/// processMemCpy - perform simplification of memcpy's. If we have memcpy A
/// which copies X to Y, and memcpy B which copies Y to Z, then we can rewrite
/// B to be a memcpy from X to Z (or potentially a memmove, depending on
/// circumstances). This allows later passes to remove the first memcpy
/// altogether.
bool MemCpyOpt::processMemCpy(MemCpyInst *M) {
// We can only optimize statically-sized memcpy's that are non-volatile.
ConstantInt *CopySize = dyn_cast<ConstantInt>(M->getLength());
if (CopySize == 0 || M->isVolatile()) return false;
// If the source and destination of the memcpy are the same, then zap it.
if (M->getSource() == M->getDest()) {
MD->removeInstruction(M);
M->eraseFromParent();
return false;
}
// If copying from a constant, try to turn the memcpy into a memset.
if (GlobalVariable *GV = dyn_cast<GlobalVariable>(M->getSource()))
if (GV->isConstant() && GV->hasDefinitiveInitializer())
if (Value *ByteVal = isBytewiseValue(GV->getInitializer())) {
IRBuilder<> Builder(M);
Builder.CreateMemSet(M->getRawDest(), ByteVal, CopySize,
M->getAlignment(), false);
MD->removeInstruction(M);
M->eraseFromParent();
++NumCpyToSet;
return true;
}
// The are two possible optimizations we can do for memcpy:
// a) memcpy-memcpy xform which exposes redundance for DSE.
// b) call-memcpy xform for return slot optimization.
MemDepResult DepInfo = MD->getDependency(M);
if (DepInfo.isClobber()) {
if (CallInst *C = dyn_cast<CallInst>(DepInfo.getInst())) {
if (performCallSlotOptzn(M, M->getDest(), M->getSource(),
CopySize->getZExtValue(), M->getAlignment(),
C)) {
MD->removeInstruction(M);
M->eraseFromParent();
return true;
}
}
}
AliasAnalysis::Location SrcLoc = AliasAnalysis::getLocationForSource(M);
MemDepResult SrcDepInfo = MD->getPointerDependencyFrom(SrcLoc, true,
M, M->getParent());
if (SrcDepInfo.isClobber()) {
if (MemCpyInst *MDep = dyn_cast<MemCpyInst>(SrcDepInfo.getInst()))
return processMemCpyMemCpyDependence(M, MDep, CopySize->getZExtValue());
}
return false;
}
/// processMemCpy - perform simplification of memcpy's. If we have memcpy A
/// which copies X to Y, and memcpy B which copies Y to Z, then we can rewrite
/// B to be a memcpy from X to Z (or potentially a memmove, depending on
/// circumstances). This allows later passes to remove the first memcpy
/// altogether.
bool MemCpyOpt::processMemCpy(MemCpyInst *M) {
// We can only optimize statically-sized memcpy's that are non-volatile.
ConstantInt *CopySize = dyn_cast<ConstantInt>(M->getLength());
if (CopySize == 0 || M->isVolatile()) return false;
// If the source and destination of the memcpy are the same, then zap it.
if (M->getSource() == M->getDest()) {
MD->removeInstruction(M);
M->eraseFromParent();
return false;
}
// The are two possible optimizations we can do for memcpy:
// a) memcpy-memcpy xform which exposes redundance for DSE.
// b) call-memcpy xform for return slot optimization.
MemDepResult DepInfo = MD->getDependency(M);
if (!DepInfo.isClobber())
return false;
if (MemCpyInst *MDep = dyn_cast<MemCpyInst>(DepInfo.getInst()))
return processMemCpyMemCpyDependence(M, MDep, CopySize->getZExtValue());
if (CallInst *C = dyn_cast<CallInst>(DepInfo.getInst())) {
if (performCallSlotOptzn(M, M->getDest(), M->getSource(),
CopySize->getZExtValue(), C)) {
M->eraseFromParent();
return true;
}
}
return false;
}
/// Handle frees of entire structures whose dependency is a store
/// to a field of that structure.
static bool handleFree(CallInst *F, AliasAnalysis *AA,
MemoryDependenceResults *MD, DominatorTree *DT,
const TargetLibraryInfo *TLI,
InstOverlapIntervalsTy &IOL,
DenseMap<Instruction*, size_t> *InstrOrdering) {
bool MadeChange = false;
MemoryLocation Loc = MemoryLocation(F->getOperand(0));
SmallVector<BasicBlock *, 16> Blocks;
Blocks.push_back(F->getParent());
const DataLayout &DL = F->getModule()->getDataLayout();
while (!Blocks.empty()) {
BasicBlock *BB = Blocks.pop_back_val();
Instruction *InstPt = BB->getTerminator();
if (BB == F->getParent()) InstPt = F;
MemDepResult Dep =
MD->getPointerDependencyFrom(Loc, false, InstPt->getIterator(), BB);
while (Dep.isDef() || Dep.isClobber()) {
Instruction *Dependency = Dep.getInst();
if (!hasMemoryWrite(Dependency, *TLI) || !isRemovable(Dependency))
break;
Value *DepPointer =
GetUnderlyingObject(getStoredPointerOperand(Dependency), DL);
// Check for aliasing.
if (!AA->isMustAlias(F->getArgOperand(0), DepPointer))
break;
DEBUG(dbgs() << "DSE: Dead Store to soon to be freed memory:\n DEAD: "
<< *Dependency << '\n');
// DCE instructions only used to calculate that store.
BasicBlock::iterator BBI(Dependency);
deleteDeadInstruction(Dependency, &BBI, *MD, *TLI, IOL, InstrOrdering);
++NumFastStores;
MadeChange = true;
// Inst's old Dependency is now deleted. Compute the next dependency,
// which may also be dead, as in
// s[0] = 0;
// s[1] = 0; // This has just been deleted.
