M,Intrinsic::stackrestore);
// Insert the llvm.stacksave.
CallInst *SavedPtr = IRBuilder<>(FirstNewBlock, FirstNewBlock->begin())
.CreateCall(StackSave, "savedstack");
// Insert a call to llvm.stackrestore before any return instructions in the
// inlined function.
for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
IRBuilder<>(Returns[i]).CreateCall(StackRestore, SavedPtr);
}
}
// If we are inlining tail call instruction through a call site that isn't
// marked 'tail', we must remove the tail marker for any calls in the inlined
// code. Also, calls inlined through a 'nounwind' call site should be marked
// 'nounwind'.
if (InlinedFunctionInfo.ContainsCalls &&
(MustClearTailCallFlags || MarkNoUnwind)) {
for (Function::iterator BB = FirstNewBlock, E = Caller->end();
BB != E; ++BB)
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
if (CallInst *CI = dyn_cast<CallInst>(I)) {
if (MustClearTailCallFlags)
CI->setTailCall(false);
if (MarkNoUnwind)
CI->setDoesNotThrow();
}
}
// 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 we cloned in _exactly one_ basic block, and if that block ends in a
// return instruction, we splice the body of the inlined callee directly into
// the calling basic block.
if (Returns.size() == 1 && std::distance(FirstNewBlock, Caller->end()) == 1) {
// Move all of the instructions right before the call.
OrigBB->getInstList().splice(TheCall, FirstNewBlock->getInstList(),
FirstNewBlock->begin(), FirstNewBlock->end());
// Remove the cloned basic block.
Caller->getBasicBlockList().pop_back();
// If the call site was an invoke instruction, add a branch to the normal
// destination.
if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall)) {
BranchInst *NewBr = BranchInst::Create(II->getNormalDest(), TheCall);
NewBr->setDebugLoc(Returns[0]->getDebugLoc());
}
// If the return instruction returned a value, replace uses of the call with
// uses of the returned value.
if (!TheCall->use_empty()) {
ReturnInst *R = Returns[0];
if (TheCall == R->getReturnValue())
TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
else
TheCall->replaceAllUsesWith(R->getReturnValue());
}
// Since we are now done with the Call/Invoke, we can delete it.
TheCall->eraseFromParent();
// Since we are now done with the return instruction, delete it also.
Returns[0]->eraseFromParent();
// We are now done with the inlining.
return true;
}
// Otherwise, we have the normal case, of more than one block to inline or
// multiple return sites.
// We want to clone the entire callee function into the hole between the
// "starter" and "ender" blocks. How we accomplish this depends on whether
// this is an invoke instruction or a call instruction.
BasicBlock *AfterCallBB;
BranchInst *CreatedBranchToNormalDest = NULL;
if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall)) {
// Add an unconditional branch to make this look like the CallInst case...
CreatedBranchToNormalDest = BranchInst::Create(II->getNormalDest(), TheCall);
// Split the basic block. This guarantees that no PHI nodes will have to be
// updated due to new incoming edges, and make the invoke case more
// symmetric to the call case.
AfterCallBB = OrigBB->splitBasicBlock(CreatedBranchToNormalDest,
CalledFunc->getName()+".exit");
} else { // It's a call
// If this is a call instruction, we need to split the basic block that
// the call lives in.
//
AfterCallBB = OrigBB->splitBasicBlock(TheCall,
CalledFunc->getName()+".exit");
}
// Change the branch that used to go to AfterCallBB to branch to the first
// basic block of the inlined function.
//
TerminatorInst *Br = OrigBB->getTerminator();
assert(Br && Br->getOpcode() == Instruction::Br &&
"splitBasicBlock broken!");
Br->setOperand(0, FirstNewBlock);
// Now that the function is correct, make it a little bit nicer. In
// particular, move the basic blocks inserted from the end of the function
// into the space made by splitting the source basic block.
Caller->getBasicBlockList().splice(AfterCallBB, Caller->getBasicBlockList(),
FirstNewBlock, Caller->end());
// Handle all of the return instructions that we just cloned in, and eliminate
// any users of the original call/invoke instruction.
Type *RTy = CalledFunc->getReturnType();
PHINode *PHI = 0;
if (Returns.size() > 1) {
// The PHI node should go at the front of the new basic block to merge all
// possible incoming values.
if (!TheCall->use_empty()) {
PHI = PHINode::Create(RTy, Returns.size(), TheCall->getName(),
AfterCallBB->begin());
// Anything that used the result of the function call should now use the
// PHI node as their operand.
