const DataLayout *DL;
private:
+ bool isDenselyPacked(Type *type);
+ bool canPaddingBeAccessed(Argument *Arg);
CallGraphNode *PromoteArguments(CallGraphNode *CGN);
bool isSafeToPromoteArgument(Argument *Arg, bool isByVal) const;
CallGraphNode *DoPromotion(Function *F,
bool ArgPromotion::runOnSCC(CallGraphSCC &SCC) {
bool Changed = false, LocalChange;
- DataLayoutPass *DLP = getAnalysisIfAvailable<DataLayoutPass>();
- DL = DLP ? &DLP->getDataLayout() : nullptr;
-
do { // Iterate until we stop promoting from this SCC.
LocalChange = false;
// Attempt to promote arguments from all functions in this SCC.
return Changed;
}
+/// \brief Checks if a type could have padding bytes.
+bool ArgPromotion::isDenselyPacked(Type *type) {
+
+ // There is no size information, so be conservative.
+ if (!type->isSized())
+ return false;
+
+ // 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))
+ return false;
+
+ if (!isa<CompositeType>(type))
+ return true;
+
+ // For homogenous sequential types, check for padding within members.
+ if (SequentialType *seqTy = dyn_cast<SequentialType>(type))
+ return isa<PointerType>(seqTy) || isDenselyPacked(seqTy->getElementType());
+
+ // Check for padding within and between elements of a struct.
+ StructType *StructTy = cast<StructType>(type);
+ 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))
+ return false;
+ if (StartPos != Layout->getElementOffsetInBits(i))
+ return false;
+ StartPos += DL->getTypeAllocSizeInBits(ElTy);
+ }
+
+ return true;
+}
+
+/// \brief Checks if the padding bytes of an argument could be accessed.
+bool ArgPromotion::canPaddingBeAccessed(Argument *arg) {
+
+ assert(arg->hasByValAttr());
+
+ // Track all the pointers to the argument to make sure they are not captured.
+ SmallPtrSet<Value *, 16> PtrValues;
+ PtrValues.insert(arg);
+
+ // Track all of the stores.
+ SmallVector<StoreInst *, 16> Stores;
+
+ // Scan through the uses recursively to make sure the pointer is always used
+ // sanely.
+ SmallVector<Value *, 16> WorkList;
+ WorkList.insert(WorkList.end(), arg->user_begin(), arg->user_end());
+ while (!WorkList.empty()) {
+ Value *V = WorkList.back();
+ WorkList.pop_back();
+ if (isa<GetElementPtrInst>(V) || isa<PHINode>(V)) {
+ if (PtrValues.insert(V).second)
+ WorkList.insert(WorkList.end(), V->user_begin(), V->user_end());
+ } else if (StoreInst *Store = dyn_cast<StoreInst>(V)) {
+ Stores.push_back(Store);
+ } else if (!isa<LoadInst>(V)) {
+ return true;
+ }
+ }
+
+// Check to make sure the pointers aren't captured
+ for (StoreInst *Store : Stores)
+ if (PtrValues.count(Store->getValueOperand()))
+ return true;
+
+ return false;
+}
+
/// PromoteArguments - This method checks the specified function to see if there
/// are any promotable arguments and if it is safe to promote the function (for
/// example, all callers are direct). If safe to promote some arguments, it
// 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)
isSelfRecursive = true;
}
+ // Don't promote arguments for variadic functions. Adding, removing, or
+ // changing non-pack parameters can change the classification of pack
+ // parameters. Frontends encode that classification at the call site in the
+ // IR, while in the callee the classification is determined dynamically based
+ // on the number of registers consumed so far.
+ if (F->isVarArg()) return nullptr;
+
// Check to see which arguments are promotable. If an argument is promotable,
// add it to ArgsToPromote.
SmallPtrSet<Argument*, 8> ArgsToPromote;
Type *AgTy = cast<PointerType>(PtrArg->getType())->getElementType();
// If this is a byval argument, and if the aggregate type is small, just
- // pass the elements, which is always safe. This does not apply to
- // inalloca.
- if (PtrArg->hasByValAttr()) {
+ // pass the elements, which is always safe, if the passed value is densely
+ // 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));
+ if (isSafeToPromote) {
if (StructType *STy = dyn_cast<StructType>(AgTy)) {
if (maxElements > 0 && STy->getNumElements() > maxElements) {
DEBUG(dbgs() << "argpromotion disable promoting argument '"
// of elements of the aggregate.
return false;
}
- ToPromote.insert(Operands);
+ ToPromote.insert(std::move(Operands));
}
}
BasicBlock *BB = Load->getParent();
AliasAnalysis::Location Loc = AA.getLocation(Load);
- if (AA.canInstructionRangeModify(BB->front(), *Load, Loc))
+ if (AA.canInstructionRangeModRef(BB->front(), *Load, Loc,
+ AliasAnalysis::Mod))
return false; // Pointer is invalidated!
// Now check every path from the entry block to the load for transparency.
// To do this, we perform a depth first search on the inverse CFG from the
// loading block.
- for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
- BasicBlock *P = *PI;
+ for (BasicBlock *P : predecessors(BB)) {
for (BasicBlock *TranspBB : inverse_depth_first_ext(P, TranspBlocks))
if (AA.canBasicBlockModify(*TranspBB, Loc))
return false;
// Simple byval argument? Just add all the struct element types.
Type *AgTy = cast<PointerType>(I->getType())->getElementType();
StructType *STy = cast<StructType>(AgTy);
- for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i)
- Params.push_back(STy->getElementType(i));
+ Params.insert(Params.end(), STy->element_begin(), STy->element_end());
++NumByValArgsPromoted;
} else if (!ArgsToPromote.count(I)) {
// Unchanged argument
if (DI != FunctionDIs.end()) {
DISubprogram SP = DI->second;
SP.replaceFunction(NF);
+ // Ensure the map is updated so it can be reused on subsequent argument
+ // promotions of the same function.
FunctionDIs.erase(DI);
FunctionDIs[NF] = SP;
}
projects / oota-llvm.git / blobdiff