/// RewriteUsesOfClonedInstructions - We just cloned the instructions from the
/// old header into the preheader. If there were uses of the values produced by
/// these instruction that were outside of the loop, we have to insert PHI nodes
/// to merge the two values. Do this now.
static void RewriteUsesOfClonedInstructions(BasicBlock *OrigHeader,
BasicBlock *OrigPreheader,
ValueToValueMapTy &ValueMap) {
// Remove PHI node entries that are no longer live.
BasicBlock::iterator I, E = OrigHeader->end();
for (I = OrigHeader->begin(); PHINode *PN = dyn_cast<PHINode>(I); ++I)
PN->removeIncomingValue(PN->getBasicBlockIndex(OrigPreheader));
// Now fix up users of the instructions in OrigHeader, inserting PHI nodes
// as necessary.
SSAUpdater SSA;
for (I = OrigHeader->begin(); I != E; ++I) {
Value *OrigHeaderVal = I;
// If there are no uses of the value (e.g. because it returns void), there
// is nothing to rewrite.
if (OrigHeaderVal->use_empty())
continue;
Value *OrigPreHeaderVal = ValueMap[OrigHeaderVal];
// The value now exits in two versions: the initial value in the preheader
// and the loop "next" value in the original header.
SSA.Initialize(OrigHeaderVal->getType(), OrigHeaderVal->getName());
SSA.AddAvailableValue(OrigHeader, OrigHeaderVal);
SSA.AddAvailableValue(OrigPreheader, OrigPreHeaderVal);
// Visit each use of the OrigHeader instruction.
for (Value::use_iterator UI = OrigHeaderVal->use_begin(),
UE = OrigHeaderVal->use_end(); UI != UE; ) {
// Grab the use before incrementing the iterator.
Use &U = UI.getUse();
// Increment the iterator before removing the use from the list.
++UI;
// SSAUpdater can't handle a non-PHI use in the same block as an
// earlier def. We can easily handle those cases manually.
Instruction *UserInst = cast<Instruction>(U.getUser());
if (!isa<PHINode>(UserInst)) {
BasicBlock *UserBB = UserInst->getParent();
// The original users in the OrigHeader are already using the
// original definitions.
if (UserBB == OrigHeader)
continue;
// Users in the OrigPreHeader need to use the value to which the
// original definitions are mapped.
if (UserBB == OrigPreheader) {
U = OrigPreHeaderVal;
continue;
}
}
// Anything else can be handled by SSAUpdater.
SSA.RewriteUse(U);
}
}
}
/// If there's a single exit block, sink any loop-invariant values that
/// were defined in the preheader but not used inside the loop into the
/// exit block to reduce register pressure in the loop.
void IndVarSimplify::SinkUnusedInvariants(Loop *L) {
BasicBlock *ExitBlock = L->getExitBlock();
if (!ExitBlock) return;
BasicBlock *Preheader = L->getLoopPreheader();
if (!Preheader) return;
Instruction *InsertPt = ExitBlock->getFirstNonPHI();
BasicBlock::iterator I = Preheader->getTerminator();
while (I != Preheader->begin()) {
--I;
// New instructions were inserted at the end of the preheader.
if (isa<PHINode>(I))
break;
// Don't move instructions which might have side effects, since the side
// effects need to complete before instructions inside the loop. Also
// don't move instructions which might read memory, since the loop may
// modify memory. Note that it's okay if the instruction might have
// undefined behavior: LoopSimplify guarantees that the preheader
// dominates the exit block.
if (I->mayHaveSideEffects() || I->mayReadFromMemory())
continue;
// Don't sink static AllocaInsts out of the entry block, which would
// turn them into dynamic allocas!
if (AllocaInst *AI = dyn_cast<AllocaInst>(I))
if (AI->isStaticAlloca())
continue;
// Determine if there is a use in or before the loop (direct or
// otherwise).
bool UsedInLoop = false;
for (Value::use_iterator UI = I->use_begin(), UE = I->use_end();
UI != UE; ++UI) {
BasicBlock *UseBB = cast<Instruction>(UI)->getParent();
if (PHINode *P = dyn_cast<PHINode>(UI)) {
unsigned i =
PHINode::getIncomingValueNumForOperand(UI.getOperandNo());
UseBB = P->getIncomingBlock(i);
}
if (UseBB == Preheader || L->contains(UseBB)) {
UsedInLoop = true;
break;
}
}
// If there is, the def must remain in the preheader.
if (UsedInLoop)
continue;
// Otherwise, sink it to the exit block.
Instruction *ToMove = I;
bool Done = false;
if (I != Preheader->begin())
--I;
else
Done = true;
ToMove->moveBefore(InsertPt);
if (Done)
break;
InsertPt = ToMove;
}
}
static Value::use_iterator skipToNextUser(Value::use_iterator I,
Value::use_iterator End) {
User *CurUser = I->getUser();
++I;
while (I != End && I->getUser() == CurUser)
++I;
return I;
}
void detail::PtrUseVisitorBase::enqueueUsers(Instruction &I) {
for (Value::use_iterator UI = I.use_begin(), UE = I.use_end();
UI != UE; ++UI) {
if (VisitedUses.insert(&UI.getUse())) {
UseToVisit NewU = {
UseToVisit::UseAndIsOffsetKnownPair(&UI.getUse(), IsOffsetKnown),
Offset
};
Worklist.push_back(llvm_move(NewU));
}
}
}
/// WriteThunk - Replace G with a simple tail call to bitcast(F). Also replace
/// direct uses of G with bitcast(F). Deletes G.
void MergeFunctions::WriteThunk(Function *F, Function *G) const {
if (!G->mayBeOverridden()) {
// Redirect direct callers of G to F.
Constant *BitcastF = ConstantExpr::getBitCast(F, G->getType());
for (Value::use_iterator UI = G->use_begin(), UE = G->use_end();
UI != UE;) {
Value::use_iterator TheIter = UI;
++UI;
CallSite CS(*TheIter);
if (CS && CS.isCallee(TheIter))
TheIter.getUse().set(BitcastF);
}
}
// If G was internal then we may have replaced all uses of G with F. If so,
// stop here and delete G. There's no need for a thunk.
if (G->hasLocalLinkage() && G->use_empty()) {
G->eraseFromParent();
return;
}
Function *NewG = Function::Create(G->getFunctionType(), G->getLinkage(), "",
G->getParent());
BasicBlock *BB = BasicBlock::Create(F->getContext(), "", NewG);
IRBuilder<false> Builder(BB);
SmallVector<Value *, 16> Args;
unsigned i = 0;
const FunctionType *FFTy = F->getFunctionType();
for (Function::arg_iterator AI = NewG->arg_begin(), AE = NewG->arg_end();
AI != AE; ++AI) {
Args.push_back(Builder.CreateBitCast(AI, FFTy->getParamType(i)));
++i;
}
CallInst *CI = Builder.CreateCall(F, Args.begin(), Args.end());
CI->setTailCall();
CI->setCallingConv(F->getCallingConv());
if (NewG->getReturnType()->isVoidTy()) {
Builder.CreateRetVoid();
} else {
Builder.CreateRet(Builder.CreateBitCast(CI, NewG->getReturnType()));
}
NewG->copyAttributesFrom(G);
NewG->takeName(G);
G->replaceAllUsesWith(NewG);
G->eraseFromParent();
DEBUG(dbgs() << "WriteThunk: " << NewG->getName() << '\n');
++NumThunksWritten;
}
// Replace direct callers of Old with New.
void MergeFunctions::replaceDirectCallers(Function *Old, Function *New) {
Constant *BitcastNew = ConstantExpr::getBitCast(New, Old->getType());
for (Value::use_iterator UI = Old->use_begin(), UE = Old->use_end();
UI != UE;) {
Value::use_iterator TheIter = UI;
++UI;
CallSite CS(*TheIter);
if (CS && CS.isCallee(TheIter)) {
remove(CS.getInstruction()->getParent()->getParent());
TheIter.getUse().set(BitcastNew);
}
}
}
/// OptimizeCmpExpression - sink the given CmpInst into user blocks to reduce
/// the number of virtual registers that must be created and coalesced. This is
/// a clear win except on targets with multiple condition code registers
/// (PowerPC), where it might lose; some adjustment may be wanted there.
