#include "llvm/Module.h"
#include "llvm/Pass.h"
#include "llvm/Analysis/ConstantFolding.h"
+#include "llvm/Analysis/MallocFreeHelper.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"
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) {}
/// 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;
} else if (isa<ConstantAggregateZero>(Agg)) {
if (const StructType *STy = dyn_cast<StructType>(Agg->getType())) {
if (IdxV < STy->getNumElements())
- return Context.getNullValue(STy->getElementType(IdxV));
+ return Constant::getNullValue(STy->getElementType(IdxV));
} else if (const SequentialType *STy =
dyn_cast<SequentialType>(Agg->getType())) {
- return Context.getNullValue(STy->getElementType());
+ return Constant::getNullValue(STy->getElementType());
}
} else if (isa<UndefValue>(Agg)) {
if (const StructType *STy = dyn_cast<StructType>(Agg->getType())) {
if (CE->getOpcode() == Instruction::GetElementPtr) {
Constant *SubInit = 0;
if (Init)
- SubInit = ConstantFoldLoadThroughGEPConstantExpr(Init, CE, Context);
+ SubInit = ConstantFoldLoadThroughGEPConstantExpr(Init, CE);
Changed |= CleanupConstantGlobalUsers(CE, SubInit, Context);
} else if (CE->getOpcode() == Instruction::BitCast &&
isa<PointerType>(CE->getType())) {
ConstantExpr *CE =
dyn_cast_or_null<ConstantExpr>(ConstantFoldInstruction(GEP, Context));
if (Init && CE && CE->getOpcode() == Instruction::GetElementPtr)
- SubInit = ConstantFoldLoadThroughGEPConstantExpr(Init, CE, Context);
+ SubInit = ConstantFoldLoadThroughGEPConstantExpr(Init, CE);
}
Changed |= CleanupConstantGlobalUsers(GEP, SubInit, Context);
// 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)
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(Context), i),
Context);
assert(In && "Couldn't get element of initializer?");
GlobalVariable *NGV = new GlobalVariable(Context,
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(Context), i),
Context);
assert(In && "Couldn't get element of initializer?");
if (NewGlobals.empty())
return 0;
- DOUT << "PERFORMING GLOBAL SRA ON: " << *GV;
+ DEBUG(errs() << "PERFORMING GLOBAL SRA ON: " << *GV);
- Constant *NullInt = Context.getNullValue(Type::Int32Ty);
+ Constant *NullInt = Constant::getNullValue(Type::getInt32Ty(Context));
// Loop over all of the uses of the global, replacing the constantexpr geps,
// with smaller constantexpr geps or direct references.
}
if (Changed) {
- DOUT << "OPTIMIZED LOADS FROM STORED ONCE POINTER: " << *GV;
+ DEBUG(errs() << "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(errs() << " *** GLOBAL NOW DEAD!\n");
CleanupConstantGlobalUsers(GV, 0, Context);
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,
- LLVMContext &Context) {
- DOUT << "PROMOTING MALLOC GLOBAL: " << *GV << " MALLOC = " << *MI;
- ConstantInt *NElements = cast<ConstantInt>(MI->getArraySize());
-
+ CallInst *CI,
+ BitCastInst *BCI,
+ LLVMContext &Context,
+ TargetData* TD) {
+ DEBUG(errs() << "PROMOTING MALLOC GLOBAL: " << *GV
+ << " CALL = " << *CI << " BCI = " << *BCI << '\n');
+
+ const Type *IntPtrTy = TD->getIntPtrType(Context);
+
+ Value* ArraySize = getMallocArraySize(CI, Context, TD);
+ assert(ArraySize && "not a malloc whose array size can be determined");
+ ConstantInt *NElements = cast<ConstantInt>(ArraySize);
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(),
+ Type *NewTy = ArrayType::get(getMallocAllocatedType(CI),
NElements->getZExtValue());
- MallocInst *NewMI =
- new MallocInst(NewTy, Context.getNullValue(Type::Int32Ty),
- MI->getAlignment(), MI->getName(), MI);
+ Value* NewM = CallInst::CreateMalloc(CI, IntPtrTy, NewTy);
+ Instruction* NewMI = cast<Instruction>(NewM);
Value* Indices[2];
- Indices[0] = Indices[1] = Context.getNullValue(Type::Int32Ty);
+ Indices[0] = Indices[1] = Constant::getNullValue(IntPtrTy);
Value *NewGEP = GetElementPtrInst::Create(NewMI, Indices, Indices + 2,
- NewMI->getName()+".el0", MI);
- MI->replaceAllUsesWith(NewGEP);
- MI->eraseFromParent();
- MI = NewMI;
+ NewMI->getName()+".el0", CI);
+ BCI->replaceAllUsesWith(NewGEP);
+ BCI->eraseFromParent();
+ CI->eraseFromParent();
+ BCI = cast<BitCastInst>(NewMI);
+ CI = extractMallocCallFromBitCast(NewMI);
}
// Create the new global variable. The contents of the malloc'd memory is
// undefined, so initialize with an undef value.
