#define DEBUG_TYPE "dse"
#include "llvm/Transforms/Scalar.h"
-#include "llvm/Constants.h"
-#include "llvm/Function.h"
-#include "llvm/GlobalVariable.h"
-#include "llvm/Instructions.h"
-#include "llvm/IntrinsicInst.h"
-#include "llvm/Pass.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/SetVector.h"
+#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/CaptureTracking.h"
#include "llvm/Analysis/Dominators.h"
#include "llvm/Analysis/MemoryBuiltins.h"
#include "llvm/Analysis/MemoryDependenceAnalysis.h"
#include "llvm/Analysis/ValueTracking.h"
-#include "llvm/Target/TargetData.h"
-#include "llvm/Transforms/Utils/Local.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/Pass.h"
#include "llvm/Support/Debug.h"
-#include "llvm/ADT/SmallPtrSet.h"
-#include "llvm/ADT/Statistic.h"
-#include "llvm/ADT/STLExtras.h"
+#include "llvm/Target/TargetLibraryInfo.h"
+#include "llvm/Transforms/Utils/Local.h"
using namespace llvm;
STATISTIC(NumFastStores, "Number of stores deleted");
AliasAnalysis *AA;
MemoryDependenceAnalysis *MD;
DominatorTree *DT;
+ const TargetLibraryInfo *TLI;
static char ID; // Pass identification, replacement for typeid
DSE() : FunctionPass(ID), AA(0), MD(0), DT(0) {
AA = &getAnalysis<AliasAnalysis>();
MD = &getAnalysis<MemoryDependenceAnalysis>();
DT = &getAnalysis<DominatorTree>();
+ TLI = AA->getTargetLibraryInfo();
bool Changed = false;
for (Function::iterator I = F.begin(), E = F.end(); I != E; ++I)
bool HandleFree(CallInst *F);
bool handleEndBlock(BasicBlock &BB);
void RemoveAccessedObjects(const AliasAnalysis::Location &LoadedLoc,
- SmallPtrSet<Value*, 16> &DeadStackObjects);
+ SmallSetVector<Value*, 16> &DeadStackObjects);
virtual void getAnalysisUsage(AnalysisUsage &AU) const {
AU.setPreservesCFG();
///
static void DeleteDeadInstruction(Instruction *I,
MemoryDependenceAnalysis &MD,
- SmallPtrSet<Value*, 16> *ValueSet = 0) {
+ const TargetLibraryInfo *TLI,
+ SmallSetVector<Value*, 16> *ValueSet = 0) {
SmallVector<Instruction*, 32> NowDeadInsts;
NowDeadInsts.push_back(I);
if (!Op->use_empty()) continue;
if (Instruction *OpI = dyn_cast<Instruction>(Op))
- if (isInstructionTriviallyDead(OpI))
+ if (isInstructionTriviallyDead(OpI, TLI))
NowDeadInsts.push_back(OpI);
}
DeadInst->eraseFromParent();
- if (ValueSet) ValueSet->erase(DeadInst);
+ if (ValueSet) ValueSet->remove(DeadInst);
} while (!NowDeadInsts.empty());
}
/// hasMemoryWrite - Does this instruction write some memory? This only returns
/// true for things that we can analyze with other helpers below.
-static bool hasMemoryWrite(Instruction *I) {
+static bool hasMemoryWrite(Instruction *I, const TargetLibraryInfo *TLI) {
if (isa<StoreInst>(I))
return true;
if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
return true;
}
}
+ if (CallSite CS = I) {
+ if (Function *F = CS.getCalledFunction()) {
+ if (TLI && TLI->has(LibFunc::strcpy) &&
+ F->getName() == TLI->getName(LibFunc::strcpy)) {
+ return true;
+ }
+ if (TLI && TLI->has(LibFunc::strncpy) &&
+ F->getName() == TLI->getName(LibFunc::strncpy)) {
+ return true;
+ }
+ if (TLI && TLI->has(LibFunc::strcat) &&
+ F->getName() == TLI->getName(LibFunc::strcat)) {
+ return true;
+ }
+ if (TLI && TLI->has(LibFunc::strncat) &&
+ F->getName() == TLI->getName(LibFunc::strncat)) {
+ return true;
+ }
+ }
+ }
return false;
}
// If we don't have target data around, an unknown size in Location means
// that we should use the size of the pointee type. This isn't valid for
// memset/memcpy, which writes more than an i8.
