/// UpdateCallGraphAfterInlining - Once we have cloned code over from a callee
/// into the caller, update the specified callgraph to reflect the changes we
/// made. Note that it's possible that not all code was copied over, so only
/// some edges of the callgraph may remain.
static void UpdateCallGraphAfterInlining(CallSite CS,
Function::iterator FirstNewBlock,
ValueToValueMapTy &VMap,
InlineFunctionInfo &IFI) {
CallGraph &CG = *IFI.CG;
const Function *Caller = CS.getInstruction()->getParent()->getParent();
const Function *Callee = CS.getCalledFunction();
CallGraphNode *CalleeNode = CG[Callee];
CallGraphNode *CallerNode = CG[Caller];
// Since we inlined some uninlined call sites in the callee into the caller,
// add edges from the caller to all of the callees of the callee.
CallGraphNode::iterator I = CalleeNode->begin(), E = CalleeNode->end();
// Consider the case where CalleeNode == CallerNode.
CallGraphNode::CalledFunctionsVector CallCache;
if (CalleeNode == CallerNode) {
CallCache.assign(I, E);
I = CallCache.begin();
E = CallCache.end();
}
for (; I != E; ++I) {
const Value *OrigCall = I->first;
ValueToValueMapTy::iterator VMI = VMap.find(OrigCall);
// Only copy the edge if the call was inlined!
if (VMI == VMap.end() || VMI->second == 0)
continue;
// If the call was inlined, but then constant folded, there is no edge to
// add. Check for this case.
Instruction *NewCall = dyn_cast<Instruction>(VMI->second);
if (NewCall == 0) continue;
// Remember that this call site got inlined for the client of
// InlineFunction.
IFI.InlinedCalls.push_back(NewCall);
// It's possible that inlining the callsite will cause it to go from an
// indirect to a direct call by resolving a function pointer. If this
// happens, set the callee of the new call site to a more precise
// destination. This can also happen if the call graph node of the caller
// was just unnecessarily imprecise.
if (I->second->getFunction() == 0)
if (Function *F = CallSite(NewCall).getCalledFunction()) {
// Indirect call site resolved to direct call.
CallerNode->addCalledFunction(CallSite(NewCall), CG[F]);
continue;
}
CallerNode->addCalledFunction(CallSite(NewCall), I->second);
}
// Update the call graph by deleting the edge from Callee to Caller. We must
// do this after the loop above in case Caller and Callee are the same.
CallerNode->removeCallEdgeFor(CS);
}
// doFinalization - Remove now-dead linkonce functions at the end of
// processing to avoid breaking the SCC traversal.
bool Inliner::doFinalization(CallGraph &CG) {
std::set<CallGraphNode*> FunctionsToRemove;
// Scan for all of the functions, looking for ones that should now be removed
// from the program. Insert the dead ones in the FunctionsToRemove set.
for (CallGraph::iterator I = CG.begin(), E = CG.end(); I != E; ++I) {
CallGraphNode *CGN = I->second;
if (Function *F = CGN ? CGN->getFunction() : 0) {
// If the only remaining users of the function are dead constants, remove
// them.
F->removeDeadConstantUsers();
if ((F->hasLinkOnceLinkage() || F->hasInternalLinkage()) &&
F->use_empty()) {
// Remove any call graph edges from the function to its callees.
while (!CGN->empty())
CGN->removeCallEdgeTo((CGN->end()-1)->second);
// Remove any edges from the external node to the function's call graph
// node. These edges might have been made irrelegant due to
// optimization of the program.
CG.getExternalCallingNode()->removeAnyCallEdgeTo(CGN);
// Removing the node for callee from the call graph and delete it.
FunctionsToRemove.insert(CGN);
}
}
}
// Now that we know which functions to delete, do so. We didn't want to do
// this inline, because that would invalidate our CallGraph::iterator
// objects. :(
bool Changed = false;
for (std::set<CallGraphNode*>::iterator I = FunctionsToRemove.begin(),
E = FunctionsToRemove.end(); I != E; ++I) {
delete CG.removeFunctionFromModule(*I);
++NumDeleted;
Changed = true;
}
return Changed;
}
Function * StructuredModuleEditor::cloneFunc(Function * Original) {
if (Original == NULL)
return NULL;
ValueMap<const Value*, WeakVH> VMap;
// Creates a clone of the function we are cloning
Function *Clone = CloneFunction(Original, VMap, false);
Clone->setName(Original->getName() + "-cloned");
// Adds the clone to the Module
M->getFunctionList().push_back(Clone);
// Adds the clone to the CFG
CG->getOrInsertFunction(Clone);
// Adds each of the original function's CFG node's interprocedural out-edges
// to the clone's node. All of the original function's intraprocedural in-edges are redirected to the cloned function.
// The clone will have no interprocedural in-edges as it
// was just created.
CallGraphNode *CloneNode = CG->getOrInsertFunction(Clone);
for (Function::iterator BBI = Clone->begin(), BBE = Clone->end();
BBI != BBE; ++BBI) {
for (BasicBlock::iterator II = BBI->begin(), IE = BBI->end(); II != IE;
++II) {
CallSite CS(cast<Value>(II));
// If this isn't a call, or it is a call to an intrinsic...
if (!CS || isa<IntrinsicInst>(II))
continue;
Function *Callee = CS.getCalledFunction();
if (Callee == Original) {
Callee = Clone;
CS.setCalledFunction(Clone);
}
CloneNode->addCalledFunction(CS, CG->getOrInsertFunction(Callee));
}
}
return Clone;
}
void CallGraph::addToCallGraph(Function *F) {
CallGraphNode *Node = getOrInsertFunction(F);
// If this function has external linkage, anything could call it.
if (!F->hasLocalLinkage()) {
ExternalCallingNode->addCalledFunction(CallSite(), Node);
// Found the entry point?
if (F->getName() == "main") {
if (Root) // Found multiple external mains? Don't pick one.
Root = ExternalCallingNode;
else
Root = Node; // Found a main, keep track of it!
}
}
// If this function has its address taken, anything could call it.
if (F->hasAddressTaken())
ExternalCallingNode->addCalledFunction(CallSite(), Node);
// If this function is not defined in this translation unit, it could call
// anything.
if (F->isDeclaration() && !F->isIntrinsic())
Node->addCalledFunction(CallSite(), CallsExternalNode.get());
// Look for calls by this function.
for (Function::iterator BB = F->begin(), BBE = F->end(); BB != BBE; ++BB)
for (BasicBlock::iterator II = BB->begin(), IE = BB->end(); II != IE;
++II) {
CallSite CS(cast<Value>(II));
if (CS) {
const Function *Callee = CS.getCalledFunction();
if (!Callee || !Intrinsic::isLeaf(Callee->getIntrinsicID()))
// Indirect calls of intrinsics are not allowed so no need to check.
// We can be more precise here by using TargetArg returned by
// Intrinsic::isLeaf.
Node->addCalledFunction(CS, CallsExternalNode.get());
else if (!Callee->isIntrinsic())
Node->addCalledFunction(CS, getOrInsertFunction(Callee));
}
}
}
void CallGraph::verify() const {
#ifndef NDEBUG
// For every function in the module, add it to our SILFunction set.
llvm::DenseSet<SILFunction *> Functions;
for (auto &F : M)
Functions.insert(&F);
// For every pair (SILFunction, CallGraphNode) in the
// function-to-node map, verify:
//
// a. The function is in the current module.
// b. The call graph node is for that same function.
//
// In addition, call the verify method for the function.
unsigned numEdges = 0;
for (auto &P : FunctionToNodeMap) {
SILFunction *F = P.first;
CallGraphNode *Node = P.second;
assert(Functions.count(F) &&
"Function in call graph but not in module!?");
assert(Node->getFunction() == F &&
"Func mapped to node, but node has different Function inside?!");
verify(F);
numEdges += Node->getCalleeEdges().size();
}
assert(InstToEdgeMap.size() == numEdges &&
"Some edges in InstToEdgeMap are not contained in any node");
// Verify the callee sets.
for (auto Iter : CalleeSetCache) {
auto *CalleeSet = Iter.second.getPointer();
for (CallGraphNode *Node : *CalleeSet) {
SILFunction *F = Node->getFunction();
assert(tryGetCallGraphNode(F) &&
"Callee set contains dangling node poiners");
}
}
#endif
}
/// DeleteBasicBlock - remove the specified basic block from the program,
/// updating the callgraph to reflect any now-obsolete edges due to calls that
/// exist in the BB.
void PruneEH::DeleteBasicBlock(BasicBlock *BB) {
assert(pred_empty(BB) && "BB is not dead!");
CallGraph &CG = getAnalysis<CallGraphWrapperPass>().getCallGraph();
Instruction *TokenInst = nullptr;
CallGraphNode *CGN = CG[BB->getParent()];
for (BasicBlock::iterator I = BB->end(), E = BB->begin(); I != E; ) {
--I;
if (I->getType()->isTokenTy()) {
TokenInst = &*I;
break;
}
if (auto CS = CallSite (&*I)) {
const Function *Callee = CS.getCalledFunction();
if (!Callee || !Intrinsic::isLeaf(Callee->getIntrinsicID()))
CGN->removeCallEdgeFor(CS);
else if (!Callee->isIntrinsic())
CGN->removeCallEdgeFor(CS);
}
if (!I->use_empty())
I->replaceAllUsesWith(UndefValue::get(I->getType()));
}
if (TokenInst) {
if (!isa<TerminatorInst>(TokenInst))
changeToUnreachable(TokenInst->getNextNode(), /*UseLLVMTrap=*/false);
} else {
// Get the list of successors of this block.
std::vector<BasicBlock *> Succs(succ_begin(BB), succ_end(BB));
for (unsigned i = 0, e = Succs.size(); i != e; ++i)
Succs[i]->removePredecessor(BB);
BB->eraseFromParent();
}
}
/// UpdateCallGraphAfterInlining - Once we have cloned code over from a callee
/// into the caller, update the specified callgraph to reflect the changes we
/// made. Note that it's possible that not all code was copied over, so only
/// some edges of the callgraph will be remain.
static void UpdateCallGraphAfterInlining(const Function *Caller,
const Function *Callee,
Function::iterator FirstNewBlock,
DenseMap<const Value*, Value*> &ValueMap,
CallGraph &CG) {
// Update the call graph by deleting the edge from Callee to Caller
CallGraphNode *CalleeNode = CG[Callee];
CallGraphNode *CallerNode = CG[Caller];
CallerNode->removeCallEdgeTo(CalleeNode);
// Since we inlined some uninlined call sites in the callee into the caller,
// add edges from the caller to all of the callees of the callee.
for (CallGraphNode::iterator I = CalleeNode->begin(),
E = CalleeNode->end(); I != E; ++I) {
const Instruction *OrigCall = I->first.getInstruction();
DenseMap<const Value*, Value*>::iterator VMI = ValueMap.find(OrigCall);
// Only copy the edge if the call was inlined!
if (VMI != ValueMap.end() && VMI->second) {
// If the call was inlined, but then constant folded, there is no edge to
// add. Check for this case.
if (Instruction *NewCall = dyn_cast<Instruction>(VMI->second))
CallerNode->addCalledFunction(CallSite::get(NewCall), I->second);
}
}
}
bool CallGraphCFG::findBBPath(CallGraphNode *n, std::vector<BasicBlock*> &path, std::string srcFile, int srcLine)
{
if (n == NULL) return false;
Function *F = n->getFunction();
std::cerr << "Processing " << F->getNameStr() << "\n";
// Are we on a leaf?
if (n->size() == 0) {
BasicBlock *bb=NULL;
if (findLineInFunction(F,&bb,srcFile,srcLine)) {
path.push_back(bb);
return true;
}
}
for (CallGraphNode::iterator it = n->begin(); it != n->end(); ++it) {
CallSite cs = it->first;
CallGraphNode *tCGN = it->second;
Instruction *tI = cs.getInstruction();
if (tI == NULL) return false;
BasicBlock *bb = tI->getParent();
Function *tF = tCGN->getFunction();
path.push_back(bb);
if (findLineInBB(bb,srcFile,srcLine))
return true;
if (tF != F) { // Dont get stuck in recursion
if (findBBPath(tCGN,path,srcFile,srcLine))
return true;
}
std::cerr << " Dead end, reverting...\n"; // FIX: This is misleading, not really correct.
path.pop_back();
}
return false;
}
// InlineCallIfPossible - If it is possible to inline the specified call site,
// do so and update the CallGraph for this operation.
static bool InlineCallIfPossible(CallSite CS, CallGraph &CG,
const std::set<Function*> &SCCFunctions,
const TargetData &TD) {
Function *Callee = CS.getCalledFunction();
if (!InlineFunction(CS, &CG, &TD)) return false;
// If we inlined the last possible call site to the function, delete the
// function body now.
if (Callee->use_empty() && Callee->hasInternalLinkage() &&
!SCCFunctions.count(Callee)) {
DOUT << " -> Deleting dead function: " << Callee->getName() << "\n";
// Remove any call graph edges from the callee to its callees.
