#include "llvm/Instructions.h"
#include "llvm/IntrinsicInst.h"
#include "llvm/Module.h"
-#include "llvm/ParameterAttributes.h"
#include "llvm/Pass.h"
#include "llvm/Analysis/ConstantFolding.h"
#include "llvm/Target/TargetData.h"
+#include "llvm/Support/CallSite.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/GetElementPtrTypeIterator.h"
+#include "llvm/Support/MathExtras.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/StringExtras.h"
#include <algorithm>
+#include <map>
#include <set>
using namespace llvm;
bool OptimizeGlobalCtorsList(GlobalVariable *&GCL);
bool ProcessInternalGlobal(GlobalVariable *GV,Module::global_iterator &GVI);
};
-
- char GlobalOpt::ID = 0;
- RegisterPass<GlobalOpt> X("globalopt", "Global Variable Optimizer");
}
+char GlobalOpt::ID = 0;
+static RegisterPass<GlobalOpt> X("globalopt", "Global Variable Optimizer");
+
ModulePass *llvm::createGlobalOptimizerPass() { return new GlobalOpt(); }
+namespace {
+
/// 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.
HasNonInstructionUser(false), HasPHIUser(false) {}
};
-
+}
/// ConstantIsDead - Return true if the specified constant is (transitively)
/// dead. The constant may be used by other constants (e.g. constant arrays and
// If this is a direct store to the global (i.e., the global is a scalar
// value, not an aggregate), keep more specific information about
// stores.
- if (GS.StoredType != GlobalStatus::isStored)
+ if (GS.StoredType != GlobalStatus::isStored) {
if (GlobalVariable *GV = dyn_cast<GlobalVariable>(SI->getOperand(1))){
Value *StoredVal = SI->getOperand(0);
if (StoredVal == GV->getInitializer()) {
} else {
GS.StoredType = GlobalStatus::isStored;
}
+ }
} else if (isa<GetElementPtrInst>(I)) {
if (AnalyzeGlobal(I, GS, PHIUsers)) return true;
} else if (isa<SelectInst>(I)) {
/// 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) {
+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;
std::vector<GlobalVariable*> NewGlobals;
Module::GlobalListType &Globals = GV->getParent()->getGlobalList();
+ // Get the alignment of the global, either explicit or target-specific.
+ unsigned StartAlignment = GV->getAlignment();
+ if (StartAlignment == 0)
+ StartAlignment = TD.getABITypeAlignment(GV->getType());
+
if (const StructType *STy = dyn_cast<StructType>(Ty)) {
NewGlobals.reserve(STy->getNumElements());
+ const StructLayout &Layout = *TD.getStructLayout(STy);
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
Constant *In = getAggregateConstantElement(Init,
ConstantInt::get(Type::Int32Ty, i));
GV->isThreadLocal());
Globals.insert(GV, NGV);
NewGlobals.push_back(NGV);
+
+ // Calculate the known alignment of the field. If the original aggregate
+ // had 256 byte alignment for example, something might depend on that:
+ // propagate info to each field.
+ uint64_t FieldOffset = Layout.getElementOffset(i);
+ unsigned NewAlign = (unsigned)MinAlign(StartAlignment, FieldOffset);
+ if (NewAlign > TD.getABITypeAlignment(STy->getElementType(i)))
+ NGV->setAlignment(NewAlign);
}
} else if (const SequentialType *STy = dyn_cast<SequentialType>(Ty)) {
unsigned NumElements = 0;
if (const ArrayType *ATy = dyn_cast<ArrayType>(STy))
NumElements = ATy->getNumElements();
- else if (const VectorType *PTy = dyn_cast<VectorType>(STy))
- NumElements = PTy->getNumElements();
else
- assert(0 && "Unknown aggregate sequential type!");
+ NumElements = cast<VectorType>(STy)->getNumElements();
if (NumElements > 16 && GV->hasNUsesOrMore(16))
return 0; // It's not worth it.