- // 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.
- Constant *Init = UndefValue::get(MI->getAllocatedType());
+ const Type *MAT = getMallocAllocatedType(CI);
+ Constant *Init = UndefValue::get(MAT);
GlobalVariable *NewGV = new GlobalVariable(*GV->getParent(),
- MI->getAllocatedType(), false,
+ MAT, false,
GlobalValue::InternalLinkage, Init,
GV->getName()+".body",
GV,
GV->isThreadLocal());
// Anything that used the malloc now uses the global directly.
- MI->replaceAllUsesWith(NewGV);
+ BCI->replaceAllUsesWith(NewGV);
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(Context, Type::Int1Ty, false,
+ new GlobalVariable(Context, Type::getInt1Ty(Context), false,
GlobalValue::InternalLinkage,
- Context.getFalse(), GV->getName()+".init",
+ ConstantInt::getFalse(Context), GV->getName()+".init",
GV->isThreadLocal());
bool InitBoolUsed = false;
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()) {
+ switch (ICI->getPredicate()) {
default: llvm_unreachable("Unknown ICmp Predicate!");
case ICmpInst::ICMP_ULT:
case ICmpInst::ICMP_SLT:
- LV = Context.getFalse(); // X < null -> always false
+ LV = ConstantInt::getFalse(Context); // X < null -> always false
break;
case ICmpInst::ICMP_ULE:
case ICmpInst::ICMP_SLE:
case ICmpInst::ICMP_EQ:
- LV = BinaryOperator::CreateNot(Context, 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(Context.getTrue(), InitBool, SI);
+ new StoreInst(ConstantInt::getTrue(Context), InitBool, SI);
SI->eraseFromParent();
}
// Now the GV is dead, nuke it and the malloc.
GV->eraseFromParent();
- MI->eraseFromParent();
+ 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
/// 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;
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.
Context);
Value *New = new ICmpInst(SCI, SCI->getPredicate(), NPtr,
- Context.getNullValue(NPtr->getType()),
+ Constant::getNullValue(NPtr->getType()),
SCI->getName());
SCI->replaceAllUsesWith(New);
SCI->eraseFromParent();
}
}
-/// 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,
- LLVMContext &Context){
- DOUT << "SROA HEAP ALLOC: " << *GV << " MALLOC = " << *MI;
- const StructType *STy = cast<StructType>(MI->getAllocatedType());
+static GlobalVariable *PerformHeapAllocSRoA(GlobalVariable *GV,
+ CallInst *CI, BitCastInst* BCI,
+ LLVMContext &Context,
+ TargetData *TD){
+ DEBUG(errs() << "SROA HEAP ALLOC: " << *GV << " MALLOC CALL = " << *CI
+ << " BITCAST = " << *BCI << '\n');
+ const Type* MAT = getMallocAllocatedType(CI);
+ const StructType *STy = cast<StructType>(MAT);
+ Value* ArraySize = getMallocArraySize(CI, Context, TD);
+ assert(ArraySize && "not a malloc whose array size can be determined");
// 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(BCI, 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(*GV->getParent(),
PFieldTy, false, GlobalValue::InternalLinkage,
- Context.getNullValue(PFieldTy),
+ Constant::getNullValue(PFieldTy),
GV->getName() + ".f" + Twine(FieldNo), GV,
GV->isThreadLocal());
FieldGlobals.push_back(NGV);
- MallocInst *NMI = new MallocInst(FieldTy, MI->getArraySize(),
- MI->getName() + ".f" + Twine(FieldNo), MI);
+ Value *NMI = CallInst::CreateMalloc(CI, TD->getIntPtrType(Context),
+ FieldTy, ArraySize,
+ BCI->getName() + ".f" + Twine(FieldNo));
