#include "llvm/DerivedTypes.h"
#include "llvm/Instructions.h"
#include "llvm/IntrinsicInst.h"
-#include "llvm/LLVMContext.h"
#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(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;
/// null/false. When the first accessing function is noticed, it is recorded.
/// When a second different accessing function is noticed,
/// HasMultipleAccessingFunctions is set to true.
- Function *AccessingFunction;
+ const Function *AccessingFunction;
bool HasMultipleAccessingFunctions;
/// HasNonInstructionUser - Set to true if this global has a user that is not
// by constants itself. Note that constants cannot be cyclic, so this test is
// pretty easy to implement recursively.
//
-static bool SafeToDestroyConstant(Constant *C) {
+static bool SafeToDestroyConstant(const 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)) {
+ for (Value::const_use_iterator UI = C->use_begin(), E = C->use_end(); UI != E; ++UI)
+ if (const Constant *CU = dyn_cast<Constant>(*UI)) {
if (!SafeToDestroyConstant(CU)) return false;
} else
return false;
/// structure. If the global has its address taken, return true to indicate we
/// can't do anything with it.
///
-static bool AnalyzeGlobal(Value *V, GlobalStatus &GS,
- SmallPtrSet<PHINode*, 16> &PHIUsers) {
- for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI != E; ++UI)
- if (ConstantExpr *CE = dyn_cast<ConstantExpr>(*UI)) {
+static bool AnalyzeGlobal(const Value *V, GlobalStatus &GS,
+ SmallPtrSet<const PHINode*, 16> &PHIUsers) {
+ for (Value::const_use_iterator UI = V->use_begin(), E = V->use_end(); UI != E; ++UI)
+ if (const ConstantExpr *CE = dyn_cast<ConstantExpr>(*UI)) {
GS.HasNonInstructionUser = true;
if (AnalyzeGlobal(CE, GS, PHIUsers)) return true;
- } else if (Instruction *I = dyn_cast<Instruction>(*UI)) {
+ } else if (const Instruction *I = dyn_cast<Instruction>(*UI)) {
if (!GS.HasMultipleAccessingFunctions) {
- Function *F = I->getParent()->getParent();
+ const Function *F = I->getParent()->getParent();
if (GS.AccessingFunction == 0)
GS.AccessingFunction = F;
else if (GS.AccessingFunction != F)
GS.HasMultipleAccessingFunctions = true;
}
- if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
+ if (const LoadInst *LI = dyn_cast<LoadInst>(I)) {
GS.isLoaded = true;
if (LI->isVolatile()) return true; // Don't hack on volatile loads.
- } else if (StoreInst *SI = dyn_cast<StoreInst>(I)) {
+ } else if (const StoreInst *SI = dyn_cast<StoreInst>(I)) {
// Don't allow a store OF the address, only stores TO the address.
if (SI->getOperand(0) == V) return true;
// value, not an aggregate), keep more specific information about
// stores.
if (GS.StoredType != GlobalStatus::isStored) {
- if (GlobalVariable *GV = dyn_cast<GlobalVariable>(SI->getOperand(1))){
+ if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(SI->getOperand(1))){
Value *StoredVal = SI->getOperand(0);
if (StoredVal == GV->getInitializer()) {
if (GS.StoredType < GlobalStatus::isInitializerStored)
GS.StoredType = GlobalStatus::isInitializerStored;
} else if (isa<LoadInst>(StoredVal) &&
cast<LoadInst>(StoredVal)->getOperand(0) == GV) {
- // G = G
if (GS.StoredType < GlobalStatus::isInitializerStored)
GS.StoredType = GlobalStatus::isInitializerStored;
} else if (GS.StoredType < GlobalStatus::isStoredOnce) {
if (AnalyzeGlobal(I, GS, PHIUsers)) return true;
} else if (isa<SelectInst>(I)) {
if (AnalyzeGlobal(I, GS, PHIUsers)) return true;
- } else if (PHINode *PN = dyn_cast<PHINode>(I)) {
+ } else if (const PHINode *PN = dyn_cast<PHINode>(I)) {
// PHI nodes we can check just like select or GEP instructions, but we
// have to be careful about infinite recursion.
if (PHIUsers.insert(PN)) // Not already visited.
} else {
return true; // Any other non-load instruction might take address!
