#include "llvm/Module.h"
#include "llvm/Pass.h"
#include "llvm/Analysis/ConstantFolding.h"
+#include "llvm/Analysis/MemoryBuiltins.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Support/CallSite.h"
-#include "llvm/Support/Compiler.h"
#include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/GetElementPtrTypeIterator.h"
#include "llvm/Support/MathExtras.h"
+#include "llvm/Support/raw_ostream.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
-#include "llvm/ADT/StringExtras.h"
#include "llvm/ADT/STLExtras.h"
#include <algorithm>
using namespace llvm;
STATISTIC(NumFastCallFns , "Number of functions converted to fastcc");
STATISTIC(NumCtorsEvaluated, "Number of static ctors evaluated");
STATISTIC(NumNestRemoved , "Number of nest attributes removed");
+STATISTIC(NumAliasesResolved, "Number of global aliases resolved");
+STATISTIC(NumAliasesRemoved, "Number of global aliases eliminated");
namespace {
- struct VISIBILITY_HIDDEN GlobalOpt : public ModulePass {
+ struct GlobalOpt : public ModulePass {
virtual void getAnalysisUsage(AnalysisUsage &AU) const {
- AU.addRequired<TargetData>();
}
static char ID; // Pass identification, replacement for typeid
GlobalOpt() : ModulePass(&ID) {}
GlobalVariable *FindGlobalCtors(Module &M);
bool OptimizeFunctions(Module &M);
bool OptimizeGlobalVars(Module &M);
- bool ResolveAliases(Module &M);
+ bool OptimizeGlobalAliases(Module &M);
bool OptimizeGlobalCtorsList(GlobalVariable *&GCL);
bool ProcessInternalGlobal(GlobalVariable *GV,Module::global_iterator &GVI);
};
/// GlobalStatus - As we analyze each global, keep track of some information
/// about it. If we find out that the address of the global is taken, none of
/// this info will be accurate.
-struct VISIBILITY_HIDDEN GlobalStatus {
+struct GlobalStatus {
/// isLoaded - True if the global is ever loaded. If the global isn't ever
/// loaded it can be deleted.
bool isLoaded;
}
-/// ConstantIsDead - Return true if the specified constant is (transitively)
-/// dead. The constant may be used by other constants (e.g. constant arrays and
-/// constant exprs) as long as they are dead, but it cannot be used by anything
-/// else.
-static bool ConstantIsDead(Constant *C) {
+// SafeToDestroyConstant - It is safe to destroy a constant iff it is only used
+// by constants itself. Note that constants cannot be cyclic, so this test is
+// pretty easy to implement recursively.
+//
+static bool SafeToDestroyConstant(Constant *C) {
if (isa<GlobalValue>(C)) return false;
for (Value::use_iterator UI = C->use_begin(), E = C->use_end(); UI != E; ++UI)
if (Constant *CU = dyn_cast<Constant>(*UI)) {
- if (!ConstantIsDead(CU)) return false;
+ if (!SafeToDestroyConstant(CU)) return false;
} else
return false;
return true;
if (AnalyzeGlobal(I, GS, PHIUsers)) return true;
GS.HasPHIUser = true;
} else if (isa<CmpInst>(I)) {
- } else if (isa<MemCpyInst>(I) || isa<MemMoveInst>(I)) {
+ } else if (isa<MemTransferInst>(I)) {
if (I->getOperand(1) == V)
GS.StoredType = GlobalStatus::isStored;
if (I->getOperand(2) == V)
} else if (Constant *C = dyn_cast<Constant>(*UI)) {
GS.HasNonInstructionUser = true;
// We might have a dead and dangling constant hanging off of here.
- if (!ConstantIsDead(C))
+ if (!SafeToDestroyConstant(C))
return true;
} else {
GS.HasNonInstructionUser = true;
} else if (Constant *C = dyn_cast<Constant>(U)) {
// If we have a chain of dead constantexprs or other things dangling from
// us, and if they are all dead, nuke them without remorse.
- if (ConstantIsDead(C)) {
+ if (SafeToDestroyConstant(C)) {
C->destroyConstant();
// This could have invalidated UI, start over from scratch.
CleanupConstantGlobalUsers(V, Init);
static bool isSafeSROAElementUse(Value *V) {
// We might have a dead and dangling constant hanging off of here.
if (Constant *C = dyn_cast<Constant>(V))
- return ConstantIsDead(C);
+ return SafeToDestroyConstant(C);
Instruction *I = dyn_cast<Instruction>(V);
if (!I) return false;
// Scalar replacing *just* the outer index of the array is probably not
// going to be a win anyway, so just give up.
for (++GEPI; // Skip array index.
- GEPI != E && (isa<ArrayType>(*GEPI) || isa<VectorType>(*GEPI));
+ GEPI != E;
++GEPI) {
uint64_t NumElements;
if (const ArrayType *SubArrayTy = dyn_cast<ArrayType>(*GEPI))
NumElements = SubArrayTy->getNumElements();
- else
- NumElements = cast<VectorType>(*GEPI)->getNumElements();
+ else if (const VectorType *SubVectorTy = dyn_cast<VectorType>(*GEPI))
+ NumElements = SubVectorTy->getNumElements();
+ else {
+ assert(isa<StructType>(*GEPI) &&
+ "Indexed GEP type is not array, vector, or struct!");
+ continue;
+ }
ConstantInt *IdxVal = dyn_cast<ConstantInt>(GEPI.getOperand());
if (!IdxVal || IdxVal->getZExtValue() >= NumElements)
if (!GlobalUsersSafeToSRA(GV))
return 0;
- assert(GV->hasInternalLinkage() && !GV->isConstant());
+ assert(GV->hasLocalLinkage() && !GV->isConstant());
Constant *Init = GV->getInitializer();
const Type *Ty = Init->getType();
const StructLayout &Layout = *TD.getStructLayout(STy);
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
Constant *In = getAggregateConstantElement(Init,
- ConstantInt::get(Type::Int32Ty, i));
+ ConstantInt::get(Type::getInt32Ty(STy->getContext()), i));
assert(In && "Couldn't get element of initializer?");
GlobalVariable *NGV = new GlobalVariable(STy->getElementType(i), false,
GlobalVariable::InternalLinkage,
- In, GV->getName()+"."+utostr(i),
- (Module *)NULL,
+ In, GV->getName()+"."+Twine(i),
GV->isThreadLocal(),
- GV->getType()->getAddressSpace());
+ GV->getType()->getAddressSpace());
Globals.insert(GV, NGV);
NewGlobals.push_back(NGV);
return 0; // It's not worth it.
NewGlobals.reserve(NumElements);
- uint64_t EltSize = TD.getTypePaddedSize(STy->getElementType());
+ uint64_t EltSize = TD.getTypeAllocSize(STy->getElementType());
unsigned EltAlign = TD.getABITypeAlignment(STy->getElementType());
for (unsigned i = 0, e = NumElements; i != e; ++i) {
Constant *In = getAggregateConstantElement(Init,
- ConstantInt::get(Type::Int32Ty, i));
+ ConstantInt::get(Type::getInt32Ty(Init->getContext()), i));
assert(In && "Couldn't get element of initializer?");
GlobalVariable *NGV = new GlobalVariable(STy->getElementType(), false,
GlobalVariable::InternalLinkage,
- In, GV->getName()+"."+utostr(i),
- (Module *)NULL,
+ In, GV->getName()+"."+Twine(i),
GV->isThreadLocal(),
- GV->getType()->getAddressSpace());
+ GV->getType()->getAddressSpace());
Globals.insert(GV, NGV);
NewGlobals.push_back(NGV);
if (NewGlobals.empty())
return 0;
- DOUT << "PERFORMING GLOBAL SRA ON: " << *GV;
+ DEBUG(dbgs() << "PERFORMING GLOBAL SRA ON: " << *GV);
- Constant *NullInt = Constant::getNullValue(Type::Int32Ty);
+ Constant *NullInt =Constant::getNullValue(Type::getInt32Ty(GV->getContext()));
// Loop over all of the uses of the global, replacing the constantexpr geps,
// with smaller constantexpr geps or direct references.
for (unsigned i = 3, e = GEPI->getNumOperands(); i != e; ++i)
Idxs.push_back(GEPI->getOperand(i));
NewPtr = GetElementPtrInst::Create(NewPtr, Idxs.begin(), Idxs.end(),
- GEPI->getName()+"."+utostr(Val), GEPI);
+ GEPI->getName()+"."+Twine(Val),GEPI);
}
}
GEP->replaceAllUsesWith(NewPtr);
} else if (PHINode *PN = dyn_cast<PHINode>(*UI)) {
// If we've already seen this phi node, ignore it, it has already been
// checked.