FieldMallocs.push_back(NMI);
- new StoreInst(NMI, NGV, MI);
+ new StoreInst(NMI, NGV, BCI);
}
// The tricky aspect of this transformation is handling the case when malloc
// }
Value *RunningOr = 0;
for (unsigned i = 0, e = FieldMallocs.size(); i != e; ++i) {
- Value *Cond = new ICmpInst(MI, ICmpInst::ICMP_EQ, FieldMallocs[i],
- Context.getNullValue(FieldMallocs[i]->getType()),
+ Value *Cond = new ICmpInst(BCI, ICmpInst::ICMP_EQ, FieldMallocs[i],
+ Constant::getNullValue(FieldMallocs[i]->getType()),
"isnull");
if (!RunningOr)
RunningOr = Cond; // First seteq
else
- RunningOr = BinaryOperator::CreateOr(RunningOr, Cond, "tmp", MI);
+ RunningOr = BinaryOperator::CreateOr(RunningOr, Cond, "tmp", BCI);
}
// Split the basic block at the old malloc.
- BasicBlock *OrigBB = MI->getParent();
- BasicBlock *ContBB = OrigBB->splitBasicBlock(MI, "malloc_cont");
+ BasicBlock *OrigBB = BCI->getParent();
+ BasicBlock *ContBB = OrigBB->splitBasicBlock(BCI, "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(Context, "malloc_ret_null",
OrigBB->getParent());
// Remove the uncond branch from OrigBB to ContBB, turning it into a cond
for (unsigned i = 0, e = FieldGlobals.size(); i != e; ++i) {
Value *GVVal = new LoadInst(FieldGlobals[i], "tmp", NullPtrBlock);
Value *Cmp = new ICmpInst(*NullPtrBlock, ICmpInst::ICMP_NE, GVVal,
- Context.getNullValue(GVVal->getType()),
+ Constant::getNullValue(GVVal->getType()),
"tmp");
- BasicBlock *FreeBlock = BasicBlock::Create("free_it", OrigBB->getParent());
- BasicBlock *NextBlock = BasicBlock::Create("next", OrigBB->getParent());
- BranchInst::Create(FreeBlock, NextBlock, Cmp, NullPtrBlock);
+ BasicBlock *FreeBlock = BasicBlock::Create(Context, "free_it",
+ OrigBB->getParent());
+ BasicBlock *NextBlock = BasicBlock::Create(Context, "next",
+ OrigBB->getParent());
+ Instruction *BI = BranchInst::Create(FreeBlock, NextBlock,
+ Cmp, NullPtrBlock);
// Fill in FreeBlock.
- new FreeInst(GVVal, FreeBlock);
- new StoreInst(Context.getNullValue(GVVal->getType()), FieldGlobals[i],
+ 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 and BCI are no longer needed, remove them.
+ BCI->eraseFromParent();
+ 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
// Insert a store of null into each global.
for (unsigned i = 0, e = FieldGlobals.size(); i != e; ++i) {
const PointerType *PT = cast<PointerType>(FieldGlobals[i]->getType());
- Constant *Null = Context.getNullValue(PT->getElementType());
+ Constant *Null = Constant::getNullValue(PT->getElementType());
new StoreInst(Null, FieldGlobals[i], SI);
}
// Erase the original store.
/// 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,
+ BitCastInst *BCI,
Module::global_iterator &GVI,
- TargetData &TD,
+ TargetData *TD,
LLVMContext &Context) {
+ // If we can't figure out the type being malloced, then we can't optimize.
+ const Type *AllocTy = getMallocAllocatedType(CI);
+ assert(AllocTy);
+
// 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(BCI, 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.getTypeAllocSize(MI->getAllocatedType()) < 2048) {
- GVI = OptimizeGlobalAddressOfMalloc(GV, MI, Context);
- return true;
- }
- }
+ Value *NElems = getMallocArraySize(CI, Context, TD);
+ // We cannot optimize the malloc if we cannot determine malloc array size.