NewGlobals.reserve(NumElements);
+
+ uint64_t EltSize = TD.getABITypeSize(STy->getElementType());
+ unsigned EltAlign = TD.getABITypeAlignment(STy->getElementType());
for (unsigned i = 0, e = NumElements; i != e; ++i) {
Constant *In = getAggregateConstantElement(Init,
ConstantInt::get(Type::Int32Ty, i));
GV->isThreadLocal());
Globals.insert(GV, NGV);
NewGlobals.push_back(NGV);
+
+ // Calculate the known alignment of the field. If the original aggregate
+ // had 256 byte alignment for example, something might depend on that:
+ // propagate info to each field.
+ unsigned NewAlign = (unsigned)MinAlign(StartAlignment, EltSize*i);
+ if (NewAlign > EltAlign)
+ NGV->setAlignment(NewAlign);
}
}
Value *NewPtr = NewGlobals[Val];
// Form a shorter GEP if needed.
- if (GEP->getNumOperands() > 3)
+ if (GEP->getNumOperands() > 3) {
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(GEP)) {
SmallVector<Constant*, 8> Idxs;
Idxs.push_back(NullInt);
Idxs.push_back(NullInt);
for (unsigned i = 3, e = GEPI->getNumOperands(); i != e; ++i)
Idxs.push_back(GEPI->getOperand(i));
- NewPtr = new GetElementPtrInst(NewPtr, Idxs.begin(), Idxs.end(),
- GEPI->getName()+"."+utostr(Val), GEPI);
+ NewPtr = GetElementPtrInst::Create(NewPtr, Idxs.begin(), Idxs.end(),
+ GEPI->getName()+"."+utostr(Val), GEPI);
}
+ }
GEP->replaceAllUsesWith(NewPtr);
if (GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(GEP))
MI->getAlignment(), MI->getName(), MI);
Value* Indices[2];
Indices[0] = Indices[1] = Constant::getNullValue(Type::Int32Ty);
- Value *NewGEP = new GetElementPtrInst(NewMI, Indices, Indices + 2,
- NewMI->getName()+".el0", MI);
+ Value *NewGEP = GetElementPtrInst::Create(NewMI, Indices, Indices + 2,
+ NewMI->getName()+".el0", MI);
MI->replaceAllUsesWith(NewGEP);
MI->eraseFromParent();
MI = NewMI;
GV->getName()+".body",
(Module *)NULL,
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.
GEPIdx.push_back(GEPI->getOperand(1));
GEPIdx.append(GEPI->op_begin()+3, GEPI->op_end());
- Value *NGEPI = new GetElementPtrInst(NewPtr, GEPIdx.begin(), GEPIdx.end(),
- GEPI->getName(), GEPI);
+ Value *NGEPI = GetElementPtrInst::Create(NewPtr, GEPIdx.begin(), GEPIdx.end(),
+ GEPI->getName(), GEPI);
GEPI->replaceAllUsesWith(NGEPI);
GEPI->eraseFromParent();
return;
for (unsigned i = 0, e = FieldGlobals.size(); i != e; ++i) {
Value *LoadV = GetHeapSROALoad(Load, i, FieldGlobals, InsertedLoadsForPtr);
- PHINode *FieldPN = new PHINode(LoadV->getType(),
- PN->getName()+"."+utostr(i), PN);
+ PHINode *FieldPN = PHINode::Create(LoadV->getType(),
+ PN->getName()+"."+utostr(i), PN);
// Fill in the predecessor values.
for (unsigned pred = 0, e = PN->getNumIncomingValues(); pred != e; ++pred) {
// Each predecessor either uses the load or the original malloc.
// 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 = new BasicBlock("malloc_ret_null",
- OrigBB->getParent());
+ BasicBlock *NullPtrBlock = BasicBlock::Create("malloc_ret_null",
+ OrigBB->getParent());
// Remove the uncond branch from OrigBB to ContBB, turning it into a cond
// branch on RunningOr.
OrigBB->getTerminator()->eraseFromParent();
- new BranchInst(NullPtrBlock, ContBB, RunningOr, OrigBB);
+ 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.