}
- } else if (Constant *C = dyn_cast<Constant>(*UI)) {
+ } else if (const Constant *C = dyn_cast<Constant>(*UI)) {
GS.HasNonInstructionUser = true;
// We might have a dead and dangling constant hanging off of here.
if (!SafeToDestroyConstant(C))
return false;
}
-static Constant *getAggregateConstantElement(Constant *Agg, Constant *Idx,
- LLVMContext* Context) {
+static Constant *getAggregateConstantElement(Constant *Agg, Constant *Idx) {
ConstantInt *CI = dyn_cast<ConstantInt>(Idx);
if (!CI) return 0;
unsigned IdxV = CI->getZExtValue();
} 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 (IdxV < STy->getNumElements())
- return Context->getUndef(STy->getElementType(IdxV));
+ return UndefValue::get(STy->getElementType(IdxV));
} else if (const SequentialType *STy =
dyn_cast<SequentialType>(Agg->getType())) {
- return Context->getUndef(STy->getElementType());
+ return UndefValue::get(STy->getElementType());
}
}
return 0;
SubInit = ConstantFoldLoadThroughGEPConstantExpr(Init, CE);
Changed |= CleanupConstantGlobalUsers(CE, SubInit);
} else if (CE->getOpcode() == Instruction::BitCast &&
- isa<PointerType>(CE->getType())) {
+ CE->getType()->isPointerTy()) {
// Pointer cast, delete any stores and memsets to the global.
Changed |= CleanupConstantGlobalUsers(CE, 0);
}
// 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((*GEPI)->isStructTy() &&
+ "Indexed GEP type is not array, vector, or struct!");
+ continue;
+ }
ConstantInt *IdxVal = dyn_cast<ConstantInt>(GEPI.getOperand());
if (!IdxVal || IdxVal->getZExtValue() >= NumElements)
/// behavior of the program in a more fine-grained way. We have determined that
/// this transformation is safe already. We return the first global variable we
/// insert so that the caller can reprocess it.
-static GlobalVariable *SRAGlobal(GlobalVariable *GV, const TargetData &TD,
- LLVMContext* Context) {
+static GlobalVariable *SRAGlobal(GlobalVariable *GV, const TargetData &TD) {
// Make sure this global only has simple uses that we can SRA.
if (!GlobalUsersSafeToSRA(GV))
return 0;
const StructLayout &Layout = *TD.getStructLayout(STy);
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
Constant *In = getAggregateConstantElement(Init,
- Context->getConstantInt(Type::Int32Ty, i),
- Context);
+ 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);
unsigned EltAlign = TD.getABITypeAlignment(STy->getElementType());
for (unsigned i = 0, e = NumElements; i != e; ++i) {
Constant *In = getAggregateConstantElement(Init,
- Context->getConstantInt(Type::Int32Ty, i),
- Context);
+ 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;
+
+ DEBUG(dbgs() << "PERFORMING GLOBAL SRA ON: " << *GV);
- DOUT << "PERFORMING GLOBAL SRA ON: " << *GV;
-
- Constant *NullInt = Context->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.
Idxs.push_back(NullInt);
for (unsigned i = 3, e = CE->getNumOperands(); i != e; ++i)
Idxs.push_back(CE->getOperand(i));
- NewPtr = Context->getConstantExprGetElementPtr(cast<Constant>(NewPtr),
+ NewPtr = ConstantExpr::getGetElementPtr(cast<Constant>(NewPtr),
&Idxs[0], Idxs.size());
} else {
GetElementPtrInst *GEPI = cast<GetElementPtrInst>(GEP);
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);
return false; // Storing the value.
}
} else if (CallInst *CI = dyn_cast<CallInst>(*UI)) {
- if (CI->getOperand(0) != V) {
+ if (CI->getCalledValue() != V) {
//cerr << "NONTRAPPING USE: " << **UI;
return false; // Not calling the ptr
}
} else if (InvokeInst *II = dyn_cast<InvokeInst>(*UI)) {
- if (II->getOperand(0) != V) {
+ if (II->getCalledValue() != V) {
//cerr << "NONTRAPPING USE: " << **UI;
return false; // Not calling the ptr
}
} 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
+ // Ignore icmp X, null
} else {
//cerr << "NONTRAPPING USE: " << **UI;
return false;
return true;
}
-static bool OptimizeAwayTrappingUsesOfValue(Value *V, Constant *NewV,
- LLVMContext* Context) {
+static bool OptimizeAwayTrappingUsesOfValue(Value *V, Constant *NewV) {
bool Changed = false;
for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI != E; ) {
Instruction *I = cast<Instruction>(*UI++);
}
} else if (CastInst *CI = dyn_cast<CastInst>(I)) {
Changed |= OptimizeAwayTrappingUsesOfValue(CI,
- Context->getConstantExprCast(CI->getOpcode(),
- NewV, CI->getType()), Context);
+ ConstantExpr::getCast(CI->getOpcode(),
+ NewV, CI->getType()));
if (CI->use_empty()) {
Changed = true;
CI->eraseFromParent();
break;
if (Idxs.size() == GEPI->getNumOperands()-1)
Changed |= OptimizeAwayTrappingUsesOfValue(GEPI,
- Context->getConstantExprGetElementPtr(NewV, &Idxs[0],
- Idxs.size()), Context);
+ ConstantExpr::getGetElementPtr(NewV, &Idxs[0],
+ Idxs.size()));
if (GEPI->use_empty()) {
Changed = true;
GEPI->eraseFromParent();
/// value stored into it. If there are uses of the loaded value that would trap
/// if the loaded value is dynamically null, then we know that they cannot be
/// reachable with a null optimize away the load.
