diff options
Diffstat (limited to 'lib/Transforms/IPO/GlobalOpt.cpp')
-rw-r--r-- | lib/Transforms/IPO/GlobalOpt.cpp | 261 |
1 files changed, 118 insertions, 143 deletions
diff --git a/lib/Transforms/IPO/GlobalOpt.cpp b/lib/Transforms/IPO/GlobalOpt.cpp index 234d0ec..442f2fb 100644 --- a/lib/Transforms/IPO/GlobalOpt.cpp +++ b/lib/Transforms/IPO/GlobalOpt.cpp @@ -20,7 +20,6 @@ #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" @@ -245,8 +244,7 @@ static bool AnalyzeGlobal(Value *V, GlobalStatus &GS, 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(); @@ -282,8 +280,7 @@ static Constant *getAggregateConstantElement(Constant *Agg, Constant *Idx, /// users of the global, cleaning up the obvious ones. This is largely just a /// quick scan over the use list to clean up the easy and obvious cruft. This /// returns true if it made a change. -static bool CleanupConstantGlobalUsers(Value *V, Constant *Init, - LLVMContext &Context) { +static bool CleanupConstantGlobalUsers(Value *V, Constant *Init) { bool Changed = false; for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI != E;) { User *U = *UI++; @@ -304,11 +301,11 @@ static bool CleanupConstantGlobalUsers(Value *V, Constant *Init, Constant *SubInit = 0; if (Init) SubInit = ConstantFoldLoadThroughGEPConstantExpr(Init, CE); - Changed |= CleanupConstantGlobalUsers(CE, SubInit, Context); + Changed |= CleanupConstantGlobalUsers(CE, SubInit); } else if (CE->getOpcode() == Instruction::BitCast && isa<PointerType>(CE->getType())) { // Pointer cast, delete any stores and memsets to the global. - Changed |= CleanupConstantGlobalUsers(CE, 0, Context); + Changed |= CleanupConstantGlobalUsers(CE, 0); } if (CE->use_empty()) { @@ -322,11 +319,11 @@ static bool CleanupConstantGlobalUsers(Value *V, Constant *Init, Constant *SubInit = 0; if (!isa<ConstantExpr>(GEP->getOperand(0))) { ConstantExpr *CE = - dyn_cast_or_null<ConstantExpr>(ConstantFoldInstruction(GEP, Context)); + dyn_cast_or_null<ConstantExpr>(ConstantFoldInstruction(GEP)); if (Init && CE && CE->getOpcode() == Instruction::GetElementPtr) SubInit = ConstantFoldLoadThroughGEPConstantExpr(Init, CE); } - Changed |= CleanupConstantGlobalUsers(GEP, SubInit, Context); + Changed |= CleanupConstantGlobalUsers(GEP, SubInit); if (GEP->use_empty()) { GEP->eraseFromParent(); @@ -344,7 +341,7 @@ static bool CleanupConstantGlobalUsers(Value *V, Constant *Init, if (SafeToDestroyConstant(C)) { C->destroyConstant(); // This could have invalidated UI, start over from scratch. - CleanupConstantGlobalUsers(V, Init, Context); + CleanupConstantGlobalUsers(V, Init); return true; } } @@ -469,8 +466,7 @@ static bool GlobalUsersSafeToSRA(GlobalValue *GV) { /// 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; @@ -492,11 +488,9 @@ static GlobalVariable *SRAGlobal(GlobalVariable *GV, const TargetData &TD, const StructLayout &Layout = *TD.getStructLayout(STy); for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { Constant *In = getAggregateConstantElement(Init, - ConstantInt::get(Type::getInt32Ty(Context), i), - Context); + ConstantInt::get(Type::getInt32Ty(STy->getContext()), i)); assert(In && "Couldn't get element of initializer?"); - GlobalVariable *NGV = new GlobalVariable(Context, - STy->getElementType(i), false, + GlobalVariable *NGV = new GlobalVariable(STy->getElementType(i), false, GlobalVariable::InternalLinkage, In, GV->getName()+"."