CallGraphNode *CalleeNode = CG[Callee];
while (!CalleeNode->empty())
CalleeNode->removeCallEdgeTo((CalleeNode->end()-1)->second);
// Removing the node for callee from the call graph and delete it.
delete CG.removeFunctionFromModule(CalleeNode);
++NumDeleted;
}
return true;
}
/// DeleteBasicBlock - remove the specified basic block from the program,
/// updating the callgraph to reflect any now-obsolete edges due to calls that
/// exist in the BB.
void PruneEH::DeleteBasicBlock(BasicBlock *BB) {
assert(pred_begin(BB) == pred_end(BB) && "BB is not dead!");
CallGraph &CG = getAnalysis<CallGraph>();
CallGraphNode *CGN = CG[BB->getParent()];
for (BasicBlock::iterator I = BB->end(), E = BB->begin(); I != E; ) {
--I;
if (CallInst *CI = dyn_cast<CallInst>(I)) {
if (!isa<IntrinsicInst>(I))
CGN->removeCallEdgeFor(CI);
} else if (InvokeInst *II = dyn_cast<InvokeInst>(I))
CGN->removeCallEdgeFor(II);
if (!I->use_empty())
I->replaceAllUsesWith(UndefValue::get(I->getType()));
}
// Get the list of successors of this block.
std::vector<BasicBlock*> Succs(succ_begin(BB), succ_end(BB));
for (unsigned i = 0, e = Succs.size(); i != e; ++i)
Succs[i]->removePredecessor(BB);
BB->eraseFromParent();
}
bool StructuredModuleEditor::removeFunc(Function *FunctionToRemove) {
// Checks to make sure the function we are trying to remove
// actually exists in the CFG
if (FunctionToRemove == NULL) {
OS << "Function does not exist in the call graph!\n";
return false;
}
CallGraphNode *NodeToRemove = (*CG)[FunctionToRemove];
// We cannot remove a node if it has any inteprocedural in-edges
for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I) {
CallGraphNode *CallingNode = (*CG)[I];
for (CallGraphNode::iterator CGNI = CallingNode->begin(), CGNE =
CallingNode->end(); CGNI != CGNE; ++CGNI) {
Function *Caller = I;
Function *Callee = CGNI->second->getFunction();
if (Callee == FunctionToRemove && Caller != Callee) {
OS << "Cannot remove " << FunctionToRemove->getName()
<< " because it has at least one interprocedural edge!\n";
OS << "It is called by " << Caller->getName() << "\n";
return false;
}
}
}
// Removes all call graph edges from the node we are removing to its callees.
NodeToRemove->removeAllCalledFunctions();
CG->getExternalCallingNode()->removeAnyCallEdgeTo(NodeToRemove);
// Removes all call graph edges from callees to the node we are removing
for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I) {
CallGraphNode *CallingNode = (*CG)[I];
CallingNode->removeAnyCallEdgeTo(NodeToRemove);
}
NodeToRemove->removeAnyCallEdgeTo(CG->getCallsExternalNode());
// Removes the function from the module and the CFG
FunctionToRemove->dropAllReferences();
// Remove the function from the module
CG->removeFunctionFromModule(NodeToRemove);
return true;
}
bool CallGraphChecker::existsInCallGraph(Instruction *Call, Function *Callee) {
CallGraph &CG = getAnalysis<CallGraph>();
assert(Call && Callee);
CallGraphNode *CallerNode = CG[Call->getParent()->getParent()];
CallGraphNode *CalleeNode = CG[Callee];
assert(CallerNode && CalleeNode);
if (find(CallerNode->begin(), CallerNode->end(),
CallGraphNode::CallRecord(Call, CalleeNode))
!= CallerNode->end()) {
return true;
}
// An instruction conservatively calls all functions by calling
// CallsExternalNode.
if (find(CallerNode->begin(), CallerNode->end(),
CallGraphNode::CallRecord(Call, CG.getCallsExternalNode()))
!= CallerNode->end()) {
return true;
}
return false;
}
/// UpdateCallGraphAfterInlining - Once we have cloned code over from a callee
/// into the caller, update the specified callgraph to reflect the changes we
/// made. Note that it's possible that not all code was copied over, so only
/// some edges of the callgraph may remain.
static void UpdateCallGraphAfterInlining(CallSite CS,
Function::iterator FirstNewBlock,
DenseMap<const Value*, Value*> &ValueMap,
CallGraph &CG) {
const Function *Caller = CS.getInstruction()->getParent()->getParent();
const Function *Callee = CS.getCalledFunction();
CallGraphNode *CalleeNode = CG[Callee];
CallGraphNode *CallerNode = CG[Caller];
// Since we inlined some uninlined call sites in the callee into the caller,
// add edges from the caller to all of the callees of the callee.
CallGraphNode::iterator I = CalleeNode->begin(), E = CalleeNode->end();
// Consider the case where CalleeNode == CallerNode.
CallGraphNode::CalledFunctionsVector CallCache;
if (CalleeNode == CallerNode) {
CallCache.assign(I, E);
I = CallCache.begin();
E = CallCache.end();
}
for (; I != E; ++I) {
const Instruction *OrigCall = I->first.getInstruction();
DenseMap<const Value*, Value*>::iterator VMI = ValueMap.find(OrigCall);
// Only copy the edge if the call was inlined!
if (VMI != ValueMap.end() && VMI->second) {
// If the call was inlined, but then constant folded, there is no edge to
// add. Check for this case.
if (Instruction *NewCall = dyn_cast<Instruction>(VMI->second))
CallerNode->addCalledFunction(CallSite::get(NewCall), I->second);
}
}
// Update the call graph by deleting the edge from Callee to Caller. We must
// do this after the loop above in case Caller and Callee are the same.
CallerNode->removeCallEdgeFor(CS);
}
bool InternalizePass::runOnModule(Module &M) {
CallGraph *CG = getAnalysisIfAvailable<CallGraph>();
CallGraphNode *ExternalNode = CG ? CG->getExternalCallingNode() : 0;
bool Changed = false;
// Never internalize functions which code-gen might insert.
// FIXME: We should probably add this (and the __stack_chk_guard) via some
// type of call-back in CodeGen.
ExternalNames.insert("__stack_chk_fail");
// Mark all functions not in the api as internal.
// FIXME: maybe use private linkage?
for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I)
if (!I->isDeclaration() && // Function must be defined here
// Available externally is really just a "declaration with a body".
!I->hasAvailableExternallyLinkage() &&
!I->hasLocalLinkage() && // Can't already have internal linkage
!ExternalNames.count(I->getName())) {// Not marked to keep external?
I->setLinkage(GlobalValue::InternalLinkage);
// Remove a callgraph edge from the external node to this function.
if (ExternalNode) ExternalNode->removeOneAbstractEdgeTo((*CG)[I]);
Changed = true;
++NumFunctions;
DEBUG(dbgs() << "Internalizing func " << I->getName() << "\n");
}
// Never internalize the llvm.used symbol. It is used to implement
// attribute((used)).
// FIXME: Shouldn't this just filter on llvm.metadata section??
ExternalNames.insert("llvm.used");
ExternalNames.insert("llvm.compiler.used");
// Never internalize anchors used by the machine module info, else the info
// won't find them. (see MachineModuleInfo.)
ExternalNames.insert("llvm.global_ctors");
ExternalNames.insert("llvm.global_dtors");
ExternalNames.insert("llvm.global.annotations");
// Never internalize symbols code-gen inserts.
ExternalNames.insert("__stack_chk_guard");
// Mark all global variables with initializers that are not in the api as
// internal as well.
// FIXME: maybe use private linkage?
for (Module::global_iterator I = M.global_begin(), E = M.global_end();
I != E; ++I)
if (!I->isDeclaration() && !I->hasLocalLinkage() &&
// Available externally is really just a "declaration with a body".
!I->hasAvailableExternallyLinkage() &&
!ExternalNames.count(I->getName())) {
I->setLinkage(GlobalValue::InternalLinkage);
Changed = true;
++NumGlobals;
DEBUG(dbgs() << "Internalized gvar " << I->getName() << "\n");
}
// Mark all aliases that are not in the api as internal as well.
for (Module::alias_iterator I = M.alias_begin(), E = M.alias_end();
I != E; ++I)
if (!I->isDeclaration() && !I->hasInternalLinkage() &&
// Available externally is really just a "declaration with a body".
!I->hasAvailableExternallyLinkage() &&
!ExternalNames.count(I->getName())) {
I->setLinkage(GlobalValue::InternalLinkage);
Changed = true;
++NumAliases;
DEBUG(dbgs() << "Internalized alias " << I->getName() << "\n");
}
return Changed;
}
// InlineFunction - This function inlines the called function into the basic
// block of the caller. This returns false if it is not possible to inline this
// call. The program is still in a well defined state if this occurs though.
//
// Note that this only does one level of inlining. For example, if the
// instruction 'call B' is inlined, and 'B' calls 'C', then the call to 'C' now
// exists in the instruction stream. Similiarly this will inline a recursive
// function by one level.
//
bool llvm::InlineFunction(CallSite CS, CallGraph *CG, const TargetData *TD) {
Instruction *TheCall = CS.getInstruction();
assert(TheCall->getParent() && TheCall->getParent()->getParent() &&
"Instruction not in function!");
const Function *CalledFunc = CS.getCalledFunction();
if (CalledFunc == 0 || // Can't inline external function or indirect
CalledFunc->isDeclaration() || // call, or call to a vararg function!
CalledFunc->getFunctionType()->isVarArg()) return false;
// If the call to the callee is not a tail call, we must clear the 'tail'
// flags on any calls that we inline.
bool MustClearTailCallFlags =
!(isa<CallInst>(TheCall) && cast<CallInst>(TheCall)->isTailCall());
// If the call to the callee cannot throw, set the 'nounwind' flag on any
// calls that we inline.
bool MarkNoUnwind = CS.doesNotThrow();
BasicBlock *OrigBB = TheCall->getParent();
Function *Caller = OrigBB->getParent();
// GC poses two hazards to inlining, which only occur when the callee has GC:
// 1. If the caller has no GC, then the callee's GC must be propagated to the
// caller.
// 2. If the caller has a differing GC, it is invalid to inline.
if (CalledFunc->hasGC()) {
if (!Caller->hasGC())
Caller->setGC(CalledFunc->getGC());
else if (CalledFunc->getGC() != Caller->getGC())
return false;
}
// Get an iterator to the last basic block in the function, which will have
// the new function inlined after it.
//
Function::iterator LastBlock = &Caller->back();
// Make sure to capture all of the return instructions from the cloned
// function.
std::vector<ReturnInst*> Returns;
ClonedCodeInfo InlinedFunctionInfo;
Function::iterator FirstNewBlock;
{ // Scope to destroy ValueMap after cloning.
DenseMap<const Value*, Value*> ValueMap;
assert(CalledFunc->arg_size() == CS.arg_size() &&
"No varargs calls can be inlined!");
// Calculate the vector of arguments to pass into the function cloner, which
// matches up the formal to the actual argument values.
CallSite::arg_iterator AI = CS.arg_begin();
unsigned ArgNo = 0;
for (Function::const_arg_iterator I = CalledFunc->arg_begin(),
E = CalledFunc->arg_end(); I != E; ++I, ++AI, ++ArgNo) {
Value *ActualArg = *AI;
// When byval arguments actually inlined, we need to make the copy implied
// by them explicit. However, we don't do this if the callee is readonly
// or readnone, because the copy would be unneeded: the callee doesn't
// modify the struct.
if (CalledFunc->paramHasAttr(ArgNo+1, Attribute::ByVal) &&
!CalledFunc->onlyReadsMemory()) {
const Type *AggTy = cast<PointerType>(I->getType())->getElementType();
const Type *VoidPtrTy = PointerType::getUnqual(Type::Int8Ty);
// Create the alloca. If we have TargetData, use nice alignment.
unsigned Align = 1;
if (TD) Align = TD->getPrefTypeAlignment(AggTy);
Value *NewAlloca = new AllocaInst(AggTy, 0, Align, I->getName(),
Caller->begin()->begin());
// Emit a memcpy.
const Type *Tys[] = { Type::Int64Ty };
Function *MemCpyFn = Intrinsic::getDeclaration(Caller->getParent(),
Intrinsic::memcpy,
Tys, 1);
Value *DestCast = new BitCastInst(NewAlloca, VoidPtrTy, "tmp", TheCall);
Value *SrcCast = new BitCastInst(*AI, VoidPtrTy, "tmp", TheCall);
Value *Size;
if (TD == 0)
Size = ConstantExpr::getSizeOf(AggTy);
else
Size = ConstantInt::get(Type::Int64Ty, TD->getTypeStoreSize(AggTy));
// Always generate a memcpy of alignment 1 here because we don't know
// the alignment of the src pointer. Other optimizations can infer
// better alignment.
Value *CallArgs[] = {
DestCast, SrcCast, Size, ConstantInt::get(Type::Int32Ty, 1)
};
CallInst *TheMemCpy =
CallInst::Create(MemCpyFn, CallArgs, CallArgs+4, "", TheCall);
// If we have a call graph, update it.
if (CG) {
CallGraphNode *MemCpyCGN = CG->getOrInsertFunction(MemCpyFn);
CallGraphNode *CallerNode = (*CG)[Caller];
CallerNode->addCalledFunction(TheMemCpy, MemCpyCGN);
}
// Uses of the argument in the function should use our new alloca
// instead.
ActualArg = NewAlloca;
}
ValueMap[I] = ActualArg;
}
// We want the inliner to prune the code as it copies. We would LOVE to
// have no dead or constant instructions leftover after inlining occurs
// (which can happen, e.g., because an argument was constant), but we'll be
// happy with whatever the cloner can do.
CloneAndPruneFunctionInto(Caller, CalledFunc, ValueMap, Returns, ".i",
&InlinedFunctionInfo, TD);
// Remember the first block that is newly cloned over.
FirstNewBlock = LastBlock; ++FirstNewBlock;
// Update the callgraph if requested.
if (CG)
UpdateCallGraphAfterInlining(CS, FirstNewBlock, ValueMap, *CG);
}
// If there are any alloca instructions in the block that used to be the entry
// block for the callee, move them to the entry block of the caller. First
// calculate which instruction they should be inserted before. We insert the
// instructions at the end of the current alloca list.