-static bool OptimizeAwayTrappingUsesOfLoads(GlobalVariable *GV, Constant *LV,
- LLVMContext* Context) {
+static bool OptimizeAwayTrappingUsesOfLoads(GlobalVariable *GV, Constant *LV) {
bool Changed = false;
// Keep track of whether we are able to remove all the uses of the global
for (Value::use_iterator GUI = GV->use_begin(), E = GV->use_end(); GUI != E;){
User *GlobalUser = *GUI++;
if (LoadInst *LI = dyn_cast<LoadInst>(GlobalUser)) {
- Changed |= OptimizeAwayTrappingUsesOfValue(LI, LV, Context);
+ Changed |= OptimizeAwayTrappingUsesOfValue(LI, LV);
// If we were able to delete all uses of the loads
if (LI->use_empty()) {
LI->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,
- LLVMContext* Context) {
- DOUT << "PROMOTING MALLOC GLOBAL: " << *GV << " MALLOC = " << *MI;
- ConstantInt *NElements = cast<ConstantInt>(MI->getArraySize());
-
- 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 = Context->getArrayType(MI->getAllocatedType(),
- NElements->getZExtValue());
- MallocInst *NewMI =
- new MallocInst(NewTy, Context->getNullValue(Type::Int32Ty),
- MI->getAlignment(), MI->getName(), MI);
- Value* Indices[2];
- Indices[0] = Indices[1] = Context->getNullValue(Type::Int32Ty);
- Value *NewGEP = GetElementPtrInst::Create(NewMI, Indices, Indices + 2,
- NewMI->getName()+".el0", MI);
- MI->replaceAllUsesWith(NewGEP);
- MI->eraseFromParent();
- MI = NewMI;
- }
+ CallInst *CI,
+ const Type *AllocTy,
+ ConstantInt *NElements,
+ TargetData* TD) {
+ DEBUG(errs() << "PROMOTING GLOBAL: " << *GV << " CALL = " << *CI << '\n');
+
+ const Type *GlobalType;
+ if (NElements->getZExtValue() == 1)
+ GlobalType = AllocTy;
+ else
+ // If we have an array allocation, the global variable is of an array.
+ GlobalType = ArrayType::get(AllocTy, NElements->getZExtValue());
// Create the new global variable. The contents of the malloc'd memory is
// undefined, so initialize with an undef value.
- Constant *Init = Context->getUndef(MI->getAllocatedType());
- GlobalVariable *NewGV = new GlobalVariable(MI->getAllocatedType(), false,
- GlobalValue::InternalLinkage, Init,
+ GlobalVariable *NewGV = new GlobalVariable(*GV->getParent(),
+ GlobalType, false,
+ GlobalValue::InternalLinkage,
+ UndefValue::get(GlobalType),
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);
-
+
+ // If there are bitcast users of the malloc (which is typical, usually we have
+ // a malloc + bitcast) then replace them with uses of the new global. Update
+ // other users to use the global as well.
+ BitCastInst *TheBC = 0;
+ while (!CI->use_empty()) {
+ Instruction *User = cast<Instruction>(CI->use_back());
+ if (BitCastInst *BCI = dyn_cast<BitCastInst>(User)) {
+ if (BCI->getType() == NewGV->getType()) {
+ BCI->replaceAllUsesWith(NewGV);
+ BCI->eraseFromParent();
+ } else {
+ BCI->setOperand(0, NewGV);
+ }
+ } else {
+ if (TheBC == 0)
+ TheBC = new BitCastInst(NewGV, CI->getType(), "newgv", CI);
+ User->replaceUsesOfWith(CI, TheBC);
+ }
+ }
+
Constant *RepValue = NewGV;
if (NewGV->getType() != GV->getType()->getElementType())
- RepValue = Context->getConstantExprBitCast(RepValue,
+ RepValue = ConstantExpr::getBitCast(RepValue,
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,
- Context->getConstantIntFalse(), 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.
