return ComputeLinearIndex(*EI, Indices+1, IndicesEnd, CurIndex);
CurIndex = ComputeLinearIndex(*EI, nullptr, nullptr, CurIndex);
}
+ assert(!Indices && "Unexpected out of bound");
return CurIndex;
}
// Given an array type, recursively traverse the elements.
unsigned NumElts = ATy->getNumElements();
// Compute the Linear offset when jumping one element of the array
unsigned EltLinearOffset = ComputeLinearIndex(EltTy, nullptr, nullptr, 0);
- if (Indices && *Indices < NumElts) {
+ if (Indices) {
+ assert(*Indices < NumElts && "Unexpected out of bound");
// If the indice is inside the array, compute the index to the requested
// elt and recurse inside the element with the end of the indices list
CurIndex += EltLinearOffset* *Indices;
return ComputeLinearIndex(EltTy, Indices+1, IndicesEnd, CurIndex);
}
- // Out of bound? Assert instead?
CurIndex += EltLinearOffset*NumElts;
return CurIndex;
}
/// If Offsets is non-null, it points to a vector to be filled in
/// with the in-memory offsets of each of the individual values.
///
-void llvm::ComputeValueVTs(const TargetLowering &TLI, Type *Ty,
- SmallVectorImpl<EVT> &ValueVTs,
+void llvm::ComputeValueVTs(const TargetLowering &TLI, const DataLayout &DL,
+ Type *Ty, SmallVectorImpl<EVT> &ValueVTs,
SmallVectorImpl<uint64_t> *Offsets,
uint64_t StartingOffset) {
// Given a struct type, recursively traverse the elements.
if (StructType *STy = dyn_cast<StructType>(Ty)) {
- const StructLayout *SL = TLI.getDataLayout()->getStructLayout(STy);
+ const StructLayout *SL = DL.getStructLayout(STy);
for (StructType::element_iterator EB = STy->element_begin(),
EI = EB,
EE = STy->element_end();
EI != EE; ++EI)
- ComputeValueVTs(TLI, *EI, ValueVTs, Offsets,
+ ComputeValueVTs(TLI, DL, *EI, ValueVTs, Offsets,
StartingOffset + SL->getElementOffset(EI - EB));
return;
}
// Given an array type, recursively traverse the elements.
if (ArrayType *ATy = dyn_cast<ArrayType>(Ty)) {
Type *EltTy = ATy->getElementType();
- uint64_t EltSize = TLI.getDataLayout()->getTypeAllocSize(EltTy);
+ uint64_t EltSize = DL.getTypeAllocSize(EltTy);
for (unsigned i = 0, e = ATy->getNumElements(); i != e; ++i)
- ComputeValueVTs(TLI, EltTy, ValueVTs, Offsets,
+ ComputeValueVTs(TLI, DL, EltTy, ValueVTs, Offsets,
StartingOffset + i * EltSize);
return;
}
if (Ty->isVoidTy())
return;
// Base case: we can get an EVT for this LLVM IR type.
- ValueVTs.push_back(TLI.getValueType(Ty));
+ ValueVTs.push_back(TLI.getValueType(DL, Ty));
if (Offsets)
Offsets->push_back(StartingOffset);
}
static const Value *getNoopInput(const Value *V,
SmallVectorImpl<unsigned> &ValLoc,
unsigned &DataBits,
- const TargetLoweringBase &TLI) {
+ const TargetLoweringBase &TLI,
+ const DataLayout &DL) {
while (true) {
// Try to look through V1; if V1 is not an instruction, it can't be looked
// through.
// Make sure this isn't a truncating or extending cast. We could
// support this eventually, but don't bother for now.
if (!isa<VectorType>(I->getType()) &&
- TLI.getPointerTy().getSizeInBits() ==
- cast<IntegerType>(Op->getType())->getBitWidth())
+ DL.getPointerSizeInBits() ==
+ cast<IntegerType>(Op->getType())->getBitWidth())
NoopInput = Op;
} else if (isa<PtrToIntInst>(I)) {
// Look through ptrtoint.