- if (PHIs.insert(PN))
- return AllUsesOfValueWillTrapIfNull(PN, PHIs);
+ if (PHIs.insert(PN) && !AllUsesOfValueWillTrapIfNull(PN, PHIs))
+ return false;
} else if (isa<ICmpInst>(*UI) &&
isa<ConstantPointerNull>(UI->getOperand(1))) {
// Ignore setcc X, null
break;
if (Idxs.size() == GEPI->getNumOperands()-1)
Changed |= OptimizeAwayTrappingUsesOfValue(GEPI,
- ConstantExpr::getGetElementPtr(NewV, &Idxs[0],
- Idxs.size()));
+ ConstantExpr::getGetElementPtr(NewV, &Idxs[0],
+ Idxs.size()));
if (GEPI->use_empty()) {
Changed = true;
GEPI->eraseFromParent();
}
if (Changed) {
- DOUT << "OPTIMIZED LOADS FROM STORED ONCE POINTER: " << *GV;
+ DEBUG(dbgs() << "OPTIMIZED LOADS FROM STORED ONCE POINTER: " << *GV);
++NumGlobUses;
}
// If we nuked all of the loads, then none of the stores are needed either,
// nor is the global.
if (AllNonStoreUsesGone) {
- DOUT << " *** GLOBAL NOW DEAD!\n";
+ DEBUG(dbgs() << " *** GLOBAL NOW DEAD!\n");
CleanupConstantGlobalUsers(GV, 0);
if (GV->use_empty()) {
GV->eraseFromParent();
/// malloc, there is no reason to actually DO the malloc. Instead, turn the
/// malloc into a global, and any loads of GV as uses of the new global.
static GlobalVariable *OptimizeGlobalAddressOfMalloc(GlobalVariable *GV,
- MallocInst *MI) {
- DOUT << "PROMOTING MALLOC GLOBAL: " << *GV << " MALLOC = " << *MI;
- ConstantInt *NElements = cast<ConstantInt>(MI->getArraySize());
+ CallInst *CI,
+ const Type *AllocTy,
+ Value* NElems,
+ TargetData* TD) {
+ DEBUG(dbgs() << "PROMOTING GLOBAL: " << *GV << " CALL = " << *CI << '\n');
+
+ const Type *IntPtrTy = TD->getIntPtrType(GV->getContext());
+
+ // CI has either 0 or 1 bitcast uses (getMallocType() would otherwise have
+ // returned NULL and we would not be here).
+ BitCastInst *BCI = NULL;
+ for (Value::use_iterator UI = CI->use_begin(), E = CI->use_end(); UI != E; )
+ if ((BCI = dyn_cast<BitCastInst>(cast<Instruction>(*UI++))))
+ break;
+ ConstantInt *NElements = cast<ConstantInt>(NElems);
if (NElements->getZExtValue() != 1) {
// If we have an array allocation, transform it to a single element
// allocation to make the code below simpler.
- Type *NewTy = ArrayType::get(MI->getAllocatedType(),
- NElements->getZExtValue());
- MallocInst *NewMI =
- new MallocInst(NewTy, Constant::getNullValue(Type::Int32Ty),
- MI->getAlignment(), MI->getName(), MI);
+ Type *NewTy = ArrayType::get(AllocTy, NElements->getZExtValue());
+ unsigned TypeSize = TD->getTypeAllocSize(NewTy);
+ if (const StructType *ST = dyn_cast<StructType>(NewTy))
+ TypeSize = TD->getStructLayout(ST)->getSizeInBytes();
+ Instruction *NewCI = CallInst::CreateMalloc(CI, IntPtrTy, NewTy,
+ ConstantInt::get(IntPtrTy, TypeSize));
Value* Indices[2];
- Indices[0] = Indices[1] = Constant::getNullValue(Type::Int32Ty);
- Value *NewGEP = GetElementPtrInst::Create(NewMI, Indices, Indices + 2,
- NewMI->getName()+".el0", MI);
- MI->replaceAllUsesWith(NewGEP);
- MI->eraseFromParent();
- MI = NewMI;
+ Indices[0] = Indices[1] = Constant::getNullValue(IntPtrTy);
+ Value *NewGEP = GetElementPtrInst::Create(NewCI, Indices, Indices + 2,
+ NewCI->getName()+".el0", CI);
+ Value *Cast = new BitCastInst(NewGEP, CI->getType(), "el0", CI);
+ if (BCI) BCI->replaceAllUsesWith(NewGEP);
+ CI->replaceAllUsesWith(Cast);
+ if (BCI) BCI->eraseFromParent();
+ CI->eraseFromParent();
+ BCI = dyn_cast<BitCastInst>(NewCI);
+ CI = BCI ? extractMallocCallFromBitCast(BCI) : cast<CallInst>(NewCI);
}
// Create the new global variable. The contents of the malloc'd memory is
// undefined, so initialize with an undef value.
- Constant *Init = UndefValue::get(MI->getAllocatedType());
- GlobalVariable *NewGV = new GlobalVariable(MI->getAllocatedType(), false,
+ const Type *MAT = getMallocAllocatedType(CI);
+ Constant *Init = UndefValue::get(MAT);
+ GlobalVariable *NewGV = new GlobalVariable(*GV->getParent(),
+ MAT, false,
GlobalValue::InternalLinkage, Init,
GV->getName()+".body",
- (Module *)NULL,
+ GV,
GV->isThreadLocal());
- // FIXME: This new global should have the alignment returned by malloc. Code
- // could depend on malloc returning large alignment (on the mac, 16 bytes) but
- // this would only guarantee some lower alignment.
- GV->getParent()->getGlobalList().insert(GV, NewGV);
-
- // Anything that used the malloc now uses the global directly.
- MI->replaceAllUsesWith(NewGV);
+
+ // Anything that used the malloc or its bitcast now uses the global directly.
+ if (BCI) BCI->replaceAllUsesWith(NewGV);
+ CI->replaceAllUsesWith(new BitCastInst(NewGV, CI->getType(), "newgv", CI));
Constant *RepValue = NewGV;
if (NewGV->getType() != GV->getType()->getElementType())
// If there is a comparison against null, we will insert a global bool to
// keep track of whether the global was initialized yet or not.
GlobalVariable *InitBool =
- new GlobalVariable(Type::Int1Ty, false, GlobalValue::InternalLinkage,
- ConstantInt::getFalse(), GV->getName()+".init",
- (Module *)NULL, GV->isThreadLocal());
+ new GlobalVariable(Type::getInt1Ty(GV->getContext()), false,
+ GlobalValue::InternalLinkage,
+ ConstantInt::getFalse(GV->getContext()),
+ GV->getName()+".init", GV->isThreadLocal());
bool InitBoolUsed = false;
// Loop over all uses of GV, processing them in turn.
if (!isa<ICmpInst>(LoadUse.getUser()))
LoadUse = RepValue;
else {
- ICmpInst *CI = cast<ICmpInst>(LoadUse.getUser());
+ ICmpInst *ICI = cast<ICmpInst>(LoadUse.getUser());
// Replace the cmp X, 0 with a use of the bool value.
- Value *LV = new LoadInst(InitBool, InitBool->getName()+".val", CI);
+ Value *LV = new LoadInst(InitBool, InitBool->getName()+".val", ICI);
InitBoolUsed = true;
- switch (CI->getPredicate()) {
- default: assert(0 && "Unknown ICmp Predicate!");
+ switch (ICI->getPredicate()) {
+ default: llvm_unreachable("Unknown ICmp Predicate!");
case ICmpInst::ICMP_ULT:
- case ICmpInst::ICMP_SLT:
- LV = ConstantInt::getFalse(); // X < null -> always false
+ case ICmpInst::ICMP_SLT: // X < null -> always false
+ LV = ConstantInt::getFalse(GV->getContext());
break;
case ICmpInst::ICMP_ULE:
case ICmpInst::ICMP_SLE:
case ICmpInst::ICMP_EQ:
- LV = BinaryOperator::CreateNot(LV, "notinit", CI);
+ LV = BinaryOperator::CreateNot(LV, "notinit", ICI);
break;
case ICmpInst::ICMP_NE:
case ICmpInst::ICMP_UGE:
case ICmpInst::ICMP_SGT:
break; // no change.