+ if (NElems) {
+ 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, BCI, Context, TD);
+ 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();
-
- 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 (!isArrayMalloc(CI, Context, TD))
+ 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, BCI)) {
+
+ // 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))) {
+ Value* NumElements = ConstantInt::get(Type::getInt32Ty(Context),
+ AT->getNumElements());
+ Value* NewMI = CallInst::CreateMalloc(CI, TD->getIntPtrType(Context),
+ AllocSTy, NumElements,
+ BCI->getName());
+ Value *Cast = new BitCastInst(NewMI, getMallocType(CI), "tmp", CI);
+ BCI->replaceAllUsesWith(Cast);
+ BCI->eraseFromParent();
+ CI->eraseFromParent();
+ BCI = cast<BitCastInst>(NewMI);
+ CI = extractMallocCallFromBitCast(NewMI);
+ }
- // 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, BCI, Context, TD);
+ return true;
}
-
- GVI = PerformHeapAllocSRoA(GV, MI, Context);
- 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, LLVMContext &Context) {
+ TargetData *TD, LLVMContext &Context) {
// Ignore no-op GEPs and bitcasts.
StoredOnceVal = StoredOnceVal->stripPointerCasts();
// Optimize away any trapping uses of the loaded value.
if (OptimizeAwayTrappingUsesOfLoads(GV, SOVC, Context))
return true;
- } else if (MallocInst *MI = dyn_cast<MallocInst>(StoredOnceVal)) {
- if (TryToOptimizeStoreOfMallocToGlobal(GV, MI, GVI, TD, Context))
- return true;
+ } else if (CallInst *CI = extractMallocCall(StoredOnceVal)) {
+ if (getMallocAllocatedType(CI)) {
+ BitCastInst* BCI = NULL;
+ for (Value::use_iterator UI = CI->use_begin(), E = CI->use_end();
+ UI != E; )
+ BCI = dyn_cast<BitCastInst>(cast<Instruction>(*UI++));
+ if (BCI &&
+ TryToOptimizeStoreOfMallocToGlobal(GV, CI, BCI, GVI, TD, Context))
+ return true;
+ }
}
}
// 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::Int1Ty || GVElType->isFloatingPoint() ||
+ if (GVElType == Type::getInt1Ty(Context) || GVElType->isFloatingPoint() ||
isa<PointerType>(GVElType) || isa<VectorType>(GVElType))
return false;
if (!isa<LoadInst>(I) && !isa<StoreInst>(I))
return false;
- DOUT << " *** SHRINKING TO BOOL: " << *GV;
+ DEBUG(errs() << " *** SHRINKING TO BOOL: " << *GV);
// Create the new global, initializing it to false.
- GlobalVariable *NewGV = new GlobalVariable(Context, Type::Int1Ty, false,
- GlobalValue::InternalLinkage, Context.getFalse(),
+ GlobalVariable *NewGV = new GlobalVariable(Context,
+ Type::getInt1Ty(Context), false,
+ GlobalValue::InternalLinkage, ConstantInt::getFalse(Context),
GV->getName()+".b",
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(Context) &&
+ "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(Context), 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(errs() << "GLOBAL DEAD: " << *GV);
GV->eraseFromParent();
++NumDeleted;
return true;
GS.AccessingFunction->getName() == "main" &&
GS.AccessingFunction->hasExternalLinkage() &&
GV->getType()->getAddressSpace() == 0) {
- DOUT << "LOCALIZING GLOBAL: " << *GV;
+ DEBUG(errs() << "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(errs() << "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(errs() << "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(errs() << " *** 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>(),
- GV->getContext())) {
- GVI = FirstNewGV; // Don't skip the newly produced globals!