///
/// Return true if any changes are made.
static bool OptimizeCmpExpression(CmpInst *CI) {
BasicBlock *DefBB = CI->getParent();
/// InsertedCmp - Only insert a cmp in each block once.
DenseMap<BasicBlock*, CmpInst*> InsertedCmps;
bool MadeChange = false;
for (Value::use_iterator UI = CI->use_begin(), E = CI->use_end();
UI != E; ) {
Use &TheUse = UI.getUse();
Instruction *User = cast<Instruction>(*UI);
// Preincrement use iterator so we don't invalidate it.
++UI;
// Don't bother for PHI nodes.
if (isa<PHINode>(User))
continue;
// Figure out which BB this cmp is used in.
BasicBlock *UserBB = User->getParent();
// If this user is in the same block as the cmp, don't change the cmp.
if (UserBB == DefBB) continue;
// If we have already inserted a cmp into this block, use it.
CmpInst *&InsertedCmp = InsertedCmps[UserBB];
if (!InsertedCmp) {
BasicBlock::iterator InsertPt = UserBB->getFirstNonPHI();
InsertedCmp =
CmpInst::Create(CI->getOpcode(),
CI->getPredicate(), CI->getOperand(0),
CI->getOperand(1), "", InsertPt);
MadeChange = true;
}
// Replace a use of the cmp with a use of the new cmp.
TheUse = InsertedCmp;
++NumCmpUses;
}
// If we removed all uses, nuke the cmp.
if (CI->use_empty())
CI->eraseFromParent();
return MadeChange;
}
// hasLifetimeMarkers - Check whether the given alloca already has
// lifetime.start or lifetime.end intrinsics.
static bool hasLifetimeMarkers(AllocaInst *AI) {
Type *Int8PtrTy = Type::getInt8PtrTy(AI->getType()->getContext());
if (AI->getType() == Int8PtrTy)
return isUsedByLifetimeMarker(AI);
// Do a scan to find all the casts to i8*.
for (Value::use_iterator I = AI->use_begin(), E = AI->use_end(); I != E;
++I) {
if (I->getType() != Int8PtrTy) continue;
if (I->stripPointerCasts() != AI) continue;
if (isUsedByLifetimeMarker(*I))
return true;
}
return false;
}
void BBCloner::UpdateSSA(Function &F) {
DominatorTree &DT = getAnalysis<DominatorTree>();
// The function has been greatly modified since the beginning.
DT.runOnFunction(F);
vector<pair<Instruction *, Use *> > ToResolve;
for (ValueToValueMapTy::iterator I = CloneMap.begin();
I != CloneMap.end();
++I) {
Value *Key = const_cast<Value *>(I->first);
if (Instruction *OldIns = dyn_cast<Instruction>(Key)) {
for (Value::use_iterator UI = OldIns->use_begin();
UI != OldIns->use_end();
++UI) {
if (Instruction *User = dyn_cast<Instruction>(*UI)) {
if (!DT.dominates(OldIns, User))
ToResolve.push_back(make_pair(OldIns, &UI.getUse()));
}
}
Instruction *NewIns = cast<Instruction>(I->second);
for (Value::use_iterator UI = NewIns->use_begin();
UI != NewIns->use_end();
++UI) {
if (Instruction *User = dyn_cast<Instruction>(*UI)) {
if (!DT.dominates(NewIns, User)) {
// Use OldIns intentionally.
ToResolve.push_back(make_pair(OldIns, &UI.getUse()));
}
}
}
}
}
for (size_t i = 0; i < ToResolve.size(); ) {
Instruction *OldIns = ToResolve[i].first;
Instruction *NewIns = cast<Instruction>(CloneMap.lookup(OldIns));
SSAUpdater SU;
SU.Initialize(OldIns->getType(), OldIns->getName());
SU.AddAvailableValue(OldIns->getParent(), OldIns);
SU.AddAvailableValue(NewIns->getParent(), NewIns);
size_t j = i;
while (j < ToResolve.size() && ToResolve[j].first == ToResolve[i].first) {
SU.RewriteUse(*ToResolve[j].second);
++j;
}
i = j;
}
}
/// FindAllMemoryUses - Recursively walk all the uses of I until we find a
/// memory use. If we find an obviously non-foldable instruction, return true.
/// Add the ultimately found memory instructions to MemoryUses.
static bool FindAllMemoryUses(Instruction *I,
SmallVectorImpl<std::pair<Instruction*,unsigned> > &MemoryUses,
SmallPtrSet<Instruction*, 16> &ConsideredInsts,
const TargetLowering &TLI) {
// If we already considered this instruction, we're done.
if (!ConsideredInsts.insert(I))
return false;
// If this is an obviously unfoldable instruction, bail out.
if (!MightBeFoldableInst(I))
return true;
// Loop over all the uses, recursively processing them.
for (Value::use_iterator UI = I->use_begin(), E = I->use_end();
UI != E; ++UI) {
User *U = *UI;
if (LoadInst *LI = dyn_cast<LoadInst>(U)) {
MemoryUses.push_back(std::make_pair(LI, UI.getOperandNo()));
continue;
}
if (StoreInst *SI = dyn_cast<StoreInst>(U)) {
unsigned opNo = UI.getOperandNo();
if (opNo == 0) return true; // Storing addr, not into addr.
MemoryUses.push_back(std::make_pair(SI, opNo));
continue;
}
if (CallInst *CI = dyn_cast<CallInst>(U)) {
InlineAsm *IA = dyn_cast<InlineAsm>(CI->getCalledValue());
if (!IA) return true;
// If this is a memory operand, we're cool, otherwise bail out.
if (!IsOperandAMemoryOperand(CI, IA, I, TLI))
return true;
continue;
}
if (FindAllMemoryUses(cast<Instruction>(U), MemoryUses, ConsideredInsts,
TLI))
return true;
}
return false;
}
void ARM64PromoteConstant::
computeInsertionPoints(Constant *Val,
InsertionPointsPerFunc &InsPtsPerFunc) {
DEBUG(dbgs() << "** Compute insertion points **\n");
for (Value::use_iterator UseIt = Val->use_begin(), EndUseIt = Val->use_end();
UseIt != EndUseIt; ++UseIt) {
// If the user is not an Instruction, we cannot modify it
if (!isa<Instruction>(*UseIt))
continue;
// Filter out uses that should not be converted
if (!shouldConvertUse(Val, cast<Instruction>(*UseIt), UseIt.getOperandNo()))
continue;
DEBUG(dbgs() << "Considered use, opidx " << UseIt.getOperandNo() << ":\n");
DEBUG(UseIt->print(dbgs()));
DEBUG(dbgs() << '\n');
Instruction *InsertionPoint = findInsertionPoint(UseIt);
DEBUG(dbgs() << "Considered insertion point:\n");
DEBUG(InsertionPoint->print(dbgs()));
DEBUG(dbgs() << '\n');
// Check if the current insertion point is useless, i.e., it is dominated
// by another one.