TheCall->replaceAllUsesWith(PHI);
}
// Loop over all of the return instructions adding entries to the PHI node
// as appropriate.
if (PHI) {
for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
ReturnInst *RI = Returns[i];
assert(RI->getReturnValue()->getType() == PHI->getType() &&
"Ret value not consistent in function!");
PHI->addIncoming(RI->getReturnValue(), RI->getParent());
}
}
// Add a branch to the merge points and remove return instructions.
DebugLoc Loc;
for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
ReturnInst *RI = Returns[i];
BranchInst* BI = BranchInst::Create(AfterCallBB, RI);
Loc = RI->getDebugLoc();
BI->setDebugLoc(Loc);
RI->eraseFromParent();
}
// We need to set the debug location to *somewhere* inside the
// inlined function. The line number may be nonsensical, but the
// instruction will at least be associated with the right
// function.
if (CreatedBranchToNormalDest)
CreatedBranchToNormalDest->setDebugLoc(Loc);
} else if (!Returns.empty()) {
// Otherwise, if there is exactly one return value, just replace anything
// using the return value of the call with the computed value.
if (!TheCall->use_empty()) {
if (TheCall == Returns[0]->getReturnValue())
TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
else
TheCall->replaceAllUsesWith(Returns[0]->getReturnValue());
}
// Update PHI nodes that use the ReturnBB to use the AfterCallBB.
BasicBlock *ReturnBB = Returns[0]->getParent();
ReturnBB->replaceAllUsesWith(AfterCallBB);
// Splice the code from the return block into the block that it will return
// to, which contains the code that was after the call.
AfterCallBB->getInstList().splice(AfterCallBB->begin(),
ReturnBB->getInstList());
if (CreatedBranchToNormalDest)
CreatedBranchToNormalDest->setDebugLoc(Returns[0]->getDebugLoc());
// Delete the return instruction now and empty ReturnBB now.
Returns[0]->eraseFromParent();
ReturnBB->eraseFromParent();
} else if (!TheCall->use_empty()) {
// No returns, but something is using the return value of the call. Just
// nuke the result.
TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
}
// Since we are now done with the Call/Invoke, we can delete it.
TheCall->eraseFromParent();
// We should always be able to fold the entry block of the function into the
// single predecessor of the block...
assert(cast<BranchInst>(Br)->isUnconditional() && "splitBasicBlock broken!");
BasicBlock *CalleeEntry = cast<BranchInst>(Br)->getSuccessor(0);
// Splice the code entry block into calling block, right before the
// unconditional branch.
CalleeEntry->replaceAllUsesWith(OrigBB); // Update PHI nodes
OrigBB->getInstList().splice(Br, CalleeEntry->getInstList());
// Remove the unconditional branch.
OrigBB->getInstList().erase(Br);
// Now we can remove the CalleeEntry block, which is now empty.
Caller->getBasicBlockList().erase(CalleeEntry);
// If we inserted a phi node, check to see if it has a single value (e.g. all
// 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.TD)) {
PHI->replaceAllUsesWith(V);
PHI->eraseFromParent();
}
}
return true;
}
/*
* Rewrite OpenMP call sites and their associated kernel functions -- the folloiwng pattern
call void @GOMP_parallel_start(void (i8*)* @_Z20initialize_variablesiPfS_.omp_fn.4, i8* %.omp_data_o.5571, i32 0) nounwind
call void @_Z20initialize_variablesiPfS_.omp_fn.4(i8* %.omp_data_o.5571) nounwind
call void @GOMP_parallel_end() nounwind
*/
void HeteroOMPTransform::rewrite_omp_call_sites(Module &M) {
SmallVector<Instruction *, 16> toDelete;
DenseMap<Value *, Value *> ValueMap;
for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I){
if (!I->isDeclaration()) {
for (Function::iterator BBI = I->begin(), BBE = I->end(); BBI != BBE; ++BBI) {
bool match = false;
for (BasicBlock::iterator INSNI = BBI->begin(), INSNE = BBI->end(); INSNI != INSNE; ++INSNI) {