}
- CI->replaceAllUsesWith(LV);
- CI->eraseFromParent();
+ ICI->replaceAllUsesWith(LV);
+ ICI->eraseFromParent();
}
}
LI->eraseFromParent();
} else {
StoreInst *SI = cast<StoreInst>(GV->use_back());
// The global is initialized when the store to it occurs.
- new StoreInst(ConstantInt::getTrue(), InitBool, SI);
+ new StoreInst(ConstantInt::getTrue(GV->getContext()), InitBool, SI);
SI->eraseFromParent();
}
GV->getParent()->getGlobalList().insert(GV, InitBool);
- // Now the GV is dead, nuke it and the malloc.
+ // Now the GV is dead, nuke it and the malloc (both CI and BCI).
GV->eraseFromParent();
- MI->eraseFromParent();
+ if (BCI) BCI->eraseFromParent();
+ CI->eraseFromParent();
// To further other optimizations, loop over all users of NewGV and try to
// constant prop them. This will promote GEP instructions with constant
GlobalVariable *GV,
SmallPtrSet<PHINode*, 8> &PHIs) {
for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI != E;++UI){
- Instruction *Inst = dyn_cast<Instruction>(*UI);
- if (Inst == 0) return false;
+ Instruction *Inst = cast<Instruction>(*UI);
if (isa<LoadInst>(Inst) || isa<CmpInst>(Inst)) {
continue; // Fine, ignore.
} else if (PHINode *PN = dyn_cast<PHINode>(U)) {
// Insert the load in the corresponding predecessor, not right before the
// PHI.
- unsigned PredNo = Alloc->use_begin().getOperandNo()/2;
- InsertPt = PN->getIncomingBlock(PredNo)->getTerminator();
+ InsertPt = PN->getIncomingBlock(Alloc->use_begin())->getTerminator();
} else if (isa<BitCastInst>(U)) {
// Must be bitcast between the malloc and store to initialize the global.
ReplaceUsesOfMallocWithGlobal(U, GV);
/// of a load) are simple enough to perform heap SRA on. This permits GEP's
/// that index through the array and struct field, icmps of null, and PHIs.
static bool LoadUsesSimpleEnoughForHeapSRA(Value *V,
- SmallPtrSet<PHINode*, 32> &LoadUsingPHIs) {
+ SmallPtrSet<PHINode*, 32> &LoadUsingPHIs,
+ SmallPtrSet<PHINode*, 32> &LoadUsingPHIsPerLoad) {
// We permit two users of the load: setcc comparing against the null
// pointer, and a getelementptr of a specific form.
for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI != E;++UI){
}
if (PHINode *PN = dyn_cast<PHINode>(User)) {
- // If we have already recursively analyzed this PHI, then it is safe.
- if (LoadUsingPHIs.insert(PN))
+ if (!LoadUsingPHIsPerLoad.insert(PN))
+ // This means some phi nodes are dependent on each other.
+ // Avoid infinite looping!
+ return false;
+ if (!LoadUsingPHIs.insert(PN))
+ // If we have already analyzed this PHI, then it is safe.
continue;
// Make sure all uses of the PHI are simple enough to transform.
- if (!LoadUsesSimpleEnoughForHeapSRA(PN, LoadUsingPHIs))
+ if (!LoadUsesSimpleEnoughForHeapSRA(PN,
+ LoadUsingPHIs, LoadUsingPHIsPerLoad))
return false;
continue;
/// AllGlobalLoadUsesSimpleEnoughForHeapSRA - If all users of values loaded from
/// GV are simple enough to perform HeapSRA, return true.
static bool AllGlobalLoadUsesSimpleEnoughForHeapSRA(GlobalVariable *GV,
- MallocInst *MI) {
+ Instruction *StoredVal) {
SmallPtrSet<PHINode*, 32> LoadUsingPHIs;
+ SmallPtrSet<PHINode*, 32> LoadUsingPHIsPerLoad;
for (Value::use_iterator UI = GV->use_begin(), E = GV->use_end(); UI != E;
++UI)
- if (LoadInst *LI = dyn_cast<LoadInst>(*UI))
- if (!LoadUsesSimpleEnoughForHeapSRA(LI, LoadUsingPHIs))
+ if (LoadInst *LI = dyn_cast<LoadInst>(*UI)) {
+ if (!LoadUsesSimpleEnoughForHeapSRA(LI, LoadUsingPHIs,
+ LoadUsingPHIsPerLoad))
return false;
+ LoadUsingPHIsPerLoad.clear();
+ }
// If we reach here, we know that all uses of the loads and transitive uses
// (through PHI nodes) are simple enough to transform. However, we don't know
Value *InVal = PN->getIncomingValue(op);
// PHI of the stored value itself is ok.
- if (InVal == MI) continue;
+ if (InVal == StoredVal) continue;
if (PHINode *InPN = dyn_cast<PHINode>(InVal)) {
// One of the PHIs in our set is (optimistically) ok.
Result = new LoadInst(GetHeapSROAValue(LI->getOperand(0), FieldNo,
InsertedScalarizedValues,
PHIsToRewrite),
- LI->getName()+".f" + utostr(FieldNo), LI);
+ LI->getName()+".f"+Twine(FieldNo), LI);
} else if (PHINode *PN = dyn_cast<PHINode>(V)) {
// PN's type is pointer to struct. Make a new PHI of pointer to struct
// field.
const StructType *ST =
cast<StructType>(cast<PointerType>(PN->getType())->getElementType());
- Result =PHINode::Create(PointerType::getUnqual(ST->getElementType(FieldNo)),
- PN->getName()+".f"+utostr(FieldNo), PN);
+ Result =
+ PHINode::Create(PointerType::getUnqual(ST->getElementType(FieldNo)),
+ PN->getName()+".f"+Twine(FieldNo), PN);
PHIsToRewrite.push_back(std::make_pair(PN, FieldNo));
} else {
- assert(0 && "Unknown usable value");
+ llvm_unreachable("Unknown usable value");
Result = 0;
}
Value *NPtr = GetHeapSROAValue(SCI->getOperand(0), 0,
InsertedScalarizedValues, PHIsToRewrite);
- Value *New = new ICmpInst(SCI->getPredicate(), NPtr,
- Constant::getNullValue(NPtr->getType()),
- SCI->getName(), SCI);
+ Value *New = new ICmpInst(SCI, SCI->getPredicate(), NPtr,
+ Constant::getNullValue(NPtr->getType()),
+ SCI->getName());
SCI->replaceAllUsesWith(New);
SCI->eraseFromParent();
return;
}
}
-/// PerformHeapAllocSRoA - MI is an allocation of an array of structures. Break
+/// PerformHeapAllocSRoA - CI is an allocation of an array of structures. Break
/// it up into multiple allocations of arrays of the fields.
-static GlobalVariable *PerformHeapAllocSRoA(GlobalVariable *GV, MallocInst *MI){
- DOUT << "SROA HEAP ALLOC: " << *GV << " MALLOC = " << *MI;
- const StructType *STy = cast<StructType>(MI->getAllocatedType());
+static GlobalVariable *PerformHeapAllocSRoA(GlobalVariable *GV, CallInst *CI,
+ Value* NElems, TargetData *TD) {
+ DEBUG(dbgs() << "SROA HEAP ALLOC: " << *GV << " MALLOC = " << *CI << '\n');
+ const Type* MAT = getMallocAllocatedType(CI);
+ const StructType *STy = cast<StructType>(MAT);
// There is guaranteed to be at least one use of the malloc (storing
// it into GV). If there are other uses, change them to be uses of
// the global to simplify later code. This also deletes the store
// into GV.
- ReplaceUsesOfMallocWithGlobal(MI, GV);
-
+ ReplaceUsesOfMallocWithGlobal(CI, GV);
+
// Okay, at this point, there are no users of the malloc. Insert N
- // new mallocs at the same place as MI, and N globals.