InsertionPoints &InsertPts =
InsPtsPerFunc[InsertionPoint->getParent()->getParent()];
if (isDominated(InsertionPoint, UseIt, InsertPts))
continue;
// This insertion point is useful, check if we can merge some insertion
// point in a common dominator or if NewPt dominates an existing one.
if (tryAndMerge(InsertionPoint, UseIt, InsertPts))
continue;
DEBUG(dbgs() << "Keep considered insertion point\n");
// It is definitely useful by its own
InsertPts[InsertionPoint].push_back(UseIt);
}
}
// For each instruction used by the value, remove() the function that contains
// the instruction. This should happen right before a call to RAUW.
void MergeFunctions::removeUsers(Value *V) {
std::vector<Value *> Worklist;
Worklist.push_back(V);
while (!Worklist.empty()) {
Value *V = Worklist.back();
Worklist.pop_back();
for (Value::use_iterator UI = V->use_begin(), UE = V->use_end();
UI != UE; ++UI) {
Use &U = UI.getUse();
if (Instruction *I = dyn_cast<Instruction>(U.getUser())) {
remove(I->getParent()->getParent());
} else if (isa<GlobalValue>(U.getUser())) {
// do nothing
} else if (Constant *C = dyn_cast<Constant>(U.getUser())) {
for (Value::use_iterator CUI = C->use_begin(), CUE = C->use_end();
CUI != CUE; ++CUI)
Worklist.push_back(*CUI);
}
}
}
}
bool HeterotbbTransform::runOnModule(Module &M) {
bool localChange = true;
TD = getAnalysisIfAvailable<DataLayout>();
//hetero_f_const = NULL;
rewrite_CPP(M);
#if 0
Function *offload_func = M.getFunction("offload");
for (Value::use_iterator i = offload_func->use_begin(), e = offload_func->use_end(); i != e; ++i) {
Instruction *call;
DEBUG(dbgs() << "Next _offload function:");
i->dump();
if ((call = dyn_cast<InvokeInst>(*i)) || (call = dyn_cast<CallInst>(*i))) {
CallSite CI(cast<Instruction>(call));
CI->dump();
}
}
#endif
//M.dump();
return localChange;
}
/// AllUsesDominatedByBlock - Return true if all uses of the specified value
/// occur in blocks dominated by the specified block.
bool Sinking::AllUsesDominatedByBlock(Instruction *Inst,
BasicBlock *BB) const {
// Ignoring debug uses is necessary so debug info doesn't affect the code.
// This may leave a referencing dbg_value in the original block, before
// the definition of the vreg. Dwarf generator handles this although the
// user might not get the right info at runtime.
for (Value::use_iterator I = Inst->use_begin(),
E = Inst->use_end(); I != E; ++I) {
// Determine the block of the use.
Instruction *UseInst = cast<Instruction>(*I);
BasicBlock *UseBlock = UseInst->getParent();
if (PHINode *PN = dyn_cast<PHINode>(UseInst)) {
// PHI nodes use the operand in the predecessor block, not the block with
// the PHI.
unsigned Num = PHINode::getIncomingValueNumForOperand(I.getOperandNo());
UseBlock = PN->getIncomingBlock(Num);
}
// Check that it dominates.
if (!DT->dominates(BB, UseBlock))
return false;
}
return true;
}
/// scanForInterest - This function decides which arguments would be worth
/// specializing on.
void PartSpec::scanForInterest(Function& F, InterestingArgVector& args) {
for(Function::arg_iterator ii = F.arg_begin(), ee = F.arg_end();
ii != ee; ++ii) {
for(Value::use_iterator ui = ii->use_begin(), ue = ii->use_end();
ui != ue; ++ui) {
bool interesting = false;
User *U = *ui;
if (isa<CmpInst>(U)) interesting = true;
else if (isa<CallInst>(U))
interesting = ui->getOperand(0) == ii;
else if (isa<InvokeInst>(U))
interesting = ui->getOperand(0) == ii;
else if (isa<SwitchInst>(U)) interesting = true;
else if (isa<BranchInst>(U)) interesting = true;
if (interesting) {
args.push_back(std::distance(F.arg_begin(), ii));
break;
}
}
}
}
/// OptimizeNoopCopyExpression - If the specified cast instruction is a noop
/// copy (e.g. it's casting from one pointer type to another, i32->i8 on PPC),
/// sink it into user blocks to reduce the number of virtual
/// registers that must be created and coalesced.
///
/// Return true if any changes are made.
///
static bool OptimizeNoopCopyExpression(CastInst *CI, const TargetLowering &TLI) {
// If this is a noop copy,
EVT SrcVT = TLI.getValueType(CI->getOperand(0)->getType());
EVT DstVT = TLI.getValueType(CI->getType());
// This is an fp<->int conversion?
if (SrcVT.isInteger() != DstVT.isInteger())
return false;
// If this is an extension, it will be a zero or sign extension, which
// isn't a noop.
if (SrcVT.bitsLT(DstVT)) return false;
// If these values will be promoted, find out what they will be promoted
// to. This helps us consider truncates on PPC as noop copies when they
// are.
if (TLI.getTypeAction(CI->getContext(), SrcVT) ==
TargetLowering::TypePromoteInteger)
SrcVT = TLI.getTypeToTransformTo(CI->getContext(), SrcVT);
if (TLI.getTypeAction(CI->getContext(), DstVT) ==
TargetLowering::TypePromoteInteger)
DstVT = TLI.getTypeToTransformTo(CI->getContext(), DstVT);
// If, after promotion, these are the same types, this is a noop copy.
if (SrcVT != DstVT)
return false;
BasicBlock *DefBB = CI->getParent();
/// InsertedCasts - Only insert a cast in each block once.
DenseMap<BasicBlock*, CastInst*> InsertedCasts;
bool MadeChange = false;
for (Value::use_iterator UI = CI->use_begin(), E = CI->use_end();
UI != E; ) {
Use &TheUse = UI.getUse();
Instruction *User = cast<Instruction>(*UI);
// Figure out which BB this cast is used in. For PHI's this is the
// appropriate predecessor block.
BasicBlock *UserBB = User->getParent();
if (PHINode *PN = dyn_cast<PHINode>(User)) {
UserBB = PN->getIncomingBlock(UI);
}
// Preincrement use iterator so we don't invalidate it.
++UI;
// If this user is in the same block as the cast, don't change the cast.
if (UserBB == DefBB) continue;
// If we have already inserted a cast into this block, use it.
CastInst *&InsertedCast = InsertedCasts[UserBB];
if (!InsertedCast) {
BasicBlock::iterator InsertPt = UserBB->getFirstInsertionPt();
InsertedCast =
CastInst::Create(CI->getOpcode(), CI->getOperand(0), CI->getType(), "",
InsertPt);
MadeChange = true;
}
// Replace a use of the cast with a use of the new cast.
TheUse = InsertedCast;
++NumCastUses;
}
// If we removed all uses, nuke the cast.
if (CI->use_empty()) {
CI->eraseFromParent();
MadeChange = true;
}
return MadeChange;
}
bool CodeGenPrepare::OptimizeExtUses(Instruction *I) {
BasicBlock *DefBB = I->getParent();
// If the result of a {s|z}ext and its source are both live out, rewrite all
// other uses of the source with result of extension.
Value *Src = I->getOperand(0);
if (Src->hasOneUse())
return false;
// Only do this xform if truncating is free.
if (TLI && !TLI->isTruncateFree(I->getType(), Src->getType()))
return false;
// Only safe to perform the optimization if the source is also defined in
// this block.
if (!isa<Instruction>(Src) || DefBB != cast<Instruction>(Src)->getParent())
return false;
bool DefIsLiveOut = false;
for (Value::use_iterator UI = I->use_begin(), E = I->use_end();
UI != E; ++UI) {
Instruction *User = cast<Instruction>(*UI);
// Figure out which BB this ext is used in.