if (isa<CallInst>(INSNI)) {
CallSite CI(cast<Instruction>(INSNI));
if (CI.getCalledFunction() != NULL){
string called_func_name = CI.getCalledFunction()->getName();
if (called_func_name == OMP_PARALLEL_START_NAME && CI.arg_size() == 3) {
// change alloc to malloc_shared
// %5 = call i8* @_Z13malloc_sharedm(i64 20) ; <i8*> [#uses=5]
// %6 = bitcast i8* %5 to float* ; <float*> [#uses=2]
AllocaInst *AllocCall;
Value *arg_0 = CI.getArgument(0); // function
Value *arg_1 = CI.getArgument(1); // context
Value *loop_ub = NULL;
Function *function;
BitCastInst* BCI;
Function *kernel_function;
BasicBlock::iterator iI(*INSNI);
//BasicBlock::iterator iJ = iI+1;
iI++; iI++;
//BasicBlock::iterator iK = iI;
CallInst /**next,*/ *next_next;
if (arg_0 != NULL && arg_1 != NULL /*&& (next = dyn_cast<CallInst>(*iJ))*/
&& (next_next = dyn_cast<CallInst>(iI)) && (next_next->getCalledFunction() != NULL)
&& (next_next->getCalledFunction()->getName() == OMP_PARALLEL_END_NAME)
&& (BCI = dyn_cast<BitCastInst>(arg_1)) && (AllocCall = dyn_cast<AllocaInst>(BCI->getOperand(0)))
&& (function = dyn_cast<Function>(arg_0)) && (loop_ub = find_loop_upper_bound (AllocCall))
&& (kernel_function=convert_to_kernel_function (M, function))){
SmallVector<Value*, 16> Args;
Args.push_back(AllocCall->getArraySize());
Instruction *MallocCall = CallInst::Create(mallocFnTy, Args, "", AllocCall);
CastInst *MallocCast = CastInst::Create(Instruction::BitCast, MallocCall, AllocCall->getType(), "", AllocCall);
ValueMap[AllocCall] = MallocCast;
//AllocCall->replaceAllUsesWith(MallocCall);
// Add offload function
Args.clear();
Args.push_back(loop_ub);
Args.push_back(BCI);
Args.push_back(kernel_function);
if (offloadFnTy == NULL) {
init_offload_type(M, kernel_function);
}
Instruction *call = CallInst::Create(offloadFnTy, Args, "", INSNI);
if (find(toDelete.begin(), toDelete.end(), AllocCall) == toDelete.end()){
toDelete.push_back(AllocCall);
}
toDelete.push_back(&(*INSNI));
match = true;
}
}
else if (called_func_name == OMP_PARALLEL_END_NAME && CI.arg_size() == 0 && match) {
toDelete.push_back(&(*INSNI));
match = false;
}
else if (match) {
toDelete.push_back(&(*INSNI));
}
}
}
}
}
}
}
/* Replace AllocCalls by MallocCalls */
for(DenseMap<Value *, Value *>::iterator I = ValueMap.begin(), E = ValueMap.end(); I != E; I++) {
I->first->replaceAllUsesWith(I->second);
}
/* delete the instructions for get_omp_num_thread and get_omp_thread_num */
while(!toDelete.empty()) {
Instruction *g = toDelete.back();
toDelete.pop_back();
g->eraseFromParent();
}
}
//
// Method: insertBadAllocationSizes()
//
// Description:
// This method will look for allocations and change their size to be
// incorrect. It does the following:
// o) Changes the number of array elements allocated by alloca and malloc.
//
// Return value:
// true - The module was modified.
// false - The module was left unmodified.
//
bool
FaultInjector::insertBadAllocationSizes (Function & F) {
// Worklist of allocation sites to rewrite
std::vector<AllocaInst * > WorkList;
for (Function::iterator fI = F.begin(), fE = F.end(); fI != fE; ++fI) {
BasicBlock & BB = *fI;
for (BasicBlock::iterator I = BB.begin(), bE = BB.end(); I != bE; ++I) {
if (AllocaInst * AI = dyn_cast<AllocaInst>(I)) {
if (AI->isArrayAllocation()) {
// Skip if we should not insert a fault.
if (!doFault()) continue;
WorkList.push_back(AI);
}
}
}
}
while (WorkList.size()) {
AllocaInst * AI = WorkList.back();
WorkList.pop_back();
//
// Print information about where the fault is being inserted.
//
printSourceInfo ("Bad allocation size", AI);
Instruction * NewAlloc = 0;
NewAlloc = new AllocaInst (AI->getAllocatedType(),
ConstantInt::get(Int32Type,0),
AI->getAlignment(),
AI->getName(),
AI);
AI->replaceAllUsesWith (NewAlloc);
AI->eraseFromParent();
++BadSizes;
}
//
// Try harder to make bad allocation sizes.