- if (Loc.Size == AliasAnalysis::UnknownSize && AA.getTargetData() == 0)
+ if (Loc.Size == AliasAnalysis::UnknownSize && AA.getDataLayout() == 0)
return AliasAnalysis::Location();
return Loc;
}
// If we don't have target data around, an unknown size in Location means
// that we should use the size of the pointee type. This isn't valid for
// init.trampoline, which writes more than an i8.
- if (AA.getTargetData() == 0) return AliasAnalysis::Location();
+ if (AA.getDataLayout() == 0) return AliasAnalysis::Location();
// FIXME: We don't know the size of the trampoline, so we can't really
// handle it here.
/// instruction if any.
static AliasAnalysis::Location
getLocForRead(Instruction *Inst, AliasAnalysis &AA) {
- assert(hasMemoryWrite(Inst) && "Unknown instruction case");
+ assert(hasMemoryWrite(Inst, AA.getTargetLibraryInfo()) &&
+ "Unknown instruction case");
// The only instructions that both read and write are the mem transfer
// instructions (memcpy/memmove).
if (StoreInst *SI = dyn_cast<StoreInst>(I))
return SI->isUnordered();
- IntrinsicInst *II = cast<IntrinsicInst>(I);
- switch (II->getIntrinsicID()) {
- default: llvm_unreachable("doesn't pass 'hasMemoryWrite' predicate");
- case Intrinsic::lifetime_end:
- // Never remove dead lifetime_end's, e.g. because it is followed by a
- // free.
- return false;
- case Intrinsic::init_trampoline:
- // Always safe to remove init_trampoline.
- return true;
+ if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
+ switch (II->getIntrinsicID()) {
+ default: llvm_unreachable("doesn't pass 'hasMemoryWrite' predicate");
+ case Intrinsic::lifetime_end:
+ // Never remove dead lifetime_end's, e.g. because it is followed by a
+ // free.
+ return false;
+ case Intrinsic::init_trampoline:
+ // Always safe to remove init_trampoline.
+ return true;
- case Intrinsic::memset:
- case Intrinsic::memmove:
- case Intrinsic::memcpy:
- // Don't remove volatile memory intrinsics.
- return !cast<MemIntrinsic>(II)->isVolatile();
+ case Intrinsic::memset:
+ case Intrinsic::memmove:
+ case Intrinsic::memcpy:
+ // Don't remove volatile memory intrinsics.
+ return !cast<MemIntrinsic>(II)->isVolatile();
+ }
}
+
+ if (CallSite CS = I)
+ return CS.getInstruction()->use_empty();
+
+ return false;
}
// Don't shorten stores for now
if (isa<StoreInst>(I))
return false;
-
- IntrinsicInst *II = cast<IntrinsicInst>(I);
- switch (II->getIntrinsicID()) {
- default: return false;
- case Intrinsic::memset:
- case Intrinsic::memcpy:
- // Do shorten memory intrinsics.
- return true;
+
+ if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
+ switch (II->getIntrinsicID()) {
+ default: return false;
+ case Intrinsic::memset:
+ case Intrinsic::memcpy:
+ // Do shorten memory intrinsics.
+ return true;
+ }
}
+
+ // Don't shorten libcalls calls for now.
+
+ return false;
}
/// getStoredPointerOperand - Return the pointer that is being written to.
if (MemIntrinsic *MI = dyn_cast<MemIntrinsic>(I))
return MI->getDest();
- IntrinsicInst *II = cast<IntrinsicInst>(I);
- switch (II->getIntrinsicID()) {
- default: llvm_unreachable("Unexpected intrinsic!");
- case Intrinsic::init_trampoline:
- return II->getArgOperand(0);
+ if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
+ switch (II->getIntrinsicID()) {
+ default: llvm_unreachable("Unexpected intrinsic!");
+ case Intrinsic::init_trampoline:
+ return II->getArgOperand(0);
+ }
}
+
+ CallSite CS = I;
+ // All the supported functions so far happen to have dest as their first
+ // argument.