- std::vector<StoreInst*> Stores;
- while (!GV->use_empty())
- if (LoadInst *LI = dyn_cast<LoadInst>(GV->use_back())) {
- while (!LI->use_empty()) {
- Use &LoadUse = LI->use_begin().getUse();
- if (!isa<ICmpInst>(LoadUse.getUser()))
- LoadUse = RepValue;
- else {
- ICmpInst *CI = 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);
- InitBoolUsed = true;
- switch (CI->getPredicate()) {
- default: assert(0 && "Unknown ICmp Predicate!");
- case ICmpInst::ICMP_ULT:
- case ICmpInst::ICMP_SLT:
- LV = Context->getConstantIntFalse(); // X < null -> always false
- break;
- case ICmpInst::ICMP_ULE:
- case ICmpInst::ICMP_SLE:
- case ICmpInst::ICMP_EQ:
- LV = BinaryOperator::CreateNot(LV, "notinit", CI);
- break;
- case ICmpInst::ICMP_NE:
- case ICmpInst::ICMP_UGE:
- case ICmpInst::ICMP_SGE:
- case ICmpInst::ICMP_UGT:
- case ICmpInst::ICMP_SGT:
- break; // no change.
- }
- CI->replaceAllUsesWith(LV);
- CI->eraseFromParent();
- }
- }
- LI->eraseFromParent();
- } else {
- StoreInst *SI = cast<StoreInst>(GV->use_back());
+ while (!GV->use_empty()) {
+ if (StoreInst *SI = dyn_cast<StoreInst>(GV->use_back())) {
// The global is initialized when the store to it occurs.
- new StoreInst(Context->getConstantIntTrue(), InitBool, SI);
+ new StoreInst(ConstantInt::getTrue(GV->getContext()), InitBool, SI);
SI->eraseFromParent();
+ continue;
}
+
+ LoadInst *LI = cast<LoadInst>(GV->use_back());
+ while (!LI->use_empty()) {
+ Use &LoadUse = LI->use_begin().getUse();
+ if (!isa<ICmpInst>(LoadUse.getUser())) {
+ LoadUse = RepValue;
+ continue;
+ }
+
+ 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", ICI);
+ InitBoolUsed = true;
+ switch (ICI->getPredicate()) {
+ default: llvm_unreachable("Unknown ICmp Predicate!");
+ case ICmpInst::ICMP_ULT:
+ 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", ICI);
+ break;
+ case ICmpInst::ICMP_NE:
+ case ICmpInst::ICMP_UGE:
+ case ICmpInst::ICMP_SGE:
+ case ICmpInst::ICMP_UGT:
+ case ICmpInst::ICMP_SGT:
+ break; // no change.
+ }
+ ICI->replaceAllUsesWith(LV);
+ ICI->eraseFromParent();
+ }
+ LI->eraseFromParent();
+ }
// If the initialization boolean was used, insert it, otherwise delete it.
if (!InitBoolUsed) {
while (!InitBool->use_empty()) // Delete initializations
- cast<Instruction>(InitBool->use_back())->eraseFromParent();
+ cast<StoreInst>(InitBool->use_back())->eraseFromParent();
delete InitBool;
} else
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..
GV->eraseFromParent();
- MI->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
/// LoadUsesSimpleEnoughForHeapSRA - Verify that all uses of V (a load, or a phi
/// 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> &LoadUsingPHIsPerLoad) {
+static bool LoadUsesSimpleEnoughForHeapSRA(const Value *V,
+ SmallPtrSet<const PHINode*, 32> &LoadUsingPHIs,
+ SmallPtrSet<const 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){
- Instruction *User = cast<Instruction>(*UI);
+ for (Value::const_use_iterator UI = V->use_begin(), E = V->use_end(); UI != E;++UI){
+ const Instruction *User = cast<Instruction>(*UI);
// Comparison against null is ok.
- if (ICmpInst *ICI = dyn_cast<ICmpInst>(User)) {
+ if (const ICmpInst *ICI = dyn_cast<ICmpInst>(User)) {
if (!isa<ConstantPointerNull>(ICI->getOperand(1)))
return false;
continue;
}
// getelementptr is also ok, but only a simple form.
- if (GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(User)) {
+ if (const GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(User)) {
// Must index into the array and into the struct.
if (GEPI->getNumOperands() < 3)
return false;
continue;
}
- if (PHINode *PN = dyn_cast<PHINode>(User)) {
+ if (const PHINode *PN = dyn_cast<PHINode>(User)) {
if (!LoadUsingPHIsPerLoad.insert(PN))
// This means some phi nodes are dependent on each other.
// Avoid infinite looping!
/// AllGlobalLoadUsesSimpleEnoughForHeapSRA - If all users of values loaded from
/// GV are simple enough to perform HeapSRA, return true.