//
WorkList.clear();
for (Function::iterator fI = F.begin(), fE = F.end(); fI != fE; ++fI) {
BasicBlock & BB = *fI;
for (BasicBlock::iterator I = BB.begin(), bE = BB.end(); I != bE; ++I) {
if (AllocaInst * AI = dyn_cast<AllocaInst>(I)) {
//
// Determine if this is a data type that we can make smaller.
//
if (((TD->getTypeAllocSize(AI->getAllocatedType())) > 4) && doFault()) {
WorkList.push_back(AI);
}
}
}
}
//
// Replace these allocations with an allocation of an integer and cast the
// result back into the appropriate type.
//
while (WorkList.size()) {
AllocaInst * AI = WorkList.back();
WorkList.pop_back();
Instruction * NewAlloc = 0;
NewAlloc = new AllocaInst (Int32Type,
AI->getArraySize(),
AI->getAlignment(),
AI->getName(),
AI);
NewAlloc = castTo (NewAlloc, AI->getType(), "", AI);
AI->replaceAllUsesWith (NewAlloc);
AI->eraseFromParent();
++BadSizes;
}
return (BadSizes > 0);
}
// 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.
//
// 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. Similiarly this will inline a recursive
// function by one level.
//
bool llvm::InlineFunction(CallSite CS, InlineFunctionInfo &IFI) {
Instruction *TheCall = CS.getInstruction();
LLVMContext &Context = TheCall->getContext();
assert(TheCall->getParent() && TheCall->getParent()->getParent() &&
"Instruction not in function!");
// If IFI has any state in it, zap it before we fill it in.
IFI.reset();
const Function *CalledFunc = CS.getCalledFunction();
if (CalledFunc == 0 || // Can't inline external function or indirect
CalledFunc->isDeclaration() || // call, or call to a vararg function!
CalledFunc->getFunctionType()->isVarArg()) return false;
// If the call to the callee is not a tail call, we must clear the 'tail'
// flags on any calls that we inline.
bool MustClearTailCallFlags =
!(isa<CallInst>(TheCall) && cast<CallInst>(TheCall)->isTailCall());
// If the call to the callee cannot throw, set the 'nounwind' flag on any
// calls that we inline.
bool MarkNoUnwind = CS.doesNotThrow();
BasicBlock *OrigBB = TheCall->getParent();
Function *Caller = OrigBB->getParent();
// GC poses two hazards to inlining, which only occur when the callee has GC:
// 1. If the caller has no GC, then the callee's GC must be propagated to the
// caller.
// 2. If the caller has a differing GC, it is invalid to inline.
if (CalledFunc->hasGC()) {
if (!Caller->hasGC())
Caller->setGC(CalledFunc->getGC());
else if (CalledFunc->getGC() != Caller->getGC())
return false;
}
// Get an iterator to the last basic block in the function, which will have
// the new function inlined after it.
//
Function::iterator LastBlock = &Caller->back();
// Make sure to capture all of the return instructions from the cloned
// function.
SmallVector<ReturnInst*, 8> Returns;
ClonedCodeInfo InlinedFunctionInfo;
Function::iterator FirstNewBlock;
{ // Scope to destroy VMap after cloning.
ValueMap<const Value*, Value*> VMap;
assert(CalledFunc->arg_size() == CS.arg_size() &&
"No varargs calls can be inlined!");
// Calculate the vector of arguments to pass into the function cloner, which
// matches up the formal to the actual argument values.
CallSite::arg_iterator AI = CS.arg_begin();
unsigned ArgNo = 0;
for (Function::const_arg_iterator I = CalledFunc->arg_begin(),
E = CalledFunc->arg_end(); I != E; ++I, ++AI, ++ArgNo) {
Value *ActualArg = *AI;
// When byval arguments actually inlined, we need to make the copy implied
// by them explicit. However, we don't do this if the callee is readonly
// or readnone, because the copy would be unneeded: the callee doesn't
// modify the struct.
if (CalledFunc->paramHasAttr(ArgNo+1, Attribute::ByVal) &&
!CalledFunc->onlyReadsMemory()) {
const Type *AggTy = cast<PointerType>(I->getType())->getElementType();
const Type *VoidPtrTy =
Type::getInt8PtrTy(Context);
// Create the alloca. If we have TargetData, use nice alignment.