+ return CS.getArgument(0);
}
static uint64_t getPointerSize(const Value *V, AliasAnalysis &AA) {
- const TargetData *TD = AA.getTargetData();
-
- if (const CallInst *CI = extractMallocCall(V)) {
- if (const ConstantInt *C = dyn_cast<ConstantInt>(CI->getArgOperand(0)))
- return C->getZExtValue();
- }
-
- if (TD == 0)
- return AliasAnalysis::UnknownSize;
-
- if (const AllocaInst *A = dyn_cast<AllocaInst>(V)) {
- // Get size information for the alloca
- if (const ConstantInt *C = dyn_cast<ConstantInt>(A->getArraySize()))
- return C->getZExtValue() * TD->getTypeAllocSize(A->getAllocatedType());
- }
-
- if (const Argument *A = dyn_cast<Argument>(V)) {
- if (A->hasByValAttr())
- if (PointerType *PT = dyn_cast<PointerType>(A->getType()))
- return TD->getTypeAllocSize(PT->getElementType());
- }
-
- if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(V)) {
- if (!GV->mayBeOverridden())
- return TD->getTypeAllocSize(GV->getType()->getElementType());
- }
-
+ uint64_t Size;
+ if (getObjectSize(V, Size, AA.getDataLayout(), AA.getTargetLibraryInfo()))
+ return Size;
return AliasAnalysis::UnknownSize;
}
/// isOverwrite - Return 'OverwriteComplete' if a store to the 'Later' location
/// completely overwrites a store to the 'Earlier' location.
-/// 'OverwriteEnd' if the end of the 'Earlier' location is completely
+/// 'OverwriteEnd' if the end of the 'Earlier' location is completely
/// overwritten by 'Later', or 'OverwriteUnknown' if nothing can be determined
static OverwriteResult isOverwrite(const AliasAnalysis::Location &Later,
const AliasAnalysis::Location &Earlier,
// comparison.
if (Later.Size == AliasAnalysis::UnknownSize ||
Earlier.Size == AliasAnalysis::UnknownSize) {
- // If we have no TargetData information around, then the size of the store
+ // If we have no DataLayout information around, then the size of the store
// is inferrable from the pointee type. If they are the same type, then
// we know that the store is safe.
- if (AA.getTargetData() == 0 &&
+ if (AA.getDataLayout() == 0 &&
Later.Ptr->getType() == Earlier.Ptr->getType())
return OverwriteComplete;
-
+
return OverwriteUnknown;
}
// larger than the earlier one.
if (Later.Size == AliasAnalysis::UnknownSize ||
Earlier.Size == AliasAnalysis::UnknownSize ||
- AA.getTargetData() == 0)
+ AA.getDataLayout() == 0)
return OverwriteUnknown;
// 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
// other store to the same object.
- const TargetData &TD = *AA.getTargetData();
+ const DataLayout &TD = *AA.getDataLayout();
const Value *UO1 = GetUnderlyingObject(P1, &TD),
*UO2 = GetUnderlyingObject(P2, &TD);
//
// We have to be careful here as *Off is signed while *.Size is unsigned.
if (EarlierOff >= LaterOff &&
- Later.Size > Earlier.Size &&
+ Later.Size >= Earlier.Size &&
uint64_t(EarlierOff - LaterOff) + Earlier.Size <= Later.Size)
return OverwriteComplete;
-
+
// The other interesting case is if the later store overwrites the end of
// the earlier store
//
Instruction *Inst = BBI++;
// Handle 'free' calls specially.
- if (CallInst *F = isFreeCall(Inst)) {
+ if (CallInst *F = isFreeCall(Inst, TLI)) {
MadeChange |= HandleFree(F);
continue;
}
// If we find something that writes memory, get its memory dependence.
- if (!hasMemoryWrite(Inst))
+ if (!hasMemoryWrite(Inst, TLI))
continue;
MemDepResult InstDep = MD->getDependency(Inst);
// in case we need it.
WeakVH NextInst(BBI);
- DeleteDeadInstruction(SI, *MD);
+ DeleteDeadInstruction(SI, *MD, TLI);
if (NextInst == 0) // Next instruction deleted.
BBI = BB.begin();
// 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) &&
+ if (isRemovable(DepWrite) &&
!isPossibleSelfRead(Inst, Loc, DepWrite, *AA)) {
- int64_t InstWriteOffset, DepWriteOffset;
- OverwriteResult OR = isOverwrite(Loc, DepLoc, *AA,
- DepWriteOffset, InstWriteOffset);
+ 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);
+ DeleteDeadInstruction(DepWrite, *MD, TLI);
++NumFastStores;
MadeChange = true;
-
+
// DeleteDeadInstruction can delete the current instruction in loop
// cases, reset BBI.