+Twine(i), GV->isThreadLocal(), @@ -527,12 +521,10 @@ static GlobalVariable *SRAGlobal(GlobalVariable *GV, const TargetData &TD, unsigned EltAlign = TD.getABITypeAlignment(STy->getElementType()); for (unsigned i = 0, e = NumElements; i != e; ++i) { Constant *In = getAggregateConstantElement(Init, - ConstantInt::get(Type::getInt32Ty(Context), i), - Context); + ConstantInt::get(Type::getInt32Ty(Init->getContext()), i)); assert(In && "Couldn't get element of initializer?"); - GlobalVariable *NGV = new GlobalVariable(Context, - STy->getElementType(), false, + GlobalVariable *NGV = new GlobalVariable(STy->getElementType(), false, GlobalVariable::InternalLinkage, In, GV->getName()+"."+Twine(i), GV->isThreadLocal(), @@ -554,7 +546,7 @@ static GlobalVariable *SRAGlobal(GlobalVariable *GV, const TargetData &TD, DEBUG(errs() << "PERFORMING GLOBAL SRA ON: " << *GV); - Constant *NullInt = Constant::getNullValue(Type::getInt32Ty(Context)); + 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. @@ -678,8 +670,7 @@ static bool AllUsesOfLoadedValueWillTrapIfNull(GlobalVariable *GV) { 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++); @@ -712,7 +703,7 @@ static bool OptimizeAwayTrappingUsesOfValue(Value *V, Constant *NewV, } else if (CastInst *CI = dyn_cast<CastInst>(I)) { Changed |= OptimizeAwayTrappingUsesOfValue(CI, ConstantExpr::getCast(CI->getOpcode(), - NewV, CI->getType()), Context); + NewV, CI->getType())); if (CI->use_empty()) { Changed = true; CI->eraseFromParent(); @@ -730,7 +721,7 @@ static bool OptimizeAwayTrappingUsesOfValue(Value *V, Constant *NewV, if (Idxs.size() == GEPI->getNumOperands()-1) Changed |= OptimizeAwayTrappingUsesOfValue(GEPI, ConstantExpr::getGetElementPtr(NewV, &Idxs[0], - Idxs.size()), Context); + Idxs.size())); if (GEPI->use_empty()) { Changed = true; GEPI->eraseFromParent(); @@ -746,8 +737,7 @@ static bool OptimizeAwayTrappingUsesOfValue(Value *V, Constant *NewV, /// 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 @@ -758,7 +748,7 @@ static bool OptimizeAwayTrappingUsesOfLoads(GlobalVariable *GV, Constant *LV, 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(); @@ -789,7 +779,7 @@ static bool OptimizeAwayTrappingUsesOfLoads(GlobalVariable *GV, Constant *LV, // nor is the global. if (AllNonStoreUsesGone) { DEBUG(errs() << " *** GLOBAL NOW DEAD!\n"); - CleanupConstantGlobalUsers(GV, 0, Context); + CleanupConstantGlobalUsers(GV, 0); if (GV->use_empty()) { GV->eraseFromParent(); ++NumDeleted; @@ -801,10 +791,10 @@ static bool OptimizeAwayTrappingUsesOfLoads(GlobalVariable *GV, Constant *LV, /// ConstantPropUsersOf - Walk the use list of V, constant folding all of the /// instructions that are foldable. -static void ConstantPropUsersOf(Value *V, LLVMContext &Context) { +static void ConstantPropUsersOf(Value *V) { for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI != E; ) if (Instruction *I = dyn_cast<Instruction>(*UI++)) - if (Constant *NewC = ConstantFoldInstruction(I, Context)) { + if (Constant *NewC = ConstantFoldInstruction(I)) { I->replaceAllUsesWith(NewC); // Advance UI to the next non-I use to avoid invalidating it! @@ -824,11 +814,10 @@ static GlobalVariable *OptimizeGlobalAddressOfMalloc(GlobalVariable *GV, CallInst *CI, const Type *AllocTy, Value* NElems, - LLVMContext &Context, TargetData* TD) { DEBUG(errs() << "PROMOTING GLOBAL: " << *GV << " CALL = " << *CI << '\n'); - const Type *IntPtrTy = TD->getIntPtrType(Context); + const Type *IntPtrTy = TD->getIntPtrType(GV->getContext()); // CI has either 0 or 1 bitcast uses (getMallocType() would otherwise have // returned NULL and we would not be here). @@ -883,10 +872,10 @@ static GlobalVariable *OptimizeGlobalAddressOfMalloc(GlobalVariable *GV, // 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::getInt1Ty(Context), false, + new