// free(s);
Dep = MD->getPointerDependencyFrom(Loc, false, BBI, BB);
}
if (Dep.isNonLocal())
findUnconditionalPreds(Blocks, BB, DT);
}
return MadeChange;
}
DataDependence::DepInfo DataDependence::getDepInfo(MemDepResult dep) {
if (dep.isClobber())
return DepInfo(dep.getInst(), Clobber);
if (dep.isDef())
return DepInfo(dep.getInst(), Def);
if (dep.isNonFuncLocal())
return DepInfo(dep.getInst(), NonFuncLocal);
if (dep.isUnknown())
return DepInfo(dep.getInst(), Unknown);
if (dep.isNonLocal())
return DepInfo(dep.getInst(), NonLocal);
llvm_unreachable("unknown dependence type");
}
/// HandleFree - Handle frees of entire structures whose dependency is a store
/// to a field of that structure.
bool DSE::HandleFree(CallInst *F) {
bool MadeChange = false;
MemoryLocation Loc = MemoryLocation(F->getOperand(0));
SmallVector<BasicBlock *, 16> Blocks;
Blocks.push_back(F->getParent());
const DataLayout &DL = F->getModule()->getDataLayout();
while (!Blocks.empty()) {
BasicBlock *BB = Blocks.pop_back_val();
Instruction *InstPt = BB->getTerminator();
if (BB == F->getParent()) InstPt = F;
MemDepResult Dep = MD->getPointerDependencyFrom(Loc, false, InstPt, BB);
while (Dep.isDef() || Dep.isClobber()) {
Instruction *Dependency = Dep.getInst();
if (!hasMemoryWrite(Dependency, *TLI) || !isRemovable(Dependency))
break;
Value *DepPointer =
GetUnderlyingObject(getStoredPointerOperand(Dependency), DL);
// Check for aliasing.
if (!AA->isMustAlias(F->getArgOperand(0), DepPointer))
break;
Instruction *Next = std::next(BasicBlock::iterator(Dependency));
// DCE instructions only used to calculate that store
DeleteDeadInstruction(Dependency, *MD, *TLI);
++NumFastStores;
MadeChange = true;
// Inst's old Dependency is now deleted. Compute the next dependency,
// which may also be dead, as in
// s[0] = 0;
// s[1] = 0; // This has just been deleted.
// free(s);
Dep = MD->getPointerDependencyFrom(Loc, false, Next, BB);
}
if (Dep.isNonLocal())
FindUnconditionalPreds(Blocks, BB, DT);
}
return MadeChange;
}
/// processByValArgument - This is called on every byval argument in call sites.
bool MemCpyOpt::processByValArgument(CallSite CS, unsigned ArgNo) {
if (TD == 0) return false;
// Find out what feeds this byval argument.
Value *ByValArg = CS.getArgument(ArgNo);
Type *ByValTy = cast<PointerType>(ByValArg->getType())->getElementType();
uint64_t ByValSize = TD->getTypeAllocSize(ByValTy);
MemDepResult DepInfo =
MD->getPointerDependencyFrom(AliasAnalysis::Location(ByValArg, ByValSize),
true, CS.getInstruction(),
CS.getInstruction()->getParent());
if (!DepInfo.isClobber())
return false;
// If the byval argument isn't fed by a memcpy, ignore it. If it is fed by
// a memcpy, see if we can byval from the source of the memcpy instead of the
// result.
MemCpyInst *MDep = dyn_cast<MemCpyInst>(DepInfo.getInst());
if (MDep == 0 || MDep->isVolatile() ||
ByValArg->stripPointerCasts() != MDep->getDest())
return false;
// The length of the memcpy must be larger or equal to the size of the byval.
ConstantInt *C1 = dyn_cast<ConstantInt>(MDep->getLength());
if (C1 == 0 || C1->getValue().getZExtValue() < ByValSize)
return false;
// Get the alignment of the byval. If the call doesn't specify the alignment,
// then it is some target specific value that we can't know.
unsigned ByValAlign = CS.getParamAlignment(ArgNo+1);
if (ByValAlign == 0) return false;
// If it is greater than the memcpy, then we check to see if we can force the
// source of the memcpy to the alignment we need. If we fail, we bail out.
if (MDep->getAlignment() < ByValAlign &&
getOrEnforceKnownAlignment(MDep->getSource(),ByValAlign, TD) < ByValAlign)
return false;
// Verify that the copied-from memory doesn't change in between the memcpy and
// the byval call.
// memcpy(a <- b)
// *b = 42;
// foo(*a)
// It would be invalid to transform the second memcpy into foo(*b).
//
// NOTE: This is conservative, it will stop on any read from the source loc,
// not just the defining memcpy.
MemDepResult SourceDep =
MD->getPointerDependencyFrom(AliasAnalysis::getLocationForSource(MDep),
false, CS.getInstruction(), MDep->getParent());
if (!SourceDep.isClobber() || SourceDep.getInst() != MDep)
return false;
Value *TmpCast = MDep->getSource();
if (MDep->getSource()->getType() != ByValArg->getType())
TmpCast = new BitCastInst(MDep->getSource(), ByValArg->getType(),
"tmpcast", CS.getInstruction());
DEBUG(dbgs() << "MemCpyOpt: Forwarding memcpy to byval:\n"
<< " " << *MDep << "\n"
<< " " << *CS.getInstruction() << "\n");
// Otherwise we're good! Update the byval argument.
CS.setArgument(ArgNo, TmpCast);
++NumMemCpyInstr;
return true;
}
/// processMemCpyMemCpyDependence - We've found that the (upward scanning)
/// memory dependence of memcpy 'M' is the memcpy 'MDep'. Try to simplify M to
/// copy from MDep's input if we can. MSize is the size of M's copy.