+ // new mallocs at the same place as CI, and N globals.
std::vector<Value*> FieldGlobals;
- std::vector<MallocInst*> FieldMallocs;
+ std::vector<Value*> FieldMallocs;
for (unsigned FieldNo = 0, e = STy->getNumElements(); FieldNo != e;++FieldNo){
const Type *FieldTy = STy->getElementType(FieldNo);
- const Type *PFieldTy = PointerType::getUnqual(FieldTy);
+ const PointerType *PFieldTy = PointerType::getUnqual(FieldTy);
GlobalVariable *NGV =
- new GlobalVariable(PFieldTy, false, GlobalValue::InternalLinkage,
+ new GlobalVariable(*GV->getParent(),
+ PFieldTy, false, GlobalValue::InternalLinkage,
Constant::getNullValue(PFieldTy),
- GV->getName() + ".f" + utostr(FieldNo), GV,
+ GV->getName() + ".f" + Twine(FieldNo), GV,
GV->isThreadLocal());
FieldGlobals.push_back(NGV);
- MallocInst *NMI = new MallocInst(FieldTy, MI->getArraySize(),
- MI->getName() + ".f" + utostr(FieldNo),MI);
- FieldMallocs.push_back(NMI);
- new StoreInst(NMI, NGV, MI);
+ unsigned TypeSize = TD->getTypeAllocSize(FieldTy);
+ if (const StructType *ST = dyn_cast<StructType>(FieldTy))
+ TypeSize = TD->getStructLayout(ST)->getSizeInBytes();
+ const Type *IntPtrTy = TD->getIntPtrType(CI->getContext());
+ Value *NMI = CallInst::CreateMalloc(CI, IntPtrTy, FieldTy,
+ ConstantInt::get(IntPtrTy, TypeSize),
+ NElems,
+ CI->getName() + ".f" + Twine(FieldNo));
+ CallInst *NCI = dyn_cast<BitCastInst>(NMI) ?
+ extractMallocCallFromBitCast(NMI) : cast<CallInst>(NMI);
+ FieldMallocs.push_back(NCI);
+ new StoreInst(NMI, NGV, CI);
}
// The tricky aspect of this transformation is handling the case when malloc
// if (F1) { free(F1); F1 = 0; }
// if (F2) { free(F2); F2 = 0; }
// }
- Value *RunningOr = 0;
+ // The malloc can also fail if its argument is too large.
+ Constant *ConstantZero = ConstantInt::get(CI->getOperand(1)->getType(), 0);
+ Value *RunningOr = new ICmpInst(CI, ICmpInst::ICMP_SLT, CI->getOperand(1),
+ ConstantZero, "isneg");
for (unsigned i = 0, e = FieldMallocs.size(); i != e; ++i) {
- Value *Cond = new ICmpInst(ICmpInst::ICMP_EQ, FieldMallocs[i],
+ Value *Cond = new ICmpInst(CI, ICmpInst::ICMP_EQ, FieldMallocs[i],
Constant::getNullValue(FieldMallocs[i]->getType()),
- "isnull", MI);
- if (!RunningOr)
- RunningOr = Cond; // First seteq
- else
- RunningOr = BinaryOperator::CreateOr(RunningOr, Cond, "tmp", MI);
+ "isnull");
+ RunningOr = BinaryOperator::CreateOr(RunningOr, Cond, "tmp", CI);
}
// Split the basic block at the old malloc.
- BasicBlock *OrigBB = MI->getParent();
- BasicBlock *ContBB = OrigBB->splitBasicBlock(MI, "malloc_cont");
+ BasicBlock *OrigBB = CI->getParent();
+ BasicBlock *ContBB = OrigBB->splitBasicBlock(CI, "malloc_cont");
// Create the block to check the first condition. Put all these blocks at the
// end of the function as they are unlikely to be executed.
- BasicBlock *NullPtrBlock = BasicBlock::Create("malloc_ret_null",
+ BasicBlock *NullPtrBlock = BasicBlock::Create(OrigBB->getContext(),
+ "malloc_ret_null",
OrigBB->getParent());
// Remove the uncond branch from OrigBB to ContBB, turning it into a cond
// pointer, because some may be null while others are not.
for (unsigned i = 0, e = FieldGlobals.size(); i != e; ++i) {
Value *GVVal = new LoadInst(FieldGlobals[i], "tmp", NullPtrBlock);
- Value *Cmp = new ICmpInst(ICmpInst::ICMP_NE, GVVal,
+ Value *Cmp = new ICmpInst(*NullPtrBlock, ICmpInst::ICMP_NE, GVVal,
Constant::getNullValue(GVVal->getType()),
- "tmp", NullPtrBlock);
- BasicBlock *FreeBlock = BasicBlock::Create("free_it", OrigBB->getParent());
- BasicBlock *NextBlock = BasicBlock::Create("next", OrigBB->getParent());
- BranchInst::Create(FreeBlock, NextBlock, Cmp, NullPtrBlock);
+ "tmp");
+ BasicBlock *FreeBlock = BasicBlock::Create(Cmp->getContext(), "free_it",
+ OrigBB->getParent());
+ BasicBlock *NextBlock = BasicBlock::Create(Cmp->getContext(), "next",
+ OrigBB->getParent());
+ Instruction *BI = BranchInst::Create(FreeBlock, NextBlock,
+ Cmp, NullPtrBlock);
// Fill in FreeBlock.
- new FreeInst(GVVal, FreeBlock);
+ CallInst::CreateFree(GVVal, BI);
new StoreInst(Constant::getNullValue(GVVal->getType()), FieldGlobals[i],
FreeBlock);
BranchInst::Create(NextBlock, FreeBlock);
}
BranchInst::Create(ContBB, NullPtrBlock);
-
- // MI is no longer needed, remove it.
- MI->eraseFromParent();
+
+ // CI is no longer needed, remove it.
+ CI->eraseFromParent();
/// InsertedScalarizedLoads - As we process loads, if we can't immediately
/// update all uses of the load, keep track of what scalarized loads are
/// pointer global variable with a single value stored it that is a malloc or
/// cast of malloc.
static bool TryToOptimizeStoreOfMallocToGlobal(GlobalVariable *GV,
- MallocInst *MI,
+ CallInst *CI,
+ const Type *AllocTy,
Module::global_iterator &GVI,
- TargetData &TD) {
+ TargetData *TD) {
// If this is a malloc of an abstract type, don't touch it.
- if (!MI->getAllocatedType()->isSized())
+ if (!AllocTy->isSized())
return false;
-
+
// We can't optimize this global unless all uses of it are *known* to be
// of the malloc value, not of the null initializer value (consider a use
// that compares the global's value against zero to see if the malloc has
// happen after the malloc.
if (!AllUsesOfLoadedValueWillTrapIfNull(GV))
return false;
-
+
// We can't optimize this if the malloc itself is used in a complex way,
// for example, being stored into multiple globals. This allows the
// malloc to be stored into the specified global, loaded setcc'd, and
// for.
{
SmallPtrSet<PHINode*, 8> PHIs;
- if (!ValueIsOnlyUsedLocallyOrStoredToOneGlobal(MI, GV, PHIs))
+ if (!ValueIsOnlyUsedLocallyOrStoredToOneGlobal(CI, GV, PHIs))
return false;
- }
-
-
+ }
+
// If we have a global that is only initialized with a fixed size malloc,
// transform the program to use global memory instead of malloc'd memory.
// This eliminates dynamic allocation, avoids an indirection accessing the
// data, and exposes the resultant global to further GlobalOpt.
- if (ConstantInt *NElements = dyn_cast<ConstantInt>(MI->getArraySize())) {
- // Restrict this transformation to only working on small allocations
- // (2048 bytes currently), as we don't want to introduce a 16M global or
- // something.
- if (NElements->getZExtValue()*
- TD.getTypePaddedSize(MI->getAllocatedType()) < 2048) {
- GVI = OptimizeGlobalAddressOfMalloc(GV, MI);
- return true;
- }
- }
-
- // If the allocation is an array of structures, consider transforming this
- // into multiple malloc'd arrays, one for each field. This is basically
- // SRoA for malloc'd memory.
- const Type *AllocTy = MI->getAllocatedType();
-
- // If this is an allocation of a fixed size array of structs, analyze as a
- // variable size array. malloc [100 x struct],1 -> malloc struct, 100
- if (!MI->isArrayAllocation())
- if (const ArrayType *AT = dyn_cast<ArrayType>(AllocTy))
- AllocTy = AT->getElementType();
+ // We cannot optimize the malloc if we cannot determine malloc array size.
+ if (Value *NElems = getMallocArraySize(CI, TD, true)) {
+ if (ConstantInt *NElements = dyn_cast<ConstantInt>(NElems))
+ // Restrict this transformation to only working on small allocations
+ // (2048 bytes currently), as we don't want to introduce a 16M global or
+ // something.
+ if (TD &&
+ NElements->getZExtValue() * TD->getTypeAllocSize(AllocTy) < 2048) {
+ GVI = OptimizeGlobalAddressOfMalloc(GV, CI, AllocTy, NElems, TD);
+ return true;
+ }
- if (const StructType *AllocSTy = dyn_cast<StructType>(AllocTy)) {
- // This the structure has an unreasonable number of fields, leave it
- // alone.