-static bool AllGlobalLoadUsesSimpleEnoughForHeapSRA(GlobalVariable *GV,
- MallocInst *MI) {
- SmallPtrSet<PHINode*, 32> LoadUsingPHIs;
- SmallPtrSet<PHINode*, 32> LoadUsingPHIsPerLoad;
- for (Value::use_iterator UI = GV->use_begin(), E = GV->use_end(); UI != E;
+static bool AllGlobalLoadUsesSimpleEnoughForHeapSRA(const GlobalVariable *GV,
+ Instruction *StoredVal) {
+ SmallPtrSet<const PHINode*, 32> LoadUsingPHIs;
+ SmallPtrSet<const PHINode*, 32> LoadUsingPHIsPerLoad;
+ for (Value::const_use_iterator UI = GV->use_begin(), E = GV->use_end(); UI != E;
++UI)
- if (LoadInst *LI = dyn_cast<LoadInst>(*UI)) {
+ if (const LoadInst *LI = dyn_cast<LoadInst>(*UI)) {
if (!LoadUsesSimpleEnoughForHeapSRA(LI, LoadUsingPHIs,
LoadUsingPHIsPerLoad))
return false;
// that all inputs the to the PHI nodes are in the same equivalence sets.
// Check to verify that all operands of the PHIs are either PHIS that can be
// transformed, loads from GV, or MI itself.
- for (SmallPtrSet<PHINode*, 32>::iterator I = LoadUsingPHIs.begin(),
+ for (SmallPtrSet<const PHINode*, 32>::const_iterator I = LoadUsingPHIs.begin(),
E = LoadUsingPHIs.end(); I != E; ++I) {
- PHINode *PN = *I;
+ const PHINode *PN = *I;
for (unsigned op = 0, e = PN->getNumIncomingValues(); op != e; ++op) {
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)) {
+ if (const PHINode *InPN = dyn_cast<PHINode>(InVal)) {
// One of the PHIs in our set is (optimistically) ok.
if (LoadUsingPHIs.count(InPN))
continue;
}
// Load from GV is ok.
- if (LoadInst *LI = dyn_cast<LoadInst>(InVal))
+ if (const LoadInst *LI = dyn_cast<LoadInst>(InVal))
if (LI->getOperand(0) == GV)
continue;
static Value *GetHeapSROAValue(Value *V, unsigned FieldNo,
DenseMap<Value*, std::vector<Value*> > &InsertedScalarizedValues,
- std::vector<std::pair<PHINode*, unsigned> > &PHIsToRewrite,
- LLVMContext* Context) {
+ std::vector<std::pair<PHINode*, unsigned> > &PHIsToRewrite) {
std::vector<Value*> &FieldVals = InsertedScalarizedValues[V];
if (FieldNo >= FieldVals.size())
// a new Load of the scalarized global.
Result = new LoadInst(GetHeapSROAValue(LI->getOperand(0), FieldNo,
InsertedScalarizedValues,
- PHIsToRewrite, Context),
- LI->getName()+".f" + utostr(FieldNo), LI);
+ PHIsToRewrite),
+ 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.
cast<StructType>(cast<PointerType>(PN->getType())->getElementType());
Result =
- PHINode::Create(Context->getPointerTypeUnqual(ST->getElementType(FieldNo)),
- PN->getName()+".f"+utostr(FieldNo), PN);
+ 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;
}
/// the load, rewrite the derived value to use the HeapSRoA'd load.
static void RewriteHeapSROALoadUser(Instruction *LoadUser,
DenseMap<Value*, std::vector<Value*> > &InsertedScalarizedValues,
- std::vector<std::pair<PHINode*, unsigned> > &PHIsToRewrite,
- LLVMContext* Context) {
+ std::vector<std::pair<PHINode*, unsigned> > &PHIsToRewrite) {
// If this is a comparison against null, handle it.
if (ICmpInst *SCI = dyn_cast<ICmpInst>(LoadUser)) {
assert(isa<ConstantPointerNull>(SCI->getOperand(1)));
// If we have a setcc of the loaded pointer, we can use a setcc of any
// field.
Value *NPtr = GetHeapSROAValue(SCI->getOperand(0), 0,
- InsertedScalarizedValues, PHIsToRewrite,
- Context);
+ InsertedScalarizedValues, PHIsToRewrite);
- Value *New = new ICmpInst(SCI->getPredicate(), NPtr,
- Context->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;
// Load the pointer for this field.
unsigned FieldNo = cast<ConstantInt>(GEPI->getOperand(2))->getZExtValue();
Value *NewPtr = GetHeapSROAValue(GEPI->getOperand(0), FieldNo,
- InsertedScalarizedValues, PHIsToRewrite,
- Context);
+ InsertedScalarizedValues, PHIsToRewrite);
// Create the new GEP idx vector.