//
{
BasicBlock::iterator InsertPoint = Caller->begin()->begin();
for (BasicBlock::iterator I = FirstNewBlock->begin(),
E = FirstNewBlock->end(); I != E; )
if (AllocaInst *AI = dyn_cast<AllocaInst>(I++)) {
// If the alloca is now dead, remove it. This often occurs due to code
// specialization.
if (AI->use_empty()) {
AI->eraseFromParent();
continue;
}
if (isa<Constant>(AI->getArraySize())) {
// Scan for the block of allocas that we can move over, and move them
// all at once.
while (isa<AllocaInst>(I) &&
isa<Constant>(cast<AllocaInst>(I)->getArraySize()))
++I;
// Transfer all of the allocas over in a block. Using splice means
// that the instructions aren't removed from the symbol table, then
// reinserted.
Caller->getEntryBlock().getInstList().splice(
InsertPoint,
FirstNewBlock->getInstList(),
AI, I);
}
}
}
// If the inlined code contained dynamic alloca instructions, wrap the inlined
// code with llvm.stacksave/llvm.stackrestore intrinsics.
if (InlinedFunctionInfo.ContainsDynamicAllocas) {
Module *M = Caller->getParent();
// Get the two intrinsics we care about.
Constant *StackSave, *StackRestore;
StackSave = Intrinsic::getDeclaration(M, Intrinsic::stacksave);
StackRestore = Intrinsic::getDeclaration(M, Intrinsic::stackrestore);
// If we are preserving the callgraph, add edges to the stacksave/restore
// functions for the calls we insert.
CallGraphNode *StackSaveCGN = 0, *StackRestoreCGN = 0, *CallerNode = 0;
if (CG) {
// We know that StackSave/StackRestore are Function*'s, because they are
// intrinsics which must have the right types.
StackSaveCGN = CG->getOrInsertFunction(cast<Function>(StackSave));
StackRestoreCGN = CG->getOrInsertFunction(cast<Function>(StackRestore));
CallerNode = (*CG)[Caller];
}
// Insert the llvm.stacksave.
CallInst *SavedPtr = CallInst::Create(StackSave, "savedstack",
FirstNewBlock->begin());
if (CG) CallerNode->addCalledFunction(SavedPtr, StackSaveCGN);
// Insert a call to llvm.stackrestore before any return instructions in the
// inlined function.
for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
CallInst *CI = CallInst::Create(StackRestore, SavedPtr, "", Returns[i]);
if (CG) CallerNode->addCalledFunction(CI, StackRestoreCGN);
}
// Count the number of StackRestore calls we insert.
unsigned NumStackRestores = Returns.size();
// If we are inlining an invoke instruction, insert restores before each
// unwind. These unwinds will be rewritten into branches later.
if (InlinedFunctionInfo.ContainsUnwinds && isa<InvokeInst>(TheCall)) {
for (Function::iterator BB = FirstNewBlock, E = Caller->end();
BB != E; ++BB)
if (UnwindInst *UI = dyn_cast<UnwindInst>(BB->getTerminator())) {
CallInst::Create(StackRestore, SavedPtr, "", UI);
++NumStackRestores;
}
}
}
// If we are inlining tail call instruction through a call site that isn't
// marked 'tail', we must remove the tail marker for any calls in the inlined
// code. Also, calls inlined through a 'nounwind' call site should be marked
// 'nounwind'.
if (InlinedFunctionInfo.ContainsCalls &&
(MustClearTailCallFlags || MarkNoUnwind)) {
for (Function::iterator BB = FirstNewBlock, E = Caller->end();
BB != E; ++BB)
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
if (CallInst *CI = dyn_cast<CallInst>(I)) {
if (MustClearTailCallFlags)
CI->setTailCall(false);
if (MarkNoUnwind)
CI->setDoesNotThrow();
}
}
// If we are inlining through a 'nounwind' call site then any inlined 'unwind'
// instructions are unreachable.
if (InlinedFunctionInfo.ContainsUnwinds && MarkNoUnwind)
for (Function::iterator BB = FirstNewBlock, E = Caller->end();
BB != E; ++BB) {
TerminatorInst *Term = BB->getTerminator();
if (isa<UnwindInst>(Term)) {
new UnreachableInst(Term);
BB->getInstList().erase(Term);
}
}
// If we are inlining for an invoke instruction, we must make sure to rewrite
// any inlined 'unwind' instructions into branches to the invoke exception
// destination, and call instructions into invoke instructions.
if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall))
HandleInlinedInvoke(II, FirstNewBlock, InlinedFunctionInfo);
// If we cloned in _exactly one_ basic block, and if that block ends in a
// return instruction, we splice the body of the inlined callee directly into
// the calling basic block.
if (Returns.size() == 1 && std::distance(FirstNewBlock, Caller->end()) == 1) {
// Move all of the instructions right before the call.
OrigBB->getInstList().splice(TheCall, FirstNewBlock->getInstList(),
FirstNewBlock->begin(), FirstNewBlock->end());
// Remove the cloned basic block.
Caller->getBasicBlockList().pop_back();
// If the call site was an invoke instruction, add a branch to the normal
// destination.
if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall))
BranchInst::Create(II->getNormalDest(), TheCall);
// If the return instruction returned a value, replace uses of the call with
// uses of the returned value.
if (!TheCall->use_empty()) {
ReturnInst *R = Returns[0];
TheCall->replaceAllUsesWith(R->getReturnValue());
}
// Since we are now done with the Call/Invoke, we can delete it.
TheCall->eraseFromParent();
// Since we are now done with the return instruction, delete it also.
Returns[0]->eraseFromParent();
// We are now done with the inlining.
return true;
}
// Otherwise, we have the normal case, of more than one block to inline or
// multiple return sites.
// We want to clone the entire callee function into the hole between the
// "starter" and "ender" blocks. How we accomplish this depends on whether
// this is an invoke instruction or a call instruction.
BasicBlock *AfterCallBB;
if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall)) {
// Add an unconditional branch to make this look like the CallInst case...
BranchInst *NewBr = BranchInst::Create(II->getNormalDest(), TheCall);
// Split the basic block. This guarantees that no PHI nodes will have to be
// updated due to new incoming edges, and make the invoke case more
// symmetric to the call case.
AfterCallBB = OrigBB->splitBasicBlock(NewBr,
CalledFunc->getName()+".exit");
} else { // It's a call
// If this is a call instruction, we need to split the basic block that
// the call lives in.
//
AfterCallBB = OrigBB->splitBasicBlock(TheCall,
CalledFunc->getName()+".exit");
}
// Change the branch that used to go to AfterCallBB to branch to the first
// basic block of the inlined function.
//
TerminatorInst *Br = OrigBB->getTerminator();
assert(Br && Br->getOpcode() == Instruction::Br &&
"splitBasicBlock broken!");
Br->setOperand(0, FirstNewBlock);
// Now that the function is correct, make it a little bit nicer. In
// particular, move the basic blocks inserted from the end of the function
// into the space made by splitting the source basic block.
Caller->getBasicBlockList().splice(AfterCallBB, Caller->getBasicBlockList(),
FirstNewBlock, Caller->end());
// Handle all of the return instructions that we just cloned in, and eliminate
// any users of the original call/invoke instruction.
const Type *RTy = CalledFunc->getReturnType();
if (Returns.size() > 1) {
// The PHI node should go at the front of the new basic block to merge all
// possible incoming values.
PHINode *PHI = 0;
if (!TheCall->use_empty()) {
PHI = PHINode::Create(RTy, TheCall->getName(),
AfterCallBB->begin());
// Anything that used the result of the function call should now use the
// PHI node as their operand.
TheCall->replaceAllUsesWith(PHI);
}
// Loop over all of the return instructions adding entries to the PHI node
// as appropriate.
if (PHI) {
for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
ReturnInst *RI = Returns[i];
assert(RI->getReturnValue()->getType() == PHI->getType() &&
"Ret value not consistent in function!");
PHI->addIncoming(RI->getReturnValue(), RI->getParent());
}
}
// Add a branch to the merge points and remove return instructions.
for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
ReturnInst *RI = Returns[i];
BranchInst::Create(AfterCallBB, RI);
RI->eraseFromParent();
}
} else if (!Returns.empty()) {
// Otherwise, if there is exactly one return value, just replace anything
// using the return value of the call with the computed value.
if (!TheCall->use_empty())
TheCall->replaceAllUsesWith(Returns[0]->getReturnValue());
// Splice the code from the return block into the block that it will return
// to, which contains the code that was after the call.
BasicBlock *ReturnBB = Returns[0]->getParent();
AfterCallBB->getInstList().splice(AfterCallBB->begin(),
ReturnBB->getInstList());
// Update PHI nodes that use the ReturnBB to use the AfterCallBB.
ReturnBB->replaceAllUsesWith(AfterCallBB);
// Delete the return instruction now and empty ReturnBB now.
Returns[0]->eraseFromParent();
ReturnBB->eraseFromParent();
} else if (!TheCall->use_empty()) {
// No returns, but something is using the return value of the call. Just
// nuke the result.
TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
}
// Since we are now done with the Call/Invoke, we can delete it.
TheCall->eraseFromParent();
// We should always be able to fold the entry block of the function into the
// single predecessor of the block...
assert(cast<BranchInst>(Br)->isUnconditional() && "splitBasicBlock broken!");
BasicBlock *CalleeEntry = cast<BranchInst>(Br)->getSuccessor(0);
// Splice the code entry block into calling block, right before the
// unconditional branch.
OrigBB->getInstList().splice(Br, CalleeEntry->getInstList());
CalleeEntry->replaceAllUsesWith(OrigBB); // Update PHI nodes
// Remove the unconditional branch.
OrigBB->getInstList().erase(Br);
// Now we can remove the CalleeEntry block, which is now empty.
Caller->getBasicBlockList().erase(CalleeEntry);
return true;
}
/// updateCallSites - Update all sites that call F to use NF.
CallGraphNode *SRETPromotion::updateCallSites(Function *F, Function *NF) {
CallGraph &CG = getAnalysis<CallGraph>();
SmallVector<Value*, 16> Args;
// Attributes - Keep track of the parameter attributes for the arguments.
SmallVector<AttributeWithIndex, 8> ArgAttrsVec;
// Get a new callgraph node for NF.
CallGraphNode *NF_CGN = CG.getOrInsertFunction(NF);
while (!F->use_empty()) {
CallSite CS(*F->use_begin());
Instruction *Call = CS.getInstruction();
const AttrListPtr &PAL = F->getAttributes();
// Add any return attributes.
if (Attributes attrs = PAL.getRetAttributes())
ArgAttrsVec.push_back(AttributeWithIndex::get(0, attrs));
// Copy arguments, however skip first one.
CallSite::arg_iterator AI = CS.arg_begin(), AE = CS.arg_end();
Value *FirstCArg = *AI;
++AI;
// 0th parameter attribute is reserved for return type.
// 1th parameter attribute is for first 1st sret argument.
unsigned ParamIndex = 2;
while (AI != AE) {
Args.push_back(*AI);
if (Attributes Attrs = PAL.getParamAttributes(ParamIndex))
ArgAttrsVec.push_back(AttributeWithIndex::get(ParamIndex - 1, Attrs));
++ParamIndex;
++AI;
}
// Add any function attributes.
if (Attributes attrs = PAL.getFnAttributes())
ArgAttrsVec.push_back(AttributeWithIndex::get(~0, attrs));
AttrListPtr NewPAL = AttrListPtr::get(ArgAttrsVec.begin(), ArgAttrsVec.end());
// Build new call instruction.
Instruction *New;
if (InvokeInst *II = dyn_cast<InvokeInst>(Call)) {
New = InvokeInst::Create(NF, II->getNormalDest(), II->getUnwindDest(),
Args.begin(), Args.end(), "", Call);
cast<InvokeInst>(New)->setCallingConv(CS.getCallingConv());
cast<InvokeInst>(New)->setAttributes(NewPAL);
} else {
New = CallInst::Create(NF, Args.begin(), Args.end(), "", Call);
cast<CallInst>(New)->setCallingConv(CS.getCallingConv());
cast<CallInst>(New)->setAttributes(NewPAL);
if (cast<CallInst>(Call)->isTailCall())
cast<CallInst>(New)->setTailCall();
}
Args.clear();
ArgAttrsVec.clear();
New->takeName(Call);
// Update the callgraph to know that the callsite has been transformed.
CallGraphNode *CalleeNode = CG[Call->getParent()->getParent()];
CalleeNode->removeCallEdgeFor(Call);
CalleeNode->addCalledFunction(New, NF_CGN);
// Update all users of sret parameter to extract value using extractvalue.
for (Value::use_iterator UI = FirstCArg->use_begin(),
UE = FirstCArg->use_end(); UI != UE; ) {
User *U2 = *UI++;
CallInst *C2 = dyn_cast<CallInst>(U2);
if (C2 && (C2 == Call))
continue;
GetElementPtrInst *UGEP = cast<GetElementPtrInst>(U2);
ConstantInt *Idx = cast<ConstantInt>(UGEP->getOperand(2));
Value *GR = ExtractValueInst::Create(New, Idx->getZExtValue(),
"evi", UGEP);
while(!UGEP->use_empty()) {
// isSafeToUpdateAllCallers has checked that all GEP uses are
// LoadInsts
LoadInst *L = cast<LoadInst>(*UGEP->use_begin());
L->replaceAllUsesWith(GR);
L->eraseFromParent();
}
UGEP->eraseFromParent();
continue;
}
Call->eraseFromParent();
}
return NF_CGN;
}
bool InternalizePass::runOnModule(Module &M) {
CallGraphWrapperPass *CGPass = getAnalysisIfAvailable<CallGraphWrapperPass>();
CallGraph *CG = CGPass ? &CGPass->getCallGraph() : 0;
CallGraphNode *ExternalNode = CG ? CG->getExternalCallingNode() : 0;
bool Changed = false;
SmallPtrSet<GlobalValue *, 8> Used;
collectUsedGlobalVariables(M, Used, false);
// We must assume that globals in llvm.used have a reference that not even
// the linker can see, so we don't internalize them.
