// Instrument memset/memmove/memcpy
bool AddressSanitizer::instrumentMemIntrinsic(AsanFunctionContext &AFC,
MemIntrinsic *MI) {
Value *Dst = MI->getDest();
MemTransferInst *MemTran = dyn_cast<MemTransferInst>(MI);
Value *Src = MemTran ? MemTran->getSource() : 0;
Value *Length = MI->getLength();
Constant *ConstLength = dyn_cast<Constant>(Length);
Instruction *InsertBefore = MI;
if (ConstLength) {
if (ConstLength->isNullValue()) return false;
} else {
// The size is not a constant so it could be zero -- check at run-time.
IRBuilder<> IRB(InsertBefore);
Value *Cmp = IRB.CreateICmpNE(Length,
Constant::getNullValue(Length->getType()));
InsertBefore = splitBlockAndInsertIfThen(Cmp, false);
}
instrumentMemIntrinsicParam(AFC, MI, Dst, Length, InsertBefore, true);
if (Src)
instrumentMemIntrinsicParam(AFC, MI, Src, Length, InsertBefore, false);
return true;
}
/// isOnlyCopiedFromConstantGlobal - Recursively walk the uses of a (derived)
/// pointer to an alloca. Ignore any reads of the pointer, return false if we
/// see any stores or other unknown uses. If we see pointer arithmetic, keep
/// track of whether it moves the pointer (with IsOffset) but otherwise traverse
/// the uses. If we see a memcpy/memmove that targets an unoffseted pointer to
/// the alloca, and if the source pointer is a pointer to a constant global, we
/// can optimize this.
static bool
isOnlyCopiedFromConstantGlobal(Value *V, MemTransferInst *&TheCopy,
SmallVectorImpl<Instruction *> &ToDelete) {
// We track lifetime intrinsics as we encounter them. If we decide to go
// ahead and replace the value with the global, this lets the caller quickly
// eliminate the markers.
SmallVector<std::pair<Value *, bool>, 35> ValuesToInspect;
ValuesToInspect.push_back(std::make_pair(V, false));
while (!ValuesToInspect.empty()) {
auto ValuePair = ValuesToInspect.pop_back_val();
const bool IsOffset = ValuePair.second;
for (auto &U : ValuePair.first->uses()) {
Instruction *I = cast<Instruction>(U.getUser());
if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
// Ignore non-volatile loads, they are always ok.
if (!LI->isSimple()) return false;
continue;
}
if (isa<BitCastInst>(I) || isa<AddrSpaceCastInst>(I)) {
// If uses of the bitcast are ok, we are ok.
ValuesToInspect.push_back(std::make_pair(I, IsOffset));
continue;
}
if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(I)) {
// If the GEP has all zero indices, it doesn't offset the pointer. If it
// doesn't, it does.
ValuesToInspect.push_back(
std::make_pair(I, IsOffset || !GEP->hasAllZeroIndices()));
continue;
}
if (auto CS = CallSite(I)) {
// If this is the function being called then we treat it like a load and
// ignore it.
if (CS.isCallee(&U))
continue;
unsigned DataOpNo = CS.getDataOperandNo(&U);
bool IsArgOperand = CS.isArgOperand(&U);
// Inalloca arguments are clobbered by the call.
if (IsArgOperand && CS.isInAllocaArgument(DataOpNo))
return false;
// If this is a readonly/readnone call site, then we know it is just a
// load (but one that potentially returns the value itself), so we can
// ignore it if we know that the value isn't captured.
if (CS.onlyReadsMemory() &&
(CS.getInstruction()->use_empty() || CS.doesNotCapture(DataOpNo)))
continue;
// If this is being passed as a byval argument, the caller is making a
// copy, so it is only a read of the alloca.
if (IsArgOperand && CS.isByValArgument(DataOpNo))
continue;
}
// Lifetime intrinsics can be handled by the caller.
if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
if (II->getIntrinsicID() == Intrinsic::lifetime_start ||
II->getIntrinsicID() == Intrinsic::lifetime_end) {
assert(II->use_empty() && "Lifetime markers have no result to use!");
ToDelete.push_back(II);
continue;
}
}
// If this is isn't our memcpy/memmove, reject it as something we can't
// handle.