unsigned Align = 1;
if (IFI.TD) Align = IFI.TD->getPrefTypeAlignment(AggTy);
Value *NewAlloca = new AllocaInst(AggTy, 0, Align,
I->getName(),
&*Caller->begin()->begin());
// Emit a memcpy.
const Type *Tys[3] = {VoidPtrTy, VoidPtrTy, Type::getInt64Ty(Context)};
Function *MemCpyFn = Intrinsic::getDeclaration(Caller->getParent(),
Intrinsic::memcpy,
Tys, 3);
Value *DestCast = new BitCastInst(NewAlloca, VoidPtrTy, "tmp", TheCall);
Value *SrcCast = new BitCastInst(*AI, VoidPtrTy, "tmp", TheCall);
Value *Size;
if (IFI.TD == 0)
Size = ConstantExpr::getSizeOf(AggTy);
else
Size = ConstantInt::get(Type::getInt64Ty(Context),
IFI.TD->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[] = {
DestCast, SrcCast, Size,
ConstantInt::get(Type::getInt32Ty(Context), 1),
ConstantInt::get(Type::getInt1Ty(Context), 0)
};
CallInst *TheMemCpy =
CallInst::Create(MemCpyFn, CallArgs, CallArgs+5, "", TheCall);
// If we have a call graph, update it.
if (CallGraph *CG = IFI.CG) {
CallGraphNode *MemCpyCGN = CG->getOrInsertFunction(MemCpyFn);
CallGraphNode *CallerNode = (*CG)[Caller];
CallerNode->addCalledFunction(TheMemCpy, MemCpyCGN);
}
// Uses of the argument in the function should use our new alloca
// instead.
ActualArg = NewAlloca;
// Calls that we inline may use the new alloca, so we need to clear
// their 'tail' flags.
MustClearTailCallFlags = true;
}
VMap[I] = ActualArg;
}
// 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, Returns, ".i",
&InlinedFunctionInfo, IFI.TD, TheCall);
// Remember the first block that is newly cloned over.
FirstNewBlock = LastBlock; ++FirstNewBlock;
// Update the callgraph if requested.
if (IFI.CG)
UpdateCallGraphAfterInlining(CS, FirstNewBlock, VMap, IFI);
}
// If there are any alloca instructions in the block that used to be the entry
// block for the callee, move them to the entry block of the caller. First
// calculate which instruction they should be inserted before. We insert the
// instructions at the end of the current alloca list.
//
{
BasicBlock::iterator InsertPoint = Caller->begin()->begin();
for (BasicBlock::iterator I = FirstNewBlock->begin(),
E = FirstNewBlock->end(); I != E; ) {
AllocaInst *AI = dyn_cast<AllocaInst>(I++);
if (AI == 0) continue;
// If the alloca is now dead, remove it. This often occurs due to code
// specialization.
if (AI->use_empty()) {
AI->eraseFromParent();
continue;
}
if (!isa<Constant>(AI->getArraySize()))
continue;
// Keep track of the static allocas that we inline into the caller if the
// StaticAllocas pointer is non-null.
IFI.StaticAllocas.push_back(AI);
// Scan for the block of allocas that we can move over, and move them
// all at once.
while (isa<AllocaInst>(I) &&
isa<Constant>(cast<AllocaInst>(I)->getArraySize())) {
IFI.StaticAllocas.push_back(cast<AllocaInst>(I));
++I;
}
// Transfer all of the allocas over in a block. Using splice means
// that the instructions aren't removed from the symbol table, then
// reinserted.
Caller->getEntryBlock().getInstList().splice(InsertPoint,
FirstNewBlock->getInstList(),
AI, I);
}
}
// If the inlined code contained dynamic alloca instructions, wrap the inlined
// code with llvm.stacksave/llvm.stackrestore intrinsics.
if (InlinedFunctionInfo.ContainsDynamicAllocas) {
Module *M = Caller->getParent();
// Get the two intrinsics we care about.
Function *StackSave = Intrinsic::getDeclaration(M, Intrinsic::stacksave);
Function *StackRestore=Intrinsic::getDeclaration(M,Intrinsic::stackrestore);
// If we are preserving the callgraph, add edges to the stacksave/restore
// functions for the calls we insert.
