RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-25 20:23:11 +01:00
Code Issues Projects Releases Wiki Activity

Try again at r138809 (make DSE more aggressive in removing dead stores at the end of a function), now with less deleting stores before memcpy's.

Browse Source 
llvm-svn: 139150
This commit is contained in:
Owen Anderson 2011-09-06 18:14:09 +00:00
parent 7994269719
commit ca0326a423
3 changed files with 192 additions and 79 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

159
lib/Transforms/Scalar/DeadStoreElimination.cpp
View File

@ -52,18 +52,18 @@ namespace {
AA = &getAnalysis<AliasAnalysis>(); AA = &getAnalysis<AliasAnalysis>();
MD = &getAnalysis<MemoryDependenceAnalysis>(); MD = &getAnalysis<MemoryDependenceAnalysis>();
DominatorTree &DT = getAnalysis<DominatorTree>(); DominatorTree &DT = getAnalysis<DominatorTree>();
bool Changed = false; bool Changed = false;
for (Function::iterator I = F.begin(), E = F.end(); I != E; ++I) for (Function::iterator I = F.begin(), E = F.end(); I != E; ++I)
// Only check non-dead blocks. Dead blocks may have strange pointer // Only check non-dead blocks. Dead blocks may have strange pointer
// cycles that will confuse alias analysis. // cycles that will confuse alias analysis.
if (DT.isReachableFromEntry(I)) if (DT.isReachableFromEntry(I))
Changed |= runOnBasicBlock(*I); Changed |= runOnBasicBlock(*I);
AA = 0; MD = 0; AA = 0; MD = 0;
return Changed; return Changed;
} }
bool runOnBasicBlock(BasicBlock &BB); bool runOnBasicBlock(BasicBlock &BB);
bool HandleFree(CallInst *F); bool HandleFree(CallInst *F);
bool handleEndBlock(BasicBlock &BB); bool handleEndBlock(BasicBlock &BB);
@ -105,34 +105,34 @@ static void DeleteDeadInstruction(Instruction *I,
MemoryDependenceAnalysis &MD, MemoryDependenceAnalysis &MD,
SmallPtrSet<Value*, 16> *ValueSet = 0) { SmallPtrSet<Value*, 16> *ValueSet = 0) {
SmallVector<Instruction*, 32> NowDeadInsts; SmallVector<Instruction*, 32> NowDeadInsts;
NowDeadInsts.push_back(I); NowDeadInsts.push_back(I);
--NumFastOther; --NumFastOther;
// Before we touch this instruction, remove it from memdep! // Before we touch this instruction, remove it from memdep!
do { do {
Instruction *DeadInst = NowDeadInsts.pop_back_val(); Instruction *DeadInst = NowDeadInsts.pop_back_val();
++NumFastOther; ++NumFastOther;
// This instruction is dead, zap it, in stages. Start by removing it from // This instruction is dead, zap it, in stages. Start by removing it from
// MemDep, which needs to know the operands and needs it to be in the // MemDep, which needs to know the operands and needs it to be in the
// function. // function.
MD.removeInstruction(DeadInst); MD.removeInstruction(DeadInst);
for (unsigned op = 0, e = DeadInst->getNumOperands(); op != e; ++op) { for (unsigned op = 0, e = DeadInst->getNumOperands(); op != e; ++op) {
Value *Op = DeadInst->getOperand(op); Value *Op = DeadInst->getOperand(op);
DeadInst->setOperand(op, 0); DeadInst->setOperand(op, 0);
// If this operand just became dead, add it to the NowDeadInsts list. // If this operand just became dead, add it to the NowDeadInsts list.
if (!Op->use_empty()) continue; if (!Op->use_empty()) continue;
if (Instruction *OpI = dyn_cast<Instruction>(Op)) if (Instruction *OpI = dyn_cast<Instruction>(Op))
if (isInstructionTriviallyDead(OpI)) if (isInstructionTriviallyDead(OpI))
NowDeadInsts.push_back(OpI); NowDeadInsts.push_back(OpI);
} }
DeadInst->eraseFromParent(); DeadInst->eraseFromParent();
if (ValueSet) ValueSet->erase(DeadInst); if (ValueSet) ValueSet->erase(DeadInst);
} while (!NowDeadInsts.empty()); } while (!NowDeadInsts.empty());
} }
@ -163,7 +163,7 @@ static AliasAnalysis::Location
getLocForWrite(Instruction *Inst, AliasAnalysis &AA) { getLocForWrite(Instruction *Inst, AliasAnalysis &AA) {
if (StoreInst *SI = dyn_cast<StoreInst>(Inst)) if (StoreInst *SI = dyn_cast<StoreInst>(Inst))
return AA.getLocation(SI); return AA.getLocation(SI);
if (MemIntrinsic *MI = dyn_cast<MemIntrinsic>(Inst)) { if (MemIntrinsic *MI = dyn_cast<MemIntrinsic>(Inst)) {
// memcpy/memmove/memset. // memcpy/memmove/memset.