SmallVector<Value*, 8> GEPIdx;
// users.
for (Value::use_iterator UI = PN->use_begin(), E = PN->use_end(); UI != E; ) {
Instruction *User = cast<Instruction>(*UI++);
- RewriteHeapSROALoadUser(User, InsertedScalarizedValues, PHIsToRewrite,
- Context);
+ RewriteHeapSROALoadUser(User, InsertedScalarizedValues, PHIsToRewrite);
}
}
/// AllGlobalLoadUsesSimpleEnoughForHeapSRA.
static void RewriteUsesOfLoadForHeapSRoA(LoadInst *Load,
DenseMap<Value*, std::vector<Value*> > &InsertedScalarizedValues,
- std::vector<std::pair<PHINode*, unsigned> > &PHIsToRewrite,
- LLVMContext* Context) {
+ std::vector<std::pair<PHINode*, unsigned> > &PHIsToRewrite) {
for (Value::use_iterator UI = Load->use_begin(), E = Load->use_end();
UI != E; ) {
Instruction *User = cast<Instruction>(*UI++);
- RewriteHeapSROALoadUser(User, InsertedScalarizedValues, PHIsToRewrite,
- Context);
+ RewriteHeapSROALoadUser(User, InsertedScalarizedValues, PHIsToRewrite);
}
if (Load->use_empty()) {
}
}
-/// 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,
+ 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 = Context->getPointerTypeUnqual(FieldTy);
+ const PointerType *PFieldTy = PointerType::getUnqual(FieldTy);
GlobalVariable *NGV =
- new GlobalVariable(PFieldTy, false, GlobalValue::InternalLinkage,
- Context->getNullValue(PFieldTy),
- GV->getName() + ".f" + utostr(FieldNo), GV,
+ new GlobalVariable(*GV->getParent(),
+ PFieldTy, false, GlobalValue::InternalLinkage,
+ 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" + utostr(FieldNo),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));
FieldMallocs.push_back(NMI);
- new StoreInst(NMI, NGV, MI);
+ 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],
- Context->getNullValue(FieldMallocs[i]->getType()),
- "isnull", MI);
- if (!RunningOr)
- RunningOr = Cond; // First seteq
- else
- RunningOr = BinaryOperator::CreateOr(RunningOr, Cond, "tmp", MI);
+ Value *Cond = new ICmpInst(CI, ICmpInst::ICMP_EQ, FieldMallocs[i],
+ Constant::getNullValue(FieldMallocs[i]->getType()),
+ "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,
- Context->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);
+ Value *Cmp = new ICmpInst(*NullPtrBlock, ICmpInst::ICMP_NE, GVVal,
+ Constant::getNullValue(GVVal->getType()),
+ "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);
- 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 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
Instruction *User = cast<Instruction>(*UI++);
if (LoadInst *LI = dyn_cast<LoadInst>(User)) {
- RewriteUsesOfLoadForHeapSRoA(LI, InsertedScalarizedValues, PHIsToRewrite,
- Context);
+ RewriteUsesOfLoadForHeapSRoA(LI, InsertedScalarizedValues, PHIsToRewrite);
continue;
}
// 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.
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
Value *InVal = PN->getIncomingValue(i);
InVal = GetHeapSROAValue(InVal, FieldNo, InsertedScalarizedValues,
- PHIsToRewrite, Context);
+ PHIsToRewrite);
FieldPN->addIncoming(InVal, PN->getIncomingBlock(i));
}
}
/// 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,
- LLVMContext* Context) {
+ 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.getTypeAllocSize(MI->getAllocatedType()) < 2048) {
- GVI = OptimizeGlobalAddressOfMalloc(GV, MI, Context);
- 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, NElements, 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,
- Context->getConstantInt(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, 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) {
// Ignore no-op GEPs and bitcasts.
StoredOnceVal = StoredOnceVal->stripPointerCasts();
// only has one (non-null) value stored into it, then we can optimize any
// users of the loaded value (often calls and loads) that would trap if the
// value was null.
- if (isa<PointerType>(GV->getInitializer()->getType()) &&
+ if (GV->getInitializer()->getType()->isPointerTy() &&
GV->getInitializer()->isNullValue()) {
if (Constant *SOVC = dyn_cast<Constant>(StoredOnceVal)) {
if (GV->getInitializer()->getType() != SOVC->getType())
SOVC =
- Context->getConstantExprBitCast(SOVC, GV->getInitializer()->getType());
+ ConstantExpr::getBitCast(SOVC, GV->getInitializer()->getType());
// Optimize away any trapping uses of the loaded value.
- if (OptimizeAwayTrappingUsesOfLoads(GV, SOVC, Context))
+ if (OptimizeAwayTrappingUsesOfLoads(GV, SOVC))
return true;
- } else if (MallocInst *MI = dyn_cast<MallocInst>(StoredOnceVal)) {
- if (TryToOptimizeStoreOfMallocToGlobal(GV, MI, GVI, TD, Context))
+ } else if (CallInst *CI = extractMallocCall(StoredOnceVal)) {
+ const Type* MallocType = getMallocAllocatedType(CI);
+ if (MallocType && TryToOptimizeStoreOfMallocToGlobal(GV, CI, MallocType,
+ GVI, TD))
return true;
}
}
/// two values ever stored into GV are its initializer and OtherVal. See if we
/// can shrink the global into a boolean and select between the two values
/// whenever it is used. This exposes the values to other scalar optimizations.