///
bool MemCpyOpt::processMemCpyMemCpyDependence(MemCpyInst *M, MemCpyInst *MDep,
uint64_t MSize) {
// We can only transforms memcpy's where the dest of one is the source of the
// other.
if (M->getSource() != MDep->getDest() || MDep->isVolatile())
return false;
// If dep instruction is reading from our current input, then it is a noop
// transfer and substituting the input won't change this instruction. Just
// ignore the input and let someone else zap MDep. This handles cases like:
// memcpy(a <- a)
// memcpy(b <- a)
if (M->getSource() == MDep->getSource())
return false;
// Second, the length of the memcpy's must be the same, or the preceding one
// must be larger than the following one.
ConstantInt *MDepLen = dyn_cast<ConstantInt>(MDep->getLength());
ConstantInt *MLen = dyn_cast<ConstantInt>(M->getLength());
if (!MDepLen || !MLen || MDepLen->getZExtValue() < MLen->getZExtValue())
return false;
AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
// Verify that the copied-from memory doesn't change in between the two
// transfers. For example, in:
// memcpy(a <- b)
// *b = 42;
// memcpy(c <- a)
// It would be invalid to transform the second memcpy into memcpy(c <- b).
//
// TODO: If the code between M and MDep is transparent to the destination "c",
// then we could still perform the xform by moving M up to the first memcpy.
//
// NOTE: This is conservative, it will stop on any read from the source loc,
// not just the defining memcpy.
MemDepResult SourceDep =
MD->getPointerDependencyFrom(AA.getLocationForSource(MDep),
false, M, M->getParent());
if (!SourceDep.isClobber() || SourceDep.getInst() != MDep)
return false;
// If the dest of the second might alias the source of the first, then the
// source and dest might overlap. We still want to eliminate the intermediate
// value, but we have to generate a memmove instead of memcpy.
bool UseMemMove = false;
if (!AA.isNoAlias(AA.getLocationForDest(M), AA.getLocationForSource(MDep)))
UseMemMove = true;
// If all checks passed, then we can transform M.
// Make sure to use the lesser of the alignment of the source and the dest
// since we're changing where we're reading from, but don't want to increase
// the alignment past what can be read from or written to.
// TODO: Is this worth it if we're creating a less aligned memcpy? For
// example we could be moving from movaps -> movq on x86.
unsigned Align = std::min(MDep->getAlignment(), M->getAlignment());
IRBuilder<> Builder(M);
if (UseMemMove)
Builder.CreateMemMove(M->getRawDest(), MDep->getRawSource(), M->getLength(),
Align, M->isVolatile());
else
Builder.CreateMemCpy(M->getRawDest(), MDep->getRawSource(), M->getLength(),
Align, M->isVolatile());
// Remove the instruction we're replacing.
MD->removeInstruction(M);
M->eraseFromParent();
++NumMemCpyInstr;
return true;
}
bool MemCpyOpt::processStore(StoreInst *SI, BasicBlock::iterator &BBI) {
if (!SI->isSimple()) return false;
if (TD == 0) return false;
// Detect cases where we're performing call slot forwarding, but
// happen to be using a load-store pair to implement it, rather than
// a memcpy.
if (LoadInst *LI = dyn_cast<LoadInst>(SI->getOperand(0))) {
if (LI->isSimple() && LI->hasOneUse() &&
LI->getParent() == SI->getParent()) {
MemDepResult ldep = MD->getDependency(LI);
CallInst *C = 0;
if (ldep.isClobber() && !isa<MemCpyInst>(ldep.getInst()))
C = dyn_cast<CallInst>(ldep.getInst());
if (C) {
// Check that nothing touches the dest of the "copy" between
// the call and the store.
AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
AliasAnalysis::Location StoreLoc = AA.getLocation(SI);
for (BasicBlock::iterator I = --BasicBlock::iterator(SI),
E = C; I != E; --I) {
if (AA.getModRefInfo(&*I, StoreLoc) != AliasAnalysis::NoModRef) {
C = 0;
break;
}
}
}
if (C) {
unsigned storeAlign = SI->getAlignment();
if (!storeAlign)
storeAlign = TD->getABITypeAlignment(SI->getOperand(0)->getType());
unsigned loadAlign = LI->getAlignment();
if (!loadAlign)
loadAlign = TD->getABITypeAlignment(LI->getType());
bool changed = performCallSlotOptzn(LI,
SI->getPointerOperand()->stripPointerCasts(),
LI->getPointerOperand()->stripPointerCasts(),
TD->getTypeStoreSize(SI->getOperand(0)->getType()),
std::min(storeAlign, loadAlign), C);
if (changed) {
MD->removeInstruction(SI);
SI->eraseFromParent();
MD->removeInstruction(LI);
LI->eraseFromParent();
++NumMemCpyInstr;
return true;
}
}
}
}
// There are two cases that are interesting for this code to handle: memcpy
// and memset. Right now we only handle memset.
// Ensure that the value being stored is something that can be memset'able a
// byte at a time like "0" or "-1" or any width, as well as things like
// 0xA0A0A0A0 and 0.0.
if (Value *ByteVal = isBytewiseValue(SI->getOperand(0)))
if (Instruction *I = tryMergingIntoMemset(SI, SI->getPointerOperand(),
ByteVal)) {
BBI = I; // Don't invalidate iterator.
return true;
}
return false;
}
/// processMemCpy - perform simplication of memcpy's. If we have memcpy A which
/// copies X to Y, and memcpy B which copies Y to Z, then we can rewrite B to be
/// a memcpy from X to Z (or potentially a memmove, depending on circumstances).