AliasAnalysis::Location Loc = AA.getLocationForDest(MI); AliasAnalysis::Location Loc = AA.getLocationForDest(MI);
@ -174,10 +174,10 @@ getLocForWrite(Instruction *Inst, AliasAnalysis &AA) {
return AliasAnalysis::Location(); return AliasAnalysis::Location();
return Loc; return Loc;
} }
IntrinsicInst *II = dyn_cast<IntrinsicInst>(Inst); IntrinsicInst *II = dyn_cast<IntrinsicInst>(Inst);
if (II == 0) return AliasAnalysis::Location(); if (II == 0) return AliasAnalysis::Location();
switch (II->getIntrinsicID()) { switch (II->getIntrinsicID()) {
default: return AliasAnalysis::Location(); // Unhandled intrinsic. default: return AliasAnalysis::Location(); // Unhandled intrinsic.
case Intrinsic::init_trampoline: case Intrinsic::init_trampoline:
@ -185,7 +185,7 @@ getLocForWrite(Instruction *Inst, AliasAnalysis &AA) {
// that we should use the size of the pointee type. This isn't valid for // that we should use the size of the pointee type. This isn't valid for
// init.trampoline, which writes more than an i8. // init.trampoline, which writes more than an i8.
if (AA.getTargetData() == 0) return AliasAnalysis::Location(); if (AA.getTargetData() == 0) return AliasAnalysis::Location();
// FIXME: We don't know the size of the trampoline, so we can't really // FIXME: We don't know the size of the trampoline, so we can't really
// handle it here. // handle it here.
return AliasAnalysis::Location(II->getArgOperand(0)); return AliasAnalysis::Location(II->getArgOperand(0));
@ -198,10 +198,10 @@ getLocForWrite(Instruction *Inst, AliasAnalysis &AA) {
/// getLocForRead - Return the location read by the specified "hasMemoryWrite" /// getLocForRead - Return the location read by the specified "hasMemoryWrite"
/// instruction if any. /// instruction if any.
static AliasAnalysis::Location static AliasAnalysis::Location
getLocForRead(Instruction *Inst, AliasAnalysis &AA) { getLocForRead(Instruction *Inst, AliasAnalysis &AA) {
assert(hasMemoryWrite(Inst) && "Unknown instruction case"); assert(hasMemoryWrite(Inst) && "Unknown instruction case");
// The only instructions that both read and write are the mem transfer // The only instructions that both read and write are the mem transfer
// instructions (memcpy/memmove). // instructions (memcpy/memmove).
if (MemTransferInst *MTI = dyn_cast<MemTransferInst>(Inst)) if (MemTransferInst *MTI = dyn_cast<MemTransferInst>(Inst))
@ -216,7 +216,7 @@ static bool isRemovable(Instruction *I) {
// Don't remove volatile/atomic stores. // Don't remove volatile/atomic stores.
if (StoreInst *SI = dyn_cast<StoreInst>(I)) if (StoreInst *SI = dyn_cast<StoreInst>(I))
return SI->isUnordered(); return SI->isUnordered();
IntrinsicInst *II = cast<IntrinsicInst>(I); IntrinsicInst *II = cast<IntrinsicInst>(I);
switch (II->getIntrinsicID()) { switch (II->getIntrinsicID()) {
default: assert(0 && "doesn't pass 'hasMemoryWrite' predicate"); default: assert(0 && "doesn't pass 'hasMemoryWrite' predicate");
@ -227,7 +227,7 @@ static bool isRemovable(Instruction *I) {
case Intrinsic::init_trampoline: case Intrinsic::init_trampoline:
// Always safe to remove init_trampoline. // Always safe to remove init_trampoline.
return true; return true;
case Intrinsic::memset: case Intrinsic::memset:
case Intrinsic::memmove: case Intrinsic::memmove:
case Intrinsic::memcpy: case Intrinsic::memcpy:
@ -255,14 +255,14 @@ static uint64_t getPointerSize(Value *V, AliasAnalysis &AA) {
const TargetData *TD = AA.getTargetData(); const TargetData *TD = AA.getTargetData();
if (TD == 0) if (TD == 0)
return AliasAnalysis::UnknownSize; return AliasAnalysis::UnknownSize;
if (AllocaInst *A = dyn_cast<AllocaInst>(V)) { if (AllocaInst *A = dyn_cast<AllocaInst>(V)) {
// Get size information for the alloca // Get size information for the alloca
if (ConstantInt *C = dyn_cast<ConstantInt>(A->getArraySize())) if (ConstantInt *C = dyn_cast<ConstantInt>(A->getArraySize()))
return C->getZExtValue() * TD->getTypeAllocSize(A->getAllocatedType()); return C->getZExtValue() * TD->getTypeAllocSize(A->getAllocatedType());
return AliasAnalysis::UnknownSize; return AliasAnalysis::UnknownSize;
} }
assert(isa<Argument>(V) && "Expected AllocaInst or Argument!"); assert(isa<Argument>(V) && "Expected AllocaInst or Argument!");
PointerType *PT = cast<PointerType>(V->getType()); PointerType *PT = cast<PointerType>(V->getType());
return TD->getTypeAllocSize(PT->getElementType()); return TD->getTypeAllocSize(PT->getElementType());
@ -287,7 +287,7 @@ static bool isCompleteOverwrite(const AliasAnalysis::Location &Later,
AliasAnalysis &AA) { AliasAnalysis &AA) {
const Value *P1 = Earlier.Ptr->stripPointerCasts(); const Value *P1 = Earlier.Ptr->stripPointerCasts();
const Value *P2 = Later.Ptr->stripPointerCasts(); const Value *P2 = Later.Ptr->stripPointerCasts();
// If the start pointers are the same, we just have to compare sizes to see if // If the start pointers are the same, we just have to compare sizes to see if
// the later store was larger than the earlier store. // the later store was larger than the earlier store.
