#include "llvm/Module.h"
#include "llvm/Pass.h"
#include "llvm/Analysis/ConstantFolding.h"
-#include "llvm/Analysis/MallocHelper.h"
+#include "llvm/Analysis/MallocFreeHelper.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Support/CallSite.h"
-#include "llvm/Support/Compiler.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/GetElementPtrTypeIterator.h"
STATISTIC(NumAliasesRemoved, "Number of global aliases eliminated");
namespace {
- struct VISIBILITY_HIDDEN GlobalOpt : public ModulePass {
+ struct GlobalOpt : public ModulePass {
virtual void getAnalysisUsage(AnalysisUsage &AU) const {
}
static char ID; // Pass identification, replacement for typeid
/// 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;
if (CE->getOpcode() == Instruction::GetElementPtr) {
Constant *SubInit = 0;
if (Init)
- SubInit = ConstantFoldLoadThroughGEPConstantExpr(Init, CE, Context);
+ SubInit = ConstantFoldLoadThroughGEPConstantExpr(Init, CE);
Changed |= CleanupConstantGlobalUsers(CE, SubInit, Context);
} else if (CE->getOpcode() == Instruction::BitCast &&
isa<PointerType>(CE->getType())) {
ConstantExpr *CE =
dyn_cast_or_null<ConstantExpr>(ConstantFoldInstruction(GEP, Context));
if (Init && CE && CE->getOpcode() == Instruction::GetElementPtr)
- SubInit = ConstantFoldLoadThroughGEPConstantExpr(Init, CE, Context);
+ SubInit = ConstantFoldLoadThroughGEPConstantExpr(Init, CE);
}
Changed |= CleanupConstantGlobalUsers(GEP, SubInit, Context);
}
}
-/// OptimizeGlobalAddressOfMalloc - This function takes the specified global
-/// variable, and transforms the program as if it always contained the result of
-/// the specified malloc. Because it is always the result of the specified
-/// 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) {
- DEBUG(errs() << "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 = ArrayType::get(MI->getAllocatedType(),
- NElements->getZExtValue());
- MallocInst *NewMI =
- new MallocInst(NewTy, Constant::getNullValue(Type::getInt32Ty(Context)),
- MI->getAlignment(), MI->getName(), MI);
- Value* Indices[2];
- Indices[0] = Indices[1] = Constant::getNullValue(Type::getInt32Ty(Context));
- Value *NewGEP = GetElementPtrInst::Create(NewMI, Indices, Indices + 2,
- NewMI->getName()+".el0", MI);
- MI->replaceAllUsesWith(NewGEP);
- MI->eraseFromParent();
- MI = NewMI;
- }
-
- // Create the new global variable. The contents of the malloc'd memory is
- // undefined, so initialize with an undef value.
- // FIXME: This new global should have the alignment returned by malloc. Code
- // could depend on malloc returning large alignment (on the mac, 16 bytes) but
- // this would only guarantee some lower alignment.
- Constant *Init = UndefValue::get(MI->getAllocatedType());
- GlobalVariable *NewGV = new GlobalVariable(*GV->getParent(),
- MI->getAllocatedType(), false,
- GlobalValue::InternalLinkage, Init,
- GV->getName()+".body",
- GV,
- GV->isThreadLocal());
-
- // Anything that used the malloc now uses the global directly.
- MI->replaceAllUsesWith(NewGV);
-
- Constant *RepValue = NewGV;
- if (NewGV->getType() != GV->getType()->getElementType())
- 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(Context, Type::getInt1Ty(Context), false,
- GlobalValue::InternalLinkage,
- ConstantInt::getFalse(Context), 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: llvm_unreachable("Unknown ICmp Predicate!");
- case ICmpInst::ICMP_ULT:
- case ICmpInst::ICMP_SLT:
- LV = ConstantInt::getFalse(Context); // 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());
- // The global is initialized when the store to it occurs.
- new StoreInst(ConstantInt::getTrue(Context), InitBool, SI);
- SI->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();
- delete InitBool;
- } else
- GV->getParent()->getGlobalList().insert(GV, InitBool);
-
-
- // Now the GV is dead, nuke it and the malloc.