CallGraphNode *StackSaveCGN = 0, *StackRestoreCGN = 0, *CallerNode = 0;
if (CallGraph *CG = IFI.CG) {
StackSaveCGN = CG->getOrInsertFunction(StackSave);
StackRestoreCGN = CG->getOrInsertFunction(StackRestore);
CallerNode = (*CG)[Caller];
}
// Insert the llvm.stacksave.
CallInst *SavedPtr = CallInst::Create(StackSave, "savedstack",
FirstNewBlock->begin());
if (IFI.CG) CallerNode->addCalledFunction(SavedPtr, StackSaveCGN);
// Insert a call to llvm.stackrestore before any return instructions in the
// inlined function.
for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
CallInst *CI = CallInst::Create(StackRestore, SavedPtr, "", Returns[i]);
if (IFI.CG) CallerNode->addCalledFunction(CI, StackRestoreCGN);
}
// Count the number of StackRestore calls we insert.
unsigned NumStackRestores = Returns.size();
// If we are inlining an invoke instruction, insert restores before each
// unwind. These unwinds will be rewritten into branches later.
if (InlinedFunctionInfo.ContainsUnwinds && isa<InvokeInst>(TheCall)) {
for (Function::iterator BB = FirstNewBlock, E = Caller->end();
BB != E; ++BB)
if (UnwindInst *UI = dyn_cast<UnwindInst>(BB->getTerminator())) {
CallInst *CI = CallInst::Create(StackRestore, SavedPtr, "", UI);
if (IFI.CG) CallerNode->addCalledFunction(CI, StackRestoreCGN);
++NumStackRestores;
}
}
}
// If we are inlining tail call instruction through a call site that isn't
// marked 'tail', we must remove the tail marker for any calls in the inlined
// code. Also, calls inlined through a 'nounwind' call site should be marked
// 'nounwind'.
if (InlinedFunctionInfo.ContainsCalls &&
(MustClearTailCallFlags || MarkNoUnwind)) {
for (Function::iterator BB = FirstNewBlock, E = Caller->end();
BB != E; ++BB)
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
if (CallInst *CI = dyn_cast<CallInst>(I)) {
if (MustClearTailCallFlags)
CI->setTailCall(false);
if (MarkNoUnwind)
CI->setDoesNotThrow();
}
}
// If we are inlining through a 'nounwind' call site then any inlined 'unwind'
// instructions are unreachable.
if (InlinedFunctionInfo.ContainsUnwinds && MarkNoUnwind)
for (Function::iterator BB = FirstNewBlock, E = Caller->end();
BB != E; ++BB) {
TerminatorInst *Term = BB->getTerminator();
if (isa<UnwindInst>(Term)) {
new UnreachableInst(Context, Term);
BB->getInstList().erase(Term);
}
}
// If we are inlining for an invoke instruction, we must make sure to rewrite
// any inlined 'unwind' instructions into branches to the invoke exception
// destination, and call instructions into invoke instructions.
if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall))
HandleInlinedInvoke(II, FirstNewBlock, InlinedFunctionInfo);
// If we cloned in _exactly one_ basic block, and if that block ends in a
// return instruction, we splice the body of the inlined callee directly into
// the calling basic block.
if (Returns.size() == 1 && std::distance(FirstNewBlock, Caller->end()) == 1) {
// Move all of the instructions right before the call.
OrigBB->getInstList().splice(TheCall, FirstNewBlock->getInstList(),
FirstNewBlock->begin(), FirstNewBlock->end());
// Remove the cloned basic block.
Caller->getBasicBlockList().pop_back();
// If the call site was an invoke instruction, add a branch to the normal
// destination.
if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall))
BranchInst::Create(II->getNormalDest(), TheCall);
// If the return instruction returned a value, replace uses of the call with
// uses of the returned value.
if (!TheCall->use_empty()) {
ReturnInst *R = Returns[0];
if (TheCall == R->getReturnValue())
TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
else
TheCall->replaceAllUsesWith(R->getReturnValue());
}
// Since we are now done with the Call/Invoke, we can delete it.
TheCall->eraseFromParent();
// Since we are now done with the return instruction, delete it also.
Returns[0]->eraseFromParent();
// We are now done with the inlining.
return true;
}
// Otherwise, we have the normal case, of more than one block to inline or
// multiple return sites.
// We want to clone the entire callee function into the hole between the
// "starter" and "ender" blocks. How we accomplish this depends on whether
// this is an invoke instruction or a call instruction.
BasicBlock *AfterCallBB;
if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall)) {
// Add an unconditional branch to make this look like the CallInst case...