if (P1 == P2) { if (P1 == P2) {
@ -302,33 +302,33 @@ static bool isCompleteOverwrite(const AliasAnalysis::Location &Later,
return Later.Ptr->getType() == Earlier.Ptr->getType(); return Later.Ptr->getType() == Earlier.Ptr->getType();
return false; return false;
} }
// Make sure that the Later size is >= the Earlier size. // Make sure that the Later size is >= the Earlier size.
if (Later.Size < Earlier.Size) if (Later.Size < Earlier.Size)
return false; return false;
return true; return true;
} }
// Otherwise, we have to have size information, and the later store has to be // Otherwise, we have to have size information, and the later store has to be
// larger than the earlier one. // larger than the earlier one.
if (Later.Size == AliasAnalysis::UnknownSize || if (Later.Size == AliasAnalysis::UnknownSize ||
Earlier.Size == AliasAnalysis::UnknownSize || Earlier.Size == AliasAnalysis::UnknownSize ||
Later.Size <= Earlier.Size || AA.getTargetData() == 0) Later.Size <= Earlier.Size || AA.getTargetData() == 0)
return false; return false;
// Check to see if the later store is to the entire object (either a global, // Check to see if the later store is to the entire object (either a global,
// an alloca, or a byval argument). If so, then it clearly overwrites any // an alloca, or a byval argument). If so, then it clearly overwrites any
// other store to the same object. // other store to the same object.
const TargetData &TD = *AA.getTargetData(); const TargetData &TD = *AA.getTargetData();
const Value *UO1 = GetUnderlyingObject(P1, &TD), const Value *UO1 = GetUnderlyingObject(P1, &TD),
*UO2 = GetUnderlyingObject(P2, &TD); *UO2 = GetUnderlyingObject(P2, &TD);
// If we can't resolve the same pointers to the same object, then we can't // If we can't resolve the same pointers to the same object, then we can't
// analyze them at all. // analyze them at all.
if (UO1 != UO2) if (UO1 != UO2)
return false; return false;
// If the "Later" store is to a recognizable object, get its size. // If the "Later" store is to a recognizable object, get its size.
if (isObjectPointerWithTrustworthySize(UO2)) { if (isObjectPointerWithTrustworthySize(UO2)) {
uint64_t ObjectSize = uint64_t ObjectSize =
@ -336,26 +336,26 @@ static bool isCompleteOverwrite(const AliasAnalysis::Location &Later,
if (ObjectSize == Later.Size) if (ObjectSize == Later.Size)
return true; return true;
} }
// Okay, we have stores to two completely different pointers. Try to // Okay, we have stores to two completely different pointers. Try to
// decompose the pointer into a "base + constant_offset" form. If the base // decompose the pointer into a "base + constant_offset" form. If the base
// pointers are equal, then we can reason about the two stores. // pointers are equal, then we can reason about the two stores.
int64_t EarlierOff = 0, LaterOff = 0; int64_t EarlierOff = 0, LaterOff = 0;
const Value *BP1 = GetPointerBaseWithConstantOffset(P1, EarlierOff, TD); const Value *BP1 = GetPointerBaseWithConstantOffset(P1, EarlierOff, TD);
const Value *BP2 = GetPointerBaseWithConstantOffset(P2, LaterOff, TD); const Value *BP2 = GetPointerBaseWithConstantOffset(P2, LaterOff, TD);
// If the base pointers still differ, we have two completely different stores. // If the base pointers still differ, we have two completely different stores.
if (BP1 != BP2) if (BP1 != BP2)
return false; return false;
// The later store completely overlaps the earlier store if: // The later store completely overlaps the earlier store if:
// //
// 1. Both start at the same offset and the later one's size is greater than // 1. Both start at the same offset and the later one's size is greater than
// or equal to the earlier one's, or // or equal to the earlier one's, or
// //
// |--earlier--| // |--earlier--|
// |-- later --| // |-- later --|
// //
// 2. The earlier store has an offset greater than the later offset, but which // 2. The earlier store has an offset greater than the later offset, but which
// still lies completely within the later store. // still lies completely within the later store.
// //
@ -373,7 +373,7 @@ static bool isCompleteOverwrite(const AliasAnalysis::Location &Later,
/// isPossibleSelfRead - If 'Inst' might be a self read (i.e. a noop copy of a /// isPossibleSelfRead - If 'Inst' might be a self read (i.e. a noop copy of a
/// memory region into an identical pointer) then it doesn't actually make its /// memory region into an identical pointer) then it doesn't actually make its
/// input dead in the traditional sense. Consider this case: /// input dead in the traditional sense. Consider this case:
/// ///
/// memcpy(A <- B) /// memcpy(A <- B)
/// memcpy(A <- A) /// memcpy(A <- A)
@ -391,10 +391,10 @@ static bool isPossibleSelfRead(Instruction *Inst,
// location read. // location read.
AliasAnalysis::Location InstReadLoc = getLocForRead(Inst, AA); AliasAnalysis::Location InstReadLoc = getLocForRead(Inst, AA);
if (InstReadLoc.Ptr == 0) return false; // Not a reading instruction. if (InstReadLoc.Ptr == 0) return false; // Not a reading instruction.