- GV->eraseFromParent();
- MI->eraseFromParent();
-
- // 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);
- if (RepValue != NewGV)
- ConstantPropUsersOf(RepValue, Context);
-
- return NewGV;
-}
-
/// OptimizeGlobalAddressOfMalloc - This function takes the specified global
/// variable, and transforms the program as if it always contained the result of
/// the specified malloc. Because it is always the result of the specified
BitCastInst *BCI,
LLVMContext &Context,
TargetData* TD) {
+ DEBUG(errs() << "PROMOTING MALLOC GLOBAL: " << *GV
+ << " CALL = " << *CI << " BCI = " << *BCI << '\n');
+
const Type *IntPtrTy = TD->getIntPtrType(Context);
- DEBUG(errs() << "PROMOTING MALLOC GLOBAL: " << *GV << " MALLOC = " << *CI);
-
- ConstantInt *NElements = cast<ConstantInt>(getMallocArraySize(CI,
- Context, TD));
+ Value* ArraySize = getMallocArraySize(CI, Context, TD);
+ assert(ArraySize && "not a malloc whose array size can be determined");
+ ConstantInt *NElements = cast<ConstantInt>(ArraySize);
if (NElements->getZExtValue() != 1) {
// If we have an array allocation, transform it to a single element
// allocation to make the code below simpler.
// Create the new global variable. The contents of the malloc'd memory is
// undefined, so initialize with an undef value.
- // FIXME: This new global should have the alignment returned by malloc. Code
- // could depend on malloc returning large alignment (on the mac, 16 bytes) but
- // this would only guarantee some lower alignment.
const Type *MAT = getMallocAllocatedType(CI);
Constant *Init = UndefValue::get(MAT);
GlobalVariable *NewGV = new GlobalVariable(*GV->getParent(),
}
}
-/// PerformHeapAllocSRoA - MI 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){
- DEBUG(errs() << "SROA HEAP ALLOC: " << *GV << " MALLOC = " << *MI);
- const StructType *STy = cast<StructType>(MI->getAllocatedType());
-
- // 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);
-
- // Okay, at this point, there are no users of the malloc. Insert N
- // new mallocs at the same place as MI, and N globals.
- std::vector<Value*> FieldGlobals;
- std::vector<MallocInst*> FieldMallocs;
-
- for (unsigned FieldNo = 0, e = STy->getNumElements(); FieldNo != e;++FieldNo){
- const Type *FieldTy = STy->getElementType(FieldNo);
- const Type *PFieldTy = PointerType::getUnqual(FieldTy);
-
- GlobalVariable *NGV =
- 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" + Twine(FieldNo), MI);
- FieldMallocs.push_back(NMI);
- new StoreInst(NMI, NGV, MI);
- }
-
- // The tricky aspect of this transformation is handling the case when malloc
- // fails. In the original code, malloc failing would set the result pointer
- // of malloc to null. In this case, some mallocs could succeed and others
- // could fail. As such, we emit code that looks like this:
- // F0 = malloc(field0)
- // F1 = malloc(field1)
- // F2 = malloc(field2)
- // if (F0 == 0 || F1 == 0 || F2 == 0) {
- // if (F0) { free(F0); F0 = 0; }
- // if (F1) { free(F1); F1 = 0; }
- // if (F2) { free(F2); F2 = 0; }
- // }
- Value *RunningOr = 0;
- for (unsigned i = 0, e = FieldMallocs.size(); i != e; ++i) {
- Value *Cond = new ICmpInst(MI, ICmpInst::ICMP_EQ, FieldMallocs[i],
- Constant::getNullValue(FieldMallocs[i]->getType()),
- "isnull");
- if (!RunningOr)
- RunningOr = Cond; // First seteq
- else
- RunningOr = BinaryOperator::CreateOr(RunningOr, Cond, "tmp", MI);
- }
-
- // Split the basic block at the old malloc.
- BasicBlock *OrigBB = MI->getParent();
- BasicBlock *ContBB = OrigBB->splitBasicBlock(MI, "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(Context, "malloc_ret_null",
- OrigBB->getParent());
-
- // Remove the uncond branch from OrigBB to ContBB, turning it into a cond
- // branch on RunningOr.
- OrigBB->getTerminator()->eraseFromParent();
- BranchInst::Create(NullPtrBlock, ContBB, RunningOr, OrigBB);
-
- // Within the NullPtrBlock, we need to emit a comparison and branch for each
- // 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(*NullPtrBlock, ICmpInst::ICMP_NE, GVVal,
- Constant::getNullValue(GVVal->getType()),
- "tmp");
- BasicBlock *FreeBlock = BasicBlock::Create(Context, "free_it",
- OrigBB->getParent());
- BasicBlock *NextBlock = BasicBlock::Create(Context, "next",
- OrigBB->getParent());
- BranchInst::Create(FreeBlock, NextBlock, Cmp, NullPtrBlock);
-
- // Fill in FreeBlock.
