bool runOnSCC(CallGraphSCC &SCC) override;
static char ID; // Pass identification, replacement for typeid
explicit ArgPromotion(unsigned maxElements = 3)
- : CallGraphSCCPass(ID), DL(nullptr), maxElements(maxElements) {
+ : CallGraphSCCPass(ID), maxElements(maxElements) {
initializeArgPromotionPass(*PassRegistry::getPassRegistry());
}
/// A vector used to hold the indices of a single GEP instruction
typedef std::vector<uint64_t> IndicesVector;
- const DataLayout *DL;
private:
- bool isDenselyPacked(Type *type);
+ bool isDenselyPacked(Type *type, const DataLayout &DL);
bool canPaddingBeAccessed(Argument *Arg);
CallGraphNode *PromoteArguments(CallGraphNode *CGN);
bool isSafeToPromoteArgument(Argument *Arg, bool isByVal) const;
}
/// \brief Checks if a type could have padding bytes.
-bool ArgPromotion::isDenselyPacked(Type *type) {
+bool ArgPromotion::isDenselyPacked(Type *type, const DataLayout &DL) {
// There is no size information, so be conservative.
if (!type->isSized())
// If the alloc size is not equal to the storage size, then there are padding
// bytes. For x86_fp80 on x86-64, size: 80 alloc size: 128.
- if (!DL || DL->getTypeSizeInBits(type) != DL->getTypeAllocSizeInBits(type))
+ if (DL.getTypeSizeInBits(type) != DL.getTypeAllocSizeInBits(type))
return false;
if (!isa<CompositeType>(type))
// For homogenous sequential types, check for padding within members.
if (SequentialType *seqTy = dyn_cast<SequentialType>(type))
- return isa<PointerType>(seqTy) || isDenselyPacked(seqTy->getElementType());
+ return isa<PointerType>(seqTy) ||
+ isDenselyPacked(seqTy->getElementType(), DL);
// Check for padding within and between elements of a struct.
StructType *StructTy = cast<StructType>(type);
- const StructLayout *Layout = DL->getStructLayout(StructTy);
+ const StructLayout *Layout = DL.getStructLayout(StructTy);
uint64_t StartPos = 0;
for (unsigned i = 0, E = StructTy->getNumElements(); i < E; ++i) {
Type *ElTy = StructTy->getElementType(i);
- if (!isDenselyPacked(ElTy))
+ if (!isDenselyPacked(ElTy, DL))
return false;
if (StartPos != Layout->getElementOffsetInBits(i))
return false;
- StartPos += DL->getTypeAllocSizeInBits(ElTy);
+ StartPos += DL.getTypeAllocSizeInBits(ElTy);
}
return true;
// Make sure that it is local to this module.
if (!F || !F->hasLocalLinkage()) return nullptr;
- DL = &F->getParent()->getDataLayout();
-
// First check: see if there are any pointer arguments! If not, quick exit.
SmallVector<Argument*, 16> PointerArgs;
for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(); I != E; ++I)
// IR, while in the callee the classification is determined dynamically based
// on the number of registers consumed so far.
if (F->isVarArg()) return nullptr;
+ const DataLayout &DL = F->getParent()->getDataLayout();
// Check to see which arguments are promotable. If an argument is promotable,
// add it to ArgsToPromote.
// packed or if we can prove the padding bytes are never accessed. This does
// not apply to inalloca.
bool isSafeToPromote =
- PtrArg->hasByValAttr() &&
- (isDenselyPacked(AgTy) || !canPaddingBeAccessed(PtrArg));
+ PtrArg->hasByValAttr() &&
+ (isDenselyPacked(AgTy, DL) || !canPaddingBeAccessed(PtrArg));
if (isSafeToPromote) {
if (StructType *STy = dyn_cast<StructType>(AgTy)) {
if (maxElements > 0 && STy->getNumElements() > maxElements) {
/// AllCallersPassInValidPointerForArgument - Return true if we can prove that
/// all callees pass in a valid pointer for the specified function argument.
-static bool AllCallersPassInValidPointerForArgument(Argument *Arg,
- const DataLayout *DL) {
+static bool AllCallersPassInValidPointerForArgument(Argument *Arg) {
Function *Callee = Arg->getParent();
+ const DataLayout &DL = Callee->getParent()->getDataLayout();
unsigned ArgNo = Arg->getArgNo();
GEPIndicesSet ToPromote;
// If the pointer is always valid, any load with first index 0 is valid.