-static bool TryToShrinkGlobalToBoolean(GlobalVariable *GV, Constant *OtherVal,
- LLVMContext* Context) {
+static bool TryToShrinkGlobalToBoolean(GlobalVariable *GV, Constant *OtherVal) {
const Type *GVElType = GV->getType()->getElementType();
// If GVElType is already i1, it is already shrunk. If the type of the GV is
// 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() ||
- isa<PointerType>(GVElType) || isa<VectorType>(GVElType))
+ if (GVElType == Type::getInt1Ty(GV->getContext()) ||
+ GVElType->isFloatingPointTy() ||
+ GVElType->isPointerTy() || GVElType->isVectorTy())
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, Context->getConstantIntFalse(),
+ 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 = Context->getConstantInt(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
/// it if possible. If we make a change, return true.
bool GlobalOpt::ProcessInternalGlobal(GlobalVariable *GV,
Module::global_iterator &GVI) {
- SmallPtrSet<PHINode*, 16> PHIUsers;
+ SmallPtrSet<const PHINode*, 16> PHIUsers;
GlobalStatus GS;
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
GS.AccessingFunction->getName() == "main" &&
GS.AccessingFunction->hasExternalLinkage() &&
GV->getType()->getAddressSpace() == 0) {
- DOUT << "LOCALIZING GLOBAL: " << *GV;
- Instruction* FirstI = GS.AccessingFunction->getEntryBlock().begin();
+ DEBUG(dbgs() << "LOCALIZING GLOBAL: " << *GV);
+ Instruction& FirstI = const_cast<Instruction&>(*GS.AccessingFunction
+ ->getEntryBlock().begin());
const Type* ElemTy = GV->getType()->getElementType();
// FIXME: Pass Global's alignment when globals have alignment
- AllocaInst* Alloca = new AllocaInst(ElemTy, NULL, GV->getName(), FirstI);
+ AllocaInst* Alloca = new AllocaInst(ElemTy, NULL, GV->getName(), &FirstI);
if (!isa<UndefValue>(GV->getInitializer()))
- new StoreInst(GV->getInitializer(), Alloca, FirstI);
+ new StoreInst(GV->getInitializer(), Alloca, &FirstI);
GV->replaceAllUsesWith(Alloca);
GV->eraseFromParent();
// 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>(),
- Context)) {
- 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>(), Context))
+ getAnalysisIfAvailable<TargetData>()))
return true;
// Otherwise, if the global was not a boolean, we can shrink it to be a
// boolean.
if (Constant *SOVConstant = dyn_cast<Constant>(GS.StoredOnceValue))
- if (TryToShrinkGlobalToBoolean(GV, SOVConstant, Context)) {
+ if (TryToShrinkGlobalToBoolean(GV, SOVConstant)) {
++NumShrunkToBool;
return true;
}
if (!F->hasName() && !F->isDeclaration())
F->setLinkage(GlobalValue::InternalLinkage);
F->removeDeadConstantUsers();
- if (F->use_empty() && (F->hasLocalLinkage() ||
- F->hasLinkOnceLinkage())) {
- M.getFunctionList().erase(F);
+ if (F->use_empty() && (F->hasLocalLinkage() || F->hasLinkOnceLinkage())) {
+ F->eraseFromParent();
Changed = true;
++NumFnDeleted;
} else if (F->hasLocalLinkage()) {
// 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)
/// InstallGlobalCtors - Given a specified llvm.global_ctors list, install the
/// specified array, returning the new global to use.
static GlobalVariable *InstallGlobalCtors(GlobalVariable *GCL,
- const std::vector<Function*> &Ctors,
- LLVMContext* Context) {
+ const std::vector<Function*> &Ctors) {
// If we made a change, reassemble the initializer list.
std::vector<Constant*> CSVals;
- CSVals.push_back(Context->getConstantInt(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 = Context->getFunctionType(Type::VoidTy, false);
- const PointerType *PFTy = Context->getPointerTypeUnqual(FTy);
- CSVals[1] = Context->getNullValue(PFTy);
- CSVals[0] = Context->getConstantInt(Type::Int32Ty, 2147483647);
+ 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::getInt32Ty(GCL->getContext()),
+ 2147483647);
}
- CAList.push_back(Context->getConstantStruct(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 = Context->getConstantArray(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);
if (!GCL->use_empty()) {
Constant *V = NGV;
if (V->getType() != GCL->getType())
- V = Context->getConstantExprBitCast(V, GCL->getType());
+ V = ConstantExpr::getBitCast(V, GCL->getType());
GCL->replaceAllUsesWith(V);
}
GCL->eraseFromParent();
/// 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->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);
+ // 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;
}
/// initializer. This returns 'Init' modified to reflect 'Val' stored into it.