- if (AllocSTy->getNumElements() <= 16 && AllocSTy->getNumElements() != 0 &&
- AllGlobalLoadUsesSimpleEnoughForHeapSRA(GV, MI)) {
+ // If the allocation is an array of structures, consider transforming this
+ // into multiple malloc'd arrays, one for each field. This is basically
+ // SRoA for malloc'd memory.
+
+ // If this is an allocation of a fixed size array of structs, analyze as a
+ // variable size array. malloc [100 x struct],1 -> malloc struct, 100
+ if (NElems == ConstantInt::get(CI->getOperand(1)->getType(), 1))
+ if (const ArrayType *AT = dyn_cast<ArrayType>(AllocTy))
+ AllocTy = AT->getElementType();
+
+ if (const StructType *AllocSTy = dyn_cast<StructType>(AllocTy)) {
+ // This the structure has an unreasonable number of fields, leave it
+ // alone.
+ if (AllocSTy->getNumElements() <= 16 && AllocSTy->getNumElements() != 0 &&
+ AllGlobalLoadUsesSimpleEnoughForHeapSRA(GV, CI)) {
+
+ // If this is a fixed size array, transform the Malloc to be an alloc of
+ // structs. malloc [100 x struct],1 -> malloc struct, 100
+ if (const ArrayType *AT =
+ dyn_cast<ArrayType>(getMallocAllocatedType(CI))) {
+ const Type *IntPtrTy = TD->getIntPtrType(CI->getContext());
+ unsigned TypeSize = TD->getStructLayout(AllocSTy)->getSizeInBytes();
+ Value *AllocSize = ConstantInt::get(IntPtrTy, TypeSize);
+ Value *NumElements = ConstantInt::get(IntPtrTy, AT->getNumElements());
+ Instruction *Malloc = CallInst::CreateMalloc(CI, IntPtrTy, AllocSTy,
+ AllocSize, NumElements,
+ CI->getName());
+ Instruction *Cast = new BitCastInst(Malloc, CI->getType(), "tmp", CI);
+ CI->replaceAllUsesWith(Cast);
+ CI->eraseFromParent();
+ CI = dyn_cast<BitCastInst>(Malloc) ?
+ extractMallocCallFromBitCast(Malloc) : cast<CallInst>(Malloc);
+ }
- // If this is a fixed size array, transform the Malloc to be an alloc of
- // structs. malloc [100 x struct],1 -> malloc struct, 100
- if (const ArrayType *AT = dyn_cast<ArrayType>(MI->getAllocatedType())) {
- MallocInst *NewMI =
- new MallocInst(AllocSTy,
- ConstantInt::get(Type::Int32Ty, AT->getNumElements()),
- "", MI);
- NewMI->takeName(MI);
- Value *Cast = new BitCastInst(NewMI, MI->getType(), "tmp", MI);
- MI->replaceAllUsesWith(Cast);
- MI->eraseFromParent();
- MI = NewMI;
+ GVI = PerformHeapAllocSRoA(GV, CI, getMallocArraySize(CI, TD, true),TD);
+ return true;
}
-
- GVI = PerformHeapAllocSRoA(GV, MI);
- return true;
}
}
// that only one value (besides its initializer) is ever stored to the global.
static bool OptimizeOnceStoredGlobal(GlobalVariable *GV, Value *StoredOnceVal,
Module::global_iterator &GVI,
- TargetData &TD) {
+ TargetData *TD) {
// Ignore no-op GEPs and bitcasts.
StoredOnceVal = StoredOnceVal->stripPointerCasts();
GV->getInitializer()->isNullValue()) {
if (Constant *SOVC = dyn_cast<Constant>(StoredOnceVal)) {
if (GV->getInitializer()->getType() != SOVC->getType())
- SOVC = ConstantExpr::getBitCast(SOVC, GV->getInitializer()->getType());
+ SOVC =
+ ConstantExpr::getBitCast(SOVC, GV->getInitializer()->getType());
// Optimize away any trapping uses of the loaded value.
if (OptimizeAwayTrappingUsesOfLoads(GV, SOVC))
return true;
- } else if (MallocInst *MI = dyn_cast<MallocInst>(StoredOnceVal)) {
- if (TryToOptimizeStoreOfMallocToGlobal(GV, MI, GVI, TD))
+ } else if (CallInst *CI = extractMallocCall(StoredOnceVal)) {
+ const Type* MallocType = getMallocAllocatedType(CI);
+ if (MallocType && TryToOptimizeStoreOfMallocToGlobal(GV, CI, MallocType,
+ GVI, TD))
return true;
}
}
const Type *GVElType = GV->getType()->getElementType();
// If GVElType is already i1, it is already shrunk. If the type of the GV is
- // an FP value or vector, don't do this optimization because a select between
- // them is very expensive and unlikely to lead to later simplification.
- if (GVElType == Type::Int1Ty || GVElType->isFloatingPoint() ||
- isa<VectorType>(GVElType))
+ // an FP value, pointer or vector, don't do this optimization because a select
+ // between them is very expensive and unlikely to lead to later
+ // simplification. In these cases, we typically end up with "cond ? v1 : v2"
+ // where v1 and v2 both require constant pool loads, a big loss.
+ if (GVElType == Type::getInt1Ty(GV->getContext()) ||
+ GVElType->isFloatingPointTy() ||
+ isa<PointerType>(GVElType) || isa<VectorType>(GVElType))
return false;
// Walk the use list of the global seeing if all the uses are load or store.
if (!isa<LoadInst>(I) && !isa<StoreInst>(I))
return false;
- DOUT << " *** SHRINKING TO BOOL: " << *GV;
+ DEBUG(dbgs() << " *** SHRINKING TO BOOL: " << *GV);
// Create the new global, initializing it to false.
- GlobalVariable *NewGV = new GlobalVariable(Type::Int1Ty, false,
- GlobalValue::InternalLinkage, ConstantInt::getFalse(),
+ GlobalVariable *NewGV = new GlobalVariable(Type::getInt1Ty(GV->getContext()),
+ false,
+ GlobalValue::InternalLinkage,
+ ConstantInt::getFalse(GV->getContext()),
GV->getName()+".b",
- (Module *)NULL,
GV->isThreadLocal());
GV->getParent()->getGlobalList().insert(GV, NewGV);
Constant *InitVal = GV->getInitializer();
- assert(InitVal->getType() != Type::Int1Ty && "No reason to shrink to bool!");
+ assert(InitVal->getType() != Type::getInt1Ty(GV->getContext()) &&
+ "No reason to shrink to bool!");
// If initialized to zero and storing one into the global, we can use a cast
// instead of a select to synthesize the desired value.
// Only do this if we weren't storing a loaded value.
Value *StoreVal;
if (StoringOther || SI->getOperand(0) == InitVal)
- StoreVal = ConstantInt::get(Type::Int1Ty, StoringOther);
+ StoreVal = ConstantInt::get(Type::getInt1Ty(GV->getContext()),
+ StoringOther);
else {
// Otherwise, we are storing a previously loaded copy. To do this,
// change the copy from copying the original value to just copying the
GV->removeDeadConstantUsers();
if (GV->use_empty()) {
- DOUT << "GLOBAL DEAD: " << *GV;
+ DEBUG(dbgs() << "GLOBAL DEAD: " << *GV);
GV->eraseFromParent();
++NumDeleted;
return true;
if (!AnalyzeGlobal(GV, GS, PHIUsers)) {
#if 0
- cerr << "Global: " << *GV;
- cerr << " isLoaded = " << GS.isLoaded << "\n";
- cerr << " StoredType = ";
+ DEBUG(dbgs() << "Global: " << *GV);
+ DEBUG(dbgs() << " isLoaded = " << GS.isLoaded << "\n");
+ DEBUG(dbgs() << " StoredType = ");
switch (GS.StoredType) {
- case GlobalStatus::NotStored: cerr << "NEVER STORED\n"; break;
- case GlobalStatus::isInitializerStored: cerr << "INIT STORED\n"; break;
- case GlobalStatus::isStoredOnce: cerr << "STORED ONCE\n"; break;
- case GlobalStatus::isStored: cerr << "stored\n"; break;
+ case GlobalStatus::NotStored: DEBUG(dbgs() << "NEVER STORED\n"); break;
+ case GlobalStatus::isInitializerStored: DEBUG(dbgs() << "INIT STORED\n");
+ break;
+ case GlobalStatus::isStoredOnce: DEBUG(dbgs() << "STORED ONCE\n"); break;
+ case GlobalStatus::isStored: DEBUG(dbgs() << "stored\n"); break;
}
if (GS.StoredType == GlobalStatus::isStoredOnce && GS.StoredOnceValue)
- cerr << " StoredOnceValue = " << *GS.StoredOnceValue << "\n";
+ DEBUG(dbgs() << " StoredOnceValue = " << *GS.StoredOnceValue << "\n");
if (GS.AccessingFunction && !GS.HasMultipleAccessingFunctions)
- cerr << " AccessingFunction = " << GS.AccessingFunction->getName()
- << "\n";
- cerr << " HasMultipleAccessingFunctions = "
- << GS.HasMultipleAccessingFunctions << "\n";
- cerr << " HasNonInstructionUser = " << GS.HasNonInstructionUser<<"\n";
- cerr << "\n";
+ DEBUG(dbgs() << " AccessingFunction = " << GS.AccessingFunction->getName()
+ << "\n");
+ DEBUG(dbgs() << " HasMultipleAccessingFunctions = "
+ << GS.HasMultipleAccessingFunctions << "\n");
+ DEBUG(dbgs() << " HasNonInstructionUser = "
+ << GS.HasNonInstructionUser<<"\n");
+ DEBUG(dbgs() << "\n");
#endif
// If this is a first class global and has only one accessing function
//
// NOTE: It doesn't make sense to promote non single-value types since we
// are just replacing static memory to stack memory.