/// This allows later passes to remove the first memcpy altogether.
bool MemCpyOpt::processMemCpy(MemCpyInst *M) {
MemoryDependenceAnalysis &MD = getAnalysis<MemoryDependenceAnalysis>();
// The are two possible optimizations we can do for memcpy:
// a) memcpy-memcpy xform which exposes redundance for DSE.
// b) call-memcpy xform for return slot optimization.
MemDepResult dep = MD.getDependency(M);
if (!dep.isClobber())
return false;
if (!isa<MemCpyInst>(dep.getInst())) {
if (CallInst *C = dyn_cast<CallInst>(dep.getInst()))
return performCallSlotOptzn(M, C);
return false;
}
MemCpyInst *MDep = cast<MemCpyInst>(dep.getInst());
// We can only transforms memcpy's where the dest of one is the source of the
// other
if (M->getSource() != MDep->getDest())
return false;
// Second, the length of the memcpy's must be the same, or the preceeding one
// must be larger than the following one.
ConstantInt *C1 = dyn_cast<ConstantInt>(MDep->getLength());
ConstantInt *C2 = dyn_cast<ConstantInt>(M->getLength());
if (!C1 || !C2)
return false;
uint64_t DepSize = C1->getValue().getZExtValue();
uint64_t CpySize = C2->getValue().getZExtValue();
if (DepSize < CpySize)
return false;
// Finally, we have to make sure that the dest of the second does not
// alias the source of the first
AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
if (AA.alias(M->getRawDest(), CpySize, MDep->getRawSource(), DepSize) !=
AliasAnalysis::NoAlias)
return false;
else if (AA.alias(M->getRawDest(), CpySize, M->getRawSource(), CpySize) !=
AliasAnalysis::NoAlias)
return false;
else if (AA.alias(MDep->getRawDest(), DepSize, MDep->getRawSource(), DepSize)
!= AliasAnalysis::NoAlias)
return false;
// If all checks passed, then we can transform these memcpy's
const Type *ArgTys[3] = { M->getRawDest()->getType(),
MDep->getRawSource()->getType(),
M->getLength()->getType() };
Function *MemCpyFun = Intrinsic::getDeclaration(
M->getParent()->getParent()->getParent(),
M->getIntrinsicID(), ArgTys, 3);
Value *Args[5] = {
M->getRawDest(), MDep->getRawSource(), M->getLength(),
M->getAlignmentCst(), M->getVolatileCst()
};
CallInst *C = CallInst::Create(MemCpyFun, Args, Args+5, "", M);
// If C and M don't interfere, then this is a valid transformation. If they
// did, this would mean that the two sources overlap, which would be bad.
if (MD.getDependency(C) == dep) {
MD.removeInstruction(M);
M->eraseFromParent();
++NumMemCpyInstr;
return true;
}
// Otherwise, there was no point in doing this, so we remove the call we
// inserted and act like nothing happened.
MD.removeInstruction(C);
C->eraseFromParent();
return false;
}
/// processMemCpy - perform simplification of memcpy's. If we have memcpy A
/// which copies X to Y, and memcpy B which copies Y to Z, then we can rewrite
/// B to be a memcpy from X to Z (or potentially a memmove, depending on
/// circumstances). This allows later passes to remove the first memcpy
/// altogether.
bool MemCpyOpt::processMemCpy(MemCpyInst *M) {
// We can only optimize non-volatile memcpy's.
if (M->isVolatile()) return false;
// If the source and destination of the memcpy are the same, then zap it.
if (M->getSource() == M->getDest()) {
MD->removeInstruction(M);
M->eraseFromParent();
return false;
}
// If copying from a constant, try to turn the memcpy into a memset.
if (GlobalVariable *GV = dyn_cast<GlobalVariable>(M->getSource()))
if (GV->isConstant() && GV->hasDefinitiveInitializer())
if (Value *ByteVal = isBytewiseValue(GV->getInitializer())) {
IRBuilder<> Builder(M);
Builder.CreateMemSet(M->getRawDest(), ByteVal, M->getLength(),
M->getAlignment(), false);
MD->removeInstruction(M);
M->eraseFromParent();
++NumCpyToSet;
return true;
}
MemDepResult DepInfo = MD->getDependency(M);
// Try to turn a partially redundant memset + memcpy into
// memcpy + smaller memset. We don't need the memcpy size for this.
if (DepInfo.isClobber())
if (MemSetInst *MDep = dyn_cast<MemSetInst>(DepInfo.getInst()))
if (processMemSetMemCpyDependence(M, MDep))
return true;
// The optimizations after this point require the memcpy size.
ConstantInt *CopySize = dyn_cast<ConstantInt>(M->getLength());
if (!CopySize) return false;
// There are four possible optimizations we can do for memcpy:
// a) memcpy-memcpy xform which exposes redundance for DSE.
// b) call-memcpy xform for return slot optimization.
// c) memcpy from freshly alloca'd space or space that has just started its
// lifetime copies undefined data, and we can therefore eliminate the
// memcpy in favor of the data that was already at the destination.