GlobalVariable(Type::getInt1Ty(GV->getContext()), false, GlobalValue::InternalLinkage, - ConstantInt::getFalse(Context), GV->getName()+".init", - GV->isThreadLocal()); + ConstantInt::getFalse(GV->getContext()), + GV->getName()+".init", GV->isThreadLocal()); bool InitBoolUsed = false; // Loop over all uses of GV, processing them in turn. @@ -905,8 +894,8 @@ static GlobalVariable *OptimizeGlobalAddressOfMalloc(GlobalVariable *GV, switch (ICI->getPredicate()) { default: llvm_unreachable("Unknown ICmp Predicate!"); case ICmpInst::ICMP_ULT: - case ICmpInst::ICMP_SLT: - LV = ConstantInt::getFalse(Context); // X < null -> always false + case ICmpInst::ICMP_SLT: // X < null -> always false + LV = ConstantInt::getFalse(GV->getContext()); break; case ICmpInst::ICMP_ULE: case ICmpInst::ICMP_SLE: @@ -928,7 +917,7 @@ static GlobalVariable *OptimizeGlobalAddressOfMalloc(GlobalVariable *GV, } else { StoreInst *SI = cast<StoreInst>(GV->use_back()); // The global is initialized when the store to it occurs. - new StoreInst(ConstantInt::getTrue(Context), InitBool, SI); + new StoreInst(ConstantInt::getTrue(GV->getContext()), InitBool, SI); SI->eraseFromParent(); } @@ -949,9 +938,9 @@ static GlobalVariable *OptimizeGlobalAddressOfMalloc(GlobalVariable *GV, // To further other optimizations, loop over all users of NewGV and try to // constant prop them. This will promote GEP instructions with constant // indices into GEP constant-exprs, which will allow global-opt to hack on it. - ConstantPropUsersOf(NewGV, Context); + ConstantPropUsersOf(NewGV); if (RepValue != NewGV) - ConstantPropUsersOf(RepValue, Context); + ConstantPropUsersOf(RepValue); return NewGV; } @@ -1153,8 +1142,7 @@ static bool AllGlobalLoadUsesSimpleEnoughForHeapSRA(GlobalVariable *GV, 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()) @@ -1172,7 +1160,7 @@ static Value *GetHeapSROAValue(Value *V, unsigned FieldNo, // a new Load of the scalarized global. Result = new LoadInst(GetHeapSROAValue(LI->getOperand(0), FieldNo, InsertedScalarizedValues, - PHIsToRewrite, Context), + 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 @@ -1196,16 +1184,14 @@ static Value *GetHeapSROAValue(Value *V, unsigned FieldNo, /// 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, SCI->getPredicate(), NPtr, Constant::getNullValue(NPtr->getType()), @@ -1223,8 +1209,7 @@ static void RewriteHeapSROALoadUser(Instruction *LoadUser, // 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; @@ -1256,8 +1241,7 @@ static void RewriteHeapSROALoadUser(Instruction *LoadUser, // 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); } } @@ -1267,13 +1251,11 @@ static void RewriteHeapSROALoadUser(Instruction *LoadUser, /// 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()) { @@ -1285,8 +1267,7 @@ static void RewriteUsesOfLoadForHeapSRoA(LoadInst *Load, /// 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, CallInst *CI, - Value* NElems, LLVMContext &Context, - TargetData *TD) { + Value* NElems, TargetData *TD) { DEBUG(errs() << "SROA HEAP ALLOC: " << *GV << " MALLOC = " << *CI << '\n'); const Type* MAT = getMallocAllocatedType(CI); const StructType *STy = cast<StructType>(MAT); @@ -1315,14 +1296,16 @@ static GlobalVariable *PerformHeapAllocSRoA(GlobalVariable *GV, CallInst *CI, FieldGlobals.push_back(NGV); unsigned TypeSize = TD->getTypeAllocSize(FieldTy); - if (const StructType* ST = dyn_cast<StructType>(FieldTy)) + if (const StructType *ST = dyn_cast<StructType>(FieldTy)) TypeSize = TD->getStructLayout(ST)->getSizeInBytes(); - const Type* IntPtrTy = TD->getIntPtrType(Context); + 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); + CallInst *NCI = dyn_cast<BitCastInst>(NMI) ? + extractMallocCallFromBitCast(NMI) : cast<CallInst>(NMI); + FieldMallocs.push_back(NCI); new