// If the read and written loc obviously don't alias, it isn't a read. // If the read and written loc obviously don't alias, it isn't a read.
if (AA.isNoAlias(InstReadLoc, InstStoreLoc)) return false; if (AA.isNoAlias(InstReadLoc, InstStoreLoc)) return false;
// Okay, 'Inst' may copy over itself. However, we can still remove a the // Okay, 'Inst' may copy over itself. However, we can still remove a the
// DepWrite instruction if we can prove that it reads from the same location // DepWrite instruction if we can prove that it reads from the same location
// as Inst. This handles useful cases like: // as Inst. This handles useful cases like:
@ -404,10 +404,10 @@ static bool isPossibleSelfRead(Instruction *Inst,
// aliases, so removing the first memcpy is safe (assuming it writes <= # // aliases, so removing the first memcpy is safe (assuming it writes <= #
// bytes as the second one. // bytes as the second one.
AliasAnalysis::Location DepReadLoc = getLocForRead(DepWrite, AA); AliasAnalysis::Location DepReadLoc = getLocForRead(DepWrite, AA);
if (DepReadLoc.Ptr && AA.isMustAlias(InstReadLoc.Ptr, DepReadLoc.Ptr)) if (DepReadLoc.Ptr && AA.isMustAlias(InstReadLoc.Ptr, DepReadLoc.Ptr))
return false; return false;
// If DepWrite doesn't read memory or if we can't prove it is a must alias, // If DepWrite doesn't read memory or if we can't prove it is a must alias,
// then it can't be considered dead. // then it can't be considered dead.
return true; return true;
@ -420,28 +420,28 @@ static bool isPossibleSelfRead(Instruction *Inst,
bool DSE::runOnBasicBlock(BasicBlock &BB) { bool DSE::runOnBasicBlock(BasicBlock &BB) {
bool MadeChange = false; bool MadeChange = false;
// Do a top-down walk on the BB. // Do a top-down walk on the BB.
for (BasicBlock::iterator BBI = BB.begin(), BBE = BB.end(); BBI != BBE; ) { for (BasicBlock::iterator BBI = BB.begin(), BBE = BB.end(); BBI != BBE; ) {
Instruction *Inst = BBI++; Instruction *Inst = BBI++;
// Handle 'free' calls specially. // Handle 'free' calls specially.
if (CallInst *F = isFreeCall(Inst)) { if (CallInst *F = isFreeCall(Inst)) {
MadeChange |= HandleFree(F); MadeChange |= HandleFree(F);
continue; continue;
} }
// If we find something that writes memory, get its memory dependence. // If we find something that writes memory, get its memory dependence.
if (!hasMemoryWrite(Inst)) if (!hasMemoryWrite(Inst))
continue; continue;
MemDepResult InstDep = MD->getDependency(Inst); MemDepResult InstDep = MD->getDependency(Inst);
// Ignore any store where we can't find a local dependence. // Ignore any store where we can't find a local dependence.
// FIXME: cross-block DSE would be fun. :) // FIXME: cross-block DSE would be fun. :)
if (InstDep.isNonLocal() || InstDep.isUnknown()) if (InstDep.isNonLocal() || InstDep.isUnknown())
continue; continue;
// If we're storing the same value back to a pointer that we just // If we're storing the same value back to a pointer that we just
// loaded from, then the store can be removed. // loaded from, then the store can be removed.
if (StoreInst *SI = dyn_cast<StoreInst>(Inst)) { if (StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
@ -450,13 +450,13 @@ bool DSE::runOnBasicBlock(BasicBlock &BB) {
SI->getOperand(0) == DepLoad && isRemovable(SI)) { SI->getOperand(0) == DepLoad && isRemovable(SI)) {
DEBUG(dbgs() << "DSE: Remove Store Of Load from same pointer:\n " DEBUG(dbgs() << "DSE: Remove Store Of Load from same pointer:\n "
<< "LOAD: " << *DepLoad << "\n STORE: " << *SI << '\n'); << "LOAD: " << *DepLoad << "\n STORE: " << *SI << '\n');
// DeleteDeadInstruction can delete the current instruction. Save BBI // DeleteDeadInstruction can delete the current instruction. Save BBI
// in case we need it. // in case we need it.
WeakVH NextInst(BBI); WeakVH NextInst(BBI);
DeleteDeadInstruction(SI, *MD); DeleteDeadInstruction(SI, *MD);
if (NextInst == 0) // Next instruction deleted. if (NextInst == 0) // Next instruction deleted.
BBI = BB.begin(); BBI = BB.begin();
else if (BBI != BB.begin()) // Revisit this instruction if possible. else if (BBI != BB.begin()) // Revisit this instruction if possible.
@ -467,14 +467,14 @@ bool DSE::runOnBasicBlock(BasicBlock &BB) {
} }
} }
} }
// Figure out what location is being stored to. // Figure out what location is being stored to.
AliasAnalysis::Location Loc = getLocForWrite(Inst, *AA); AliasAnalysis::Location Loc = getLocForWrite(Inst, *AA);
// If we didn't get a useful location, fail. // If we didn't get a useful location, fail.
if (Loc.Ptr == 0) if (Loc.Ptr == 0)
continue; continue;
while (!InstDep.isNonLocal() && !InstDep.isUnknown()) { while (!InstDep.isNonLocal() && !InstDep.isUnknown()) {
// Get the memory clobbered by the instruction we depend on. MemDep will // Get the memory clobbered by the instruction we depend on. MemDep will
// skip any instructions that 'Loc' clearly doesn't interact with. If we // skip any instructions that 'Loc' clearly doesn't interact with. If we
@ -496,12 +496,12 @@ bool DSE::runOnBasicBlock(BasicBlock &BB) {
!isPossibleSelfRead(Inst, Loc, DepWrite, *AA)) { !isPossibleSelfRead(Inst, Loc, DepWrite, *AA)) {
DEBUG(dbgs() << "DSE: Remove Dead Store:\n DEAD: " DEBUG(dbgs() << "DSE: Remove Dead Store:\n DEAD: "
<< *DepWrite << "\n KILLER: " << *Inst << '\n'); << *DepWrite << "\n KILLER: " << *Inst << '\n');
// Delete the store and now-dead instructions that feed it. // Delete the store and now-dead instructions that feed it.