BasicBlock *UserBB = User->getParent();
if (UserBB == DefBB) continue;
DefIsLiveOut = true;
break;
}
if (!DefIsLiveOut)
return false;
// Make sure non of the uses are PHI nodes.
for (Value::use_iterator UI = Src->use_begin(), E = Src->use_end();
UI != E; ++UI) {
Instruction *User = cast<Instruction>(*UI);
BasicBlock *UserBB = User->getParent();
if (UserBB == DefBB) continue;
// Be conservative. We don't want this xform to end up introducing
// reloads just before load / store instructions.
if (isa<PHINode>(User) || isa<LoadInst>(User) || isa<StoreInst>(User))
return false;
}
// InsertedTruncs - Only insert one trunc in each block once.
DenseMap<BasicBlock*, Instruction*> InsertedTruncs;
bool MadeChange = false;
for (Value::use_iterator UI = Src->use_begin(), E = Src->use_end();
UI != E; ++UI) {
Use &TheUse = UI.getUse();
Instruction *User = cast<Instruction>(*UI);
// Figure out which BB this ext is used in.
BasicBlock *UserBB = User->getParent();
if (UserBB == DefBB) continue;
// Both src and def are live in this block. Rewrite the use.
Instruction *&InsertedTrunc = InsertedTruncs[UserBB];
if (!InsertedTrunc) {
BasicBlock::iterator InsertPt = UserBB->getFirstInsertionPt();
InsertedTrunc = new TruncInst(I, Src->getType(), "", InsertPt);
}
// Replace a use of the {s|z}ext source with a use of the result.
TheUse = InsertedTrunc;
++NumExtUses;
MadeChange = true;
}
return MadeChange;
}
//
// Method: runOnModule()
//
// Description:
// Entry point for this LLVM pass. Search for functions which could be called
// indirectly and create clones for them which are only called by direct
// calls.
//
// Inputs:
// M - A reference to the LLVM module to transform.
//
// Outputs:
// M - The transformed LLVM module.
//
// Return value:
// true - The module was modified.
// false - The module was not modified.
//
bool
IndClone::runOnModule(Module& M) {
// Set of functions to clone
std::vector<Function*> toClone;
//
// Check all of the functions in the module. If the function could be called
// by an indirect function call, add it to our worklist of functions to
// clone.
//
for (Module::iterator I = M.begin(); I != M.end(); ++I) {
// Flag whether the function should be cloned
bool pleaseCloneTheFunction = false;
//
// Only clone functions which are defined and cannot be replaced by another
// function by the linker.
//
if (!I->isDeclaration() && !I->mayBeOverridden()) {
for (Value::use_iterator ui = I->use_begin(), ue = I->use_end();
ui != ue; ++ui) {
if (!isa<CallInst>(*ui) && !isa<InvokeInst>(*ui)) {
if(!ui->use_empty())
//
// If this function is used for anything other than a direct function
// call, then we want to clone it.
//
pleaseCloneTheFunction = true;
} else {
//
// This is a call instruction, but hold up ranger! We need to make
// sure that the function isn't passed as an argument to *another*
// function. That would make the function usable in an indirect
// function call.
//
for (unsigned index = 1; index < ui->getNumOperands(); ++index) {
if (ui->getOperand(index)->stripPointerCasts() == I) {
pleaseCloneTheFunction = true;
break;
}
}
}
//
// If we've discovered that the function could be used by an indirect
// call site, schedule it for cloning.
//
if (pleaseCloneTheFunction) {
toClone.push_back(I);
break;
}
}
}
}
//
// Update the statistics on the number of functions we'll be cloning.
// We only update the statistic if we want to clone one or more functions;
// due to the magic of how statistics work, avoiding assignment prevents it
// from needlessly showing up.
//
if (toClone.size())
numCloned += toClone.size();
//
// Go through the worklist and clone each function. After cloning a
// function, change all direct calls to use the clone instead of using the
// original function.
//
for (unsigned index = 0; index < toClone.size(); ++index) {
//
// Clone the function and give it a name indicating that it is a clone to
// be used for direct function calls.
//
Function * Original = toClone[index];
Function* DirectF = CloneFunction(Original);
DirectF->setName(Original->getName() + "_DIRECT");
//
// Make the clone internal; external code can use the original function.
//
DirectF->setLinkage(GlobalValue::InternalLinkage);
//
// Link the cloned function into the set of functions belonging to the
// module.
//
Original->getParent()->getFunctionList().push_back(DirectF);
//
// Find all uses of the function that use it as a direct call. Change
// them to use the clone.
//
for (Value::use_iterator ui = Original->use_begin(),
ue = Original->use_end();
ui != ue; ) {
CallInst *CI = dyn_cast<CallInst>(*ui);
ui++;
if (CI) {
if (CI->getCalledFunction() == Original) {
++numReplaced;
CI->setCalledFunction(DirectF);
}
}
}
}
//
// Assume that we've cloned at least one function.
//
return true;
}
/// SplitCriticalEdge - If this edge is a critical edge, insert a new node to
/// split the critical edge. This will update DominatorTree and
/// DominatorFrontier information if it is available, thus calling this pass
/// will not invalidate either of them. This returns the new block if the edge
/// was split, null otherwise.
///
/// If MergeIdenticalEdges is true (not the default), *all* edges from TI to the
/// specified successor will be merged into the same critical edge block.
/// This is most commonly interesting with switch instructions, which may
/// have many edges to any one destination. This ensures that all edges to that
/// dest go to one block instead of each going to a different block, but isn't
/// the standard definition of a "critical edge".
///
/// It is invalid to call this function on a critical edge that starts at an
/// IndirectBrInst. Splitting these edges will almost always create an invalid
/// program because the address of the new block won't be the one that is jumped
/// to.
///
BasicBlock *llvm::SplitCriticalEdge(TerminatorInst *TI, unsigned SuccNum,
Pass *P, bool MergeIdenticalEdges) {
if (!isCriticalEdge(TI, SuccNum, MergeIdenticalEdges)) return 0;
assert(!isa<IndirectBrInst>(TI) &&
"Cannot split critical edge from IndirectBrInst");
BasicBlock *TIBB = TI->getParent();
BasicBlock *DestBB = TI->getSuccessor(SuccNum);
// Create a new basic block, linking it into the CFG.
BasicBlock *NewBB = BasicBlock::Create(TI->getContext(),
TIBB->getName() + "." + DestBB->getName() + "_crit_edge");
// Create our unconditional branch.
BranchInst::Create(DestBB, NewBB);
// Branch to the new block, breaking the edge.
TI->setSuccessor(SuccNum, NewBB);
// Insert the block into the function... right after the block TI lives in.
Function &F = *TIBB->getParent();
Function::iterator FBBI = TIBB;
F.getBasicBlockList().insert(++FBBI, NewBB);
// If there are any PHI nodes in DestBB, we need to update them so that they
// merge incoming values from NewBB instead of from TIBB.
if (PHINode *APHI = dyn_cast<PHINode>(DestBB->begin())) {
// This conceptually does:
// foreach (PHINode *PN in DestBB)
// PN->setIncomingBlock(PN->getIncomingBlock(TIBB), NewBB);
// but is optimized for two cases.
if (APHI->getNumIncomingValues() <= 8) { // Small # preds case.
unsigned BBIdx = 0;
for (BasicBlock::iterator I = DestBB->begin(); isa<PHINode>(I); ++I) {
// We no longer enter through TIBB, now we come in through NewBB.
// Revector exactly one entry in the PHI node that used to come from
// TIBB to come from NewBB.