StoreInst(NMI, NGV, CI); } @@ -1338,15 +1321,15 @@ static GlobalVariable *PerformHeapAllocSRoA(GlobalVariable *GV, CallInst *CI, // 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(CI, ICmpInst::ICMP_EQ, FieldMallocs[i], Constant::getNullValue(FieldMallocs[i]->getType()), "isnull"); - if (!RunningOr) - RunningOr = Cond; // First seteq - else - RunningOr = BinaryOperator::CreateOr(RunningOr, Cond, "tmp", CI); + RunningOr = BinaryOperator::CreateOr(RunningOr, Cond, "tmp", CI); } // Split the basic block at the old malloc. @@ -1355,7 +1338,8 @@ static GlobalVariable *PerformHeapAllocSRoA(GlobalVariable *GV, CallInst *CI, // 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(Context, "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 @@ -1370,9 +1354,9 @@ static GlobalVariable *PerformHeapAllocSRoA(GlobalVariable *GV, CallInst *CI, Value *Cmp = new ICmpInst(*NullPtrBlock, ICmpInst::ICMP_NE, GVVal, Constant::getNullValue(GVVal->getType()), "tmp"); - BasicBlock *FreeBlock = BasicBlock::Create(Context, "free_it", + BasicBlock *FreeBlock = BasicBlock::Create(Cmp->getContext(), "free_it", OrigBB->getParent()); - BasicBlock *NextBlock = BasicBlock::Create(Context, "next", + BasicBlock *NextBlock = BasicBlock::Create(Cmp->getContext(), "next", OrigBB->getParent()); Instruction *BI = BranchInst::Create(FreeBlock, NextBlock, Cmp, NullPtrBlock); @@ -1406,8 +1390,7 @@ static GlobalVariable *PerformHeapAllocSRoA(GlobalVariable *GV, CallInst *CI, Instruction *User = cast<Instruction>(*UI++); if (LoadInst *LI = dyn_cast<LoadInst>(User)) { - RewriteUsesOfLoadForHeapSRoA(LI, InsertedScalarizedValues, PHIsToRewrite, - Context); + RewriteUsesOfLoadForHeapSRoA(LI, InsertedScalarizedValues, PHIsToRewrite); continue; } @@ -1438,7 +1421,7 @@ static GlobalVariable *PerformHeapAllocSRoA(GlobalVariable *GV, CallInst *CI, 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)); } } @@ -1477,8 +1460,7 @@ static bool TryToOptimizeStoreOfMallocToGlobal(GlobalVariable *GV, 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 (!AllocTy->isSized()) return false; @@ -1508,15 +1490,14 @@ static bool TryToOptimizeStoreOfMallocToGlobal(GlobalVariable *GV, // This eliminates dynamic allocation, avoids an indirection accessing the // data, and exposes the resultant global to further GlobalOpt. // We cannot optimize the malloc if we cannot determine malloc array size. - if (Value *NElems = getMallocArraySize(CI, Context, TD)) { + if (Value *NElems = getMallocArraySize(CI, TD, true)) { if (ConstantInt *NElements = dyn_cast<ConstantInt>(NElems)) // Restrict this transformation to only working on small allocations // (2048 bytes currently), as we don't want to introduce a 16M global or // something. if (TD && NElements->getZExtValue() * TD->getTypeAllocSize(AllocTy) < 2048) { - GVI = OptimizeGlobalAddressOfMalloc(GV, CI, AllocTy, NElems, - Context, TD); + GVI = OptimizeGlobalAddressOfMalloc(GV, CI, AllocTy, NElems, TD); return true; } @@ -1540,7 +1521,7 @@ static bool TryToOptimizeStoreOfMallocToGlobal(GlobalVariable *GV, // structs. malloc [100 x struct],1 -> malloc struct, 100 if (const ArrayType *AT = dyn_cast<ArrayType>(getMallocAllocatedType(CI))) { - const Type *IntPtrTy = TD->getIntPtrType(Context); + 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()); @@ -1551,12 +1532,10 @@ static bool TryToOptimizeStoreOfMallocToGlobal(GlobalVariable *GV, CI->replaceAllUsesWith(Cast); CI->eraseFromParent(); CI = dyn_cast<BitCastInst>(Malloc) ? - extractMallocCallFromBitCast(Malloc): - cast<CallInst>(Malloc); + extractMallocCallFromBitCast(Malloc) : cast<CallInst>(Malloc); } - GVI = PerformHeapAllocSRoA(GV, CI, getMallocArraySize(CI, Context, TD), - Context, TD); + GVI = PerformHeapAllocSRoA(GV, CI, getMallocArraySize(CI, TD, true),TD); return true; } } @@ -1569,7 +1548,7 @@ static bool TryToOptimizeStoreOfMallocToGlobal(GlobalVariable *GV, // 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(); @@ -1585,12 +1564,12 @@ static bool OptimizeOnceStoredGlobal(GlobalVariable *GV, Value *StoredOnceVal, 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 (CallInst *CI = extractMallocCall(StoredOnceVal)) { const Type* MallocType = getMallocAllocatedType(CI); if (MallocType && TryToOptimizeStoreOfMallocToGlobal(GV, CI, MallocType, - GVI, TD, Context)) + GVI, TD)) return true; } } @@ -1602,8 +1581,7 @@ static bool OptimizeOnceStoredGlobal(GlobalVariable *GV, Value *StoredOnceVal, /// 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 @@ -1611,7 +1589,8 @@ static bool TryToShrinkGlobalToBoolean(GlobalVariable *GV, Constant *OtherVal, // between them is very expensive and unlikely to lead to later // simplification. In these cases, we typically end up with "cond ? v1 : v2" // where v1 and v2 both require constant pool loads, a big loss. - if (GVElType == Type::getInt1Ty(Context) || GVElType->isFloatingPoint() || + if (GVElType == Type::getInt1Ty(GV->getContext()) || + GVElType->isFloatingPoint() || isa<PointerType>(GVElType) || isa<VectorType>(GVElType)) return false; @@ -1624,15 +1603,16 @@ static bool TryToShrinkGlobalToBoolean(GlobalVariable *GV, Constant *OtherVal, DEBUG(errs() << " *** SHRINKING TO BOOL: " << *GV); // Create the new global, initializing it to false. - GlobalVariable *NewGV = new GlobalVariable(Context, - Type::getInt1Ty(Context), false, - GlobalValue::InternalLinkage, ConstantInt::getFalse(Context), + GlobalVariable *NewGV = new GlobalVariable(Type::getInt1Ty(GV->getContext()), + false, + GlobalValue::InternalLinkage, + ConstantInt::getFalse(GV->getContext()), GV->getName()+".b", GV->isThreadLocal()); GV->getParent()->getGlobalList().insert(GV, NewGV); Constant *InitVal = GV->getInitializer(); - assert(InitVal->getType() != Type::getInt1Ty(Context) && + 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 @@ -1649,7 +1629,8 @@ static bool TryToShrinkGlobalToBoolean(GlobalVariable *GV, Constant *OtherVal, // Only do this if we weren't storing a loaded value. Value *StoreVal; if (StoringOther || SI->getOperand(0) == InitVal) - StoreVal = ConstantInt::get(Type::getInt1Ty(Context), 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 @@ -1708,24 +1689,26 @@ bool GlobalOpt::ProcessInternalGlobal(GlobalVariable *GV, if (!AnalyzeGlobal(GV, GS, PHIUsers)) { #if 0 - cerr << "Global: " << *GV; - cerr << " isLoaded = " << GS.isLoaded << "\n"; - cerr << " StoredType = "; + DEBUG(errs() << "Global: " << *GV); + DEBUG(errs() << " isLoaded = " << GS.isLoaded << "\n"); + DEBUG(errs() << " 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(errs() << "NEVER STORED\n"); break; + case GlobalStatus::isInitializerStored: DEBUG(errs() << "INIT STORED\n"); + break; + case GlobalStatus::isStoredOnce: DEBUG(errs() << "STORED ONCE\n"); break; + case GlobalStatus::isStored: DEBUG(errs() << "stored\n"); break; } if (GS.StoredType == GlobalStatus::isStoredOnce && GS.StoredOnceValue) - cerr << " StoredOnceValue = " << *GS.StoredOnceValue << "\n"; + DEBUG(errs() << " 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(errs() << " AccessingFunction = " << GS.AccessingFunction->getName() + << "\n"); + DEBUG(errs() << " HasMultipleAccessingFunctions = " + << GS.HasMultipleAccessingFunctions << "\n"); + DEBUG(errs() << " HasNonInstructionUser = " + << GS.HasNonInstructionUser<<"\n"); + DEBUG(errs() << "\n"); #endif // If this is a first class global and has only one accessing function @@ -1764,8 +1747,7 @@ bool GlobalOpt::ProcessInternalGlobal(GlobalVariable *GV, // Delete any stores we can find to the global. We may not be able to // make it completely dead though. - bool Changed = CleanupConstantGlobalUsers(GV, GV->getInitializer(), - GV->getContext()); + bool Changed = CleanupConstantGlobalUsers(GV, GV->getInitializer()); // If the global is dead now, delete it. if (GV->use_empty()) { @@ -1780,7 +1762,7 @@ bool GlobalOpt::ProcessInternalGlobal(GlobalVariable *GV, GV->setConstant(true); // Clean up any obviously simplifiable users now. - CleanupConstantGlobalUsers(GV, GV->getInitializer(), GV->getContext()); + CleanupConstantGlobalUsers(GV, GV->getInitializer()); // If the global is dead now, just nuke it. if (GV->use_empty()) { @@ -1794,8 +1776,7 @@ bool GlobalOpt::ProcessInternalGlobal(GlobalVariable *GV, return true; } else if (!GV->getInitializer()->getType()->isSingleValueType()) { if (TargetData *TD = getAnalysisIfAvailable<TargetData>()) - if (GlobalVariable *FirstNewGV = SRAGlobal(GV, *TD, - GV->getContext())) { + if (GlobalVariable *FirstNewGV = SRAGlobal(GV, *TD)) { GVI = FirstNewGV; // Don't skip the newly produced globals! return true; } @@ -1810,8 +1791,7 @@ bool GlobalOpt::ProcessInternalGlobal(GlobalVariable *GV, GV->setInitializer(SOVConstant); // Clean up any obviously simplifiable users now. - CleanupConstantGlobalUsers(GV, GV->getInitializer(), - GV->getContext()); + CleanupConstantGlobalUsers(GV, GV->getInitializer()); if (GV->use_empty()) { DEBUG(errs() << " *** Substituting initializer allowed us to " @@ -1828,14 +1808,13 @@ bool GlobalOpt::ProcessInternalGlobal(GlobalVariable *GV, // 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, - getAnalysisIfAvailable<TargetData>(), - GV->getContext())) + 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, GV->getContext())) { + if (TryToShrinkGlobalToBoolean(GV, SOVConstant)) { ++NumShrunkToBool; return true; } @@ -1987,11 +1966,10 @@ static std::vector<Function*> ParseGlobalCtors(GlobalVariable *GV) { /// 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(ConstantInt::get(Type::getInt32Ty(Context), 65535)); + CSVals.push_back(ConstantInt::get(Type::getInt32Ty(GCL->getContext()),65535)); CSVals.push_back(0); // Create the new init list. @@ -2000,12 +1978,14 @@ static GlobalVariable *InstallGlobalCtors(GlobalVariable *GCL, if (Ctors[i]) { CSVals[1] = Ctors[i]; } else { - const Type *FTy = FunctionType::get(Type::getVoidTy(Context), false); + const Type *FTy = FunctionType::get(Type::getVoidTy(GCL->getContext()), + false); const PointerType *PFTy = PointerType::getUnqual(FTy); CSVals[1] = Constant::getNullValue(PFTy); - CSVals[0] = ConstantInt::get(Type::getInt32Ty(Context), 2147483647); + CSVals[0] = ConstantInt::get(Type::getInt32Ty(GCL->getContext()), + 2147483647); } - CAList.push_back(ConstantStruct::get(Context, CSVals, false)); + CAList.push_back(ConstantStruct::get(GCL->getContext(), CSVals, false)); } // Create the array initializer. @@ -2021,8 +2001,7 @@ static GlobalVariable *InstallGlobalCtors(GlobalVariable *GCL, } // Create the new global and insert it next to the existing list. - GlobalVariable *NGV = new GlobalVariable(Context, CA->getType(), - GCL->isConstant(), + GlobalVariable *NGV = new GlobalVariable(CA->getType(), GCL->isConstant(), GCL->getLinkage(), CA, "", GCL->isThreadLocal()); GCL->getParent()->getGlobalList().insert(GCL, NGV); @@ -2056,7 +2035,7 @@ static Constant *getVal(DenseMap<Value*, Constant*> &ComputedValues, /// enough for us to understand. In particular, if it is a cast of something, /// 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) { +static bool isSimpleEnoughPointerToCommit(Constant *C) { // 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()) @@ -2096,8 +2075,7 @@ static bool isSimpleEnoughPointerToCommit(Constant *C, LLVMContext &Context) { /// 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!"); @@ -2126,10 +2104,11 @@ static Constant *EvaluateStoreInto(Constant *Init, Constant *Val, 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 ConstantStruct::get(Context, &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()); @@ -2152,15 +2131,14 @@ static Constant *EvaluateStoreInto(Constant *Init, Constant *Val, assert(CI->getZExtValue() < ATy->getNumElements()); Elts[CI->getZExtValue()] = - EvaluateStoreInto(Elts[CI->getZExtValue()], Val, Addr, OpNo+1, Context); + EvaluateStoreInto(Elts[CI->getZExtValue()], Val, Addr, OpNo+1); return ConstantArray::get(ATy, Elts); } } /// 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); @@ -2171,7 +2149,7 @@ static void CommitValueTo(Constant *Val, Constant *Addr, GlobalVariable *GV = cast<GlobalVariable>(CE->getOperand(0)); Constant *Init = GV->getInitializer(); - Init = EvaluateStoreInto(Init, Val, CE, 2, Context); + Init = EvaluateStoreInto(Init, Val, CE, 2); GV->setInitializer(Init); } @@ -2179,8 +2157,7 @@ static void CommitValueTo(Constant *Val, Constant *Addr, /// P after the stores reflected by 'memory' have been performed. If we can't /// decide, return null. static Constant *ComputeLoadResult(Constant *P, - const DenseMap<Constant*, Constant*> &Memory, - LLVMContext &Context) { + const DenseMap<Constant*, Constant*> &Memory) { // If this memory location has been recently stored, use the stored value: it // is the most up-to-date. DenseMap<Constant*, Constant*>::const_iterator I = Memory.find(P); @@ -2218,8 +2195,6 @@ static bool EvaluateFunction(Function *F, Constant *&RetVal, 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. @@ -2246,7 +2221,7 @@ static bool EvaluateFunction(Function *F, Constant *&RetVal, if (StoreInst *SI = dyn_cast<StoreInst>(CurInst)) { if (SI->isVolatile()) return false; // no volatile accesses. Constant *Ptr = getVal(Values, SI->getOperand(1)); - if (!isSimpleEnoughPointerToCommit(Ptr, Context)) + if (!isSimpleEnoughPointerToCommit(Ptr)) // If this is too complex for us to commit, reject it. return false; Constant *Val = getVal(Values, SI->getOperand(0)); @@ -2280,12 +2255,12 @@ static bool EvaluateFunction(Function *F, Constant *&RetVal, } else if (LoadInst *LI = dyn_cast<LoadInst>(CurInst)) { if (LI->isVolatile()) return false; // no volatile accesses. InstResult = ComputeLoadResult(getVal(Values, LI->getOperand(0)), - MutatedMemory, Context); + MutatedMemory); if (InstResult == 0) return false; // Could not evaluate load. } else if (AllocaInst *AI = dyn_cast<AllocaInst>(CurInst)) { if (AI->isArrayAllocation()) return false; // Cannot handle array allocs. const Type *Ty = AI->getType()->getElementType(); - AllocaTmps.push_back(new GlobalVariable(Context, Ty, false, + AllocaTmps.push_back(new GlobalVariable(Ty, false, GlobalValue::InternalLinkage, UndefValue::get(Ty), AI->getName())); @@ -2423,7 +2398,7 @@ static bool EvaluateStaticConstructor(Function *F) { << " 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' @@ -2480,7 +2455,7 @@ bool GlobalOpt::OptimizeGlobalCtorsList(GlobalVariable *&GCL) { if (!MadeChange) return false; - GCL = InstallGlobalCtors(GCL, Ctors, GCL->getContext()); + GCL = InstallGlobalCtors(GCL, Ctors); return true; } |