- new FreeInst(GVVal, FreeBlock);
- new StoreInst(Constant::getNullValue(GVVal->getType()), FieldGlobals[i],
- FreeBlock);
- BranchInst::Create(NextBlock, FreeBlock);
-
- NullPtrBlock = NextBlock;
- }
-
- BranchInst::Create(ContBB, NullPtrBlock);
-
- // MI is no longer needed, remove it.
- MI->eraseFromParent();
-
- /// InsertedScalarizedLoads - As we process loads, if we can't immediately
- /// update all uses of the load, keep track of what scalarized loads are
- /// inserted for a given load.
- DenseMap<Value*, std::vector<Value*> > InsertedScalarizedValues;
- InsertedScalarizedValues[GV] = FieldGlobals;
-
- std::vector<std::pair<PHINode*, unsigned> > PHIsToRewrite;
-
- // Okay, the malloc site is completely handled. All of the uses of GV are now
- // loads, and all uses of those loads are simple. Rewrite them to use loads
- // of the per-field globals instead.
- for (Value::use_iterator UI = GV->use_begin(), E = GV->use_end(); UI != E;) {
- Instruction *User = cast<Instruction>(*UI++);
-
- if (LoadInst *LI = dyn_cast<LoadInst>(User)) {
- RewriteUsesOfLoadForHeapSRoA(LI, InsertedScalarizedValues, PHIsToRewrite,
- Context);
- continue;
- }
-
- // Must be a store of null.
- StoreInst *SI = cast<StoreInst>(User);
- assert(isa<ConstantPointerNull>(SI->getOperand(0)) &&
- "Unexpected heap-sra user!");
-
- // 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 = Constant::getNullValue(PT->getElementType());
- new StoreInst(Null, FieldGlobals[i], SI);
- }
- // Erase the original store.
- SI->eraseFromParent();
- }
-
- // While we have PHIs that are interesting to rewrite, do it.
- while (!PHIsToRewrite.empty()) {
- PHINode *PN = PHIsToRewrite.back().first;
- unsigned FieldNo = PHIsToRewrite.back().second;
- PHIsToRewrite.pop_back();
- PHINode *FieldPN = cast<PHINode>(InsertedScalarizedValues[PN][FieldNo]);
- assert(FieldPN->getNumIncomingValues() == 0 &&"Already processed this phi");
-
- // Add all the incoming values. This can materialize more phis.
- for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
- Value *InVal = PN->getIncomingValue(i);
- InVal = GetHeapSROAValue(InVal, FieldNo, InsertedScalarizedValues,
- PHIsToRewrite, Context);
- FieldPN->addIncoming(InVal, PN->getIncomingBlock(i));
- }
- }
-
- // Drop all inter-phi links and any loads that made it this far.
- for (DenseMap<Value*, std::vector<Value*> >::iterator
- I = InsertedScalarizedValues.begin(), E = InsertedScalarizedValues.end();
- I != E; ++I) {
- if (PHINode *PN = dyn_cast<PHINode>(I->first))
- PN->dropAllReferences();
- else if (LoadInst *LI = dyn_cast<LoadInst>(I->first))
- LI->dropAllReferences();
- }
-
- // Delete all the phis and loads now that inter-references are dead.
- for (DenseMap<Value*, std::vector<Value*> >::iterator
- I = InsertedScalarizedValues.begin(), E = InsertedScalarizedValues.end();
- I != E; ++I) {
- if (PHINode *PN = dyn_cast<PHINode>(I->first))
- PN->eraseFromParent();
- else if (LoadInst *LI = dyn_cast<LoadInst>(I->first))
- LI->eraseFromParent();
- }
-
- // The old global is now dead, remove it.