DeleteDeadInstruction(DepWrite, *MD); DeleteDeadInstruction(DepWrite, *MD);
++NumFastStores; ++NumFastStores;
MadeChange = true; MadeChange = true;
// DeleteDeadInstruction can delete the current instruction in loop // DeleteDeadInstruction can delete the current instruction in loop
// cases, reset BBI. // cases, reset BBI.
BBI = Inst; BBI = Inst;
@ -509,7 +509,7 @@ bool DSE::runOnBasicBlock(BasicBlock &BB) {
--BBI; --BBI;
break; break;
} }
// If this is a may-aliased store that is clobbering the store value, we // 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 // can keep searching past it for another must-aliased pointer that stores
// to the same location. For example, in: // to the same location. For example, in:
@ -519,20 +519,20 @@ bool DSE::runOnBasicBlock(BasicBlock &BB) {
// we can remove the first store to P even though we don't know if P and Q // we can remove the first store to P even though we don't know if P and Q
// alias. // alias.
if (DepWrite == &BB.front()) break; if (DepWrite == &BB.front()) break;
// Can't look past this instruction if it might read 'Loc'. // Can't look past this instruction if it might read 'Loc'.
if (AA->getModRefInfo(DepWrite, Loc) & AliasAnalysis::Ref) if (AA->getModRefInfo(DepWrite, Loc) & AliasAnalysis::Ref)
break; break;
InstDep = MD->getPointerDependencyFrom(Loc, false, DepWrite, &BB); InstDep = MD->getPointerDependencyFrom(Loc, false, DepWrite, &BB);
} }
} }
// If this block ends in a return, unwind, or unreachable, all allocas are // 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. // dead at its end, which means stores to them are also dead.
if (BB.getTerminator()->getNumSuccessors() == 0) if (BB.getTerminator()->getNumSuccessors() == 0)
MadeChange |= handleEndBlock(BB); MadeChange |= handleEndBlock(BB);
return MadeChange; return MadeChange;
} }
@ -547,14 +547,14 @@ bool DSE::HandleFree(CallInst *F) {
Instruction *Dependency = Dep.getInst(); Instruction *Dependency = Dep.getInst();
if (!hasMemoryWrite(Dependency) || !isRemovable(Dependency)) if (!hasMemoryWrite(Dependency) || !isRemovable(Dependency))
return MadeChange; return MadeChange;
Value *DepPointer = Value *DepPointer =
GetUnderlyingObject(getStoredPointerOperand(Dependency)); GetUnderlyingObject(getStoredPointerOperand(Dependency));
// Check for aliasing. // Check for aliasing.
if (!AA->isMustAlias(F->getArgOperand(0), DepPointer)) if (!AA->isMustAlias(F->getArgOperand(0), DepPointer))
return MadeChange; return MadeChange;
// DCE instructions only used to calculate that store // DCE instructions only used to calculate that store
DeleteDeadInstruction(Dependency, *MD); DeleteDeadInstruction(Dependency, *MD);
++NumFastStores; ++NumFastStores;
@ -567,7 +567,7 @@ bool DSE::HandleFree(CallInst *F) {
// free(s); // free(s);
Dep = MD->getDependency(F); Dep = MD->getDependency(F);
}; };
return MadeChange; return MadeChange;
} }
@ -579,28 +579,28 @@ bool DSE::HandleFree(CallInst *F) {
/// ret void /// ret void
bool DSE::handleEndBlock(BasicBlock &BB) { bool DSE::handleEndBlock(BasicBlock &BB) {
bool MadeChange = false; bool MadeChange = false;
// Keep track of all of the stack objects that are dead at the end of the // Keep track of all of the stack objects that are dead at the end of the
// function. // function.
SmallPtrSet<Value*, 16> DeadStackObjects; SmallPtrSet<Value*, 16> DeadStackObjects;
// Find all of the alloca'd pointers in the entry block. // Find all of the alloca'd pointers in the entry block.
BasicBlock *Entry = BB.getParent()->begin(); BasicBlock *Entry = BB.getParent()->begin();
for (BasicBlock::iterator I = Entry->begin(), E = Entry->end(); I != E; ++I) for (BasicBlock::iterator I = Entry->begin(), E = Entry->end(); I != E; ++I)
if (AllocaInst *AI = dyn_cast<AllocaInst>(I)) if (AllocaInst *AI = dyn_cast<AllocaInst>(I))
DeadStackObjects.insert(AI); DeadStackObjects.insert(AI);
// Treat byval arguments the same, stores to them are dead at the end of the // Treat byval arguments the same, stores to them are dead at the end of the
// function. // function.