Value *Cmp = new ICmpInst(ICmpInst::ICMP_NE, GVVal,
Constant::getNullValue(GVVal->getType()),
"tmp", NullPtrBlock);
- BasicBlock *FreeBlock = new BasicBlock("free_it", OrigBB->getParent());
- BasicBlock *NextBlock = new BasicBlock("next", OrigBB->getParent());
- new BranchInst(FreeBlock, NextBlock, Cmp, NullPtrBlock);
+ BasicBlock *FreeBlock = BasicBlock::Create("free_it", OrigBB->getParent());
+ BasicBlock *NextBlock = BasicBlock::Create("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);
- new BranchInst(NextBlock, FreeBlock);
+ BranchInst::Create(NextBlock, FreeBlock);
NullPtrBlock = NextBlock;
}
- new BranchInst(ContBB, NullPtrBlock);
-
+ BranchInst::Create(ContBB, NullPtrBlock);
// MI is no longer needed, remove it.
MI->eraseFromParent();
if (IsOneZero)
NSI = new ZExtInst(NLI, LI->getType(), "", LI);
else
- NSI = new SelectInst(NLI, OtherVal, InitVal, "", LI);
+ NSI = SelectInst::Create(NLI, OtherVal, InitVal, "", LI);
NSI->takeName(LI);
LI->replaceAllUsesWith(NSI);
}
++NumMarked;
return true;
} else if (!GV->getInitializer()->getType()->isFirstClassType()) {
- if (GlobalVariable *FirstNewGV = SRAGlobal(GV)) {
+ if (GlobalVariable *FirstNewGV = SRAGlobal(GV,
+ getAnalysis<TargetData>())) {
GVI = FirstNewGV; // Don't skip the newly produced globals!
return true;
}
/// function, changing them to FastCC.
static void ChangeCalleesToFastCall(Function *F) {
for (Value::use_iterator UI = F->use_begin(), E = F->use_end(); UI != E;++UI){
- Instruction *User = cast<Instruction>(*UI);
- if (CallInst *CI = dyn_cast<CallInst>(User))
- CI->setCallingConv(CallingConv::Fast);
- else
- cast<InvokeInst>(User)->setCallingConv(CallingConv::Fast);
+ CallSite User(cast<Instruction>(*UI));
+ User.setCallingConv(CallingConv::Fast);
}
}
-static const ParamAttrsList *StripNest(const ParamAttrsList *Attrs) {
- if (Attrs) {
- for (unsigned i = 0, e = Attrs->size(); i != e; ++i) {
- uint16_t A = Attrs->getParamAttrsAtIndex(i);
- if (A & ParamAttr::Nest) {
- Attrs = ParamAttrsList::excludeAttrs(Attrs, Attrs->getParamIndex(i),
- ParamAttr::Nest);
- // There can be only one.
- break;
- }
- }
+static PAListPtr StripNest(const PAListPtr &Attrs) {
+ for (unsigned i = 0, e = Attrs.getNumSlots(); i != e; ++i) {
+ if ((Attrs.getSlot(i).Attrs & ParamAttr::Nest) == 0)
+ continue;
+
+ // There can be only one.
+ return Attrs.removeAttr(Attrs.getSlot(i).Index, ParamAttr::Nest);
}
return Attrs;
static void RemoveNestAttribute(Function *F) {
F->setParamAttrs(StripNest(F->getParamAttrs()));
for (Value::use_iterator UI = F->use_begin(), E = F->use_end(); UI != E;++UI){
- Instruction *User = cast<Instruction>(*UI);
- if (CallInst *CI = dyn_cast<CallInst>(User)) {
- CI->setParamAttrs(StripNest(CI->getParamAttrs()));
- } else {
- InvokeInst *II = cast<InvokeInst>(User);
- II->setParamAttrs(StripNest(II->getParamAttrs()));
- }
+ CallSite User(cast<Instruction>(*UI));
+ User.setParamAttrs(StripNest(User.getParamAttrs()));
}
}
Changed = true;
}
- if (F->getParamAttrs() &&
- F->getParamAttrs()->hasAttrSomewhere(ParamAttr::Nest) &&
+ if (F->getParamAttrs().hasAttrSomewhere(ParamAttr::Nest) &&
OnlyCalledDirectly(F)) {
// The function is not used by a trampoline intrinsic, so it is safe
// to remove the 'nest' attribute.
projects / oota-llvm.git / blobdiff