BBI = Inst;
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 << ", "
+ << *DepWrite << "\n KILLER (offset "
+ << InstWriteOffset << ", "
<< DepLoc.Size << ")"
<< *Inst << '\n');
-
+
Value* DepWriteLength = DepIntrinsic->getLength();
Value* TrimmedLength = ConstantInt::get(DepWriteLength->getType(),
- InstWriteOffset -
+ InstWriteOffset -
DepWriteOffset);
DepIntrinsic->setLength(TrimmedLength);
MadeChange = true;
MemDepResult Dep = MD->getPointerDependencyFrom(Loc, false, InstPt, BB);
while (Dep.isDef() || Dep.isClobber()) {
Instruction *Dependency = Dep.getInst();
- if (!hasMemoryWrite(Dependency) || !isRemovable(Dependency))
+ if (!hasMemoryWrite(Dependency, TLI) || !isRemovable(Dependency))
break;
Value *DepPointer =
Instruction *Next = llvm::next(BasicBlock::iterator(Dependency));
// DCE instructions only used to calculate that store
- DeleteDeadInstruction(Dependency, *MD);
+ DeleteDeadInstruction(Dependency, *MD, TLI);
++NumFastStores;
MadeChange = true;
return MadeChange;
}
+namespace {
+ struct CouldRef {
+ typedef Value *argument_type;
+ const CallSite CS;
+ AliasAnalysis *AA;
+
+ bool operator()(Value *I) {
+ // See if the call site touches the value.
+ AliasAnalysis::ModRefResult A =
+ AA->getModRefInfo(CS, I, getPointerSize(I, *AA));
+
+ return A == AliasAnalysis::ModRef || A == AliasAnalysis::Ref;
+ }
+ };
+}
+
/// handleEndBlock - Remove dead stores to stack-allocated locations in the
/// function end block. Ex:
/// %A = alloca i32
// Keep track of all of the stack objects that are dead at the end of the
// function.
- SmallPtrSet<Value*, 16> DeadStackObjects;
+ SmallSetVector<Value*, 16> DeadStackObjects;
// Find all of the alloca'd pointers in the entry block.
BasicBlock *Entry = BB.getParent()->begin();
for (BasicBlock::iterator I = Entry->begin(), E = Entry->end(); I != E; ++I) {
- if (AllocaInst *AI = dyn_cast<AllocaInst>(I))
- DeadStackObjects.insert(AI);
+ if (isa<AllocaInst>(I))
+ DeadStackObjects.insert(I);
// Okay, so these are dead heap objects, but if the pointer never escapes
// then it's leaked by this function anyways.
- if (CallInst *CI = extractMallocCall(I))
- if (!PointerMayBeCaptured(CI, true, true))
- DeadStackObjects.insert(CI);
+ else if (isAllocLikeFn(I, TLI) && !PointerMayBeCaptured(I, true, true))
+ DeadStackObjects.insert(I);
}
// Treat byval arguments the same, stores to them are dead at the end of the
--BBI;
// If we find a store, check to see if it points into a dead stack value.
- if (hasMemoryWrite(BBI) && isRemovable(BBI)) {
+ if (hasMemoryWrite(BBI, TLI) && isRemovable(BBI)) {
// See through pointer-to-pointer bitcasts
- Value *Pointer = GetUnderlyingObject(getStoredPointerOperand(BBI));
+ SmallVector<Value *, 4> Pointers;
+ GetUnderlyingObjects(getStoredPointerOperand(BBI), Pointers);
// Stores to stack values are valid candidates for removal.
- if (DeadStackObjects.count(Pointer)) {
+ bool AllDead = true;
+ for (SmallVectorImpl<Value *>::iterator I = Pointers.begin(),
+ E = Pointers.end(); I != E; ++I)
+ if (!DeadStackObjects.count(*I)) {
+ AllDead = false;
+ break;
+ }
+
+ if (AllDead) {
Instruction *Dead = BBI++;
DEBUG(dbgs() << "DSE: Dead Store at End of Block:\n DEAD: "
- << *Dead << "\n Object: " << *Pointer << '\n');
+ << *Dead << "\n Objects: ";
+ for (SmallVectorImpl<Value *>::iterator I = Pointers.begin(),
+ E = Pointers.end(); I != E; ++I) {
+ dbgs() << **I;
+ if (llvm::next(I) != E)
+ dbgs() << ", ";
+ }
+ dbgs() << '\n');
// DCE instructions only used to calculate that store.
- DeleteDeadInstruction(Dead, *MD, &DeadStackObjects);
+ DeleteDeadInstruction(Dead, *MD, TLI, &DeadStackObjects);
++NumFastStores;
MadeChange = true;
continue;
}
// Remove any dead non-memory-mutating instructions.