PHINode *PN = cast<PHINode>(I);
// Reuse the previous value of BBIdx if it lines up. In cases where we
// have multiple phi nodes with *lots* of predecessors, this is a speed
// win because we don't have to scan the PHI looking for TIBB. This
// happens because the BB list of PHI nodes are usually in the same
// order.
if (PN->getIncomingBlock(BBIdx) != TIBB)
BBIdx = PN->getBasicBlockIndex(TIBB);
PN->setIncomingBlock(BBIdx, NewBB);
}
} else {
// However, the foreach loop is slow for blocks with lots of predecessors
// because PHINode::getIncomingBlock is O(n) in # preds. Instead, walk
// the user list of TIBB to find the PHI nodes.
SmallPtrSet<PHINode*, 16> UpdatedPHIs;
for (Value::use_iterator UI = TIBB->use_begin(), E = TIBB->use_end();
UI != E; ) {
Value::use_iterator Use = UI++;
if (PHINode *PN = dyn_cast<PHINode>(Use)) {
// Remove one entry from each PHI.
if (PN->getParent() == DestBB && UpdatedPHIs.insert(PN))
PN->setOperand(Use.getOperandNo(), NewBB);
}
}
}
}
// If there are any other edges from TIBB to DestBB, update those to go
// through the split block, making those edges non-critical as well (and
// reducing the number of phi entries in the DestBB if relevant).
if (MergeIdenticalEdges) {
for (unsigned i = SuccNum+1, e = TI->getNumSuccessors(); i != e; ++i) {
if (TI->getSuccessor(i) != DestBB) continue;
// Remove an entry for TIBB from DestBB phi nodes.
DestBB->removePredecessor(TIBB);
// We found another edge to DestBB, go to NewBB instead.
TI->setSuccessor(i, NewBB);
}
}
// If we don't have a pass object, we can't update anything...
if (P == 0) return NewBB;
DominatorTree *DT = P->getAnalysisIfAvailable<DominatorTree>();
DominanceFrontier *DF = P->getAnalysisIfAvailable<DominanceFrontier>();
LoopInfo *LI = P->getAnalysisIfAvailable<LoopInfo>();
ProfileInfo *PI = P->getAnalysisIfAvailable<ProfileInfo>();
// If we have nothing to update, just return.
if (DT == 0 && DF == 0 && LI == 0 && PI == 0)
return NewBB;
// Now update analysis information. Since the only predecessor of NewBB is
// the TIBB, TIBB clearly dominates NewBB. TIBB usually doesn't dominate
// anything, as there are other successors of DestBB. However, if all other
// predecessors of DestBB are already dominated by DestBB (e.g. DestBB is a
// loop header) then NewBB dominates DestBB.
SmallVector<BasicBlock*, 8> OtherPreds;
// If there is a PHI in the block, loop over predecessors with it, which is
// faster than iterating pred_begin/end.
if (PHINode *PN = dyn_cast<PHINode>(DestBB->begin())) {
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
if (PN->getIncomingBlock(i) != NewBB)
OtherPreds.push_back(PN->getIncomingBlock(i));
} else {
for (pred_iterator I = pred_begin(DestBB), E = pred_end(DestBB);
I != E; ++I)
if (*I != NewBB)
OtherPreds.push_back(*I);
}
bool NewBBDominatesDestBB = true;
// Should we update DominatorTree information?
if (DT) {
DomTreeNode *TINode = DT->getNode(TIBB);
// The new block is not the immediate dominator for any other nodes, but
// TINode is the immediate dominator for the new node.
//
if (TINode) { // Don't break unreachable code!
DomTreeNode *NewBBNode = DT->addNewBlock(NewBB, TIBB);
DomTreeNode *DestBBNode = 0;
// If NewBBDominatesDestBB hasn't been computed yet, do so with DT.
if (!OtherPreds.empty()) {
DestBBNode = DT->getNode(DestBB);
while (!OtherPreds.empty() && NewBBDominatesDestBB) {
if (DomTreeNode *OPNode = DT->getNode(OtherPreds.back()))
NewBBDominatesDestBB = DT->dominates(DestBBNode, OPNode);
OtherPreds.pop_back();
}
OtherPreds.clear();
}
// If NewBBDominatesDestBB, then NewBB dominates DestBB, otherwise it
// doesn't dominate anything.
if (NewBBDominatesDestBB) {
if (!DestBBNode) DestBBNode = DT->getNode(DestBB);
DT->changeImmediateDominator(DestBBNode, NewBBNode);
}
}
}
// Should we update DominanceFrontier information?
if (DF) {
// If NewBBDominatesDestBB hasn't been computed yet, do so with DF.
if (!OtherPreds.empty()) {
// FIXME: IMPLEMENT THIS!
llvm_unreachable("Requiring domfrontiers but not idom/domtree/domset."
" not implemented yet!");
}
// Since the new block is dominated by its only predecessor TIBB,
// it cannot be in any block's dominance frontier. If NewBB dominates
// DestBB, its dominance frontier is the same as DestBB's, otherwise it is
// just {DestBB}.
DominanceFrontier::DomSetType NewDFSet;
if (NewBBDominatesDestBB) {
DominanceFrontier::iterator I = DF->find(DestBB);
if (I != DF->end()) {
DF->addBasicBlock(NewBB, I->second);
if (I->second.count(DestBB)) {
// However NewBB's frontier does not include DestBB.
DominanceFrontier::iterator NF = DF->find(NewBB);
DF->removeFromFrontier(NF, DestBB);
}
}
else
DF->addBasicBlock(NewBB, DominanceFrontier::DomSetType());
} else {
DominanceFrontier::DomSetType NewDFSet;
NewDFSet.insert(DestBB);
DF->addBasicBlock(NewBB, NewDFSet);
}
}
// Update LoopInfo if it is around.
if (LI) {
if (Loop *TIL = LI->getLoopFor(TIBB)) {
// If one or the other blocks were not in a loop, the new block is not
// either, and thus LI doesn't need to be updated.
if (Loop *DestLoop = LI->getLoopFor(DestBB)) {
if (TIL == DestLoop) {
// Both in the same loop, the NewBB joins loop.
DestLoop->addBasicBlockToLoop(NewBB, LI->getBase());
} else if (TIL->contains(DestLoop)) {
// Edge from an outer loop to an inner loop. Add to the outer loop.
TIL->addBasicBlockToLoop(NewBB, LI->getBase());
} else if (DestLoop->contains(TIL)) {
// Edge from an inner loop to an outer loop. Add to the outer loop.
DestLoop->addBasicBlockToLoop(NewBB, LI->getBase());
} else {
// Edge from two loops with no containment relation. Because these
// are natural loops, we know that the destination block must be the
// header of its loop (adding a branch into a loop elsewhere would
// create an irreducible loop).
assert(DestLoop->getHeader() == DestBB &&
"Should not create irreducible loops!");
if (Loop *P = DestLoop->getParentLoop())
P->addBasicBlockToLoop(NewBB, LI->getBase());
}
}
// If TIBB is in a loop and DestBB is outside of that loop, split the
// other exit blocks of the loop that also have predecessors outside
// the loop, to maintain a LoopSimplify guarantee.
if (!TIL->contains(DestBB) &&
P->mustPreserveAnalysisID(LoopSimplifyID)) {
assert(!TIL->contains(NewBB) &&
"Split point for loop exit is contained in loop!");
// Update LCSSA form in the newly created exit block.
if (P->mustPreserveAnalysisID(LCSSAID)) {
SmallVector<BasicBlock *, 1> OrigPred;
OrigPred.push_back(TIBB);
CreatePHIsForSplitLoopExit(OrigPred, NewBB, DestBB);
}
// For each unique exit block...