for (Function::arg_iterator AI = BB.getParent()->arg_begin(), for (Function::arg_iterator AI = BB.getParent()->arg_begin(),
AE = BB.getParent()->arg_end(); AI != AE; ++AI) AE = BB.getParent()->arg_end(); AI != AE; ++AI)
if (AI->hasByValAttr()) if (AI->hasByValAttr())
DeadStackObjects.insert(AI); DeadStackObjects.insert(AI);
// Scan the basic block backwards // Scan the basic block backwards
for (BasicBlock::iterator BBI = BB.end(); BBI != BB.begin(); ){ for (BasicBlock::iterator BBI = BB.end(); BBI != BB.begin(); ){
--BBI; --BBI;
// If we find a store, check to see if it points into a dead stack value. // If we find a store, check to see if it points into a dead stack value.
if (hasMemoryWrite(BBI) && isRemovable(BBI)) { if (hasMemoryWrite(BBI) && isRemovable(BBI)) {
// See through pointer-to-pointer bitcasts // See through pointer-to-pointer bitcasts
@ -609,10 +609,10 @@ bool DSE::handleEndBlock(BasicBlock &BB) {
// Stores to stack values are valid candidates for removal. // Stores to stack values are valid candidates for removal.
if (DeadStackObjects.count(Pointer)) { if (DeadStackObjects.count(Pointer)) {
Instruction *Dead = BBI++; Instruction *Dead = BBI++;
DEBUG(dbgs() << "DSE: Dead Store at End of Block:\n DEAD: " DEBUG(dbgs() << "DSE: Dead Store at End of Block:\n DEAD: "
<< *Dead << "\n Object: " << *Pointer << '\n'); << *Dead << "\n Object: " << *Pointer << '\n');
// DCE instructions only used to calculate that store. // DCE instructions only used to calculate that store.
DeleteDeadInstruction(Dead, *MD, &DeadStackObjects); DeleteDeadInstruction(Dead, *MD, &DeadStackObjects);
++NumFastStores; ++NumFastStores;
@ -620,7 +620,7 @@ bool DSE::handleEndBlock(BasicBlock &BB) {
continue; continue;
} }
} }
// Remove any dead non-memory-mutating instructions. // Remove any dead non-memory-mutating instructions.
if (isInstructionTriviallyDead(BBI)) { if (isInstructionTriviallyDead(BBI)) {
Instruction *Inst = BBI++; Instruction *Inst = BBI++;
@ -629,45 +629,45 @@ bool DSE::handleEndBlock(BasicBlock &BB) {
MadeChange = true; MadeChange = true;
continue; continue;
} }
if (AllocaInst *A = dyn_cast<AllocaInst>(BBI)) { if (AllocaInst *A = dyn_cast<AllocaInst>(BBI)) {
DeadStackObjects.erase(A); DeadStackObjects.erase(A);
continue; continue;
} }
if (CallSite CS = cast<Value>(BBI)) { if (CallSite CS = cast<Value>(BBI)) {
// If this call does not access memory, it can't be loading any of our // If this call does not access memory, it can't be loading any of our
// pointers. // pointers.
if (AA->doesNotAccessMemory(CS)) if (AA->doesNotAccessMemory(CS))
continue; continue;
// If the call might load from any of our allocas, then any store above // If the call might load from any of our allocas, then any store above
// the call is live. // the call is live.
SmallVector<Value*, 8> LiveAllocas; SmallVector<Value*, 8> LiveAllocas;
for (SmallPtrSet<Value*, 16>::iterator I = DeadStackObjects.begin(), for (SmallPtrSet<Value*, 16>::iterator I = DeadStackObjects.begin(),
E = DeadStackObjects.end(); I != E; ++I) { E = DeadStackObjects.end(); I != E; ++I) {
// See if the call site touches it. // See if the call site touches it.
AliasAnalysis::ModRefResult A = AliasAnalysis::ModRefResult A =
AA->getModRefInfo(CS, *I, getPointerSize(*I, *AA)); AA->getModRefInfo(CS, *I, getPointerSize(*I, *AA));
if (A == AliasAnalysis::ModRef || A == AliasAnalysis::Ref) if (A == AliasAnalysis::ModRef || A == AliasAnalysis::Ref)
LiveAllocas.push_back(*I); LiveAllocas.push_back(*I);
} }
for (SmallVector<Value*, 8>::iterator I = LiveAllocas.begin(), for (SmallVector<Value*, 8>::iterator I = LiveAllocas.begin(),
E = LiveAllocas.end(); I != E; ++I) E = LiveAllocas.end(); I != E; ++I)
DeadStackObjects.erase(*I); DeadStackObjects.erase(*I);
// If all of the allocas were clobbered by the call then we're not going // If all of the allocas were clobbered by the call then we're not going
// to find anything else to process. // to find anything else to process.
if (DeadStackObjects.empty()) if (DeadStackObjects.empty())
return MadeChange; return MadeChange;
continue; continue;
} }
AliasAnalysis::Location LoadedLoc; AliasAnalysis::Location LoadedLoc;
// If we encounter a use of the pointer, it is no longer considered dead // If we encounter a use of the pointer, it is no longer considered dead
if (LoadInst *L = dyn_cast<LoadInst>(BBI)) { if (LoadInst *L = dyn_cast<LoadInst>(BBI)) {
if (!L->isUnordered()) // Be conservative with atomic/volatile load if (!L->isUnordered()) // Be conservative with atomic/volatile load
@ -677,8 +677,9 @@ bool DSE::handleEndBlock(BasicBlock &BB) {
LoadedLoc = AA->getLocation(V); LoadedLoc = AA->getLocation(V);
} else if (MemTransferInst *MTI = dyn_cast<MemTransferInst>(BBI)) { } else if (MemTransferInst *MTI = dyn_cast<MemTransferInst>(BBI)) {
LoadedLoc = AA->getLocationForSource(MTI); LoadedLoc = AA->getLocationForSource(MTI);
} else if (!BBI->mayReadOrWriteMemory()) { } else if (!BBI->mayReadFromMemory()) {
// Instruction doesn't touch memory. // Instruction doesn't read memory. Note that stores that weren't removed
// above will hit this case.
continue; continue;
} else { } else {
// Unknown inst; assume it clobbers everything. // Unknown inst; assume it clobbers everything.