// d) memcpy from a just-memset'd source can be turned into memset.
if (DepInfo.isClobber()) {
if (CallInst *C = dyn_cast<CallInst>(DepInfo.getInst())) {
if (performCallSlotOptzn(M, M->getDest(), M->getSource(),
CopySize->getZExtValue(), M->getAlignment(),
C)) {
MD->removeInstruction(M);
M->eraseFromParent();
return true;
}
}
}
AliasAnalysis::Location SrcLoc = AliasAnalysis::getLocationForSource(M);
MemDepResult SrcDepInfo = MD->getPointerDependencyFrom(SrcLoc, true,
M, M->getParent());
if (SrcDepInfo.isClobber()) {
if (MemCpyInst *MDep = dyn_cast<MemCpyInst>(SrcDepInfo.getInst()))
return processMemCpyMemCpyDependence(M, MDep);
} else if (SrcDepInfo.isDef()) {
Instruction *I = SrcDepInfo.getInst();
bool hasUndefContents = false;
if (isa<AllocaInst>(I)) {
hasUndefContents = true;
} else if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
if (II->getIntrinsicID() == Intrinsic::lifetime_start)
if (ConstantInt *LTSize = dyn_cast<ConstantInt>(II->getArgOperand(0)))
if (LTSize->getZExtValue() >= CopySize->getZExtValue())
hasUndefContents = true;
}
if (hasUndefContents) {
MD->removeInstruction(M);
M->eraseFromParent();
++NumMemCpyInstr;
return true;
}
}
if (SrcDepInfo.isClobber())
if (MemSetInst *MDep = dyn_cast<MemSetInst>(SrcDepInfo.getInst()))
if (performMemCpyToMemSetOptzn(M, MDep)) {
MD->removeInstruction(M);
M->eraseFromParent();
++NumCpyToSet;
return true;
}
return false;
}
bool DSE::runOnBasicBlock(BasicBlock &BB) {
bool MadeChange = false;
// Do a top-down walk on the BB.
for (BasicBlock::iterator BBI = BB.begin(), BBE = BB.end(); BBI != BBE; ) {
Instruction *Inst = BBI++;
// Handle 'free' calls specially.
if (CallInst *F = isFreeCall(Inst, TLI)) {
MadeChange |= HandleFree(F);
continue;
}
// If we find something that writes memory, get its memory dependence.
if (!hasMemoryWrite(Inst, TLI))
continue;
MemDepResult InstDep = MD->getDependency(Inst);
// Ignore any store where we can't find a local dependence.
// FIXME: cross-block DSE would be fun. :)
if (!InstDep.isDef() && !InstDep.isClobber())
continue;
// If we're storing the same value back to a pointer that we just
// loaded from, then the store can be removed.
if (StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
if (LoadInst *DepLoad = dyn_cast<LoadInst>(InstDep.getInst())) {
if (SI->getPointerOperand() == DepLoad->getPointerOperand() &&
SI->getOperand(0) == DepLoad && isRemovable(SI)) {
DEBUG(dbgs() << "DSE: Remove Store Of Load from same pointer:\n "
<< "LOAD: " << *DepLoad << "\n STORE: " << *SI << '\n');
// DeleteDeadInstruction can delete the current instruction. Save BBI
// in case we need it.
WeakVH NextInst(BBI);
DeleteDeadInstruction(SI, *MD, TLI);
if (!NextInst) // Next instruction deleted.
BBI = BB.begin();
else if (BBI != BB.begin()) // Revisit this instruction if possible.
--BBI;
++NumFastStores;
MadeChange = true;
continue;
}
}
}
// Figure out what location is being stored to.
AliasAnalysis::Location Loc = getLocForWrite(Inst, *AA);
// If we didn't get a useful location, fail.
if (!Loc.Ptr)
continue;
while (InstDep.isDef() || InstDep.isClobber()) {
// Get the memory clobbered by the instruction we depend on. MemDep will
// skip any instructions that 'Loc' clearly doesn't interact with. If we
// end up depending on a may- or must-aliased load, then we can't optimize
// away the store and we bail out. However, if we depend on on something
// that overwrites the memory location we *can* potentially optimize it.
//
// Find out what memory location the dependent instruction stores.
Instruction *DepWrite = InstDep.getInst();
AliasAnalysis::Location DepLoc = getLocForWrite(DepWrite, *AA);
// If we didn't get a useful location, or if it isn't a size, bail out.
if (!DepLoc.Ptr)
break;
// If we find a write that is a) removable (i.e., non-volatile), b) is
// completely obliterated by the store to 'Loc', and c) which we know that
// 'Inst' doesn't load from, then we can remove it.
if (isRemovable(DepWrite) &&
!isPossibleSelfRead(Inst, Loc, DepWrite, *AA)) {
int64_t InstWriteOffset, DepWriteOffset;
OverwriteResult OR = isOverwrite(Loc, DepLoc, *AA,
DepWriteOffset, InstWriteOffset);
if (OR == OverwriteComplete) {
DEBUG(dbgs() << "DSE: Remove Dead Store:\n DEAD: "
<< *DepWrite << "\n KILLER: " << *Inst << '\n');
// Delete the store and now-dead instructions that feed it.
DeleteDeadInstruction(DepWrite, *MD, TLI);
++NumFastStores;
MadeChange = true;
// DeleteDeadInstruction can delete the current instruction in loop
// cases, reset BBI.
BBI = Inst;
if (BBI != BB.begin())
--BBI;
break;
} else if (OR == OverwriteEnd && isShortenable(DepWrite)) {
// TODO: base this on the target vector size so that if the earlier
// store was too small to get vector writes anyway then its likely
// a good idea to shorten it
// Power of 2 vector writes are probably always a bad idea to optimize
// as any store/memset/memcpy is likely using vector instructions so
// shortening it to not vector size is likely to be slower
MemIntrinsic* DepIntrinsic = cast<MemIntrinsic>(DepWrite);
unsigned DepWriteAlign = DepIntrinsic->getAlignment();
if (llvm::isPowerOf2_64(InstWriteOffset) ||
((DepWriteAlign != 0) && InstWriteOffset % DepWriteAlign == 0)) {
DEBUG(dbgs() << "DSE: Remove Dead Store:\n OW END: "
<< *DepWrite << "\n KILLER (offset "
<< InstWriteOffset << ", "
<< DepLoc.Size << ")"
<< *Inst << '\n');
Value* DepWriteLength = DepIntrinsic->getLength();
Value* TrimmedLength = ConstantInt::get(DepWriteLength->getType(),
InstWriteOffset -
DepWriteOffset);
DepIntrinsic->setLength(TrimmedLength);
MadeChange = true;
}
}
}
// If this is a may-aliased store that is clobbering the store value, we
// can keep searching past it for another must-aliased pointer that stores
// to the same location. For example, in:
// store -> P
// store -> Q
// store -> P
// we can remove the first store to P even though we don't know if P and Q
// alias.