- if (isInstructionTriviallyDead(BBI)) {
+ if (isInstructionTriviallyDead(BBI, TLI)) {
Instruction *Inst = BBI++;
- DeleteDeadInstruction(Inst, *MD, &DeadStackObjects);
+ DeleteDeadInstruction(Inst, *MD, TLI, &DeadStackObjects);
++NumFastOther;
MadeChange = true;
continue;
}
- if (AllocaInst *A = dyn_cast<AllocaInst>(BBI)) {
- DeadStackObjects.erase(A);
- continue;
- }
-
- if (CallInst *CI = extractMallocCall(BBI)) {
- DeadStackObjects.erase(CI);
+ if (isa<AllocaInst>(BBI)) {
+ // Remove allocas from the list of dead stack objects; there can't be
+ // any references before the definition.
+ DeadStackObjects.remove(BBI);
continue;
}
if (CallSite CS = cast<Value>(BBI)) {
+ // Remove allocation function calls from the list of dead stack objects;
+ // there can't be any references before the definition.
+ if (isAllocLikeFn(BBI, TLI))
+ DeadStackObjects.remove(BBI);
+
// If this call does not access memory, it can't be loading any of our
// pointers.
if (AA->doesNotAccessMemory(CS))
// If the call might load from any of our allocas, then any store above
// the call is live.
- SmallVector<Value*, 8> LiveAllocas;
- for (SmallPtrSet<Value*, 16>::iterator I = DeadStackObjects.begin(),
- E = DeadStackObjects.end(); I != E; ++I) {
- // See if the call site touches it.
- AliasAnalysis::ModRefResult A =
- AA->getModRefInfo(CS, *I, getPointerSize(*I, *AA));
-
- if (A == AliasAnalysis::ModRef || A == AliasAnalysis::Ref)
- LiveAllocas.push_back(*I);
- }
-
- for (SmallVector<Value*, 8>::iterator I = LiveAllocas.begin(),
- E = LiveAllocas.end(); I != E; ++I)
- DeadStackObjects.erase(*I);
+ CouldRef Pred = { CS, AA };
+ DeadStackObjects.remove_if(Pred);
// If all of the allocas were clobbered by the call then we're not going
// to find anything else to process.
if (DeadStackObjects.empty())
- return MadeChange;
+ break;
continue;
}
return MadeChange;
}
+namespace {
+ struct CouldAlias {
+ typedef Value *argument_type;
+ const AliasAnalysis::Location &LoadedLoc;
+ AliasAnalysis *AA;
+
+ bool operator()(Value *I) {
+ // See if the loaded location could alias the stack location.
+ AliasAnalysis::Location StackLoc(I, getPointerSize(I, *AA));
+ return !AA->isNoAlias(StackLoc, LoadedLoc);
+ }
+ };
+}
+
/// RemoveAccessedObjects - Check to see if the specified location may alias any
/// of the stack objects in the DeadStackObjects set. If so, they become live
/// because the location is being loaded.
void DSE::RemoveAccessedObjects(const AliasAnalysis::Location &LoadedLoc,
- SmallPtrSet<Value*, 16> &DeadStackObjects) {
+ SmallSetVector<Value*, 16> &DeadStackObjects) {
const Value *UnderlyingPointer = GetUnderlyingObject(LoadedLoc.Ptr);
// A constant can't be in the dead pointer set.
// If the kill pointer can be easily reduced to an alloca, don't bother doing
// extraneous AA queries.
if (isa<AllocaInst>(UnderlyingPointer) || isa<Argument>(UnderlyingPointer)) {
- DeadStackObjects.erase(const_cast<Value*>(UnderlyingPointer));
+ DeadStackObjects.remove(const_cast<Value*>(UnderlyingPointer));
return;
}
- SmallVector<Value*, 16> NowLive;
- for (SmallPtrSet<Value*, 16>::iterator I = DeadStackObjects.begin(),
- E = DeadStackObjects.end(); I != E; ++I) {
- // See if the loaded location could alias the stack location.
- AliasAnalysis::Location StackLoc(*I, getPointerSize(*I, *AA));
- if (!AA->isNoAlias(StackLoc, LoadedLoc))
- NowLive.push_back(*I);
- }
-
- for (SmallVector<Value*, 16>::iterator I = NowLive.begin(), E = NowLive.end();
- I != E; ++I)
- DeadStackObjects.erase(*I);
+ // Remove objects that could alias LoadedLoc.
+ CouldAlias Pred = { LoadedLoc, AA };
+ DeadStackObjects.remove_if(Pred);
}