- GV->eraseFromParent();
-
- ++NumHeapSRA;
- return cast<GlobalVariable>(FieldGlobals[0]);
-}
-
/// 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,
<< " BITCAST = " << *BCI << '\n');
const Type* MAT = getMallocAllocatedType(CI);
const StructType *STy = cast<StructType>(MAT);
+ Value* ArraySize = getMallocArraySize(CI, Context, TD);
+ assert(ArraySize && "not a malloc whose array size can be determined");
// There is guaranteed to be at least one use of the malloc (storing
// it into GV). If there are other uses, change them to be uses of
GV->isThreadLocal());
FieldGlobals.push_back(NGV);
- Value *NMI = CallInst::CreateMalloc(CI, TD->getIntPtrType(Context), FieldTy,
- getMallocArraySize(CI, Context, TD),
+ Value *NMI = CallInst::CreateMalloc(CI, TD->getIntPtrType(Context),
+ FieldTy, ArraySize,
BCI->getName() + ".f" + Twine(FieldNo));
FieldMallocs.push_back(NMI);
new StoreInst(NMI, NGV, BCI);
OrigBB->getParent());
BasicBlock *NextBlock = BasicBlock::Create(Context, "next",
OrigBB->getParent());
- BranchInst::Create(FreeBlock, NextBlock, Cmp, NullPtrBlock);
+ Instruction *BI = BranchInst::Create(FreeBlock, NextBlock,
+ Cmp, NullPtrBlock);
// Fill in FreeBlock.
- new FreeInst(GVVal, FreeBlock);
+ CallInst::CreateFree(GVVal, BI);
new StoreInst(Constant::getNullValue(GVVal->getType()), FieldGlobals[i],
FreeBlock);
BranchInst::Create(NextBlock, FreeBlock);
return cast<GlobalVariable>(FieldGlobals[0]);
}
-/// TryToOptimizeStoreOfMallocToGlobal - This function is called when we see a
-/// pointer global variable with a single value stored it that is a malloc or
-/// cast of malloc.
-static bool TryToOptimizeStoreOfMallocToGlobal(GlobalVariable *GV,
- MallocInst *MI,
- Module::global_iterator &GVI,
- TargetData *TD,
- LLVMContext &Context) {
- // If this is a malloc of an abstract type, don't touch it.
- if (!MI->getAllocatedType()->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
- // been reached). To do this, we check to see if all uses of the global
- // would trap if the global were null: this proves that they must all
- // 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
- // GEP'd. These are all things we could transform to using the global
- // for.
- {
- SmallPtrSet<PHINode*, 8> PHIs;
- if (!ValueIsOnlyUsedLocallyOrStoredToOneGlobal(MI, 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 (TD &&
- 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();
-
- 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 this is a fixed size array, transform the Malloc to be an alloc of
- // structs. malloc [100 x struct],1 -> malloc struct, 100
- if (const ArrayType *AT = dyn_cast<ArrayType>(MI->getAllocatedType())) {
- MallocInst *NewMI =
- new MallocInst(AllocSTy,
- ConstantInt::get(Type::getInt32Ty(Context),
- 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, MI, Context);
- return true;
- }
- }
-
- return false;
-}
-
/// TryToOptimizeStoreOfMallocToGlobal - This function is called when we see a
/// pointer global variable with a single value stored it that is a malloc or
/// cast of 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>(getMallocArraySize(CI, Context, TD))) {
- // Restrict this transformation to only working on small allocations
- // (2048 bytes currently), as we don't want to introduce a 16M global or
- // something.
- if (TD &&
- NElements->getZExtValue() * TD->getTypeAllocSize(AllocTy) < 2048) {
- GVI = OptimizeGlobalAddressOfMalloc(GV, CI, BCI, Context, TD);
- return true;
- }
- }
-
- // If the allocation is an array of structures, consider transforming this
- // into multiple malloc'd arrays, one for each field. This is basically
- // SRoA for malloc'd memory.
-
- // If this is an allocation of a fixed size array of structs, analyze as a
- // variable size array. malloc [100 x struct],1 -> malloc struct, 100
- if (!isArrayMalloc(CI, Context, TD))
- if (const ArrayType *AT = dyn_cast<ArrayType>(AllocTy))
- AllocTy = AT->getElementType();
-
- if (const StructType *AllocSTy = dyn_cast<StructType>(AllocTy)) {
- // This the structure has an unreasonable number of fields, leave it
- // alone.
- if (AllocSTy->getNumElements() <= 16 && AllocSTy->getNumElements() != 0 &&
- AllGlobalLoadUsesSimpleEnoughForHeapSRA(GV, BCI)) {
-
- // If this is a fixed size array, transform the Malloc to be an alloc of
- // structs. malloc [100 x struct],1 -> malloc struct, 100
- if (const ArrayType *AT = dyn_cast<ArrayType>(getMallocAllocatedType(CI))) {
- Value* NumElements = ConstantInt::get(Type::getInt32Ty(Context),
- AT->getNumElements());
- Value* NewMI = CallInst::CreateMalloc(CI, TD->getIntPtrType(Context),
- AllocSTy, NumElements,
- BCI->getName());
- Value *Cast = new BitCastInst(NewMI, getMallocType(CI), "tmp", CI);
- BCI->replaceAllUsesWith(Cast);
- BCI->eraseFromParent();
- CI->eraseFromParent();
- BCI = cast<BitCastInst>(NewMI);
- CI = extractMallocCallFromBitCast(NewMI);
+ Value *NElems = getMallocArraySize(CI, Context, TD);
+ // We cannot optimize the malloc if we cannot determine malloc array size.