MemTransferInst *MI = dyn_cast<MemTransferInst>(I);
if (!MI)
return false;
// If the transfer is using the alloca as a source of the transfer, then
// ignore it since it is a load (unless the transfer is volatile).
if (U.getOperandNo() == 1) {
if (MI->isVolatile()) return false;
continue;
}
// If we already have seen a copy, reject the second one.
if (TheCopy) return false;
// If the pointer has been offset from the start of the alloca, we can't
// safely handle this.
if (IsOffset) return false;
// If the memintrinsic isn't using the alloca as the dest, reject it.
if (U.getOperandNo() != 0) return false;
// If the source of the memcpy/move is not a constant global, reject it.
if (!pointsToConstantGlobal(MI->getSource()))
return false;
// Otherwise, the transform is safe. Remember the copy instruction.
TheCopy = MI;
}
}
return true;
}
bool NVPTXLowerAggrCopies::runOnFunction(Function &F) {
SmallVector<LoadInst *, 4> aggrLoads;
SmallVector<MemTransferInst *, 4> aggrMemcpys;
SmallVector<MemSetInst *, 4> aggrMemsets;
DataLayout *TD = &getAnalysis<DataLayout>();
LLVMContext &Context = F.getParent()->getContext();
//
// Collect all the aggrLoads, aggrMemcpys and addrMemsets.
//
//const BasicBlock *firstBB = &F.front(); // first BB in F
for (Function::iterator BI = F.begin(), BE = F.end(); BI != BE; ++BI) {
//BasicBlock *bb = BI;
for (BasicBlock::iterator II = BI->begin(), IE = BI->end(); II != IE;
++II) {
if (LoadInst * load = dyn_cast<LoadInst>(II)) {
if (load->hasOneUse() == false) continue;
if (TD->getTypeStoreSize(load->getType()) < MaxAggrCopySize) continue;
User *use = *(load->use_begin());
if (StoreInst * store = dyn_cast<StoreInst>(use)) {
if (store->getOperand(0) != load) //getValueOperand
continue;
aggrLoads.push_back(load);
}
} else if (MemTransferInst * intr = dyn_cast<MemTransferInst>(II)) {
Value *len = intr->getLength();
// If the number of elements being copied is greater
// than MaxAggrCopySize, lower it to a loop
if (ConstantInt * len_int = dyn_cast < ConstantInt > (len)) {
if (len_int->getZExtValue() >= MaxAggrCopySize) {
aggrMemcpys.push_back(intr);
}
} else {
// turn variable length memcpy/memmov into loop
aggrMemcpys.push_back(intr);
}
} else if (MemSetInst * memsetintr = dyn_cast<MemSetInst>(II)) {
Value *len = memsetintr->getLength();
if (ConstantInt * len_int = dyn_cast<ConstantInt>(len)) {
if (len_int->getZExtValue() >= MaxAggrCopySize) {
aggrMemsets.push_back(memsetintr);
}
} else {
// turn variable length memset into loop
aggrMemsets.push_back(memsetintr);
}
}
}
}
if ((aggrLoads.size() == 0) && (aggrMemcpys.size() == 0)
&& (aggrMemsets.size() == 0)) return false;
//
// Do the transformation of an aggr load/copy/set to a loop
//
for (unsigned i = 0, e = aggrLoads.size(); i != e; ++i) {
LoadInst *load = aggrLoads[i];
StoreInst *store = dyn_cast<StoreInst>(*load->use_begin());
Value *srcAddr = load->getOperand(0);
Value *dstAddr = store->getOperand(1);
unsigned numLoads = TD->getTypeStoreSize(load->getType());
Value *len = ConstantInt::get(Type::getInt32Ty(Context), numLoads);
convertTransferToLoop(store, srcAddr, dstAddr, len, load->isVolatile(),
store->isVolatile(), Context, F);
store->eraseFromParent();
load->eraseFromParent();
}
for (unsigned i = 0, e = aggrMemcpys.size(); i != e; ++i) {
MemTransferInst *cpy = aggrMemcpys[i];
Value *len = cpy->getLength();
// llvm 2.7 version of memcpy does not have volatile
// operand yet. So always making it non-volatile
// optimistically, so that we don't see unnecessary
// st.volatile in ptx
convertTransferToLoop(cpy, cpy->getSource(), cpy->getDest(), len, false,
false, Context, F);
cpy->eraseFromParent();
}
for (unsigned i = 0, e = aggrMemsets.size(); i != e; ++i) {
MemSetInst *memsetinst = aggrMemsets[i];
Value *len = memsetinst->getLength();
Value *val = memsetinst->getValue();
convertMemSetToLoop(memsetinst, memsetinst->getDest(), len, val, Context,
F);
memsetinst->eraseFromParent();
}
return true;
}
/// isOnlyCopiedFromConstantGlobal - Recursively walk the uses of a (derived)
/// pointer to an alloca. Ignore any reads of the pointer, return false if we
/// see any stores or other unknown uses. If we see pointer arithmetic, keep
/// track of whether it moves the pointer (with IsOffset) but otherwise traverse
/// the uses. If we see a memcpy/memmove that targets an unoffseted pointer to
/// the alloca, and if the source pointer is a pointer to a constant global, we
/// can optimize this.