if (DepWrite == &BB.front()) break;
// Can't look past this instruction if it might read 'Loc'.
if (AA->getModRefInfo(DepWrite, Loc) & AliasAnalysis::Ref)
break;
InstDep = MD->getPointerDependencyFrom(Loc, false, DepWrite, &BB);
}
}
// If this block ends in a return, unwind, or unreachable, all allocas are
// dead at its end, which means stores to them are also dead.
if (BB.getTerminator()->getNumSuccessors() == 0)
MadeChange |= handleEndBlock(BB);
return MadeChange;
}
bool MemCpyOpt::processStore(StoreInst *SI, BasicBlock::iterator &BBI) {
if (!SI->isSimple()) return false;
// Avoid merging nontemporal stores since the resulting
// memcpy/memset would not be able to preserve the nontemporal hint.
// In theory we could teach how to propagate the !nontemporal metadata to
// memset calls. However, that change would force the backend to
// conservatively expand !nontemporal memset calls back to sequences of
// store instructions (effectively undoing the merging).
if (SI->getMetadata(LLVMContext::MD_nontemporal))
return false;
const DataLayout &DL = SI->getModule()->getDataLayout();
// Detect cases where we're performing call slot forwarding, but
// happen to be using a load-store pair to implement it, rather than
// a memcpy.
if (LoadInst *LI = dyn_cast<LoadInst>(SI->getOperand(0))) {
if (LI->isSimple() && LI->hasOneUse() &&
LI->getParent() == SI->getParent()) {
MemDepResult ldep = MD->getDependency(LI);
CallInst *C = nullptr;
if (ldep.isClobber() && !isa<MemCpyInst>(ldep.getInst()))
C = dyn_cast<CallInst>(ldep.getInst());
if (C) {
// Check that nothing touches the dest of the "copy" between
// the call and the store.
AliasAnalysis &AA = getAnalysis<AAResultsWrapperPass>().getAAResults();
MemoryLocation StoreLoc = MemoryLocation::get(SI);
for (BasicBlock::iterator I = --SI->getIterator(), E = C->getIterator();
I != E; --I) {
if (AA.getModRefInfo(&*I, StoreLoc) != MRI_NoModRef) {
C = nullptr;
break;
}
}
}
if (C) {
unsigned storeAlign = SI->getAlignment();
if (!storeAlign)
storeAlign = DL.getABITypeAlignment(SI->getOperand(0)->getType());
unsigned loadAlign = LI->getAlignment();
if (!loadAlign)
loadAlign = DL.getABITypeAlignment(LI->getType());
bool changed = performCallSlotOptzn(
LI, SI->getPointerOperand()->stripPointerCasts(),
LI->getPointerOperand()->stripPointerCasts(),
DL.getTypeStoreSize(SI->getOperand(0)->getType()),
std::min(storeAlign, loadAlign), C);
if (changed) {
MD->removeInstruction(SI);
SI->eraseFromParent();
MD->removeInstruction(LI);
LI->eraseFromParent();
++NumMemCpyInstr;
return true;
}
}
}
}
// There are two cases that are interesting for this code to handle: memcpy
// and memset. Right now we only handle memset.
// Ensure that the value being stored is something that can be memset'able a
// byte at a time like "0" or "-1" or any width, as well as things like
// 0xA0A0A0A0 and 0.0.
if (Value *ByteVal = isBytewiseValue(SI->getOperand(0)))
if (Instruction *I = tryMergingIntoMemset(SI, SI->getPointerOperand(),
ByteVal)) {
BBI = I->getIterator(); // Don't invalidate iterator.
return true;
}
return false;
}
bool MemCpyOpt::processStore(StoreInst *SI, BasicBlock::iterator &BBI) {
if (!SI->isSimple()) return false;
// Avoid merging nontemporal stores since the resulting
// memcpy/memset would not be able to preserve the nontemporal hint.
// In theory we could teach how to propagate the !nontemporal metadata to
// memset calls. However, that change would force the backend to
// conservatively expand !nontemporal memset calls back to sequences of
// store instructions (effectively undoing the merging).
if (SI->getMetadata(LLVMContext::MD_nontemporal))
return false;
const DataLayout &DL = SI->getModule()->getDataLayout();
// Load to store forwarding can be interpreted as memcpy.
if (LoadInst *LI = dyn_cast<LoadInst>(SI->getOperand(0))) {
if (LI->isSimple() && LI->hasOneUse() &&
LI->getParent() == SI->getParent()) {
auto *T = LI->getType();
if (T->isAggregateType()) {
AliasAnalysis &AA = getAnalysis<AAResultsWrapperPass>().getAAResults();
MemoryLocation LoadLoc = MemoryLocation::get(LI);
// We use alias analysis to check if an instruction may store to
// the memory we load from in between the load and the store. If
// such an instruction is found, we store it in AI.