SmallVector<BasicBlock *, 4> ExitBlocks;
TIL->getExitBlocks(ExitBlocks);
for (unsigned i = 0, e = ExitBlocks.size(); i != e; ++i) {
// Collect all the preds that are inside the loop, and note
// whether there are any preds outside the loop.
SmallVector<BasicBlock *, 4> Preds;
bool HasPredOutsideOfLoop = false;
BasicBlock *Exit = ExitBlocks[i];
for (pred_iterator I = pred_begin(Exit), E = pred_end(Exit);
I != E; ++I)
if (TIL->contains(*I))
Preds.push_back(*I);
else
HasPredOutsideOfLoop = true;
// If there are any preds not in the loop, we'll need to split
// the edges. The Preds.empty() check is needed because a block
// may appear multiple times in the list. We can't use
// getUniqueExitBlocks above because that depends on LoopSimplify
// form, which we're in the process of restoring!
if (!Preds.empty() && HasPredOutsideOfLoop) {
BasicBlock *NewExitBB =
SplitBlockPredecessors(Exit, Preds.data(), Preds.size(),
"split", P);
if (P->mustPreserveAnalysisID(LCSSAID))
CreatePHIsForSplitLoopExit(Preds, NewExitBB, Exit);
}
}
}
// LCSSA form was updated above for the case where LoopSimplify is
// available, which means that all predecessors of loop exit blocks
// are within the loop. Without LoopSimplify form, it would be
// necessary to insert a new phi.
assert((!P->mustPreserveAnalysisID(LCSSAID) ||
P->mustPreserveAnalysisID(LoopSimplifyID)) &&
"SplitCriticalEdge doesn't know how to update LCCSA form "
"without LoopSimplify!");
}
}
// Update ProfileInfo if it is around.
if (PI)
PI->splitEdge(TIBB, DestBB, NewBB, MergeIdenticalEdges);
return NewBB;
}
/// sink - When an instruction is found to only be used outside of the loop,
/// this function moves it to the exit blocks and patches up SSA form as needed.
/// This method is guaranteed to remove the original instruction from its
/// position, and may either delete it or move it to outside of the loop.
///
void LICM::sink(Instruction &I) {
DEBUG(dbgs() << "LICM sinking instruction: " << I << "\n");
SmallVector<BasicBlock*, 8> ExitBlocks;
CurLoop->getUniqueExitBlocks(ExitBlocks);
if (isa<LoadInst>(I)) ++NumMovedLoads;
else if (isa<CallInst>(I)) ++NumMovedCalls;
++NumSunk;
Changed = true;
// The case where there is only a single exit node of this loop is common
// enough that we handle it as a special (more efficient) case. It is more
// efficient to handle because there are no PHI nodes that need to be placed.
if (ExitBlocks.size() == 1) {
if (!DT->dominates(I.getParent(), ExitBlocks[0])) {
// Instruction is not used, just delete it.
CurAST->deleteValue(&I);
// If I has users in unreachable blocks, eliminate.
// If I is not void type then replaceAllUsesWith undef.
// This allows ValueHandlers and custom metadata to adjust itself.
if (!I.use_empty())
I.replaceAllUsesWith(UndefValue::get(I.getType()));
I.eraseFromParent();
} else {
// Move the instruction to the start of the exit block, after any PHI
// nodes in it.
I.moveBefore(ExitBlocks[0]->getFirstInsertionPt());
// This instruction is no longer in the AST for the current loop, because
// we just sunk it out of the loop. If we just sunk it into an outer
// loop, we will rediscover the operation when we process it.
CurAST->deleteValue(&I);
}
return;
}
if (ExitBlocks.empty()) {
// The instruction is actually dead if there ARE NO exit blocks.
CurAST->deleteValue(&I);
// If I has users in unreachable blocks, eliminate.
// If I is not void type then replaceAllUsesWith undef.
// This allows ValueHandlers and custom metadata to adjust itself.
if (!I.use_empty())
I.replaceAllUsesWith(UndefValue::get(I.getType()));
I.eraseFromParent();
return;
}
// Otherwise, if we have multiple exits, use the SSAUpdater to do all of the
// hard work of inserting PHI nodes as necessary.
SmallVector<PHINode*, 8> NewPHIs;
SSAUpdater SSA(&NewPHIs);
if (!I.use_empty())
SSA.Initialize(I.getType(), I.getName());
// Insert a copy of the instruction in each exit block of the loop that is
// dominated by the instruction. Each exit block is known to only be in the
// ExitBlocks list once.
BasicBlock *InstOrigBB = I.getParent();
unsigned NumInserted = 0;
for (unsigned i = 0, e = ExitBlocks.size(); i != e; ++i) {
BasicBlock *ExitBlock = ExitBlocks[i];
if (!DT->dominates(InstOrigBB, ExitBlock))
continue;
// Insert the code after the last PHI node.
BasicBlock::iterator InsertPt = ExitBlock->getFirstInsertionPt();
// If this is the first exit block processed, just move the original
// instruction, otherwise clone the original instruction and insert
// the copy.
Instruction *New;
if (NumInserted++ == 0) {
I.moveBefore(InsertPt);
New = &I;
} else {
New = I.clone();
if (!I.getName().empty())
New->setName(I.getName()+".le");
ExitBlock->getInstList().insert(InsertPt, New);
}
// Now that we have inserted the instruction, inform SSAUpdater.
if (!I.use_empty())
SSA.AddAvailableValue(ExitBlock, New);
}
// If the instruction doesn't dominate any exit blocks, it must be dead.
if (NumInserted == 0) {
CurAST->deleteValue(&I);
if (!I.use_empty())
I.replaceAllUsesWith(UndefValue::get(I.getType()));
I.eraseFromParent();
return;
}
// Next, rewrite uses of the instruction, inserting PHI nodes as needed.
for (Value::use_iterator UI = I.use_begin(), UE = I.use_end(); UI != UE; ) {
// Grab the use before incrementing the iterator.
Use &U = UI.getUse();
// Increment the iterator before removing the use from the list.
++UI;
SSA.RewriteUseAfterInsertions(U);
}
// Update CurAST for NewPHIs if I had pointer type.
if (I.getType()->isPointerTy())
for (unsigned i = 0, e = NewPHIs.size(); i != e; ++i)
CurAST->copyValue(&I, NewPHIs[i]);
// Finally, remove the instruction from CurAST. It is no longer in the loop.