+ //
+ // If the global is in different address space, don't bring it to stack.
if (!GS.HasMultipleAccessingFunctions &&
GS.AccessingFunction && !GS.HasNonInstructionUser &&
GV->getType()->getElementType()->isSingleValueType() &&
GS.AccessingFunction->getName() == "main" &&
- GS.AccessingFunction->hasExternalLinkage()) {
- DOUT << "LOCALIZING GLOBAL: " << *GV;
+ GS.AccessingFunction->hasExternalLinkage() &&
+ GV->getType()->getAddressSpace() == 0) {
+ DEBUG(dbgs() << "LOCALIZING GLOBAL: " << *GV);
Instruction* FirstI = GS.AccessingFunction->getEntryBlock().begin();
const Type* ElemTy = GV->getType()->getElementType();
// FIXME: Pass Global's alignment when globals have alignment
// If the global is never loaded (but may be stored to), it is dead.
// Delete it now.
if (!GS.isLoaded) {
- DOUT << "GLOBAL NEVER LOADED: " << *GV;
+ DEBUG(dbgs() << "GLOBAL NEVER LOADED: " << *GV);
// Delete any stores we can find to the global. We may not be able to
// make it completely dead though.
return Changed;
} else if (GS.StoredType <= GlobalStatus::isInitializerStored) {
- DOUT << "MARKING CONSTANT: " << *GV;
+ DEBUG(dbgs() << "MARKING CONSTANT: " << *GV);
GV->setConstant(true);
// Clean up any obviously simplifiable users now.
// If the global is dead now, just nuke it.
if (GV->use_empty()) {
- DOUT << " *** Marking constant allowed us to simplify "
- << "all users and delete global!\n";
+ DEBUG(dbgs() << " *** Marking constant allowed us to simplify "
+ << "all users and delete global!\n");
GV->eraseFromParent();
++NumDeleted;
}
++NumMarked;
return true;
} else if (!GV->getInitializer()->getType()->isSingleValueType()) {
- if (GlobalVariable *FirstNewGV = SRAGlobal(GV,
- getAnalysis<TargetData>())) {
- GVI = FirstNewGV; // Don't skip the newly produced globals!
- return true;
- }
+ if (TargetData *TD = getAnalysisIfAvailable<TargetData>())
+ if (GlobalVariable *FirstNewGV = SRAGlobal(GV, *TD)) {
+ GVI = FirstNewGV; // Don't skip the newly produced globals!
+ return true;
+ }
} else if (GS.StoredType == GlobalStatus::isStoredOnce) {
// If the initial value for the global was an undef value, and if only
// one other value was stored into it, we can just change the
CleanupConstantGlobalUsers(GV, GV->getInitializer());
if (GV->use_empty()) {
- DOUT << " *** Substituting initializer allowed us to "
- << "simplify all users and delete global!\n";
+ DEBUG(dbgs() << " *** Substituting initializer allowed us to "
+ << "simplify all users and delete global!\n");
GV->eraseFromParent();
++NumDeleted;
} else {
// Try to optimize globals based on the knowledge that only one value
// (besides its initializer) is ever stored to the global.
if (OptimizeOnceStoredGlobal(GV, GS.StoredOnceValue, GVI,
- getAnalysis<TargetData>()))
+ getAnalysisIfAvailable<TargetData>()))
return true;
// Otherwise, if the global was not a boolean, we can shrink it to be a
return false;
}
-/// OnlyCalledDirectly - Return true if the specified function is only called
-/// directly. In other words, its address is never taken.
-static bool OnlyCalledDirectly(Function *F) {
- for (Value::use_iterator UI = F->use_begin(), E = F->use_end(); UI != E;++UI){
- Instruction *User = dyn_cast<Instruction>(*UI);
- if (!User) return false;
- if (!isa<CallInst>(User) && !isa<InvokeInst>(User)) return false;
-
- // See if the function address is passed as an argument.
- for (User::op_iterator i = User->op_begin() + 1, e = User->op_end();
- i != e; ++i)
- if (*i == F) return false;
- }
- return true;
-}
-
/// ChangeCalleesToFastCall - Walk all of the direct calls of the specified
/// function, changing them to FastCC.
static void ChangeCalleesToFastCall(Function *F) {
// Optimize functions.
for (Module::iterator FI = M.begin(), E = M.end(); FI != E; ) {
Function *F = FI++;
+ // Functions without names cannot be referenced outside this module.
+ if (!F->hasName() && !F->isDeclaration())
+ F->setLinkage(GlobalValue::InternalLinkage);
F->removeDeadConstantUsers();
- if (F->use_empty() && (F->hasInternalLinkage() ||
- F->hasLinkOnceLinkage())) {
- M.getFunctionList().erase(F);
+ if (F->use_empty() && (F->hasLocalLinkage() || F->hasLinkOnceLinkage())) {
+ F->eraseFromParent();
Changed = true;
++NumFnDeleted;
- } else if (F->hasInternalLinkage()) {
+ } else if (F->hasLocalLinkage()) {
if (F->getCallingConv() == CallingConv::C && !F->isVarArg() &&
- OnlyCalledDirectly(F)) {
+ !F->hasAddressTaken()) {
// If this function has C calling conventions, is not a varargs
// function, and is only called directly, promote it to use the Fast
// calling convention.
}
if (F->getAttributes().hasAttrSomewhere(Attribute::Nest) &&
- OnlyCalledDirectly(F)) {
+ !F->hasAddressTaken()) {
// The function is not used by a trampoline intrinsic, so it is safe
// to remove the 'nest' attribute.
RemoveNestAttribute(F);
for (Module::global_iterator GVI = M.global_begin(), E = M.global_end();
GVI != E; ) {
GlobalVariable *GV = GVI++;
- if (!GV->isConstant() && GV->hasInternalLinkage() &&
+ // Global variables without names cannot be referenced outside this module.
+ if (!GV->hasName() && !GV->isDeclaration())
+ GV->setLinkage(GlobalValue::InternalLinkage);
+ // Simplify the initializer.
+ if (GV->hasInitializer())
+ if (ConstantExpr *CE = dyn_cast<ConstantExpr>(GV->getInitializer())) {
+ TargetData *TD = getAnalysisIfAvailable<TargetData>();
+ Constant *New = ConstantFoldConstantExpression(CE, TD);
+ if (New && New != CE)
+ GV->setInitializer(New);
+ }
+ // Do more involved optimizations if the global is internal.
+ if (!GV->isConstant() && GV->hasLocalLinkage() &&
GV->hasInitializer())
Changed |= ProcessInternalGlobal(GV, GVI);
}
if (!ATy) return 0;
const StructType *STy = dyn_cast<StructType>(ATy->getElementType());
if (!STy || STy->getNumElements() != 2 ||
- STy->getElementType(0) != Type::Int32Ty) return 0;
+ !STy->getElementType(0)->isIntegerTy(32)) return 0;
const PointerType *PFTy = dyn_cast<PointerType>(STy->getElementType(1));
if (!PFTy) return 0;
const FunctionType *FTy = dyn_cast<FunctionType>(PFTy->getElementType());
- if (!FTy || FTy->getReturnType() != Type::VoidTy || FTy->isVarArg() ||
- FTy->getNumParams() != 0)
+ if (!FTy || !FTy->getReturnType()->isVoidTy() ||
+ FTy->isVarArg() || FTy->getNumParams() != 0)
return 0;
// Verify that the initializer is simple enough for us to handle.