// For llvm.compiler.used the situation is a bit fuzzy. The assembler and
// linker can drop those symbols. If this pass is running as part of LTO,
// one might think that it could just drop llvm.compiler.used. The problem
// is that even in LTO llvm doesn't see every reference. For example,
// we don't see references from function local inline assembly. To be
// conservative, we internalize symbols in llvm.compiler.used, but we
// keep llvm.compiler.used so that the symbol is not deleted by llvm.
for (SmallPtrSet<GlobalValue *, 8>::iterator I = Used.begin(), E = Used.end();
I != E; ++I) {
GlobalValue *V = *I;
ExternalNames.insert(V->getName());
}
// Mark all functions not in the api as internal.
for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) {
if (!shouldInternalize(*I, ExternalNames, OnlyHidden))
continue;
I->setLinkage(GlobalValue::InternalLinkage);
if (ExternalNode)
// Remove a callgraph edge from the external node to this function.
ExternalNode->removeOneAbstractEdgeTo((*CG)[I]);
Changed = true;
++NumFunctions;
DEBUG(dbgs() << "Internalizing func " << I->getName() << "\n");
}
// Never internalize the llvm.used symbol. It is used to implement
// attribute((used)).
// FIXME: Shouldn't this just filter on llvm.metadata section??
ExternalNames.insert("llvm.used");
ExternalNames.insert("llvm.compiler.used");
// Never internalize anchors used by the machine module info, else the info
// won't find them. (see MachineModuleInfo.)
ExternalNames.insert("llvm.global_ctors");
ExternalNames.insert("llvm.global_dtors");
ExternalNames.insert("llvm.global.annotations");
// Never internalize symbols code-gen inserts.
// FIXME: We should probably add this (and the __stack_chk_guard) via some
// type of call-back in CodeGen.
ExternalNames.insert("__stack_chk_fail");
ExternalNames.insert("__stack_chk_guard");
// Mark all global variables with initializers that are not in the api as
// internal as well.
for (Module::global_iterator I = M.global_begin(), E = M.global_end();
I != E; ++I) {
if (!shouldInternalize(*I, ExternalNames, OnlyHidden))
continue;
I->setLinkage(GlobalValue::InternalLinkage);
Changed = true;
++NumGlobals;
DEBUG(dbgs() << "Internalized gvar " << I->getName() << "\n");
}
// Mark all aliases that are not in the api as internal as well.
for (Module::alias_iterator I = M.alias_begin(), E = M.alias_end();
I != E; ++I) {
if (!shouldInternalize(*I, ExternalNames, OnlyHidden))
continue;
I->setLinkage(GlobalValue::InternalLinkage);
Changed = true;
++NumAliases;
DEBUG(dbgs() << "Internalized alias " << I->getName() << "\n");
}
return Changed;
}
void StructuredModuleEditor::instrumentCallsToFunction(Function *Callee) {
if (Callee == NULL) {
OS << "Function not found!\n";
return;
}
InstList Calls = getCallsToFunction(Callee);
FuncList Callers;
for (InstList::iterator II = Calls.begin(), IE = Calls.end(); II != IE;
++II) {
Function *Caller = (*II)->getParent()->getParent();
if (std::find(Callers.begin(), Callers.end(), Caller) == Callers.end())
Callers.push_back(Caller);
}
OS << Callers.size() << " functions call '" << Callee->getName()
<< "'...\n";
OS << "=================================\n";
for (FuncList::iterator FI = Callers.begin(), FE = Callers.end(); FI != FE;
++FI) {
OS << (*FI)->getName() << "\n";
}
OS << "=================================\n";
std::vector<Value*> PreArgs;
std::vector<Type*> PreArgTypes;
for (Function::arg_iterator I = Callee->arg_begin(), E = Callee->arg_end();
I != E; ++I) {
PreArgTypes.push_back(I->getType());
PreArgs.push_back(I);
}
std::vector<Type*> PostArgTypes;
if (!Callee->getReturnType()->isVoidTy()) {
PostArgTypes.push_back(Callee->getReturnType());
}
FuncList Clones;
Clones.push_back(Callee);
for (uint64_t i = 0; i < Callers.size() - 1; i++) {
Function *Clone = cloneFunc(Callee);
Clones.push_back(Clone);
}
for (uint64_t i = 0; i < Clones.size(); i++) {
Constant *PreConst = M->getOrInsertFunction("",
FunctionType::get(Type::getVoidTy(getGlobalContext()),
PreArgTypes, false));
Function *Pre = cast<Function>(PreConst);
Pre->setName("pre");
CG->getOrInsertFunction(Pre);
Constant *PostConst = M->getOrInsertFunction("",
FunctionType::get(Type::getVoidTy(getGlobalContext()),
PostArgTypes, false));
Function *Post = cast<Function>(PostConst);
Post->setName("post");
CG->getOrInsertFunction(Post);
/*
OS << "\n";
OS << "Wrapping '" << Clones.at(i)->getName() << "'...\n\n";
OS << "Pre-invocation function = " << Pre->getName() << "\n";
OS << *Pre;
OS << "Post-invocation function = " << Post->getName() << "\n";
OS << *Post;
OS << "**************************************\n";*/
Function *Wrapper = wrapFunc(Clones.at(i), Pre, Post);
if (i == 0)
Callee = Wrapper;
Function *Caller = Callers.at(i);
for (Function::iterator BBI = Caller->begin(), BBE = Caller->end();
BBI != BBE; ++BBI) {
for (BasicBlock::iterator II = BBI->begin(), IE = BBI->end();
II != IE; ++II) {
CallSite CS(cast<Value>(II));
// If this isn't a call, or it is a call to an intrinsic...
if (!CS || isa<IntrinsicInst>(II))
continue;
if (Callee == CS.getCalledFunction()) {
CS.setCalledFunction(Wrapper);
// Creates an edge from the calling node to its new destination node
CallGraphNode *CallingNode = (*CG)[CS.getCaller()];
CallGraphNode *NewCalleeNode = (*CG)[Wrapper];
CallingNode->replaceCallEdge(CS, CS, NewCalleeNode);
}
}
}
}
OS << "Functions successfully wrapped!\n";
}
/// DoPromotion - This method actually performs the promotion of the specified
/// arguments, and returns the new function. At this point, we know that it's
/// safe to do so.
CallGraphNode *ArgPromotion::DoPromotion(Function *F,
SmallPtrSet<Argument*, 8> &ArgsToPromote,
SmallPtrSet<Argument*, 8> &ByValArgsToTransform) {
// Start by computing a new prototype for the function, which is the same as
// the old function, but has modified arguments.
const FunctionType *FTy = F->getFunctionType();
std::vector<const Type*> Params;
typedef std::set<IndicesVector> ScalarizeTable;
// ScalarizedElements - If we are promoting a pointer that has elements
// accessed out of it, keep track of which elements are accessed so that we
// can add one argument for each.
//
// Arguments that are directly loaded will have a zero element value here, to
// handle cases where there are both a direct load and GEP accesses.
//
std::map<Argument*, ScalarizeTable> ScalarizedElements;
// OriginalLoads - Keep track of a representative load instruction from the
// original function so that we can tell the alias analysis implementation
// what the new GEP/Load instructions we are inserting look like.
std::map<IndicesVector, LoadInst*> OriginalLoads;
// Attributes - Keep track of the parameter attributes for the arguments
// that we are *not* promoting. For the ones that we do promote, the parameter
// attributes are lost
SmallVector<AttributeWithIndex, 8> AttributesVec;
const AttrListPtr &PAL = F->getAttributes();
// Add any return attributes.
if (Attributes attrs = PAL.getRetAttributes())
AttributesVec.push_back(AttributeWithIndex::get(0, attrs));
// First, determine the new argument list
unsigned ArgIndex = 1;
for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(); I != E;
++I, ++ArgIndex) {
if (ByValArgsToTransform.count(I)) {
// Simple byval argument? Just add all the struct element types.
const Type *AgTy = cast<PointerType>(I->getType())->getElementType();
const StructType *STy = cast<StructType>(AgTy);
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i)
Params.push_back(STy->getElementType(i));
++NumByValArgsPromoted;
} else if (!ArgsToPromote.count(I)) {
// Unchanged argument
Params.push_back(I->getType());
if (Attributes attrs = PAL.getParamAttributes(ArgIndex))
AttributesVec.push_back(AttributeWithIndex::get(Params.size(), attrs));
} else if (I->use_empty()) {
// Dead argument (which are always marked as promotable)
++NumArgumentsDead;
} else {
// Okay, this is being promoted. This means that the only uses are loads
// or GEPs which are only used by loads
// In this table, we will track which indices are loaded from the argument
// (where direct loads are tracked as no indices).
ScalarizeTable &ArgIndices = ScalarizedElements[I];
for (Value::use_iterator UI = I->use_begin(), E = I->use_end(); UI != E;
++UI) {
Instruction *User = cast<Instruction>(*UI);
assert(isa<LoadInst>(User) || isa<GetElementPtrInst>(User));
IndicesVector Indices;
Indices.reserve(User->getNumOperands() - 1);
// Since loads will only have a single operand, and GEPs only a single
// non-index operand, this will record direct loads without any indices,
// and gep+loads with the GEP indices.
for (User::op_iterator II = User->op_begin() + 1, IE = User->op_end();
II != IE; ++II)
Indices.push_back(cast<ConstantInt>(*II)->getSExtValue());
// GEPs with a single 0 index can be merged with direct loads
if (Indices.size() == 1 && Indices.front() == 0)
Indices.clear();
ArgIndices.insert(Indices);
LoadInst *OrigLoad;
if (LoadInst *L = dyn_cast<LoadInst>(User))
OrigLoad = L;
else
// Take any load, we will use it only to update Alias Analysis
OrigLoad = cast<LoadInst>(User->use_back());
OriginalLoads[Indices] = OrigLoad;
}
// Add a parameter to the function for each element passed in.
for (ScalarizeTable::iterator SI = ArgIndices.begin(),
E = ArgIndices.end(); SI != E; ++SI) {
// not allowed to dereference ->begin() if size() is 0
Params.push_back(GetElementPtrInst::getIndexedType(I->getType(),
SI->begin(),
SI->end()));
assert(Params.back());
}
if (ArgIndices.size() == 1 && ArgIndices.begin()->empty())
++NumArgumentsPromoted;
else
++NumAggregatesPromoted;
}
}
// Add any function attributes.
if (Attributes attrs = PAL.getFnAttributes())
AttributesVec.push_back(AttributeWithIndex::get(~0, attrs));
const Type *RetTy = FTy->getReturnType();
// Work around LLVM bug PR56: the CWriter cannot emit varargs functions which
// have zero fixed arguments.
bool ExtraArgHack = false;
if (Params.empty() && FTy->isVarArg()) {
ExtraArgHack = true;
Params.push_back(Type::getInt32Ty(F->getContext()));
}
// Construct the new function type using the new arguments.
FunctionType *NFTy = FunctionType::get(RetTy, Params, FTy->isVarArg());
// Create the new function body and insert it into the module.
Function *NF = Function::Create(NFTy, F->getLinkage(), F->getName());
NF->copyAttributesFrom(F);
DEBUG(dbgs() << "ARG PROMOTION: Promoting to:" << *NF << "\n"
<< "From: " << *F);
// Recompute the parameter attributes list based on the new arguments for
// the function.
NF->setAttributes(AttrListPtr::get(AttributesVec.begin(),
AttributesVec.end()));
AttributesVec.clear();
F->getParent()->getFunctionList().insert(F, NF);
NF->takeName(F);
// Get the alias analysis information that we need to update to reflect our
// changes.
AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
// Get the callgraph information that we need to update to reflect our
// changes.
CallGraph &CG = getAnalysis<CallGraph>();
// Get a new callgraph node for NF.
CallGraphNode *NF_CGN = CG.getOrInsertFunction(NF);
// Loop over all of the callers of the function, transforming the call sites
// to pass in the loaded pointers.
//
SmallVector<Value*, 16> Args;
while (!F->use_empty()) {
CallSite CS = CallSite::get(F->use_back());
assert(CS.getCalledFunction() == F);
Instruction *Call = CS.getInstruction();
const AttrListPtr &CallPAL = CS.getAttributes();
// Add any return attributes.
if (Attributes attrs = CallPAL.getRetAttributes())
AttributesVec.push_back(AttributeWithIndex::get(0, attrs));
// Loop over the operands, inserting GEP and loads in the caller as
// appropriate.
CallSite::arg_iterator AI = CS.arg_begin();
ArgIndex = 1;
for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end();
I != E; ++I, ++AI, ++ArgIndex)
if (!ArgsToPromote.count(I) && !ByValArgsToTransform.count(I)) {
Args.push_back(*AI); // Unmodified argument
if (Attributes Attrs = CallPAL.getParamAttributes(ArgIndex))
AttributesVec.push_back(AttributeWithIndex::get(Args.size(), Attrs));
} else if (ByValArgsToTransform.count(I)) {
// Emit a GEP and load for each element of the struct.
const Type *AgTy = cast<PointerType>(I->getType())->getElementType();
const StructType *STy = cast<StructType>(AgTy);
Value *Idxs[2] = {
ConstantInt::get(Type::getInt32Ty(F->getContext()), 0), 0 };
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
Idxs[1] = ConstantInt::get(Type::getInt32Ty(F->getContext()), i);
Value *Idx = GetElementPtrInst::Create(*AI, Idxs, Idxs+2,
(*AI)->getName()+"."+utostr(i),
Call);
// TODO: Tell AA about the new values?