- return true;
- }
+ if (TargetData *TD = getAnalysisIfAvailable<TargetData>())
+ if (GlobalVariable *FirstNewGV = SRAGlobal(GV, *TD,
+ GV->getContext())) {
+ 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
GV->getContext());
if (GV->use_empty()) {
- DOUT << " *** Substituting initializer allowed us to "
- << "simplify all users and delete global!\n";
+ DEBUG(errs() << " *** 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>(), GV->getContext()))
+ getAnalysisIfAvailable<TargetData>(),
+ GV->getContext()))
return true;
// Otherwise, if the global was not a boolean, we can shrink it to be a
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) != Type::getInt32Ty(M.getContext())) 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() != Type::getVoidTy(M.getContext()) ||
+ 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)
LLVMContext &Context) {
// 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(Context), 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, false);
+ const Type *FTy = FunctionType::get(Type::getVoidTy(Context), false);
const PointerType *PFTy = PointerType::getUnqual(FTy);
- CSVals[1] = Context.getNullValue(PFTy);
- CSVals[0] = ConstantInt::get(Type::Int32Ty, 2147483647);
+ CSVals[1] = Constant::getNullValue(PFTy);
+ CSVals[0] = ConstantInt::get(Type::getInt32Ty(Context), 2147483647);
}
- CAList.push_back(ConstantStruct::get(CSVals));
+ CAList.push_back(ConstantStruct::get(Context, CSVals, false));
}
// Create the array initializer.
const Type *StructTy =
- cast<ArrayType>(GCL->getType()->getElementType())->getElementType();
+ cast<ArrayType>(GCL->getType()->getElementType())->getElementType();
Constant *CA = ConstantArray::get(ArrayType::get(StructTy,
- CAList.size()), CAList);
+ CAList.size()), CAList);
// If we didn't change the number of elements, don't create a new GV.
if (CA->getType() == GCL->getInitializer()->getType()) {
/// 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, LLVMContext &Context) {
- if (GlobalVariable *GV = dyn_cast<GlobalVariable>(C)) {
- if (!GV->hasExternalLinkage() && !GV->hasLocalLinkage())
- 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->hasLocalLinkage())
- return false; // do not allow weak/linkonce/dllimport/dllexport linkage.
- return GV->hasInitializer() &&
- ConstantFoldLoadThroughGEPConstantExpr(GV->getInitializer(), CE,
- Context);
+ // 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;
}
Elts.push_back(cast<Constant>(*i));
} else if (isa<ConstantAggregateZero>(Init)) {
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i)
- Elts.push_back(Context.getNullValue(STy->getElementType(i)));
+ Elts.push_back(Constant::getNullValue(STy->getElementType(i)));
} else if (isa<UndefValue>(Init)) {
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i)
Elts.push_back(UndefValue::get(STy->getElementType(i)));
Elts[Idx] = EvaluateStoreInto(Elts[Idx], Val, Addr, OpNo+1, Context);
// Return the modified struct.
- return ConstantStruct::get(&Elts[0], Elts.size(), STy->isPacked());
+ return ConstantStruct::get(Context, &Elts[0], Elts.size(), STy->isPacked());
} else {
ConstantInt *CI = cast<ConstantInt>(Addr->getOperand(OpNo));
const ArrayType *ATy = cast<ArrayType>(Init->getType());
for (User::op_iterator i = CA->op_begin(), e = CA->op_end(); i != e; ++i)
Elts.push_back(cast<Constant>(*i));
} else if (isa<ConstantAggregateZero>(Init)) {
- Constant *Elt = Context.getNullValue(ATy->getElementType());
+ Constant *Elt = Constant::getNullValue(ATy->getElementType());
Elts.assign(ATy->getNumElements(), Elt);
} else if (isa<UndefValue>(Init)) {
Constant *Elt = UndefValue::get(ATy->getElementType());
// 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())
- return ConstantFoldLoadThroughGEPConstantExpr(GV->getInitializer(), CE,
- Context);
+ if (GV->hasDefinitiveInitializer())
+ return ConstantFoldLoadThroughGEPConstantExpr(GV->getInitializer(), CE);
}
return 0; // don't know how to evaluate.
/// 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) {
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)),
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 {
// 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.
DEBUG(errs() << "FULLY EVALUATED GLOBAL CTOR FUNCTION '"
// silly, e.g. storing the address of the alloca somewhere and using it
// later. Since this is undefined, we'll just make it be null.
if (!Tmp->use_empty())
- Tmp->replaceAllUsesWith(F->getContext().getNullValue(Tmp->getType()));
+ Tmp->replaceAllUsesWith(Constant::getNullValue(Tmp->getType()));
delete Tmp;
}