- if (!I->hasInitializer()) return 0;
+ if (!I->hasDefinitiveInitializer()) return 0;
ConstantArray *CA = dyn_cast<ConstantArray>(I->getInitializer());
if (!CA) return 0;
for (User::op_iterator i = CA->op_begin(), e = CA->op_end(); i != e; ++i)
const std::vector<Function*> &Ctors) {
// If we made a change, reassemble the initializer list.
std::vector<Constant*> CSVals;
- CSVals.push_back(ConstantInt::get(Type::Int32Ty, 65535));
+ CSVals.push_back(ConstantInt::get(Type::getInt32Ty(GCL->getContext()),65535));
CSVals.push_back(0);
// Create the new init list.
if (Ctors[i]) {
CSVals[1] = Ctors[i];
} else {
- const Type *FTy = FunctionType::get(Type::VoidTy,
- std::vector<const Type*>(), false);
+ const Type *FTy = FunctionType::get(Type::getVoidTy(GCL->getContext()),
+ false);
const PointerType *PFTy = PointerType::getUnqual(FTy);
CSVals[1] = Constant::getNullValue(PFTy);
- CSVals[0] = ConstantInt::get(Type::Int32Ty, 2147483647);
+ CSVals[0] = ConstantInt::get(Type::getInt32Ty(GCL->getContext()),
+ 2147483647);
}
- CAList.push_back(ConstantStruct::get(CSVals));
+ CAList.push_back(ConstantStruct::get(GCL->getContext(), CSVals, false));
}
// Create the array initializer.
const Type *StructTy =
- cast<ArrayType>(GCL->getType()->getElementType())->getElementType();
- Constant *CA = ConstantArray::get(ArrayType::get(StructTy, CAList.size()),
- CAList);
+ cast<ArrayType>(GCL->getType()->getElementType())->getElementType();
+ Constant *CA = ConstantArray::get(ArrayType::get(StructTy,
+ CAList.size()), CAList);
// If we didn't change the number of elements, don't create a new GV.
if (CA->getType() == GCL->getInitializer()->getType()) {
// Create the new global and insert it next to the existing list.
GlobalVariable *NGV = new GlobalVariable(CA->getType(), GCL->isConstant(),
GCL->getLinkage(), CA, "",
- (Module *)NULL,
GCL->isThreadLocal());
GCL->getParent()->getGlobalList().insert(GCL, NGV);
NGV->takeName(GCL);
/// we punt. We basically just support direct accesses to globals and GEP's of
/// globals. This should be kept up to date with CommitValueTo.
static bool isSimpleEnoughPointerToCommit(Constant *C) {
- if (GlobalVariable *GV = dyn_cast<GlobalVariable>(C)) {
- if (!GV->hasExternalLinkage() && !GV->hasInternalLinkage())
- return false; // do not allow weak/linkonce/dllimport/dllexport linkage.
- return !GV->isDeclaration(); // reject external globals.
- }
+ // Conservatively, avoid aggregate types. This is because we don't
+ // want to worry about them partially overlapping other stores.
+ if (!cast<PointerType>(C->getType())->getElementType()->isSingleValueType())
+ return false;
+
+ if (GlobalVariable *GV = dyn_cast<GlobalVariable>(C))
+ // Do not allow weak/linkonce/dllimport/dllexport linkage or
+ // external globals.
+ return GV->hasDefinitiveInitializer();
+
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C))
// Handle a constantexpr gep.
if (CE->getOpcode() == Instruction::GetElementPtr &&
- isa<GlobalVariable>(CE->getOperand(0))) {
+ isa<GlobalVariable>(CE->getOperand(0)) &&
+ cast<GEPOperator>(CE)->isInBounds()) {
GlobalVariable *GV = cast<GlobalVariable>(CE->getOperand(0));
- if (!GV->hasExternalLinkage() && !GV->hasInternalLinkage())
- return false; // do not allow weak/linkonce/dllimport/dllexport linkage.
- return GV->hasInitializer() &&
- ConstantFoldLoadThroughGEPConstantExpr(GV->getInitializer(), CE);
+ // Do not allow weak/linkonce/dllimport/dllexport linkage or
+ // external globals.
+ if (!GV->hasDefinitiveInitializer())
+ return false;
+
+ // The first index must be zero.
+ ConstantInt *CI = dyn_cast<ConstantInt>(*next(CE->op_begin()));
+ if (!CI || !CI->isZero()) return false;
+
+ // The remaining indices must be compile-time known integers within the
+ // notional bounds of the corresponding static array types.
+ if (!CE->isGEPWithNoNotionalOverIndexing())
+ return false;
+
+ return ConstantFoldLoadThroughGEPConstantExpr(GV->getInitializer(), CE);
}
return false;
}
return Val;
}
+ std::vector<Constant*> Elts;
if (const StructType *STy = dyn_cast<StructType>(Init->getType())) {
- std::vector<Constant*> Elts;
// Break up the constant into its elements.
if (ConstantStruct *CS = dyn_cast<ConstantStruct>(Init)) {
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i)
Elts.push_back(UndefValue::get(STy->getElementType(i)));
} else {
- assert(0 && "This code is out of sync with "
+ llvm_unreachable("This code is out of sync with "
" ConstantFoldLoadThroughGEPConstantExpr");
}
Elts[Idx] = EvaluateStoreInto(Elts[Idx], Val, Addr, OpNo+1);
// Return the modified struct.
- return ConstantStruct::get(&Elts[0], Elts.size(), STy->isPacked());
+ return ConstantStruct::get(Init->getContext(), &Elts[0], Elts.size(),
+ STy->isPacked());
} else {
ConstantInt *CI = cast<ConstantInt>(Addr->getOperand(OpNo));
- const ArrayType *ATy = cast<ArrayType>(Init->getType());
+ const SequentialType *InitTy = cast<SequentialType>(Init->getType());
+ uint64_t NumElts;
+ if (const ArrayType *ATy = dyn_cast<ArrayType>(InitTy))
+ NumElts = ATy->getNumElements();
+ else
+ NumElts = cast<VectorType>(InitTy)->getNumElements();
+
+
// Break up the array into elements.
- std::vector<Constant*> Elts;
if (ConstantArray *CA = dyn_cast<ConstantArray>(Init)) {
for (User::op_iterator i = CA->op_begin(), e = CA->op_end(); i != e; ++i)
Elts.push_back(cast<Constant>(*i));
+ } else if (ConstantVector *CV = dyn_cast<ConstantVector>(Init)) {
+ for (User::op_iterator i = CV->op_begin(), e = CV->op_end(); i != e; ++i)
+ Elts.push_back(cast<Constant>(*i));
} else if (isa<ConstantAggregateZero>(Init)) {
- Constant *Elt = Constant::getNullValue(ATy->getElementType());
- Elts.assign(ATy->getNumElements(), Elt);
- } else if (isa<UndefValue>(Init)) {
- Constant *Elt = UndefValue::get(ATy->getElementType());
- Elts.assign(ATy->getNumElements(), Elt);
+ Elts.assign(NumElts, Constant::getNullValue(InitTy->getElementType()));
} else {
- assert(0 && "This code is out of sync with "
+ assert(isa<UndefValue>(Init) && "This code is out of sync with "
" ConstantFoldLoadThroughGEPConstantExpr");
+ Elts.assign(NumElts, UndefValue::get(InitTy->getElementType()));
}
- assert(CI->getZExtValue() < ATy->getNumElements());
+ assert(CI->getZExtValue() < NumElts);
Elts[CI->getZExtValue()] =
EvaluateStoreInto(Elts[CI->getZExtValue()], Val, Addr, OpNo+1);
- return ConstantArray::get(ATy, Elts);
+
+ if (isa<ArrayType>(Init->getType()))
+ return ConstantArray::get(cast<ArrayType>(InitTy), Elts);
+ else
+ return ConstantVector::get(&Elts[0], Elts.size());
}
}
GV->setInitializer(Val);
return;
}
-
+
ConstantExpr *CE = cast<ConstantExpr>(Addr);
GlobalVariable *GV = cast<GlobalVariable>(CE->getOperand(0));
-
- Constant *Init = GV->getInitializer();
- Init = EvaluateStoreInto(Init, Val, CE, 2);
- GV->setInitializer(Init);
+ GV->setInitializer(EvaluateStoreInto(GV->getInitializer(), Val, CE, 2));
}
/// ComputeLoadResult - Return the value that would be computed by a load from
// Access it.