Instruction *AI = nullptr;
for (BasicBlock::iterator I = ++LI->getIterator(), E = SI->getIterator();
I != E; ++I) {
if (AA.getModRefInfo(&*I, LoadLoc) & MRI_Mod) {
AI = &*I;
break;
}
}
// If no aliasing instruction is found, then we can promote the
// load/store pair to a memcpy at the store loaction.
if (!AI) {
// If we load from memory that may alias the memory we store to,
// memmove must be used to preserve semantic. If not, memcpy can
// be used.
bool UseMemMove = false;
if (!AA.isNoAlias(MemoryLocation::get(SI), LoadLoc))
UseMemMove = true;
unsigned Align = findCommonAlignment(DL, SI, LI);
uint64_t Size = DL.getTypeStoreSize(T);
IRBuilder<> Builder(SI);
Instruction *M;
if (UseMemMove)
M = Builder.CreateMemMove(SI->getPointerOperand(),
LI->getPointerOperand(), Size,
Align, SI->isVolatile());
else
M = Builder.CreateMemCpy(SI->getPointerOperand(),
LI->getPointerOperand(), Size,
Align, SI->isVolatile());
DEBUG(dbgs() << "Promoting " << *LI << " to " << *SI
<< " => " << *M << "\n");
MD->removeInstruction(SI);
SI->eraseFromParent();
MD->removeInstruction(LI);
LI->eraseFromParent();
++NumMemCpyInstr;
// Make sure we do not invalidate the iterator.
BBI = M->getIterator();
return true;
}
}
// Detect cases where we're performing call slot forwarding, but
// happen to be using a load-store pair to implement it, rather than
// a memcpy.
MemDepResult ldep = MD->getDependency(LI);
CallInst *C = nullptr;
if (ldep.isClobber() && !isa<MemCpyInst>(ldep.getInst()))
C = dyn_cast<CallInst>(ldep.getInst());
if (C) {
// Check that nothing touches the dest of the "copy" between
// the call and the store.
AliasAnalysis &AA = getAnalysis<AAResultsWrapperPass>().getAAResults();
MemoryLocation StoreLoc = MemoryLocation::get(SI);
for (BasicBlock::iterator I = --SI->getIterator(), E = C->getIterator();
I != E; --I) {
if (AA.getModRefInfo(&*I, StoreLoc) != MRI_NoModRef) {
C = nullptr;
break;
}
}
}
if (C) {
bool changed = performCallSlotOptzn(
LI, SI->getPointerOperand()->stripPointerCasts(),
LI->getPointerOperand()->stripPointerCasts(),
DL.getTypeStoreSize(SI->getOperand(0)->getType()),
findCommonAlignment(DL, SI, LI), C);
if (changed) {
MD->removeInstruction(SI);
SI->eraseFromParent();
MD->removeInstruction(LI);
LI->eraseFromParent();
++NumMemCpyInstr;
return true;
}
}
}
}
// There are two cases that are interesting for this code to handle: memcpy
// and memset. Right now we only handle memset.
// Ensure that the value being stored is something that can be memset'able a
// byte at a time like "0" or "-1" or any width, as well as things like
// 0xA0A0A0A0 and 0.0.
if (Value *ByteVal = isBytewiseValue(SI->getOperand(0)))
if (Instruction *I = tryMergingIntoMemset(SI, SI->getPointerOperand(),
ByteVal)) {
BBI = I->getIterator(); // Don't invalidate iterator.
return true;
}
return false;
}
/// processStore - When GVN is scanning forward over instructions, we look for
/// some other patterns to fold away. In particular, this looks for stores to
/// neighboring locations of memory. If it sees enough consequtive ones
/// (currently 4) it attempts to merge them together into a memcpy/memset.
bool MemCpyOpt::processStore(StoreInst *SI, BasicBlock::iterator &BBI) {
if (SI->isVolatile()) return false;
TargetData *TD = getAnalysisIfAvailable<TargetData>();
if (!TD) return false;
// Detect cases where we're performing call slot forwarding, but
// happen to be using a load-store pair to implement it, rather than
// a memcpy.
if (LoadInst *LI = dyn_cast<LoadInst>(SI->getOperand(0))) {
if (!LI->isVolatile() && LI->hasOneUse()) {
MemDepResult dep = MD->getDependency(LI);
CallInst *C = 0;
if (dep.isClobber() && !isa<MemCpyInst>(dep.getInst()))
C = dyn_cast<CallInst>(dep.getInst());
if (C) {
bool changed = performCallSlotOptzn(LI,
SI->getPointerOperand()->stripPointerCasts(),
LI->getPointerOperand()->stripPointerCasts(),
TD->getTypeStoreSize(SI->getOperand(0)->getType()), C);
if (changed) {
MD->removeInstruction(SI);
SI->eraseFromParent();
LI->eraseFromParent();
++NumMemCpyInstr;
return true;
}
}
}
}
LLVMContext &Context = SI->getContext();
// There are two cases that are interesting for this code to handle: memcpy
// and memset. Right now we only handle memset.
// Ensure that the value being stored is something that can be memset'able a
// byte at a time like "0" or "-1" or any width, as well as things like
// 0xA0A0A0A0 and 0.0.
Value *ByteVal = isBytewiseValue(SI->getOperand(0));
if (!ByteVal)
return false;
AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
Module *M = SI->getParent()->getParent()->getParent();
// Okay, so we now have a single store that can be splatable. Scan to find
// all subsequent stores of the same value to offset from the same pointer.
// Join these together into ranges, so we can decide whether contiguous blocks
// are stored.
MemsetRanges Ranges(*TD);
Value *StartPtr = SI->getPointerOperand();
BasicBlock::iterator BI = SI;
for (++BI; !isa<TerminatorInst>(BI); ++BI) {
if (isa<CallInst>(BI) || isa<InvokeInst>(BI)) {
// If the call is readnone, ignore it, otherwise bail out. We don't even
// allow readonly here because we don't want something like:
// A[1] = 2; strlen(A); A[2] = 2; -> memcpy(A, ...); strlen(A).
if (AA.getModRefBehavior(CallSite(BI)) ==
AliasAnalysis::DoesNotAccessMemory)
continue;
// TODO: If this is a memset, try to join it in.
break;
} else if (isa<VAArgInst>(BI) || isa<LoadInst>(BI))
break;
// If this is a non-store instruction it is fine, ignore it.