@ -694,7 +695,7 @@ bool DSE::handleEndBlock(BasicBlock &BB) {
if (DeadStackObjects.empty()) if (DeadStackObjects.empty())
break; break;
} }
return MadeChange; return MadeChange;
} }
@ -708,14 +709,14 @@ void DSE::RemoveAccessedObjects(const AliasAnalysis::Location &LoadedLoc,
// A constant can't be in the dead pointer set. // A constant can't be in the dead pointer set.
if (isa<Constant>(UnderlyingPointer)) if (isa<Constant>(UnderlyingPointer))
return; return;
// If the kill pointer can be easily reduced to an alloca, don't bother doing // If the kill pointer can be easily reduced to an alloca, don't bother doing
// extraneous AA queries. // extraneous AA queries.
if (isa<AllocaInst>(UnderlyingPointer) || isa<Argument>(UnderlyingPointer)) { if (isa<AllocaInst>(UnderlyingPointer) || isa<Argument>(UnderlyingPointer)) {
DeadStackObjects.erase(const_cast<Value*>(UnderlyingPointer)); DeadStackObjects.erase(const_cast<Value*>(UnderlyingPointer));
return; return;
} }
SmallVector<Value*, 16> NowLive; SmallVector<Value*, 16> NowLive;
for (SmallPtrSet<Value*, 16>::iterator I = DeadStackObjects.begin(), for (SmallPtrSet<Value*, 16>::iterator I = DeadStackObjects.begin(),
E = DeadStackObjects.end(); I != E; ++I) { E = DeadStackObjects.end(); I != E; ++I) {

27
test/Transforms/DeadStoreElimination/2011-09-06-EndOfFunction.ll Normal file
View File

@ -0,0 +1,27 @@
; RUN: opt -dse -S < %s | FileCheck %s
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
target triple = "x86_64-apple-darwin"
%"class.std::auto_ptr" = type { i32* }
; CHECK: @_Z3foov
define void @_Z3foov(%"class.std::auto_ptr"* noalias nocapture sret %agg.result) uwtable ssp {
_ZNSt8auto_ptrIiED1Ev.exit:
%temp.lvalue = alloca %"class.std::auto_ptr", align 8
call void @_Z3barv(%"class.std::auto_ptr"* sret %temp.lvalue)
%_M_ptr.i.i = getelementptr inbounds %"class.std::auto_ptr"* %temp.lvalue, i64 0, i32 0
%tmp.i.i = load i32** %_M_ptr.i.i, align 8, !tbaa !0
; CHECK-NOT: store i32* null
store i32* null, i32** %_M_ptr.i.i, align 8, !tbaa !0
%_M_ptr.i.i4 = getelementptr inbounds %"class.std::auto_ptr"* %agg.result, i64 0, i32 0
store i32* %tmp.i.i, i32** %_M_ptr.i.i4, align 8, !tbaa !0
; CHECK: ret void
ret void
}
declare void @_Z3barv(%"class.std::auto_ptr"* sret)
!0 = metadata !{metadata !"any pointer", metadata !1}
!1 = metadata !{metadata !"omnipotent char", metadata !2}
!2 = metadata !{metadata !"Simple C/C++ TBAA", null}

85
test/Transforms/DeadStoreElimination/2011-09-06-MemCpy.ll Normal file
View File

@ -0,0 +1,85 @@
; RUN: opt -dse -S < %s | FileCheck %s
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-f128:128:128-n8:16:32:64"
target triple = "x86_64-unknown-linux-gnu"
%struct.pair.162 = type { %struct.BasicBlock*, i32, [4 x i8] }
%struct.BasicBlock = type { %struct.Value, %struct.ilist_node.24, %struct.iplist.22, %struct.Function* }
%struct.Value = type { i32 (...)**, i8, i8, i16, %struct.Type*, %struct.Use*, %struct.StringMapEntry* }
%struct.Type = type { %struct.LLVMContext*, i8, [3 x i8], i32, {}* }
%struct.LLVMContext = type { %struct.LLVMContextImpl* }
%struct.LLVMContextImpl = type opaque
%struct.Use = type { %struct.Value*, %struct.Use*, %struct.PointerIntPair }
%struct.PointerIntPair = type { i64 }
%struct.StringMapEntry = type opaque
%struct.ilist_node.24 = type { %struct.ilist_half_node.23, %struct.BasicBlock* }
%struct.ilist_half_node.23 = type { %struct.BasicBlock* }
%struct.iplist.22 = type { %struct.ilist_traits.21, %struct.Instruction* }
%struct.ilist_traits.21 = type { %struct.ilist_half_node.25 }
%struct.ilist_half_node.25 = type { %struct.Instruction* }
%struct.Instruction = type { [52 x i8], %struct.ilist_node.26, %struct.BasicBlock*, %struct.DebugLoc }
%struct.ilist_node.26 = type { %struct.ilist_half_node.25, %struct.Instruction* }
%struct.DebugLoc = type { i32, i32 }
%struct.Function = type { %struct.GlobalValue, %struct.ilist_node.14, %struct.iplist.4, %struct.iplist, %struct.ValueSymbolTable*, %struct.AttrListPtr }