static bool
isOnlyCopiedFromConstantGlobal(Value *V, MemTransferInst *&TheCopy,
SmallVectorImpl<Instruction *> &ToDelete,
bool IsOffset = false) {
// We track lifetime intrinsics as we encounter them. If we decide to go
// ahead and replace the value with the global, this lets the caller quickly
// eliminate the markers.
for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI!=E; ++UI) {
User *U = cast<Instruction>(*UI);
if (LoadInst *LI = dyn_cast<LoadInst>(U)) {
// Ignore non-volatile loads, they are always ok.
if (!LI->isSimple()) return false;
continue;
}
if (BitCastInst *BCI = dyn_cast<BitCastInst>(U)) {
// If uses of the bitcast are ok, we are ok.
if (!isOnlyCopiedFromConstantGlobal(BCI, TheCopy, ToDelete, IsOffset))
return false;
continue;
}
if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(U)) {
// If the GEP has all zero indices, it doesn't offset the pointer. If it
// doesn't, it does.
if (!isOnlyCopiedFromConstantGlobal(GEP, TheCopy, ToDelete,
IsOffset || !GEP->hasAllZeroIndices()))
return false;
continue;
}
if (CallSite CS = U) {
// If this is the function being called then we treat it like a load and
// ignore it.
if (CS.isCallee(UI))
continue;
// If this is a readonly/readnone call site, then we know it is just a
// load (but one that potentially returns the value itself), so we can
// ignore it if we know that the value isn't captured.
unsigned ArgNo = CS.getArgumentNo(UI);
if (CS.onlyReadsMemory() &&
(CS.getInstruction()->use_empty() || CS.doesNotCapture(ArgNo)))
continue;
// If this is being passed as a byval argument, the caller is making a
// copy, so it is only a read of the alloca.
if (CS.isByValArgument(ArgNo))
continue;
}
// Lifetime intrinsics can be handled by the caller.
if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(U)) {
if (II->getIntrinsicID() == Intrinsic::lifetime_start ||
II->getIntrinsicID() == Intrinsic::lifetime_end) {
assert(II->use_empty() && "Lifetime markers have no result to use!");
ToDelete.push_back(II);
continue;
}
}
// If this is isn't our memcpy/memmove, reject it as something we can't
// handle.
MemTransferInst *MI = dyn_cast<MemTransferInst>(U);
if (MI == 0)
return false;
// If the transfer is using the alloca as a source of the transfer, then
// ignore it since it is a load (unless the transfer is volatile).
if (UI.getOperandNo() == 1) {
if (MI->isVolatile()) return false;
continue;
}
// If we already have seen a copy, reject the second one.
if (TheCopy) return false;
// If the pointer has been offset from the start of the alloca, we can't
// safely handle this.
if (IsOffset) return false;
// If the memintrinsic isn't using the alloca as the dest, reject it.
if (UI.getOperandNo() != 0) return false;
// If the source of the memcpy/move is not a constant global, reject it.
if (!pointsToConstantGlobal(MI->getSource()))
return false;
// Otherwise, the transform is safe. Remember the copy instruction.
TheCopy = MI;
}
return true;
}
void PropagateJuliaAddrspaces::visitMemTransferInst(MemTransferInst &MTI) {
unsigned DestAS = MTI.getDestAddressSpace();
unsigned SrcAS = MTI.getSourceAddressSpace();
if (!isSpecialAS(DestAS) && !isSpecialAS(SrcAS))
return;
Value *Dest = MTI.getRawDest();
if (isSpecialAS(DestAS)) {
Value *Replacement = LiftPointer(Dest, cast<PointerType>(Dest->getType())->getElementType(), &MTI);
if (Replacement)
Dest = Replacement;
}
Value *Src = MTI.getRawSource();
if (isSpecialAS(SrcAS)) {
Value *Replacement = LiftPointer(Src, cast<PointerType>(Src->getType())->getElementType(), &MTI);
if (Replacement)
Src = Replacement;
}
if (Dest == MTI.getRawDest() && Src == MTI.getRawSource())
return;
Value *TheFn = Intrinsic::getDeclaration(MTI.getModule(), MTI.getIntrinsicID(),
{Dest->getType(), Src->getType(),
MTI.getOperand(2)->getType()});
MTI.setCalledFunction(TheFn);
MTI.setArgOperand(0, Dest);
MTI.setArgOperand(1, Src);
}