StoreInst *NextStore = dyn_cast<StoreInst>(BI);
if (NextStore == 0) continue;
// If this is a store, see if we can merge it in.
if (NextStore->isVolatile()) break;
// Check to see if this stored value is of the same byte-splattable value.
if (ByteVal != isBytewiseValue(NextStore->getOperand(0)))
break;
// Check to see if this store is to a constant offset from the start ptr.
int64_t Offset;
if (!IsPointerOffset(StartPtr, NextStore->getPointerOperand(), Offset, *TD))
break;
Ranges.addStore(Offset, NextStore);
}
// If we have no ranges, then we just had a single store with nothing that
// could be merged in. This is a very common case of course.
if (Ranges.empty())
return false;
// If we had at least one store that could be merged in, add the starting
// store as well. We try to avoid this unless there is at least something
// interesting as a small compile-time optimization.
Ranges.addStore(0, SI);
// Now that we have full information about ranges, loop over the ranges and
// emit memset's for anything big enough to be worthwhile.
bool MadeChange = false;
for (MemsetRanges::const_iterator I = Ranges.begin(), E = Ranges.end();
I != E; ++I) {
const MemsetRange &Range = *I;
if (Range.TheStores.size() == 1) continue;
// If it is profitable to lower this range to memset, do so now.
if (!Range.isProfitableToUseMemset(*TD))
continue;
// Otherwise, we do want to transform this! Create a new memset. We put
// the memset right before the first instruction that isn't part of this
// memset block. This ensure that the memset is dominated by any addressing
// instruction needed by the start of the block.
BasicBlock::iterator InsertPt = BI;
// Get the starting pointer of the block.
StartPtr = Range.StartPtr;
// Determine alignment
unsigned Alignment = Range.Alignment;
if (Alignment == 0) {
const Type *EltType =
cast<PointerType>(StartPtr->getType())->getElementType();
Alignment = TD->getABITypeAlignment(EltType);
}
// Cast the start ptr to be i8* as memset requires.
const PointerType* StartPTy = cast<PointerType>(StartPtr->getType());
const PointerType *i8Ptr = Type::getInt8PtrTy(Context,
StartPTy->getAddressSpace());
if (StartPTy!= i8Ptr)
StartPtr = new BitCastInst(StartPtr, i8Ptr, StartPtr->getName(),
InsertPt);
Value *Ops[] = {
StartPtr, ByteVal, // Start, value
// size
ConstantInt::get(Type::getInt64Ty(Context), Range.End-Range.Start),
// align
ConstantInt::get(Type::getInt32Ty(Context), Alignment),
// volatile
ConstantInt::getFalse(Context),
};
const Type *Tys[] = { Ops[0]->getType(), Ops[2]->getType() };
Function *MemSetF = Intrinsic::getDeclaration(M, Intrinsic::memset, Tys, 2);
Value *C = CallInst::Create(MemSetF, Ops, Ops+5, "", InsertPt);
DEBUG(dbgs() << "Replace stores:\n";
for (unsigned i = 0, e = Range.TheStores.size(); i != e; ++i)
dbgs() << *Range.TheStores[i] << '\n';
dbgs() << "With: " << *C << '\n'); C=C;
// Don't invalidate the iterator
BBI = BI;
// Zap all the stores.
for (SmallVector<StoreInst*, 16>::const_iterator
SI = Range.TheStores.begin(),
SE = Range.TheStores.end(); SI != SE; ++SI)
(*SI)->eraseFromParent();
++NumMemSetInfer;
MadeChange = true;
}
return MadeChange;
}
bool MemCpyOpt::processStore(StoreInst *SI, BasicBlock::iterator &BBI) {
if (SI->isVolatile()) return false;
if (TD == 0) return false;
// Detect cases where we're performing call slot forwarding, but
// happen to be using a load-store pair to implement it, rather than
// a memcpy.
if (LoadInst *LI = dyn_cast<LoadInst>(SI->getOperand(0))) {
if (!LI->isVolatile() && LI->hasOneUse()) {
MemDepResult ldep = MD->getDependency(LI);
CallInst *C = 0;
if (ldep.isClobber() && !isa<MemCpyInst>(ldep.getInst()))
C = dyn_cast<CallInst>(ldep.getInst());
if (C) {
// Check that nothing touches the dest of the "copy" between
// the call and the store.
MemDepResult sdep = MD->getDependency(SI);
if (!sdep.isNonLocal()) {
bool FoundCall = false;
for (BasicBlock::iterator I = SI, E = sdep.getInst(); I != E; --I) {
if (&*I == C) {
FoundCall = true;
break;
}
}
if (!FoundCall)
C = 0;
}
}
if (C) {
bool changed = performCallSlotOptzn(LI,
SI->getPointerOperand()->stripPointerCasts(),
LI->getPointerOperand()->stripPointerCasts(),
TD->getTypeStoreSize(SI->getOperand(0)->getType()), C);
if (changed) {
MD->removeInstruction(SI);
SI->eraseFromParent();
MD->removeInstruction(LI);
LI->eraseFromParent();
++NumMemCpyInstr;
return true;
}
}
}
}
// There are two cases that are interesting for this code to handle: memcpy
// and memset. Right now we only handle memset.
// Ensure that the value being stored is something that can be memset'able a
// byte at a time like "0" or "-1" or any width, as well as things like
// 0xA0A0A0A0 and 0.0.
if (Value *ByteVal = isBytewiseValue(SI->getOperand(0)))
if (Instruction *I = tryMergingIntoMemset(SI, SI->getPointerOperand(),
ByteVal)) {
BBI = I; // Don't invalidate iterator.
return true;
}
return false;
}