%struct.GlobalValue = type <{ [52 x i8], [4 x i8], %struct.Module*, i8, i16, [5 x i8], %struct.basic_string }>
%struct.Module = type { %struct.LLVMContext*, %struct.iplist.20, %struct.iplist.16, %struct.iplist.12, %struct.vector.2, %struct.ilist, %struct.basic_string, %struct.ValueSymbolTable*, %struct.OwningPtr, %struct.basic_string, %struct.basic_string, %struct.basic_string, i8* }
%struct.iplist.20 = type { %struct.ilist_traits.19, %struct.GlobalVariable* }
%struct.ilist_traits.19 = type { %struct.ilist_node.18 }
%struct.ilist_node.18 = type { %struct.ilist_half_node.17, %struct.GlobalVariable* }
%struct.ilist_half_node.17 = type { %struct.GlobalVariable* }
%struct.GlobalVariable = type { %struct.GlobalValue, %struct.ilist_node.18, i8, [7 x i8] }
%struct.iplist.16 = type { %struct.ilist_traits.15, %struct.Function* }
%struct.ilist_traits.15 = type { %struct.ilist_node.14 }
%struct.ilist_node.14 = type { %struct.ilist_half_node.13, %struct.Function* }
%struct.ilist_half_node.13 = type { %struct.Function* }
%struct.iplist.12 = type { %struct.ilist_traits.11, %struct.GlobalAlias* }
%struct.ilist_traits.11 = type { %struct.ilist_node.10 }
%struct.ilist_node.10 = type { %struct.ilist_half_node.9, %struct.GlobalAlias* }
%struct.ilist_half_node.9 = type { %struct.GlobalAlias* }
%struct.GlobalAlias = type { %struct.GlobalValue, %struct.ilist_node.10 }
%struct.vector.2 = type { %struct._Vector_base.1 }
%struct._Vector_base.1 = type { %struct._Vector_impl.0 }
%struct._Vector_impl.0 = type { %struct.basic_string*, %struct.basic_string*, %struct.basic_string* }
%struct.basic_string = type { %struct._Alloc_hider }
%struct._Alloc_hider = type { i8* }
%struct.ilist = type { %struct.iplist.8 }
%struct.iplist.8 = type { %struct.ilist_traits.7, %struct.NamedMDNode* }
%struct.ilist_traits.7 = type { %struct.ilist_node.6 }
%struct.ilist_node.6 = type { %struct.ilist_half_node.5, %struct.NamedMDNode* }
%struct.ilist_half_node.5 = type { %struct.NamedMDNode* }
%struct.NamedMDNode = type { %struct.ilist_node.6, %struct.basic_string, %struct.Module*, i8* }
%struct.ValueSymbolTable = type opaque
%struct.OwningPtr = type { %struct.GVMaterializer* }
%struct.GVMaterializer = type opaque
%struct.iplist.4 = type { %struct.ilist_traits.3, %struct.BasicBlock* }
%struct.ilist_traits.3 = type { %struct.ilist_half_node.23 }
%struct.iplist = type { %struct.ilist_traits, %struct.Argument* }
%struct.ilist_traits = type { %struct.ilist_half_node }
%struct.ilist_half_node = type { %struct.Argument* }
%struct.Argument = type { %struct.Value, %struct.ilist_node, %struct.Function* }
%struct.ilist_node = type { %struct.ilist_half_node, %struct.Argument* }
%struct.AttrListPtr = type { %struct.AttributeListImpl* }
%struct.AttributeListImpl = type opaque
declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i32, i1) nounwind
; CHECK: _ZSt9iter_swapIPSt4pairIPN4llvm10BasicBlockEjES5_EvT_T0_
; CHECK: store
; CHECK: ret void
define void @_ZSt9iter_swapIPSt4pairIPN4llvm10BasicBlockEjES5_EvT_T0_(%struct.pair.162* %__a, %struct.pair.162* %__b) nounwind uwtable inlinehint {
entry:
%memtmp = alloca %struct.pair.162, align 8
%0 = getelementptr inbounds %struct.pair.162* %memtmp, i64 0, i32 0
%1 = getelementptr inbounds %struct.pair.162* %__a, i64 0, i32 0
%2 = load %struct.BasicBlock** %1, align 8
store %struct.BasicBlock* %2, %struct.BasicBlock** %0, align 8
%3 = getelementptr inbounds %struct.pair.162* %memtmp, i64 0, i32 1
%4 = getelementptr inbounds %struct.pair.162* %__a, i64 0, i32 1
%5 = load i32* %4, align 4
store i32 %5, i32* %3, align 8
%6 = bitcast %struct.pair.162* %__a to i8*
%7 = bitcast %struct.pair.162* %__b to i8*
call void @llvm.memcpy.p0i8.p0i8.i64(i8* %6, i8* %7, i64 12, i32 1, i1 false)
%8 = bitcast %struct.pair.162* %memtmp to i8*
call void @llvm.memcpy.p0i8.p0i8.i64(i8* %7, i8* %8, i64 12, i32 1, i1 false)
ret void
}