Args.push_back(new LoadInst(Idx, Idx->getName()+".val", Call));
}
} else if (!I->use_empty()) {
// Non-dead argument: insert GEPs and loads as appropriate.
ScalarizeTable &ArgIndices = ScalarizedElements[I];
// Store the Value* version of the indices in here, but declare it now
// for reuse.
std::vector<Value*> Ops;
for (ScalarizeTable::iterator SI = ArgIndices.begin(),
E = ArgIndices.end(); SI != E; ++SI) {
Value *V = *AI;
LoadInst *OrigLoad = OriginalLoads[*SI];
if (!SI->empty()) {
Ops.reserve(SI->size());
const Type *ElTy = V->getType();
for (IndicesVector::const_iterator II = SI->begin(),
IE = SI->end(); II != IE; ++II) {
// Use i32 to index structs, and i64 for others (pointers/arrays).
// This satisfies GEP constraints.
const Type *IdxTy = (ElTy->isStructTy() ?
Type::getInt32Ty(F->getContext()) :
Type::getInt64Ty(F->getContext()));
Ops.push_back(ConstantInt::get(IdxTy, *II));
// Keep track of the type we're currently indexing.
ElTy = cast<CompositeType>(ElTy)->getTypeAtIndex(*II);
}
// And create a GEP to extract those indices.
V = GetElementPtrInst::Create(V, Ops.begin(), Ops.end(),
V->getName()+".idx", Call);
Ops.clear();
AA.copyValue(OrigLoad->getOperand(0), V);
}
// Since we're replacing a load make sure we take the alignment
// of the previous load.
LoadInst *newLoad = new LoadInst(V, V->getName()+".val", Call);
newLoad->setAlignment(OrigLoad->getAlignment());
Args.push_back(newLoad);
AA.copyValue(OrigLoad, Args.back());
}
}
if (ExtraArgHack)
Args.push_back(Constant::getNullValue(Type::getInt32Ty(F->getContext())));
// Push any varargs arguments on the list.
for (; AI != CS.arg_end(); ++AI, ++ArgIndex) {
Args.push_back(*AI);
if (Attributes Attrs = CallPAL.getParamAttributes(ArgIndex))
AttributesVec.push_back(AttributeWithIndex::get(Args.size(), Attrs));
}
// Add any function attributes.
if (Attributes attrs = CallPAL.getFnAttributes())
AttributesVec.push_back(AttributeWithIndex::get(~0, attrs));
Instruction *New;
if (InvokeInst *II = dyn_cast<InvokeInst>(Call)) {
New = InvokeInst::Create(NF, II->getNormalDest(), II->getUnwindDest(),
Args.begin(), Args.end(), "", Call);
cast<InvokeInst>(New)->setCallingConv(CS.getCallingConv());
cast<InvokeInst>(New)->setAttributes(AttrListPtr::get(AttributesVec.begin(),
AttributesVec.end()));
} else {
New = CallInst::Create(NF, Args.begin(), Args.end(), "", Call);
cast<CallInst>(New)->setCallingConv(CS.getCallingConv());
cast<CallInst>(New)->setAttributes(AttrListPtr::get(AttributesVec.begin(),
AttributesVec.end()));
if (cast<CallInst>(Call)->isTailCall())
cast<CallInst>(New)->setTailCall();
}
Args.clear();
AttributesVec.clear();
// Update the alias analysis implementation to know that we are replacing
// the old call with a new one.
AA.replaceWithNewValue(Call, New);
// Update the callgraph to know that the callsite has been transformed.
CallGraphNode *CalleeNode = CG[Call->getParent()->getParent()];
CalleeNode->replaceCallEdge(Call, New, NF_CGN);
if (!Call->use_empty()) {
Call->replaceAllUsesWith(New);
New->takeName(Call);
}
// Finally, remove the old call from the program, reducing the use-count of
// F.
Call->eraseFromParent();
}
// Since we have now created the new function, splice the body of the old
// function right into the new function, leaving the old rotting hulk of the
// function empty.
NF->getBasicBlockList().splice(NF->begin(), F->getBasicBlockList());
// Loop over the argument list, transfering uses of the old arguments over to
// the new arguments, also transfering over the names as well.
//
for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(),
I2 = NF->arg_begin(); I != E; ++I) {
if (!ArgsToPromote.count(I) && !ByValArgsToTransform.count(I)) {
// If this is an unmodified argument, move the name and users over to the
// new version.
I->replaceAllUsesWith(I2);
I2->takeName(I);
AA.replaceWithNewValue(I, I2);
++I2;
continue;
}
if (ByValArgsToTransform.count(I)) {
// In the callee, we create an alloca, and store each of the new incoming
// arguments into the alloca.
Instruction *InsertPt = NF->begin()->begin();
// Just add all the struct element types.
const Type *AgTy = cast<PointerType>(I->getType())->getElementType();
Value *TheAlloca = new AllocaInst(AgTy, 0, "", InsertPt);
const StructType *STy = cast<StructType>(AgTy);
Value *Idxs[2] = {
ConstantInt::get(Type::getInt32Ty(F->getContext()), 0), 0 };
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
Idxs[1] = ConstantInt::get(Type::getInt32Ty(F->getContext()), i);
Value *Idx =
GetElementPtrInst::Create(TheAlloca, Idxs, Idxs+2,
TheAlloca->getName()+"."+Twine(i),
InsertPt);
I2->setName(I->getName()+"."+Twine(i));
new StoreInst(I2++, Idx, InsertPt);
}
// Anything that used the arg should now use the alloca.
I->replaceAllUsesWith(TheAlloca);
TheAlloca->takeName(I);
AA.replaceWithNewValue(I, TheAlloca);
continue;
}
if (I->use_empty()) {
AA.deleteValue(I);
continue;
}
// Otherwise, if we promoted this argument, then all users are load
// instructions (or GEPs with only load users), and all loads should be
// using the new argument that we added.
ScalarizeTable &ArgIndices = ScalarizedElements[I];
while (!I->use_empty()) {
if (LoadInst *LI = dyn_cast<LoadInst>(I->use_back())) {
assert(ArgIndices.begin()->empty() &&
"Load element should sort to front!");
I2->setName(I->getName()+".val");
LI->replaceAllUsesWith(I2);
AA.replaceWithNewValue(LI, I2);
LI->eraseFromParent();
DEBUG(dbgs() << "*** Promoted load of argument '" << I->getName()
<< "' in function '" << F->getName() << "'\n");
} else {
GetElementPtrInst *GEP = cast<GetElementPtrInst>(I->use_back());
IndicesVector Operands;
Operands.reserve(GEP->getNumIndices());
for (User::op_iterator II = GEP->idx_begin(), IE = GEP->idx_end();
II != IE; ++II)
Operands.push_back(cast<ConstantInt>(*II)->getSExtValue());
// GEPs with a single 0 index can be merged with direct loads
if (Operands.size() == 1 && Operands.front() == 0)
Operands.clear();
Function::arg_iterator TheArg = I2;
for (ScalarizeTable::iterator It = ArgIndices.begin();
*It != Operands; ++It, ++TheArg) {
assert(It != ArgIndices.end() && "GEP not handled??");
}
std::string NewName = I->getName();
for (unsigned i = 0, e = Operands.size(); i != e; ++i) {
NewName += "." + utostr(Operands[i]);
}
NewName += ".val";
TheArg->setName(NewName);
DEBUG(dbgs() << "*** Promoted agg argument '" << TheArg->getName()
<< "' of function '" << NF->getName() << "'\n");
// All of the uses must be load instructions. Replace them all with
// the argument specified by ArgNo.
while (!GEP->use_empty()) {
LoadInst *L = cast<LoadInst>(GEP->use_back());
L->replaceAllUsesWith(TheArg);
AA.replaceWithNewValue(L, TheArg);
L->eraseFromParent();
}
AA.deleteValue(GEP);
GEP->eraseFromParent();
}
}
// Increment I2 past all of the arguments added for this promoted pointer.
for (unsigned i = 0, e = ArgIndices.size(); i != e; ++i)
++I2;
}
// Notify the alias analysis implementation that we inserted a new argument.
if (ExtraArgHack)
AA.copyValue(Constant::getNullValue(Type::getInt32Ty(F->getContext())),
NF->arg_begin());
// Tell the alias analysis that the old function is about to disappear.
AA.replaceWithNewValue(F, NF);
NF_CGN->stealCalledFunctionsFrom(CG[F]);
// Now that the old function is dead, delete it. If there is a dangling
// reference to the CallgraphNode, just leave the dead function around for
// someone else to nuke.
CallGraphNode *CGN = CG[F];
if (CGN->getNumReferences() == 0)
delete CG.removeFunctionFromModule(CGN);
else
F->setLinkage(Function::ExternalLinkage);
return NF_CGN;
}
bool Inliner::runOnSCC(CallGraphSCC &SCC) {
CallGraph &CG = getAnalysis<CallGraphWrapperPass>().getCallGraph();
AssumptionCacheTracker *ACT = &getAnalysis<AssumptionCacheTracker>();
auto *TLIP = getAnalysisIfAvailable<TargetLibraryInfoWrapperPass>();
const TargetLibraryInfo *TLI = TLIP ? &TLIP->getTLI() : nullptr;
AliasAnalysis *AA = &getAnalysis<AliasAnalysis>();
SmallPtrSet<Function*, 8> SCCFunctions;
DEBUG(dbgs() << "Inliner visiting SCC:");
for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I) {
Function *F = (*I)->getFunction();
if (F) SCCFunctions.insert(F);
DEBUG(dbgs() << " " << (F ? F->getName() : "INDIRECTNODE"));
}
// Scan through and identify all call sites ahead of time so that we only
// inline call sites in the original functions, not call sites that result
// from inlining other functions.
SmallVector<std::pair<CallSite, int>, 16> CallSites;
// When inlining a callee produces new call sites, we want to keep track of
// the fact that they were inlined from the callee. This allows us to avoid
// infinite inlining in some obscure cases. To represent this, we use an
// index into the InlineHistory vector.
SmallVector<std::pair<Function*, int>, 8> InlineHistory;
for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I) {
Function *F = (*I)->getFunction();
if (!F) continue;
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
CallSite CS(cast<Value>(I));
// If this isn't a call, or it is a call to an intrinsic, it can
// never be inlined.
if (!CS || isa<IntrinsicInst>(I))
continue;
// If this is a direct call to an external function, we can never inline
// it. If it is an indirect call, inlining may resolve it to be a
// direct call, so we keep it.
if (CS.getCalledFunction() && CS.getCalledFunction()->isDeclaration())
continue;
CallSites.push_back(std::make_pair(CS, -1));
}
}
DEBUG(dbgs() << ": " << CallSites.size() << " call sites.\n");
// If there are no calls in this function, exit early.
if (CallSites.empty())
return false;
// Now that we have all of the call sites, move the ones to functions in the
// current SCC to the end of the list.
unsigned FirstCallInSCC = CallSites.size();
for (unsigned i = 0; i < FirstCallInSCC; ++i)
if (Function *F = CallSites[i].first.getCalledFunction())
if (SCCFunctions.count(F))
std::swap(CallSites[i--], CallSites[--FirstCallInSCC]);
InlinedArrayAllocasTy InlinedArrayAllocas;
InlineFunctionInfo InlineInfo(&CG, AA, ACT);
// Now that we have all of the call sites, loop over them and inline them if
// it looks profitable to do so.
bool Changed = false;
bool LocalChange;
do {
LocalChange = false;
// Iterate over the outer loop because inlining functions can cause indirect
// calls to become direct calls.
for (unsigned CSi = 0; CSi != CallSites.size(); ++CSi) {
CallSite CS = CallSites[CSi].first;
Function *Caller = CS.getCaller();
Function *Callee = CS.getCalledFunction();
// If this call site is dead and it is to a readonly function, we should
// just delete the call instead of trying to inline it, regardless of
// size. This happens because IPSCCP propagates the result out of the
// call and then we're left with the dead call.
if (isInstructionTriviallyDead(CS.getInstruction(), TLI)) {
DEBUG(dbgs() << " -> Deleting dead call: "
<< *CS.getInstruction() << "\n");
// Update the call graph by deleting the edge from Callee to Caller.
CG[Caller]->removeCallEdgeFor(CS);
CS.getInstruction()->eraseFromParent();
++NumCallsDeleted;
} else {
// We can only inline direct calls to non-declarations.
if (!Callee || Callee->isDeclaration()) continue;
// If this call site was obtained by inlining another function, verify
// that the include path for the function did not include the callee
// itself. If so, we'd be recursively inlining the same function,
// which would provide the same callsites, which would cause us to
// infinitely inline.
int InlineHistoryID = CallSites[CSi].second;
if (InlineHistoryID != -1 &&
InlineHistoryIncludes(Callee, InlineHistoryID, InlineHistory))
continue;
LLVMContext &CallerCtx = Caller->getContext();
// Get DebugLoc to report. CS will be invalid after Inliner.