- if (isByValOrInAlloca || AllCallersPassInValidPointerForArgument(Arg, DL))
+ if (isByValOrInAlloca || AllCallersPassInValidPointerForArgument(Arg))
SafeToUnconditionallyLoad.insert(IndicesVector(1, 0));
// First, iterate the entry block and mark loads of (geps of) arguments as
FunctionType *FTy = F->getFunctionType();
std::vector<Type*> Params;
- typedef std::set<IndicesVector> ScalarizeTable;
+ typedef std::set<std::pair<Type*, IndicesVector>> ScalarizeTable;
// ScalarizedElements - If we are promoting a pointer that has elements
// accessed out of it, keep track of which elements are accessed so that we
ScalarizeTable &ArgIndices = ScalarizedElements[I];
for (User *U : I->users()) {
Instruction *UI = cast<Instruction>(U);
- assert(isa<LoadInst>(UI) || isa<GetElementPtrInst>(UI));
+ Type *SrcTy;
+ if (LoadInst *L = dyn_cast<LoadInst>(UI))
+ SrcTy = L->getType();
+ else
+ SrcTy = cast<GetElementPtrInst>(UI)->getSourceElementType();
IndicesVector Indices;
Indices.reserve(UI->getNumOperands() - 1);
// Since loads will only have a single operand, and GEPs only a single
// GEPs with a single 0 index can be merged with direct loads
if (Indices.size() == 1 && Indices.front() == 0)
Indices.clear();
- ArgIndices.insert(Indices);
+ ArgIndices.insert(std::make_pair(SrcTy, Indices));
LoadInst *OrigLoad;
if (LoadInst *L = dyn_cast<LoadInst>(UI))
OrigLoad = L;
for (ScalarizeTable::iterator SI = ArgIndices.begin(),
E = ArgIndices.end(); SI != E; ++SI) {
// not allowed to dereference ->begin() if size() is 0
- Params.push_back(GetElementPtrInst::getIndexedType(I->getType(), *SI));
+ Params.push_back(GetElementPtrInst::getIndexedType(I->getType(), SI->second));
assert(Params.back());
}
- if (ArgIndices.size() == 1 && ArgIndices.begin()->empty())
+ if (ArgIndices.size() == 1 && ArgIndices.begin()->second.empty())
++NumArgumentsPromoted;
else
++NumAggregatesPromoted;
ConstantInt::get(Type::getInt32Ty(F->getContext()), 0), nullptr };
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
Idxs[1] = ConstantInt::get(Type::getInt32Ty(F->getContext()), i);
- Value *Idx = GetElementPtrInst::Create(*AI, Idxs,
- (*AI)->getName()+"."+utostr(i),
- Call);
+ Value *Idx = GetElementPtrInst::Create(
+ STy, *AI, Idxs, (*AI)->getName() + "." + utostr(i), Call);
// TODO: Tell AA about the new values?
Args.push_back(new LoadInst(Idx, Idx->getName()+".val", Call));
}
for (ScalarizeTable::iterator SI = ArgIndices.begin(),
E = ArgIndices.end(); SI != E; ++SI) {
Value *V = *AI;
- LoadInst *OrigLoad = OriginalLoads[std::make_pair(I, *SI)];
- if (!SI->empty()) {
- Ops.reserve(SI->size());
+ LoadInst *OrigLoad = OriginalLoads[std::make_pair(I, SI->second)];
+ if (!SI->second.empty()) {
+ Ops.reserve(SI->second.size());
Type *ElTy = V->getType();
- for (IndicesVector::const_iterator II = SI->begin(),
- IE = SI->end(); II != IE; ++II) {
+ for (IndicesVector::const_iterator II = SI->second.begin(),
+ IE = SI->second.end(); II != IE; ++II) {
// Use i32 to index structs, and i64 for others (pointers/arrays).
// This satisfies GEP constraints.
Type *IdxTy = (ElTy->isStructTy() ?
ElTy = cast<CompositeType>(ElTy)->getTypeAtIndex(*II);
}
// And create a GEP to extract those indices.
- V = GetElementPtrInst::Create(V, Ops, V->getName()+".idx", Call);
+ V = GetElementPtrInst::Create(SI->first, V, Ops, V->getName()+".idx", Call);
Ops.clear();
AA.copyValue(OrigLoad->getOperand(0), V);
}
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
Idxs[1] = ConstantInt::get(Type::getInt32Ty(F->getContext()), i);
- Value *Idx =
- GetElementPtrInst::Create(TheAlloca, Idxs,
- TheAlloca->getName()+"."+Twine(i),
- InsertPt);
+ Value *Idx = GetElementPtrInst::Create(
+ AgTy, TheAlloca, Idxs, TheAlloca->getName() + "." + Twine(i),
+ InsertPt);
I2->setName(I->getName()+"."+Twine(i));
new StoreInst(I2++, Idx, InsertPt);
}
while (!I->use_empty()) {
if (LoadInst *LI = dyn_cast<LoadInst>(I->user_back())) {
- assert(ArgIndices.begin()->empty() &&
+ assert(ArgIndices.begin()->second.empty() &&
"Load element should sort to front!");
I2->setName(I->getName()+".val");
LI->replaceAllUsesWith(I2);
Function::arg_iterator TheArg = I2;
for (ScalarizeTable::iterator It = ArgIndices.begin();
- *It != Operands; ++It, ++TheArg) {
+ It->second != Operands; ++It, ++TheArg) {
assert(It != ArgIndices.end() && "GEP not handled??");
}
projects / oota-llvm.git / blobdiff