+ if (NElems) {
+ if (ConstantInt *NElements = dyn_cast<ConstantInt>(NElems))
+ // Restrict this transformation to only working on small allocations
+ // (2048 bytes currently), as we don't want to introduce a 16M global or
+ // something.
+ if (TD &&
+ NElements->getZExtValue() * TD->getTypeAllocSize(AllocTy) < 2048) {
+ GVI = OptimizeGlobalAddressOfMalloc(GV, CI, BCI, Context, TD);
+ return true;
}
+
+ // If the allocation is an array of structures, consider transforming this
+ // into multiple malloc'd arrays, one for each field. This is basically
+ // SRoA for malloc'd memory.
+
+ // If this is an allocation of a fixed size array of structs, analyze as a
+ // variable size array. malloc [100 x struct],1 -> malloc struct, 100
+ if (!isArrayMalloc(CI, Context, TD))
+ if (const ArrayType *AT = dyn_cast<ArrayType>(AllocTy))
+ AllocTy = AT->getElementType();
+
+ if (const StructType *AllocSTy = dyn_cast<StructType>(AllocTy)) {
+ // This the structure has an unreasonable number of fields, leave it
+ // alone.
+ if (AllocSTy->getNumElements() <= 16 && AllocSTy->getNumElements() != 0 &&
+ AllGlobalLoadUsesSimpleEnoughForHeapSRA(GV, BCI)) {
+
+ // If this is a fixed size array, transform the Malloc to be an alloc of
+ // structs. malloc [100 x struct],1 -> malloc struct, 100
+ if (const ArrayType *AT =
+ dyn_cast<ArrayType>(getMallocAllocatedType(CI))) {
+ Value* NumElements = ConstantInt::get(Type::getInt32Ty(Context),
+ AT->getNumElements());
+ Value* NewMI = CallInst::CreateMalloc(CI, TD->getIntPtrType(Context),
+ AllocSTy, NumElements,
+ BCI->getName());
+ Value *Cast = new BitCastInst(NewMI, getMallocType(CI), "tmp", CI);
+ BCI->replaceAllUsesWith(Cast);
+ BCI->eraseFromParent();
+ CI->eraseFromParent();
+ BCI = cast<BitCastInst>(NewMI);
+ CI = extractMallocCallFromBitCast(NewMI);
+ }
- GVI = PerformHeapAllocSRoA(GV, CI, BCI, Context, TD);
- return true;
+ GVI = PerformHeapAllocSRoA(GV, CI, BCI, Context, TD);
+ return true;
+ }
}
}
// Optimize away any trapping uses of the loaded value.
if (OptimizeAwayTrappingUsesOfLoads(GV, SOVC, Context))
return true;
- } else if (MallocInst *MI = dyn_cast<MallocInst>(StoredOnceVal)) {
- if (TryToOptimizeStoreOfMallocToGlobal(GV, MI, GVI, TD, Context))
- return true;
} else if (CallInst *CI = extractMallocCall(StoredOnceVal)) {
if (getMallocAllocatedType(CI)) {
BitCastInst* BCI = NULL;
if (!CE->isGEPWithNoNotionalOverIndexing())
return false;
- return ConstantFoldLoadThroughGEPConstantExpr(GV->getInitializer(), CE,
- Context);
+ return ConstantFoldLoadThroughGEPConstantExpr(GV->getInitializer(), CE);
}
return false;
}
isa<GlobalVariable>(CE->getOperand(0))) {
GlobalVariable *GV = cast<GlobalVariable>(CE->getOperand(0));
if (GV->hasDefinitiveInitializer())
- return ConstantFoldLoadThroughGEPConstantExpr(GV->getInitializer(), CE,
- Context);
+ return ConstantFoldLoadThroughGEPConstantExpr(GV->getInitializer(), CE);
}
return 0; // don't know how to evaluate.
projects / oota-llvm.git / blobdiff