DebugLoc DLoc = CS.getInstruction()->getDebugLoc();
// If the policy determines that we should inline this function,
// try to do so.
if (!shouldInline(CS)) {
emitOptimizationRemarkMissed(CallerCtx, DEBUG_TYPE, *Caller, DLoc,
Twine(Callee->getName() +
" will not be inlined into " +
Caller->getName()));
continue;
}
// Attempt to inline the function.
if (!InlineCallIfPossible(CS, InlineInfo, InlinedArrayAllocas,
InlineHistoryID, InsertLifetime)) {
emitOptimizationRemarkMissed(CallerCtx, DEBUG_TYPE, *Caller, DLoc,
Twine(Callee->getName() +
" will not be inlined into " +
Caller->getName()));
continue;
}
++NumInlined;
// Report the inline decision.
emitOptimizationRemark(
CallerCtx, DEBUG_TYPE, *Caller, DLoc,
Twine(Callee->getName() + " inlined into " + Caller->getName()));
// If inlining this function gave us any new call sites, throw them
// onto our worklist to process. They are useful inline candidates.
if (!InlineInfo.InlinedCalls.empty()) {
// Create a new inline history entry for this, so that we remember
// that these new callsites came about due to inlining Callee.
int NewHistoryID = InlineHistory.size();
InlineHistory.push_back(std::make_pair(Callee, InlineHistoryID));
for (unsigned i = 0, e = InlineInfo.InlinedCalls.size();
i != e; ++i) {
Value *Ptr = InlineInfo.InlinedCalls[i];
CallSites.push_back(std::make_pair(CallSite(Ptr), NewHistoryID));
}
}
}
// If we inlined or deleted the last possible call site to the function,
// delete the function body now.
if (Callee && Callee->use_empty() && Callee->hasLocalLinkage() &&
// TODO: Can remove if in SCC now.
!SCCFunctions.count(Callee) &&
// The function may be apparently dead, but if there are indirect
// callgraph references to the node, we cannot delete it yet, this
// could invalidate the CGSCC iterator.
CG[Callee]->getNumReferences() == 0) {
DEBUG(dbgs() << " -> Deleting dead function: "
<< Callee->getName() << "\n");
CallGraphNode *CalleeNode = CG[Callee];
// Remove any call graph edges from the callee to its callees.
CalleeNode->removeAllCalledFunctions();
// Removing the node for callee from the call graph and delete it.
delete CG.removeFunctionFromModule(CalleeNode);
++NumDeleted;
}
// Remove this call site from the list. If possible, use
// swap/pop_back for efficiency, but do not use it if doing so would
// move a call site to a function in this SCC before the
// 'FirstCallInSCC' barrier.
if (SCC.isSingular()) {
CallSites[CSi] = CallSites.back();
CallSites.pop_back();
} else {
CallSites.erase(CallSites.begin()+CSi);
}
--CSi;
Changed = true;
LocalChange = true;
}
} while (LocalChange);
return Changed;
}
bool PathList::runOnModule(Module &M) {
module = &M;
llvm::dbgs() << "[runOnModule]: Moduel M has " << M.getFunctionList().size() << " Functions in all.\n";
// for test
Function *f1 = M.getFunction("fprintf");
if (!f1)
dbgs() << "[Test]: can not find function fprintf.\n";
else
dbgs() << "[Test]: find function fprintf.\n";
CallGraph &CG = getAnalysis<CallGraph>();
// CG.dump();
CallGraphNode *cgNode = CG.getRoot();
cgNode->dump();
// errs()<<node->getFunction()->getName()<<'\n';
Function *startFunc;
Function *endFunc;
startFunc = M.getFunction("__user_main");
//std::string fileName("/home/xqx/data/xqx/projects/benckmarks-klee/texinfo-4.8/build-shit/makeinfo/../../makeinfo/insertion.c");
//int lineNo = 407;
BB = getBB(fileName, lineNo);
*targetBbpp = getBB(fileName, lineNo);
if (BB) {
endFunc = BB->getParent();
if (!endFunc) {
errs()<<"Error: get endFunc failed.\n";
return false;
}
if (!startFunc) {
errs()<<"Error: get startFunc failed.\n";
return false;
}
errs()<<startFunc->getName()<<'\n';
}
else {
errs()<<"Error: get BB failed.\n";
return false;
}
//read start and end from xml files
// defectList enStart, enEnd;
// getEntryList("/tmp/entrys.xml", &enStart, "start");
// getEntryList("/tmp/entrys.xml", &enEnd, "end");
// getEntryList("/tmp/entrys.xml", &dl, "end");
// dumpEntryList(&enStart);
// dumpEntryList(&enEnd);
// dumpEntryList(&dl);
//read bug information from xml file
/* for (defectList::iterator dit = dl.begin(); dit != dl.end(); dit++) {
StringRef file(dit->first.c_str());
std::vector<int> lines = dit->second;
BasicBlock *BB = getBB(file, *(lines.begin()));
if (BB) {
endFunc = BB->getParent();
}
}
*/
//to store temporary path
std::vector<BasicBlock*> p;
// a counter
int map_count = 0;
for (Module::iterator i = M.begin(), e = M.end(); i != e; ++i) {
Function *F = i;
if (!F) {
llvm::errs() << "***NULL Function***\n";
continue;
}
cgNode = CG.getOrInsertFunction(F);
F = cgNode->getFunction();
//
for (CallGraphNode::iterator I = cgNode->begin(), E = cgNode->end();
I != E; ++I){
CallGraphNode::CallRecord *cr = &*I;
// llvm::errs() << "\tCS<" << cr->first << "> calls";
// check if the CallInst is existed
if(cr->first){
Instruction *TmpIns = dyn_cast<Instruction>(cr->first);
if(TmpIns) {
// errs() << "\t" << *TmpIns << "\n";
//unsigned int l, c;
//std::string cfi_path = getInstPath(TmpIns, l, c);
//if (!cfi_path.empty()) {
// if (cfi_path.find("uclibc") != std::string::npos) {
// dbgs() << "[Filter Uclib]: find an instruction from uclibc.\n";
// continue;
// } else if (cfi_path.find("POSIX") != std::string::npos) {
// dbgs() << "[Filter Uclib]: find an instruction from POSIX.\n";
// continue;
// }
//}
} else
continue;
}
// get the funciton pointer which is called by current CallRecord cr
Function *FI = cr->second->getFunction();
if (!FI)
continue;
// create a new CalledFunctions element and push it into calledFunctionMap.
calledFunctionMap[FI].push_back(std::make_pair(F, dyn_cast<Instruction>(cr->first)));
// for debuging
map_count++;
}
}
dbgs() << "[Count Number of calledFunctionMap]: "<< calledFunctionMap.size() <<'\n';
// analyze the global function pointer table
if(function_pointer_analysis()) {
errs() << "[Analyze global function pointer table success]\n";
} else {
errs() << "[Analyze global function pointer table failed]\n";
}
dbgs() << "[Count Number of calledFunctionMap]: "<< calledFunctionMap.size() <<'\n';
// filter the instructions from uclibc
//filter_uclibc();
llvm::errs() << "=================================hh\n";
llvm::errs() << "get Function Path: " << endFunc->getName()
<< " to " << startFunc->getName() << " \n";
// printCalledFuncAndCFGPath(endFunc, startFunc, BB, p);
// modification by wh
evo_paths = new entire_path;
//filter_paths = new func_bbs_type;
//BB_paths_map = new std::map<std::pair<Function*, BasicBlock*>, std::vector<BasicBlock*> >;
std::vector<std::pair< Function*, Instruction*> > tmp_func_path;
// std::vector<BasicBlock*> tmp_bb_path;
// explore_function_paths(endFunc, startFunc, bug_Inst, &tmp_func_path);
collect_funcitons(endFunc, startFunc, bug_Inst, &tmp_func_path);
// dbgs() << "++++++Found " << evo_paths->size() << " function paths.\n";
// for (entire_path::iterator ep_it = evo_paths->begin(); ep_it != evo_paths->end(); ep_it++) {
// for (std::vector<std::pair< Function*, Instruction*> >::iterator pair_it = ep_it->begin(); pair_it != ep_it->end(); pair_it++) {
// if (filter_paths->size() != 0) {
// std::vector<Instruction*>::iterator inst_it = std::find((*filter_paths)[pair_it->first].begin(), (*filter_paths)[pair_it->first].end(), pair_it->second);
// if (inst_it != (*filter_paths)[pair_it->first].end()) {
// continue;
// }
// }
// (*filter_paths)[pair_it->first].push_back(pair_it->second);
// }
// }
dbgs() << "[filter_paths]: contain " << filter_paths->size() << " functions in all.\n";
for (func_bbs_type::iterator fbs_it = filter_paths->begin(); fbs_it != filter_paths->end(); fbs_it++) {
for (std::vector<Instruction*>::iterator bb_it2 = fbs_it->second.begin(); bb_it2 != fbs_it->second.end(); bb_it2++) {
dbgs() << "^^^^^^ " << fbs_it->first->getName() << ": " << (*bb_it2)->getParent()->getName() << '\n';
// to expand functions
call_insts.push_back((*bb_it2));
explore_basicblock_paths(fbs_it->first, (*bb_it2)->getParent(), &(*BB_paths_map)[std::make_pair(fbs_it->first, *bb_it2)]);
dbgs() << "^^^^^^ found " << (*BB_paths_map)[std::make_pair(fbs_it->first, *bb_it2)].size() << " basicblocks.\n";
}
}
llvm::dbgs() << "!!!!!!!! Found " << call_insts.size() << " call instructions.\n";
llvm::dbgs() << "!!!!!!!! Found " << path_basicblocks.size() << " path basicblocks.\n";
// expand functions
for (std::vector<Instruction*>::iterator ci_it = call_insts.begin(); ci_it != call_insts.end(); ci_it++) {
BasicBlock *call_bb = (*ci_it)->getParent();
if (!call_bb) {
continue;
}
for (BasicBlock::iterator inst = call_bb->begin(); inst != call_bb->end(); inst++) {
if (&*inst == *ci_it) {
break;
}
if (isa<CallInst>(&*inst)) {
std::vector<Instruction*>::iterator ci = std::find(path_call_insts.begin(), path_call_insts.end(), &*inst);
if (ci != path_call_insts.end())
continue;
path_call_insts.push_back(&*inst);
}
}
}
llvm::dbgs() << "@@@@@@@@ After search call_insts, found " << path_call_insts.size() << " call instructions.\n";
for (std::vector<BasicBlock*>::iterator p_bb_it = path_basicblocks.begin(); p_bb_it != path_basicblocks.end(); p_bb_it++) {
for (BasicBlock::iterator inst = (*p_bb_it)->begin(); inst != (*p_bb_it)->end(); inst++) {
if (isa<CallInst>(&*inst)) {
std::vector<Instruction*>::iterator ci = std::find(path_call_insts.begin(), path_call_insts.end(), &*inst);
if (ci != path_call_insts.end())
continue;
path_call_insts.push_back(&*inst);
}
}
}
llvm::dbgs() << "@@@@@@@@ After search path_basicblocks, found " << path_call_insts.size() << " call instructions.\n";
for (std::vector<Instruction*>::iterator iit = path_call_insts.begin(); iit != path_call_insts.end(); iit++) {
CallInst *ci = dyn_cast<CallInst>(*iit);
if (!ci)
continue;
Function *ff = ci->getCalledFunction();
if (!ff) {
//ci->dump();
//dbgs() << "\t[called value] " << ci->getOperand(0)->getName() << '\n';
continue;
}
std::vector<Function*>::iterator fit = std::find(otherCalledFuncs->begin(), otherCalledFuncs->end(), ff);
if (fit == otherCalledFuncs->end())
otherCalledFuncs->push_back(ff);
}
llvm::dbgs() << "((((((((( Found " << otherCalledFuncs->size() << " functions.\n";
for (int index = 0; index < otherCalledFuncs->size(); index++) {
Function *f = otherCalledFuncs->at(index);
/* if (!f) {
//f->dump();
llvm::dbgs() << "?????? index = " << index << " size = " << otherCalledFuncs->size()<< '\n';
continue;
}
*/ for (inst_iterator f_it = inst_begin(f); f_it != inst_end(f); f_it++) {
CallInst *ci = dyn_cast<CallInst>(&*f_it);
if (!ci)
continue;
if (!ci->getCalledFunction()) {
//ci->dump();
continue;
}
std::vector<Function*>::iterator fit = std::find(otherCalledFuncs->begin(), otherCalledFuncs->end(), ci->getCalledFunction());
if (fit == otherCalledFuncs->end())
otherCalledFuncs->push_back(ci->getCalledFunction());
}
}
llvm::dbgs() << "((((((((( Found " << otherCalledFuncs->size() << " functions.\n";
//This should be just for statistic.
int tmp_funcNum_in_filter_notIn_other = 0;
for (func_bbs_type::iterator fbs_it = filter_paths->begin(); fbs_it != filter_paths->end(); fbs_it++) {
if (!fbs_it->first) {
llvm::dbgs() << "[Warning]: Found a null Function pointer in filter_paths.\n";
continue;
}
std::vector<Function*>::iterator fit = std::find(otherCalledFuncs->begin(), otherCalledFuncs->end(), fbs_it->first);
if (fit == otherCalledFuncs->end())
//otherCalledFuncs->push_back(fbs_it->first);
tmp_funcNum_in_filter_notIn_other ++;
}
llvm::dbgs() << "<><><><> After searching filter_paths, found " << otherCalledFuncs->size() + tmp_funcNum_in_filter_notIn_other << " functions.\n";
/* for (entire_path::iterator ep_it = evo_paths->begin(); ep_it != evo_paths->end(); ep_it++) {
dbgs() << "Path length is: " << ep_it->size() << '\n';
for (std::vector<std::pair< Function*, BasicBlock*> >::iterator pair_it = ep_it->begin(); pair_it != ep_it->end(); pair_it++) {
dbgs() << "^^^^^^ " << pair_it->first->getName() << ": " << pair_it->second->getName() << '\n';
explore_basicblock_paths(pair_it->first, pair_it->second, &(*BB_paths_map)[*pair_it]);
dbgs() << "^^^^^^ found " << (*BB_paths_map)[*pair_it].size() << " basicblocks.\n";
}
}
*/
llvm::errs() << "on-end\n";
llvm::errs() << "=================================\n";
// output all of the paths
/* errs()<<"Find "<<paths_found->size()<<" paths in all.\n";
for(paths::iterator ips = paths_found->begin();ips != paths_found->end();ips++) {
// std::vector<BasicBlock*> *tmpP = dyn_cast<std::vector<BasicBlock*>*>(&*ips);
dbgs() << "=========A Path Start============\n";
for(std::vector<BasicBlock*>::iterator ps = ips->begin(), pe = ips->end(); ps != pe; ps++) {
BasicBlock *tmpStr = *ps;
errs()<<"\t"<<tmpStr->getParent()->getName()<<": "<<tmpStr->getName()<<" -> \n";
}
errs()<<"=================================\n";
}
*/
return false;
}
CallGraphNode* ArgumentRecovery::recoverArguments(llvm::CallGraphNode *node)
{
Function* fn = node->getFunction();
if (fn == nullptr)
{
// "theoretical nodes", whatever that is
return nullptr;
}
// quick exit if there isn't exactly one argument
if (fn->arg_size() != 1)
{
return nullptr;
}
Argument* fnArg = fn->arg_begin();
if (!isStructType(fnArg))
{
return nullptr;
}
// This is a nasty NASTY hack that relies on the AA pass being RegisterUse.