if (GlobalVariable *GV = dyn_cast<GlobalVariable>(P)) {
- if (GV->hasInitializer())
+ if (GV->hasDefinitiveInitializer())
return GV->getInitializer();
return 0;
}
if (CE->getOpcode() == Instruction::GetElementPtr &&
isa<GlobalVariable>(CE->getOperand(0))) {
GlobalVariable *GV = cast<GlobalVariable>(CE->getOperand(0));
- if (GV->hasInitializer())
+ if (GV->hasDefinitiveInitializer())
return ConstantFoldLoadThroughGEPConstantExpr(GV->getInitializer(), CE);
}
/// successful, false if we can't evaluate it. ActualArgs contains the formal
/// arguments for the function.
static bool EvaluateFunction(Function *F, Constant *&RetVal,
- const std::vector<Constant*> &ActualArgs,
+ const SmallVectorImpl<Constant*> &ActualArgs,
std::vector<Function*> &CallStack,
DenseMap<Constant*, Constant*> &MutatedMemory,
std::vector<GlobalVariable*> &AllocaTmps) {
getVal(Values, CI->getOperand(0)),
CI->getType());
} else if (SelectInst *SI = dyn_cast<SelectInst>(CurInst)) {
- InstResult = ConstantExpr::getSelect(getVal(Values, SI->getOperand(0)),
+ InstResult =
+ ConstantExpr::getSelect(getVal(Values, SI->getOperand(0)),
getVal(Values, SI->getOperand(1)),
getVal(Values, SI->getOperand(2)));
} else if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(CurInst)) {
for (User::op_iterator i = GEP->op_begin() + 1, e = GEP->op_end();
i != e; ++i)
GEPOps.push_back(getVal(Values, *i));
- InstResult = ConstantExpr::getGetElementPtr(P, &GEPOps[0], GEPOps.size());
+ InstResult = cast<GEPOperator>(GEP)->isInBounds() ?
+ ConstantExpr::getInBoundsGetElementPtr(P, &GEPOps[0], GEPOps.size()) :
+ ConstantExpr::getGetElementPtr(P, &GEPOps[0], GEPOps.size());
} else if (LoadInst *LI = dyn_cast<LoadInst>(CurInst)) {
if (LI->isVolatile()) return false; // no volatile accesses.
InstResult = ComputeLoadResult(getVal(Values, LI->getOperand(0)),
AI->getName()));
InstResult = AllocaTmps.back();
} else if (CallInst *CI = dyn_cast<CallInst>(CurInst)) {
+
+ // Debug info can safely be ignored here.
+ if (isa<DbgInfoIntrinsic>(CI)) {
+ ++CurInst;
+ continue;
+ }
+
// Cannot handle inline asm.
if (isa<InlineAsm>(CI->getOperand(0))) return false;
Function *Callee = dyn_cast<Function>(getVal(Values, CI->getOperand(0)));
if (!Callee) return false; // Cannot resolve.
- std::vector<Constant*> Formals;
+ SmallVector<Constant*, 8> Formals;
for (User::op_iterator i = CI->op_begin() + 1, e = CI->op_end();
i != e; ++i)
Formals.push_back(getVal(Values, *i));
-
+
if (Callee->isDeclaration()) {
// If this is a function we can constant fold, do it.
- if (Constant *C = ConstantFoldCall(Callee, &Formals[0],
+ if (Constant *C = ConstantFoldCall(Callee, Formals.data(),
Formals.size())) {
InstResult = C;
} else {
return false;
Constant *RetVal;
-
// Execute the call, if successful, use the return value.
if (!EvaluateFunction(Callee, RetVal, Formals, CallStack,
MutatedMemory, AllocaTmps))
dyn_cast<ConstantInt>(getVal(Values, SI->getCondition()));
if (!Val) return false; // Cannot determine.
NewBB = SI->getSuccessor(SI->findCaseValue(Val));
+ } else if (IndirectBrInst *IBI = dyn_cast<IndirectBrInst>(CurInst)) {
+ Value *Val = getVal(Values, IBI->getAddress())->stripPointerCasts();
+ if (BlockAddress *BA = dyn_cast<BlockAddress>(Val))
+ NewBB = BA->getBasicBlock();
+ else
+ return false; // Cannot determine.
} else if (ReturnInst *RI = dyn_cast<ReturnInst>(CurInst)) {
if (RI->getNumOperands())
RetVal = getVal(Values, RI->getOperand(0));
// Call the function.
Constant *RetValDummy;
- bool EvalSuccess = EvaluateFunction(F, RetValDummy, std::vector<Constant*>(),
- CallStack, MutatedMemory, AllocaTmps);
+ bool EvalSuccess = EvaluateFunction(F, RetValDummy,
+ SmallVector<Constant*, 0>(), CallStack,
+ MutatedMemory, AllocaTmps);
if (EvalSuccess) {
// We succeeded at evaluation: commit the result.
- DOUT << "FULLY EVALUATED GLOBAL CTOR FUNCTION '"
- << F->getName() << "' to " << MutatedMemory.size()
- << " stores.\n";
+ DEBUG(dbgs() << "FULLY EVALUATED GLOBAL CTOR FUNCTION '"
+ << F->getName() << "' to " << MutatedMemory.size()
+ << " stores.\n");
for (DenseMap<Constant*, Constant*>::iterator I = MutatedMemory.begin(),
E = MutatedMemory.end(); I != E; ++I)
CommitValueTo(I->second, I->first);
return true;
}
-bool GlobalOpt::ResolveAliases(Module &M) {
+bool GlobalOpt::OptimizeGlobalAliases(Module &M) {
bool Changed = false;
for (Module::alias_iterator I = M.alias_begin(), E = M.alias_end();
- I != E; ++I) {
- if (I->use_empty())
+ I != E;) {
+ Module::alias_iterator J = I++;
+ // Aliases without names cannot be referenced outside this module.
+ if (!J->hasName() && !J->isDeclaration())
+ J->setLinkage(GlobalValue::InternalLinkage);
+ // If the aliasee may change at link time, nothing can be done - bail out.
+ if (J->mayBeOverridden())
continue;
- if (const GlobalValue *GV = I->resolveAliasedGlobal())
- if (GV != I) {
- I->replaceAllUsesWith(const_cast<GlobalValue*>(GV));
- Changed = true;
- }
+ Constant *Aliasee = J->getAliasee();
+ GlobalValue *Target = cast<GlobalValue>(Aliasee->stripPointerCasts());
+ Target->removeDeadConstantUsers();
+ bool hasOneUse = Target->hasOneUse() && Aliasee->hasOneUse();
+
+ // Make all users of the alias use the aliasee instead.
+ if (!J->use_empty()) {
+ J->replaceAllUsesWith(Aliasee);
+ ++NumAliasesResolved;
+ Changed = true;
+ }
+
+ // If the alias is externally visible, we may still be able to simplify it.
+ if (!J->hasLocalLinkage()) {
+ // If the aliasee has internal linkage, give it the name and linkage
+ // of the alias, and delete the alias. This turns:
+ // define internal ... @f(...)
+ // @a = alias ... @f
+ // into:
+ // define ... @a(...)
+ if (!Target->hasLocalLinkage())
+ continue;
+
+ // Do not perform the transform if multiple aliases potentially target the
+ // aliasee. This check also ensures that it is safe to replace the section
+ // and other attributes of the aliasee with those of the alias.
+ if (!hasOneUse)
+ continue;
+
+ // Give the aliasee the name, linkage and other attributes of the alias.
+ Target->takeName(J);
+ Target->setLinkage(J->getLinkage());
+ Target->GlobalValue::copyAttributesFrom(J);
+ }
+
+ // Delete the alias.
+ M.getAliasList().erase(J);
+ ++NumAliasesRemoved;
+ Changed = true;
}
return Changed;
LocalChange |= OptimizeGlobalVars(M);
// Resolve aliases, when possible.
- LocalChange |= ResolveAliases(M);
+ LocalChange |= OptimizeGlobalAliases(M);
Changed |= LocalChange;
}