CurAST->deleteValue(&I);
}
void DuettoNativeRewriter::rewriteConstructorImplementation(Module& M, Function& F)
{
//Copy the code in a function with the right signature
Function* newFunc=getReturningConstructor(M, &F);
if(!newFunc->empty())
return;
//Visit each instruction and take note of the ones that needs to be replaced
Function::const_iterator B=F.begin();
Function::const_iterator BE=F.end();
ValueToValueMapTy valueMap;
CallInst* lowerConstructor = NULL;
const CallInst* oldLowerConstructor = NULL;
for(;B!=BE;++B)
{
BasicBlock::const_iterator I=B->begin();
BasicBlock::const_iterator IE=B->end();
for(;I!=IE;++I)
{
if(I->getOpcode()!=Instruction::Call)
continue;
const CallInst* callInst=cast<CallInst>(&(*I));
Function* f=callInst->getCalledFunction();
if(!f)
continue;
const char* startOfType;
const char* endOfType;
if(!DuettoNativeRewriter::isBuiltinConstructor(f->getName().data(), startOfType, endOfType))
continue;
//Check that the constructor is for 'this'
if(callInst->getOperand(0)!=F.arg_begin())
continue;
//If this is another constructor for the same type, change it to a
//returning constructor and use it as the 'this' argument
Function* newFunc = getReturningConstructor(M, f);
llvm::SmallVector<Value*, 4> newArgs;
for(unsigned i=1;i<callInst->getNumArgOperands();i++)
newArgs.push_back(callInst->getArgOperand(i));
lowerConstructor = CallInst::Create(newFunc, newArgs);
oldLowerConstructor = callInst;
break;
}
if(lowerConstructor)
break;
}
//Clone the linkage first
newFunc->setLinkage(F.getLinkage());
Function::arg_iterator origArg=++F.arg_begin();
Function::arg_iterator newArg=newFunc->arg_begin();
valueMap.insert(make_pair(F.arg_begin(), lowerConstructor));
for(unsigned i=1;i<F.arg_size();i++)
{
valueMap.insert(make_pair(&(*origArg), &(*newArg)));
++origArg;
++newArg;
}
SmallVector<ReturnInst*, 4> returns;
CloneFunctionInto(newFunc, &F, valueMap, false, returns);
//Find the right place to add the base construtor call
assert(lowerConstructor->getNumArgOperands()<=1 && "Native constructors with multiple args are not supported");
Instruction* callPred = NULL;
if (lowerConstructor->getNumArgOperands()==1 && Instruction::classof(lowerConstructor->getArgOperand(0)))
{
//Switch the argument to the one in the new func
lowerConstructor->setArgOperand(0, valueMap[lowerConstructor->getArgOperand(0)]);
callPred = cast<Instruction>(lowerConstructor->getArgOperand(0));
}
else
callPred = &newFunc->getEntryBlock().front();
//Add add it
lowerConstructor->insertAfter(callPred);
//Override the returs values
for(unsigned i=0;i<returns.size();i++)
{
Instruction* newInst = ReturnInst::Create(M.getContext(),lowerConstructor);
newInst->insertBefore(returns[i]);
returns[i]->removeFromParent();
}
//Recursively move all the users of the lower constructor after the call itself
Instruction* insertPoint = lowerConstructor->getNextNode();
SmallVector<Value*, 4> usersQueue(lowerConstructor->getNumUses());
unsigned int i;
Value::use_iterator it;
for(i=usersQueue.size()-1,it=lowerConstructor->use_begin();it!=lowerConstructor->use_end();++it,i--)
usersQueue[i]=it->getUser();
SmallSet<Instruction*, 4> movedInstructions;
while(!usersQueue.empty())
{
Instruction* cur=dyn_cast<Instruction>(usersQueue.pop_back_val());
if(!cur)
continue;
if(movedInstructions.count(cur))
continue;
movedInstructions.insert(cur);
cur->moveBefore(insertPoint);
//Add users of this instrucution as well
usersQueue.resize(usersQueue.size()+cur->getNumUses());
for(i=usersQueue.size()-1,it=cur->use_begin();it!=cur->use_end();++it,i--)
usersQueue[i]=it->getUser();
}
cast<Instruction>(valueMap[oldLowerConstructor])->eraseFromParent();
}
/// SurveyUse - This looks at a single use of an argument or return value
/// and determines if it should be alive or not. Adds this use to MaybeLiveUses
/// if it causes the used value to become MaybeAlive.
///
/// RetValNum is the return value number to use when this use is used in a
/// return instruction. This is used in the recursion, you should always leave
/// it at 0.
DAE::Liveness DAE::SurveyUse(Value::use_iterator U, UseVector &MaybeLiveUses,
unsigned RetValNum) {
Value *V = *U;
if (ReturnInst *RI = dyn_cast<ReturnInst>(V)) {
// The value is returned from a function. It's only live when the
// function's return value is live. We use RetValNum here, for the case
// that U is really a use of an insertvalue instruction that uses the
// orginal Use.
RetOrArg Use = CreateRet(RI->getParent()->getParent(), RetValNum);
// We might be live, depending on the liveness of Use.
return MarkIfNotLive(Use, MaybeLiveUses);
}
if (InsertValueInst *IV = dyn_cast<InsertValueInst>(V)) {
if (U.getOperandNo() != InsertValueInst::getAggregateOperandIndex()
&& IV->hasIndices())
// The use we are examining is inserted into an aggregate. Our liveness
// depends on all uses of that aggregate, but if it is used as a return
// value, only index at which we were inserted counts.
RetValNum = *IV->idx_begin();
// Note that if we are used as the aggregate operand to the insertvalue,
// we don't change RetValNum, but do survey all our uses.
Liveness Result = MaybeLive;
for (Value::use_iterator I = IV->use_begin(),
E = V->use_end(); I != E; ++I) {
Result = SurveyUse(I, MaybeLiveUses, RetValNum);
if (Result == Live)
break;
}
return Result;
}
CallSite CS = CallSite::get(V);
if (CS.getInstruction()) {
Function *F = CS.getCalledFunction();
if (F) {
// Used in a direct call.
// Find the argument number. We know for sure that this use is an
// argument, since if it was the function argument this would be an
// indirect call and the we know can't be looking at a value of the
// label type (for the invoke instruction).
unsigned ArgNo = CS.getArgumentNo(U.getOperandNo());
if (ArgNo >= F->getFunctionType()->getNumParams())
// The value is passed in through a vararg! Must be live.
return Live;
assert(CS.getArgument(ArgNo)
== CS.getInstruction()->getOperand(U.getOperandNo())
&& "Argument is not where we expected it");
// Value passed to a normal call. It's only live when the corresponding
// argument to the called function turns out live.
RetOrArg Use = CreateArg(F, ArgNo);
return MarkIfNotLive(Use, MaybeLiveUses);
}
}
// Used in any other way? Value must be live.
return Live;
}
AllocaInst* Variables::changeLocal(Value* value, PointerType* newType) {
AllocaInst *oldTarget = dyn_cast<AllocaInst>(value);
PointerType* oldPointerType = dyn_cast<PointerType>(oldTarget->getType());
PointerType *oldType = dyn_cast<PointerType>(oldPointerType->getElementType());
AllocaInst *newTarget = NULL;
errs() << "Changing the precision of pointer variable \"" << oldTarget->getName() << "\" from " << *oldType
<< " to " << *newType << ".\n";
if (diffTypes(newType, oldType)) {
newTarget = new AllocaInst(newType, getInt32(1), "", oldTarget);
// we are not calling getAlignment because in this case double requires 16. Investigate further.
unsigned alignment;
switch(newType->getElementType()->getTypeID()) {
case Type::FloatTyID:
alignment = 4;
break;
case Type::DoubleTyID:
alignment = 8;
break;
case Type::X86_FP80TyID:
alignment = 16;
break;
default:
alignment = 0;
}
newTarget->setAlignment(alignment); // depends on type? 8 for float? 16 for double?
newTarget->takeName(oldTarget);
// iterating through instructions using old AllocaInst
vector<Instruction*> erase;
Value::use_iterator it = oldTarget->use_begin();
#ifdef DEBUG
errs() << "\nOld target: ";
oldTarget->dump();
#endif
for(; it != oldTarget->use_end(); it++) {
#ifdef DEBUG
errs() << "\nA use: ";
it->dump();
errs() << "\n===============================\n";
errs() << "\nTransforming use\n";
#endif
bool is_erased = Transformer::transform(it, newTarget, oldTarget, newType, oldType, alignment);
if (!is_erased) {
erase.push_back(dyn_cast<Instruction>(*it));
}
#ifdef DEBUG
errs() << "\nDone transforming use\n";
#endif
}
// erasing uses of old instructions
for(unsigned int i = 0; i < erase.size(); i++) {
erase[i]->eraseFromParent();
}
// erase old instruction
//oldTarget->eraseFromParent();
#ifdef DEBUG
errs() << "DONE ALL TRANSFORMATION FOR POINTER\n";
#endif
} else {
errs() << "\tNo changes required.\n";
}
return newTarget;
}
/// ProcessInstruction - Given an instruction in the loop, check to see if it
/// has any uses that are outside the current loop. If so, insert LCSSA PHI
/// nodes and rewrite the uses.