// The data should be moved to a separate helper pass that can be queried from both the AA pass and this one.
RegisterUse& regUse = getAnalysis<RegisterUse>();
CallGraph& cg = getAnalysis<CallGraphWrapperPass>().getCallGraph();
const auto* modRefInfo = regUse.getModRefInfo(fn);
assert(modRefInfo != nullptr);
// At this point we pretty much know that we're going to modify the function, so start doing that.
// Get register offsets from the old function before we start mutilating it.
auto& registerMap = exposeAllRegisters(fn);
// Create a new function prototype, asking RegisterUse for which registers should be passed in, and how.
LLVMContext& ctx = fn->getContext();
SmallVector<pair<const char*, Type*>, 16> parameters;
Type* int64 = Type::getInt64Ty(ctx);
Type* int64ptr = Type::getInt64PtrTy(ctx);
for (const auto& pair : *modRefInfo)
{
if (pair.second != RegisterUse::NoModRef)
{
Type* paramType = (pair.second & RegisterUse::Mod) == RegisterUse::Mod ? int64ptr : int64;
parameters.push_back({pair.first, paramType});
}
}
// Order parameters.
// FIXME: This could use an ABI-specific sort routine. For now, use a lexicographical sort.
sort(parameters.begin(), parameters.end(), [](const pair<const char*, Type*>& a, const pair<const char*, Type*>& b) {
return strcmp(a.first, b.first) < 0;
});
// Extract parameter types.
SmallVector<Type*, 16> parameterTypes;
for (const auto& pair : parameters)
{
parameterTypes.push_back(pair.second);
}
// Ideally, we would also do caller analysis here to figure out which output registers are never read, such that
// we can either eliminate them from the parameter list or pass them by value instead of by address.
// We would also pick a return value.
FunctionType* newFunctionType = FunctionType::get(Type::getVoidTy(ctx), parameterTypes, false);
Function* newFunc = Function::Create(newFunctionType, fn->getLinkage());
newFunc->copyAttributesFrom(fn);
fn->getParent()->getFunctionList().insert(fn, newFunc);
newFunc->takeName(fn);
fn->setName("__hollow_husk__" + newFunc->getName());
// Set argument names
size_t i = 0;
for (Argument& arg : newFunc->args())
{
arg.setName(parameters[i].first);
i++;
}
// update call graph
CallGraphNode* newFuncNode = cg.getOrInsertFunction(newFunc);
CallGraphNode* oldFuncNode = cg[fn];
// loop over callers and transform call sites.
while (!fn->use_empty())
{
CallSite cs(fn->user_back());
Instruction* call = cast<CallInst>(cs.getInstruction());
Function* caller = call->getParent()->getParent();
auto& registerPositions = exposeAllRegisters(caller);
SmallVector<Value*, 16> callParameters;
for (const auto& pair : parameters)
{
// HACKHACK: find a pointer to a 64-bit int in the set.
Value* registerPointer = nullptr;
auto range = registerPositions.equal_range(pair.first);
for (auto iter = range.first; iter != range.second; iter++)
{
if (auto gep = dyn_cast<GetElementPtrInst>(iter->second))
if (gep->getResultElementType() == int64)
{
registerPointer = gep;
break;
}
}
assert(registerPointer != nullptr);
if (isa<PointerType>(pair.second))
{
callParameters.push_back(registerPointer);
}
else
{
// Create a load instruction. GVN will get rid of it if it's unnecessary.
LoadInst* load = new LoadInst(registerPointer, pair.first, call);
callParameters.push_back(load);
}
}
CallInst* newCall = CallInst::Create(newFunc, callParameters, "", call);
// Update AA
regUse.replaceWithNewValue(call, newCall);
// Update call graph
CallGraphNode* calleeNode = cg[caller];
calleeNode->replaceCallEdge(cs, CallSite(newCall), newFuncNode);
// Finish replacing
if (!call->use_empty())
{
call->replaceAllUsesWith(newCall);
newCall->takeName(call);
}
call->eraseFromParent();
}
// Do not fix functions without a body.
if (!fn->isDeclaration())
{
// Fix up function code. Start by moving everything into the new function.
newFunc->getBasicBlockList().splice(newFunc->begin(), fn->getBasicBlockList());
newFuncNode->stealCalledFunctionsFrom(oldFuncNode);
// Change register uses
size_t argIndex = 0;
auto& argList = newFunc->getArgumentList();
// Create a temporary insertion point. We don't want an existing instruction since chances are that we'll remove it.
Instruction* insertionPoint = BinaryOperator::CreateAdd(ConstantInt::get(int64, 0), ConstantInt::get(int64, 0), "noop", newFunc->begin()->begin());
for (auto iter = argList.begin(); iter != argList.end(); iter++, argIndex++)
{
Value* replaceWith = iter;
const auto& paramTuple = parameters[argIndex];
if (!isa<PointerType>(paramTuple.second))
{
// Create an alloca, copy value from parameter, replace GEP with alloca.
// This is ugly code gen, but it will optimize easily, and still work if
// we need a pointer reference to the register.
auto alloca = new AllocaInst(paramTuple.second, paramTuple.first, insertionPoint);
new StoreInst(iter, alloca, insertionPoint);
replaceWith = alloca;
}
// Replace all uses with new instance.
auto iterPair = registerMap.equal_range(paramTuple.first);
for (auto registerMapIter = iterPair.first; registerMapIter != iterPair.second; registerMapIter++)
{
auto& registerValue = registerMapIter->second;
registerValue->replaceAllUsesWith(replaceWith);
cast<Instruction>(registerValue)->eraseFromParent();
registerValue = replaceWith;
}
}
// At this point, the uses of the argument struct left should be:
// * preserved registers
// * indirect jumps
const auto& target = getAnalysis<TargetInfo>();
while (!fnArg->use_empty())
{
auto lastUser = fnArg->user_back();
if (auto user = dyn_cast<GetElementPtrInst>(lastUser))
{
// Promote register to alloca.
const char* maybeName = target.registerName(*user);
const char* regName = target.largestOverlappingRegister(maybeName);
assert(regName != nullptr);
auto alloca = new AllocaInst(user->getResultElementType(), regName, insertionPoint);
user->replaceAllUsesWith(alloca);
user->eraseFromParent();
}
else
{
auto call = cast<CallInst>(lastUser);
Function* intrin = nullptr;
StringRef intrinName = call->getCalledFunction()->getName();
if (intrinName == "x86_jump_intrin")
{
intrin = indirectJump;
}
else if (intrinName == "x86_call_intrin")
{
intrin = indirectCall;
}
else
{
assert(false);
// Can't decompile this function. Delete its body.
newFunc->deleteBody();
insertionPoint = nullptr;
break;
}
// Replace intrinsic with another intrinsic.
Value* jumpTarget = call->getOperand(2);
SmallVector<Value*, 16> callArgs;
callArgs.push_back(jumpTarget);
for (Argument& arg : argList)
{
callArgs.push_back(&arg);
}
CallInst* varargCall = CallInst::Create(intrin, callArgs, "", call);
newFuncNode->replaceCallEdge(CallSite(call), CallSite(varargCall), cg[intrin]);
regUse.replaceWithNewValue(call, varargCall);
varargCall->takeName(call);
call->eraseFromParent();
}
}
if (insertionPoint != nullptr)
{
// no longer needed
insertionPoint->eraseFromParent();
}
}
// At this point nothing should be using the old register argument anymore. (Pray!)
// Leave the hollow husk of the old function in place to be erased by global DCE.
registerAddresses[newFunc] = move(registerMap);
registerAddresses.erase(fn);
// Should be all.
return newFuncNode;
}
//Inliner handles erasing functions since it also maintains a copy of the callgraph
//that needs to be kept up to date with the functions in the module
void ManualInliner::eraseFunction(Function * F) {
CallGraphNode * n = CG->getOrInsertFunction(F);
n->removeAllCalledFunctions();
CG->removeFunctionFromModule(n);
delete F;
}
bool InlineModule::runOnModule( Module & M ) {
std::vector<std::string> leafNames;
//File *file = fopen("inline_info.txt", "r");
//if (!file) {
// errs() << "Error: Could not open inline_info file.\n";
// retrun true;
//}
std::string line;
std::ifstream file ("inline_info.txt");
if(file.is_open()) {
while(std::getline(file, line))
leafNames.push_back(line);
file.close();
}
else
errs() << "Error: Could not open inline_info file.\n";
//makeLeaf.push_back(M.getFunction("ORACLE_0"));
//makeLeaf.push_back(M.getFunction("ORACLE_1"));
//makeLeaf.push_back(M.getFunction("ORACLE_2"));
//makeLeaf.push_back(M.getFunction("ORACLE_3"));
for (std::vector<std::string>::iterator i = leafNames.begin(), e = leafNames.end();
i!=e; ++i) {
if (debugInlining)
errs() << "inline_info: " << *i << "\n";
makeLeaf.push_back(M.getFunction(*i));
}
// First, get a pointer to previous analysis results
CallGraph & CG = getAnalysis<CallGraph>();
CallGraphNode * entry = CG.getRoot();
if( entry && entry->getFunction() && debugInlining)
errs() << "Entry is function: " << entry->getFunction()->getName() << "\n";
// Iterate over all SCCs in the module in bottom-up order
for( scc_iterator<CallGraph*>
si=scc_begin( &CG ), se=scc_end( &CG ); si != se; ++si ) {
runOnSCC( *si );
}
//reverse the vector for preorder
std::reverse(vectPostOrder.begin(),vectPostOrder.end());
for(std::vector<Function*>::iterator vit = vectPostOrder.begin(), vitE = vectPostOrder.end();
vit!=vitE; ++vit) {
Function *f = *vit;
runOnFunction(*f);
}
// now we have all the call sites which need to be inlined
// inline from the leaves all the way up
const TargetData *TD = getAnalysisIfAvailable<TargetData>();
InlineFunctionInfo InlineInfo(&CG, TD);
std::reverse(inlineCallInsts.begin(),inlineCallInsts.end());
for (std::vector<CallInst*>::iterator i = inlineCallInsts.begin(), e = inlineCallInsts.end();
i!=e; ++i) {
CallInst* CI = *i;
bool success = InlineFunction(CI, InlineInfo, false);
if(!success) {
if (debugInlining)
errs() << "Error: Could not inline callee function " << CI->getCalledFunction()->getName()
<< " into caller function " << "\n";
continue;
}
if (debugInlining)
errs() << "Successfully inlined callee function " << CI->getCalledFunction()->getName()
<< "into caller function " << "\n";
}
return false;
}
bool Strator::runOnModule(Module &m) {
errs() << "Strator started!\n";
/// Get previous analysis results
// multithreadedFunctionMap = &(getAnalysis<MultithreadFinder>().multithreadedFunctionMap);
// if(UseLocalValueInfo)
// localValueInfo = &(getAnalysis<LocalIdentifier>().localValueInfo);
if(UseAliasAnalysis || DebugAliasAnalysis)
aa = &getAnalysis<AliasAnalysis>();
//aa = getAnalysis<LocalIdentifier>().aa;
//useDefMap = &(getAnalysis<UseDefBuilder>().useDefMap);
CallGraph& callGraph = getAnalysis<CallGraph>();
GEPFactory = new GEPValueFactory(&m);
if(PrintPrevPasses){
printLocalModificationInfo();
}
CallGraphNode* externalNode = callGraph.getExternalCallingNode();
if (DirectedRaceDetection){
/// This is the first and default race detection strategy
/// that only tracks a main function (like the
/// unit test case). We call this directed race detection
/// There is no thread pool in this case but a single worker
threadPoolSize = 0;
/// We use this simply to mark all the external entry point functions as initially not analyzed
CallGraphNode::iterator it;
for(it = externalNode->begin(); it != externalNode->end(); ++it){
Function* f = 0;
if((f = it->second->getFunction()))
if(f->size() > 0){
functionMap[f->getName().str()] = false;
++functionCount;
}
}
// cerr << "Total entry point count: " << functionCount << endl;
CallGraphNode* root = callGraph.getRoot();
assert(root->size() && "Size of the call graph root is 0! Something is wrong");
cerr << "Root calls: " << root->size() << " functions"<< endl;
if(!root->getFunction()){
cerr << "The root represents an external node" << endl;
assert(false && "You need to switch to global race detection!");
}
// cerr << root->getFunction()->getName().str() << endl;
/// Initialize all functions to not-multithreaded
workers.push_back(new StratorWorker(this));
for(it = root->begin(); it != root->end(); ++it){
Function* f = 0;
if((f = it->second->getFunction())){
if(f->size() > 0){
workers[0]->multithreadedFunctionMap[it->second->getFunction()->getName().str()] = false;
}
}
}
Strator::StratorWorker::LockSet* lockSet = new Strator::StratorWorker::LockSet();
workers[0]->traverseFunction(*(root->getFunction()), *lockSet);
} else {
/// This is the second variant of race detection. Here all the external
/// functions with definitions are considered as root and race detection
/// is performed over all their childeren. this variant is called global
/// race detection
pthread_t tids[50];
threadPoolSize = 50;
if(externalNode){
/// TODO: We should investigate the following: Some static functions (thus
/// having internal linkage) are considered as external nodes as they are
/// used as parameters to some functions (like pthread_create). We should
/// understand if it is necessary to add such functions as external nodes.