BranchInst *NewBr = BranchInst::Create(II->getNormalDest(), TheCall);
// Split the basic block. This guarantees that no PHI nodes will have to be
// updated due to new incoming edges, and make the invoke case more
// symmetric to the call case.
AfterCallBB = OrigBB->splitBasicBlock(NewBr,
CalledFunc->getName()+".exit");
} else { // It's a call
// If this is a call instruction, we need to split the basic block that
// the call lives in.
//
AfterCallBB = OrigBB->splitBasicBlock(TheCall,
CalledFunc->getName()+".exit");
}
// Change the branch that used to go to AfterCallBB to branch to the first
// basic block of the inlined function.
//
TerminatorInst *Br = OrigBB->getTerminator();
assert(Br && Br->getOpcode() == Instruction::Br &&
"splitBasicBlock broken!");
Br->setOperand(0, FirstNewBlock);
// Now that the function is correct, make it a little bit nicer. In
// particular, move the basic blocks inserted from the end of the function
// into the space made by splitting the source basic block.
Caller->getBasicBlockList().splice(AfterCallBB, Caller->getBasicBlockList(),
FirstNewBlock, Caller->end());
// Handle all of the return instructions that we just cloned in, and eliminate
// any users of the original call/invoke instruction.
const Type *RTy = CalledFunc->getReturnType();
if (Returns.size() > 1) {
// The PHI node should go at the front of the new basic block to merge all
// possible incoming values.
PHINode *PHI = 0;
if (!TheCall->use_empty()) {
PHI = PHINode::Create(RTy, TheCall->getName(),
AfterCallBB->begin());
// Anything that used the result of the function call should now use the
// PHI node as their operand.
TheCall->replaceAllUsesWith(PHI);
}
// Loop over all of the return instructions adding entries to the PHI node
// as appropriate.
if (PHI) {
for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
ReturnInst *RI = Returns[i];
assert(RI->getReturnValue()->getType() == PHI->getType() &&
"Ret value not consistent in function!");
PHI->addIncoming(RI->getReturnValue(), RI->getParent());
}
// Now that we inserted the PHI, check to see if it has a single value
// (e.g. all the entries are the same or undef). If so, remove the PHI so
// it doesn't block other optimizations.
if (Value *V = PHI->hasConstantValue()) {
PHI->replaceAllUsesWith(V);
PHI->eraseFromParent();
}
}
// Add a branch to the merge points and remove return instructions.
for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
ReturnInst *RI = Returns[i];
BranchInst::Create(AfterCallBB, RI);
RI->eraseFromParent();
}
} else if (!Returns.empty()) {
// Otherwise, if there is exactly one return value, just replace anything
// using the return value of the call with the computed value.
if (!TheCall->use_empty()) {
if (TheCall == Returns[0]->getReturnValue())
TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
else
TheCall->replaceAllUsesWith(Returns[0]->getReturnValue());
}
// Splice the code from the return block into the block that it will return
// to, which contains the code that was after the call.
BasicBlock *ReturnBB = Returns[0]->getParent();
AfterCallBB->getInstList().splice(AfterCallBB->begin(),
ReturnBB->getInstList());
// Update PHI nodes that use the ReturnBB to use the AfterCallBB.
ReturnBB->replaceAllUsesWith(AfterCallBB);
// Delete the return instruction now and empty ReturnBB now.
Returns[0]->eraseFromParent();
ReturnBB->eraseFromParent();
} else if (!TheCall->use_empty()) {
// No returns, but something is using the return value of the call. Just
// nuke the result.
TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
}
// Since we are now done with the Call/Invoke, we can delete it.
TheCall->eraseFromParent();
// We should always be able to fold the entry block of the function into the
// single predecessor of the block...
assert(cast<BranchInst>(Br)->isUnconditional() && "splitBasicBlock broken!");
BasicBlock *CalleeEntry = cast<BranchInst>(Br)->getSuccessor(0);
// Splice the code entry block into calling block, right before the
// unconditional branch.
OrigBB->getInstList().splice(Br, CalleeEntry->getInstList());
CalleeEntry->replaceAllUsesWith(OrigBB); // Update PHI nodes
// Remove the unconditional branch.
OrigBB->getInstList().erase(Br);
// Now we can remove the CalleeEntry block, which is now empty.
Caller->getBasicBlockList().erase(CalleeEntry);
return true;
}