/// At this point, the GEP operands of Addr [0, OpNo) have been stepped into.
static Constant *EvaluateStoreInto(Constant *Init, Constant *Val,
- ConstantExpr *Addr, unsigned OpNo,
- LLVMContext* Context) {
+ ConstantExpr *Addr, unsigned OpNo) {
// Base case of the recursion.
if (OpNo == Addr->getNumOperands()) {
assert(Val->getType() == Init->getType() && "Type mismatch!");
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)) {
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(Context->getUndef(STy->getElementType(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");
}
ConstantInt *CU = cast<ConstantInt>(Addr->getOperand(OpNo));
unsigned Idx = CU->getZExtValue();
assert(Idx < STy->getNumElements() && "Struct index out of range!");
- Elts[Idx] = EvaluateStoreInto(Elts[Idx], Val, Addr, OpNo+1, Context);
+ Elts[Idx] = EvaluateStoreInto(Elts[Idx], Val, Addr, OpNo+1);
// Return the modified struct.
- return Context->getConstantStruct(&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 = Context->getNullValue(ATy->getElementType());
- Elts.assign(ATy->getNumElements(), Elt);
- } else if (isa<UndefValue>(Init)) {
- Constant *Elt = Context->getUndef(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, Context);
- return Context->getConstantArray(ATy, Elts);
+ EvaluateStoreInto(Elts[CI->getZExtValue()], Val, Addr, OpNo+1);
+
+ if (Init->getType()->isArrayTy())
+ return ConstantArray::get(cast<ArrayType>(InitTy), Elts);
+ else
+ return ConstantVector::get(&Elts[0], Elts.size());
}
}
/// CommitValueTo - We have decided that Addr (which satisfies the predicate
/// isSimpleEnoughPointerToCommit) should get Val as its value. Make it happen.
-static void CommitValueTo(Constant *Val, Constant *Addr,
- LLVMContext* Context) {
+static void CommitValueTo(Constant *Val, Constant *Addr) {
if (GlobalVariable *GV = dyn_cast<GlobalVariable>(Addr)) {
assert(GV->hasInitializer());
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, Context);
- 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) {
if (std::find(CallStack.begin(), CallStack.end(), F) != CallStack.end())
return false;
- LLVMContext* Context = F->getContext();
-
CallStack.push_back(F);
/// Values - As we compute SSA register values, we store their contents here.
Constant *Val = getVal(Values, SI->getOperand(0));
MutatedMemory[Ptr] = Val;
} else if (BinaryOperator *BO = dyn_cast<BinaryOperator>(CurInst)) {
- InstResult = Context->getConstantExpr(BO->getOpcode(),
+ InstResult = ConstantExpr::get(BO->getOpcode(),
getVal(Values, BO->getOperand(0)),
getVal(Values, BO->getOperand(1)));
} else if (CmpInst *CI = dyn_cast<CmpInst>(CurInst)) {
- InstResult = Context->getConstantExprCompare(CI->getPredicate(),
+ InstResult = ConstantExpr::getCompare(CI->getPredicate(),
getVal(Values, CI->getOperand(0)),
getVal(Values, CI->getOperand(1)));
} else if (CastInst *CI = dyn_cast<CastInst>(CurInst)) {
- InstResult = Context->getConstantExprCast(CI->getOpcode(),
+ InstResult = ConstantExpr::getCast(CI->getOpcode(),
getVal(Values, CI->getOperand(0)),
CI->getType());
} else if (SelectInst *SI = dyn_cast<SelectInst>(CurInst)) {
InstResult =
- Context->getConstantExprSelect(getVal(Values, SI->getOperand(0)),
+ 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 =
- Context->getConstantExprGetElementPtr(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)),
const Type *Ty = AI->getType()->getElementType();
AllocaTmps.push_back(new GlobalVariable(Ty, false,
GlobalValue::InternalLinkage,
- Context->getUndef(Ty),
+ UndefValue::get(Ty),
AI->getName()));
InstResult = AllocaTmps.back();
} else if (CallInst *CI = dyn_cast<CallInst>(CurInst)) {
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 {
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, F->getContext());
+ CommitValueTo(I->second, I->first);
}
// At this point, we are done interpreting. If we created any 'alloca'
// 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;
}
if (!MadeChange) return false;
- GCL = InstallGlobalCtors(GCL, Ctors, Context);
+ GCL = InstallGlobalCtors(GCL, Ctors);
return true;
}
Changed = true;
}
- // 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;
-
- // The transform is only useful if the alias does not have internal linkage.
- if (J->hasLocalLinkage())
- continue;
+ // 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;
+ // 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);
+ // 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);