// Make sure this isn't a truncating or extending cast. We could
// support this eventually, but don't bother for now.
if (!isa<VectorType>(I->getType()) &&
- TLI.getPointerTy().getSizeInBits() ==
- cast<IntegerType>(I->getType())->getBitWidth())
+ DL.getPointerSizeInBits() ==
+ cast<IntegerType>(I->getType())->getBitWidth())
NoopInput = Op;
} else if (isa<TruncInst>(I) &&
TLI.allowTruncateForTailCall(Op->getType(), I->getType())) {
} else if (const InsertValueInst *IVI = dyn_cast<InsertValueInst>(V)) {
// Value may come from either the aggregate or the scalar
ArrayRef<unsigned> InsertLoc = IVI->getIndices();
- if (std::equal(InsertLoc.rbegin(), InsertLoc.rend(),
- ValLoc.rbegin())) {
+ if (ValLoc.size() >= InsertLoc.size() &&
+ std::equal(InsertLoc.begin(), InsertLoc.end(), ValLoc.rbegin())) {
// The type being inserted is a nested sub-type of the aggregate; we
// have to remove those initial indices to get the location we're
// interested in for the operand.
// previous aggregate. Combine the two paths to obtain the true address of
// our element.
ArrayRef<unsigned> ExtractLoc = EVI->getIndices();
- std::copy(ExtractLoc.rbegin(), ExtractLoc.rend(),
- std::back_inserter(ValLoc));
+ ValLoc.append(ExtractLoc.rbegin(), ExtractLoc.rend());
NoopInput = Op;
}
// Terminate if we couldn't find anything to look through.
SmallVectorImpl<unsigned> &RetIndices,
SmallVectorImpl<unsigned> &CallIndices,
bool AllowDifferingSizes,
- const TargetLoweringBase &TLI) {
+ const TargetLoweringBase &TLI,
+ const DataLayout &DL) {
// Trace the sub-value needed by the return value as far back up the graph as
// possible, in the hope that it will intersect with the value produced by the
// call. In the simple case with no "returned" attribute, the hope is actually
// that we end up back at the tail call instruction itself.
unsigned BitsRequired = UINT_MAX;
- RetVal = getNoopInput(RetVal, RetIndices, BitsRequired, TLI);
+ RetVal = getNoopInput(RetVal, RetIndices, BitsRequired, TLI, DL);
// If this slot in the value returned is undef, it doesn't matter what the
// call puts there, it'll be fine.
// a "returned" attribute, the search will be blocked immediately and the loop
// a Noop.
unsigned BitsProvided = UINT_MAX;
- CallVal = getNoopInput(CallVal, CallIndices, BitsProvided, TLI);
+ CallVal = getNoopInput(CallVal, CallIndices, BitsProvided, TLI, DL);
// There's no hope if we can't actually trace them to (the same part of!) the
// same value.
return false;
}
+ const Function *F = ExitBB->getParent();
return returnTypeIsEligibleForTailCall(
- ExitBB->getParent(), I, Ret, *TM.getSubtargetImpl()->getTargetLowering());
+ F, I, Ret, *TM.getSubtargetImpl(*F)->getTargetLowering());
}
bool llvm::returnTypeIsEligibleForTailCall(const Function *F,
// The manipulations performed when we're looking through an insertvalue or
// an extractvalue would happen at the front of the RetPath list, so since
// we have to copy it anyway it's more efficient to create a reversed copy.
- using std::copy;
- SmallVector<unsigned, 4> TmpRetPath, TmpCallPath;
- copy(RetPath.rbegin(), RetPath.rend(), std::back_inserter(TmpRetPath));
- copy(CallPath.rbegin(), CallPath.rend(), std::back_inserter(TmpCallPath));
+ SmallVector<unsigned, 4> TmpRetPath(RetPath.rbegin(), RetPath.rend());
+ SmallVector<unsigned, 4> TmpCallPath(CallPath.rbegin(), CallPath.rend());
// Finally, we can check whether the value produced by the tail call at this
// index is compatible with the value we return.
if (!slotOnlyDiscardsData(RetVal, CallVal, TmpRetPath, TmpCallPath,
- AllowDifferingSizes, TLI))
+ AllowDifferingSizes, TLI,
+ F->getParent()->getDataLayout()))
return false;
CallEmpty = !nextRealType(CallSubTypes, CallPath);
projects / oota-llvm.git / blobdiff