bool LCSSA::ProcessInstruction(Instruction *Inst,
const SmallVectorImpl<BasicBlock*> &ExitBlocks) {
SmallVector<Use*, 16> UsesToRewrite;
BasicBlock *InstBB = Inst->getParent();
for (Value::use_iterator UI = Inst->use_begin(), E = Inst->use_end();
UI != E; ++UI) {
User *U = *UI;
BasicBlock *UserBB = cast<Instruction>(U)->getParent();
if (PHINode *PN = dyn_cast<PHINode>(U))
UserBB = PN->getIncomingBlock(UI);
if (InstBB != UserBB && !inLoop(UserBB))
UsesToRewrite.push_back(&UI.getUse());
}
// If there are no uses outside the loop, exit with no change.
if (UsesToRewrite.empty()) return false;
++NumLCSSA; // We are applying the transformation
// Invoke instructions are special in that their result value is not available
// along their unwind edge. The code below tests to see whether DomBB dominates
// the value, so adjust DomBB to the normal destination block, which is
// effectively where the value is first usable.
BasicBlock *DomBB = Inst->getParent();
if (InvokeInst *Inv = dyn_cast<InvokeInst>(Inst))
DomBB = Inv->getNormalDest();
DomTreeNode *DomNode = DT->getNode(DomBB);
SSAUpdater SSAUpdate;
SSAUpdate.Initialize(Inst->getType(), Inst->getName());
// Insert the LCSSA phi's into all of the exit blocks dominated by the
// value, and add them to the Phi's map.
for (SmallVectorImpl<BasicBlock*>::const_iterator BBI = ExitBlocks.begin(),
BBE = ExitBlocks.end(); BBI != BBE; ++BBI) {
BasicBlock *ExitBB = *BBI;
if (!DT->dominates(DomNode, DT->getNode(ExitBB))) continue;
// If we already inserted something for this BB, don't reprocess it.
if (SSAUpdate.HasValueForBlock(ExitBB)) continue;
PHINode *PN = PHINode::Create(Inst->getType(), Inst->getName()+".lcssa",
ExitBB->begin());
PN->reserveOperandSpace(PredCache.GetNumPreds(ExitBB));
// Add inputs from inside the loop for this PHI.
for (BasicBlock **PI = PredCache.GetPreds(ExitBB); *PI; ++PI) {
PN->addIncoming(Inst, *PI);
// If the exit block has a predecessor not within the loop, arrange for
// the incoming value use corresponding to that predecessor to be
// rewritten in terms of a different LCSSA PHI.
if (!inLoop(*PI))
UsesToRewrite.push_back(
&PN->getOperandUse(
PN->getOperandNumForIncomingValue(PN->getNumIncomingValues()-1)));
}
// Remember that this phi makes the value alive in this block.
SSAUpdate.AddAvailableValue(ExitBB, PN);
}
// Rewrite all uses outside the loop in terms of the new PHIs we just
// inserted.
for (unsigned i = 0, e = UsesToRewrite.size(); i != e; ++i) {
// If this use is in an exit block, rewrite to use the newly inserted PHI.
// This is required for correctness because SSAUpdate doesn't handle uses in
// the same block. It assumes the PHI we inserted is at the end of the
// block.
Instruction *User = cast<Instruction>(UsesToRewrite[i]->getUser());
BasicBlock *UserBB = User->getParent();
if (PHINode *PN = dyn_cast<PHINode>(User))
UserBB = PN->getIncomingBlock(*UsesToRewrite[i]);
if (isa<PHINode>(UserBB->begin()) &&
isExitBlock(UserBB, ExitBlocks)) {
UsesToRewrite[i]->set(UserBB->begin());
continue;
}
// Otherwise, do full PHI insertion.
SSAUpdate.RewriteUse(*UsesToRewrite[i]);
}
return true;
}
/// isOnlyCopiedFromConstantGlobal - Recursively walk the uses of a (derived)
/// pointer to an alloca. Ignore any reads of the pointer, return false if we
/// see any stores or other unknown uses. If we see pointer arithmetic, keep
/// track of whether it moves the pointer (with IsOffset) but otherwise traverse
/// the uses. If we see a memcpy/memmove that targets an unoffseted pointer to
/// the alloca, and if the source pointer is a pointer to a constant global, we
/// can optimize this.
static bool
isOnlyCopiedFromConstantGlobal(Value *V, MemTransferInst *&TheCopy,
SmallVectorImpl<Instruction *> &ToDelete,
bool IsOffset = false) {
// We track lifetime intrinsics as we encounter them. If we decide to go
// ahead and replace the value with the global, this lets the caller quickly
// eliminate the markers.
for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI!=E; ++UI) {
User *U = cast<Instruction>(*UI);
if (LoadInst *LI = dyn_cast<LoadInst>(U)) {
// Ignore non-volatile loads, they are always ok.
if (!LI->isSimple()) return false;
continue;
}
if (BitCastInst *BCI = dyn_cast<BitCastInst>(U)) {
// If uses of the bitcast are ok, we are ok.
if (!isOnlyCopiedFromConstantGlobal(BCI, TheCopy, ToDelete, IsOffset))
return false;
continue;
}
if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(U)) {
// If the GEP has all zero indices, it doesn't offset the pointer. If it
// doesn't, it does.
if (!isOnlyCopiedFromConstantGlobal(GEP, TheCopy, ToDelete,
IsOffset || !GEP->hasAllZeroIndices()))
return false;
continue;
}
if (CallSite CS = U) {
// If this is the function being called then we treat it like a load and
// ignore it.
if (CS.isCallee(UI))
continue;
// If this is a readonly/readnone call site, then we know it is just a
// load (but one that potentially returns the value itself), so we can
// ignore it if we know that the value isn't captured.
unsigned ArgNo = CS.getArgumentNo(UI);
if (CS.onlyReadsMemory() &&
(CS.getInstruction()->use_empty() || CS.doesNotCapture(ArgNo)))
continue;
// If this is being passed as a byval argument, the caller is making a
// copy, so it is only a read of the alloca.
if (CS.isByValArgument(ArgNo))
continue;
}
// Lifetime intrinsics can be handled by the caller.
if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(U)) {
if (II->getIntrinsicID() == Intrinsic::lifetime_start ||
II->getIntrinsicID() == Intrinsic::lifetime_end) {
assert(II->use_empty() && "Lifetime markers have no result to use!");
ToDelete.push_back(II);
continue;
}
}
// If this is isn't our memcpy/memmove, reject it as something we can't
// handle.
MemTransferInst *MI = dyn_cast<MemTransferInst>(U);
if (MI == 0)
return false;
// If the transfer is using the alloca as a source of the transfer, then
// ignore it since it is a load (unless the transfer is volatile).
if (UI.getOperandNo() == 1) {
if (MI->isVolatile()) return false;
continue;
}
// If we already have seen a copy, reject the second one.
if (TheCopy) return false;
// If the pointer has been offset from the start of the alloca, we can't
// safely handle this.
if (IsOffset) return false;
// If the memintrinsic isn't using the alloca as the dest, reject it.
if (UI.getOperandNo() != 0) return false;
// If the source of the memcpy/move is not a constant global, reject it.
if (!pointsToConstantGlobal(MI->getSource()))
return false;
// Otherwise, the transform is safe. Remember the copy instruction.
TheCopy = MI;
}
return true;
}