CallGraphNode::iterator it;
for(it = externalNode->begin(); it != externalNode->end(); ++it){
Function* f = 0;
if((f = it->second->getFunction()))
if(f->size() > 0){
cerr << "adding function \"" << it->second->getFunction()->getName().str() << "\" to task list" << endl;
tasks.push_back(f);
++functionCount;
}
}
/// Create the thread pool
threadPoolSize = (threadPoolSize < functionCount) ? threadPoolSize : functionCount;
/// Create as many workers as the pool size
// threadPoolSize = 1;
cerr << "Thread pool size:" << threadPoolSize << endl;
for(unsigned i=0; i<threadPoolSize; ++i)
workers.push_back(new StratorWorker(this));
/// trigger the
for(unsigned i=0; i<threadPoolSize; ++i){
int retVal = pthread_create(&tids[i], 0, stratorWorkerHelper, workers[i]);
assert(retVal == 0 && "Problem with creating the threads");
}
/// Synchronize the threads
for(unsigned i=0; i<threadPoolSize; ++i){
int retVal = pthread_join(tids[i], 0);
assert(retVal == 0 && "Problem with joining the threads");
}
}
}
if(ReportUnprocessed){
/// list all the unprocessed entry point functions
cerr << "Following entry point functions were not processed: " << endl;
FunctionMap::iterator fMIt;
for(fMIt = functionMap.begin(); fMIt != functionMap.end(); ++fMIt){
if(!fMIt->second)
cerr << fMIt->first << endl;
}
}
/// We have gathered race detection data, now report
reportRaces();
/// We did not modify the module, return false
return false;
}
/// Remove dead functions that are not included in DNR (Do Not Remove) list.
bool Inliner::removeDeadFunctions(CallGraph &CG, bool AlwaysInlineOnly) {
SmallVector<CallGraphNode*, 16> FunctionsToRemove;
SmallVector<CallGraphNode *, 16> DeadFunctionsInComdats;
SmallDenseMap<const Comdat *, int, 16> ComdatEntriesAlive;
auto RemoveCGN = [&](CallGraphNode *CGN) {
// Remove any call graph edges from the function to its callees.
CGN->removeAllCalledFunctions();
// Remove any edges from the external node to the function's call graph
// node. These edges might have been made irrelegant due to
// optimization of the program.
CG.getExternalCallingNode()->removeAnyCallEdgeTo(CGN);
// Removing the node for callee from the call graph and delete it.
FunctionsToRemove.push_back(CGN);
};
// Scan for all of the functions, looking for ones that should now be removed
// from the program. Insert the dead ones in the FunctionsToRemove set.
for (CallGraph::iterator I = CG.begin(), E = CG.end(); I != E; ++I) {
CallGraphNode *CGN = I->second;
Function *F = CGN->getFunction();
if (!F || F->isDeclaration())
continue;
// Handle the case when this function is called and we only want to care
// about always-inline functions. This is a bit of a hack to share code
// between here and the InlineAlways pass.
if (AlwaysInlineOnly && !F->hasFnAttribute(Attribute::AlwaysInline))
continue;
// If the only remaining users of the function are dead constants, remove
// them.
F->removeDeadConstantUsers();
if (!F->isDefTriviallyDead())
continue;
// It is unsafe to drop a function with discardable linkage from a COMDAT
// without also dropping the other members of the COMDAT.
// The inliner doesn't visit non-function entities which are in COMDAT
// groups so it is unsafe to do so *unless* the linkage is local.
if (!F->hasLocalLinkage()) {
if (const Comdat *C = F->getComdat()) {
--ComdatEntriesAlive[C];
DeadFunctionsInComdats.push_back(CGN);
continue;
}
}
RemoveCGN(CGN);
}
if (!DeadFunctionsInComdats.empty()) {
// Count up all the entities in COMDAT groups
auto ComdatGroupReferenced = [&](const Comdat *C) {
auto I = ComdatEntriesAlive.find(C);
if (I != ComdatEntriesAlive.end())
++(I->getSecond());
};
for (const Function &F : CG.getModule())
if (const Comdat *C = F.getComdat())
ComdatGroupReferenced(C);
for (const GlobalVariable &GV : CG.getModule().globals())
if (const Comdat *C = GV.getComdat())
ComdatGroupReferenced(C);
for (const GlobalAlias &GA : CG.getModule().aliases())
if (const Comdat *C = GA.getComdat())
ComdatGroupReferenced(C);
for (CallGraphNode *CGN : DeadFunctionsInComdats) {
Function *F = CGN->getFunction();
const Comdat *C = F->getComdat();
int NumAlive = ComdatEntriesAlive[C];
// We can remove functions in a COMDAT group if the entire group is dead.
assert(NumAlive >= 0);
if (NumAlive > 0)
continue;
RemoveCGN(CGN);
}
}
if (FunctionsToRemove.empty())
return false;
// Now that we know which functions to delete, do so. We didn't want to do
// this inline, because that would invalidate our CallGraph::iterator
// objects. :(
//
// Note that it doesn't matter that we are iterating over a non-stable order
// here to do this, it doesn't matter which order the functions are deleted
// in.
array_pod_sort(FunctionsToRemove.begin(), FunctionsToRemove.end());
FunctionsToRemove.erase(std::unique(FunctionsToRemove.begin(),
FunctionsToRemove.end()),
FunctionsToRemove.end());
for (SmallVectorImpl<CallGraphNode *>::iterator I = FunctionsToRemove.begin(),
E = FunctionsToRemove.end();
I != E; ++I) {
delete CG.removeFunctionFromModule(*I);
++NumDeleted;
}
return true;
}
static bool
inlineCallsImpl(CallGraphSCC &SCC, CallGraph &CG,
std::function<AssumptionCache &(Function &)> GetAssumptionCache,
ProfileSummaryInfo *PSI, TargetLibraryInfo &TLI,
bool InsertLifetime,
function_ref<InlineCost(CallSite CS)> GetInlineCost,
function_ref<AAResults &(Function &)> AARGetter,
ImportedFunctionsInliningStatistics &ImportedFunctionsStats) {
SmallPtrSet<Function *, 8> SCCFunctions;
LLVM_DEBUG(dbgs() << "Inliner visiting SCC:");
for (CallGraphNode *Node : SCC) {
Function *F = Node->getFunction();
if (F)
SCCFunctions.insert(F);
LLVM_DEBUG(dbgs() << " " << (F ? F->getName() : "INDIRECTNODE"));
}
// Scan through and identify all call sites ahead of time so that we only
// inline call sites in the original functions, not call sites that result
// from inlining other functions.
SmallVector<std::pair<CallSite, int>, 16> CallSites;
// When inlining a callee produces new call sites, we want to keep track of
// the fact that they were inlined from the callee. This allows us to avoid
// infinite inlining in some obscure cases. To represent this, we use an
// index into the InlineHistory vector.
SmallVector<std::pair<Function *, int>, 8> InlineHistory;
for (CallGraphNode *Node : SCC) {
Function *F = Node->getFunction();
if (!F || F->isDeclaration())
continue;
OptimizationRemarkEmitter ORE(F);
for (BasicBlock &BB : *F)
for (Instruction &I : BB) {
CallSite CS(cast<Value>(&I));
// If this isn't a call, or it is a call to an intrinsic, it can
// never be inlined.
if (!CS || isa<IntrinsicInst>(I))
continue;
// If this is a direct call to an external function, we can never inline
// it. If it is an indirect call, inlining may resolve it to be a
// direct call, so we keep it.
if (Function *Callee = CS.getCalledFunction())
if (Callee->isDeclaration()) {
using namespace ore;
ORE.emit([&]() {
return OptimizationRemarkMissed(DEBUG_TYPE, "NoDefinition", &I)
<< NV("Callee", Callee) << " will not be inlined into "
<< NV("Caller", CS.getCaller())
<< " because its definition is unavailable"
<< setIsVerbose();
});
continue;
}
CallSites.push_back(std::make_pair(CS, -1));
}
}
LLVM_DEBUG(dbgs() << ": " << CallSites.size() << " call sites.\n");
// If there are no calls in this function, exit early.
if (CallSites.empty())
return false;
// Now that we have all of the call sites, move the ones to functions in the
// current SCC to the end of the list.
unsigned FirstCallInSCC = CallSites.size();
for (unsigned i = 0; i < FirstCallInSCC; ++i)
if (Function *F = CallSites[i].first.getCalledFunction())
if (SCCFunctions.count(F))
std::swap(CallSites[i--], CallSites[--FirstCallInSCC]);
InlinedArrayAllocasTy InlinedArrayAllocas;
InlineFunctionInfo InlineInfo(&CG, &GetAssumptionCache, PSI);
// Now that we have all of the call sites, loop over them and inline them if
// it looks profitable to do so.
bool Changed = false;
bool LocalChange;
do {
LocalChange = false;
// Iterate over the outer loop because inlining functions can cause indirect
// calls to become direct calls.
// CallSites may be modified inside so ranged for loop can not be used.
for (unsigned CSi = 0; CSi != CallSites.size(); ++CSi) {
CallSite CS = CallSites[CSi].first;
Function *Caller = CS.getCaller();
Function *Callee = CS.getCalledFunction();
// We can only inline direct calls to non-declarations.
if (!Callee || Callee->isDeclaration())
continue;
Instruction *Instr = CS.getInstruction();
bool IsTriviallyDead = isInstructionTriviallyDead(Instr, &TLI);
int InlineHistoryID;
if (!IsTriviallyDead) {
// If this call site was obtained by inlining another function, verify
// that the include path for the function did not include the callee
// itself. If so, we'd be recursively inlining the same function,
// which would provide the same callsites, which would cause us to
// infinitely inline.
InlineHistoryID = CallSites[CSi].second;
if (InlineHistoryID != -1 &&
InlineHistoryIncludes(Callee, InlineHistoryID, InlineHistory))
continue;
}
// FIXME for new PM: because of the old PM we currently generate ORE and
// in turn BFI on demand. With the new PM, the ORE dependency should
// just become a regular analysis dependency.
OptimizationRemarkEmitter ORE(Caller);
Optional<InlineCost> OIC = shouldInline(CS, GetInlineCost, ORE);
// If the policy determines that we should inline this function,
// delete the call instead.
if (!OIC)
continue;
// If this call site is dead and it is to a readonly function, we should
// just delete the call instead of trying to inline it, regardless of
// size. This happens because IPSCCP propagates the result out of the
// call and then we're left with the dead call.
if (IsTriviallyDead) {
LLVM_DEBUG(dbgs() << " -> Deleting dead call: " << *Instr << "\n");
// Update the call graph by deleting the edge from Callee to Caller.
CG[Caller]->removeCallEdgeFor(CS);
Instr->eraseFromParent();
++NumCallsDeleted;
} else {
// Get DebugLoc to report. CS will be invalid after Inliner.
DebugLoc DLoc = CS->getDebugLoc();
BasicBlock *Block = CS.getParent();
// Attempt to inline the function.
using namespace ore;
if (!InlineCallIfPossible(CS, InlineInfo, InlinedArrayAllocas,
InlineHistoryID, InsertLifetime, AARGetter,
ImportedFunctionsStats)) {
ORE.emit([&]() {
return OptimizationRemarkMissed(DEBUG_TYPE, "NotInlined", DLoc,
Block)
<< NV("Callee", Callee) << " will not be inlined into "
<< NV("Caller", Caller);
});
continue;
}
++NumInlined;
ORE.emit([&]() {
bool AlwaysInline = OIC->isAlways();
StringRef RemarkName = AlwaysInline ? "AlwaysInline" : "Inlined";
OptimizationRemark R(DEBUG_TYPE, RemarkName, DLoc, Block);
R << NV("Callee", Callee) << " inlined into ";
R << NV("Caller", Caller);
if (AlwaysInline)
R << " with cost=always";
else {
R << " with cost=" << NV("Cost", OIC->getCost());
R << " (threshold=" << NV("Threshold", OIC->getThreshold());
R << ")";
}
return R;
});
// If inlining this function gave us any new call sites, throw them
// onto our worklist to process. They are useful inline candidates.
if (!InlineInfo.InlinedCalls.empty()) {
// Create a new inline history entry for this, so that we remember
// that these new callsites came about due to inlining Callee.
int NewHistoryID = InlineHistory.size();
InlineHistory.push_back(std::make_pair(Callee, InlineHistoryID));
for (Value *Ptr : InlineInfo.InlinedCalls)
CallSites.push_back(std::make_pair(CallSite(Ptr), NewHistoryID));
}
}
// If we inlined or deleted the last possible call site to the function,
// delete the function body now.
if (Callee && Callee->use_empty() && Callee->hasLocalLinkage() &&
// TODO: Can remove if in SCC now.
!SCCFunctions.count(Callee) &&
// The function may be apparently dead, but if there are indirect
// callgraph references to the node, we cannot delete it yet, this
// could invalidate the CGSCC iterator.
CG[Callee]->getNumReferences() == 0) {
LLVM_DEBUG(dbgs() << " -> Deleting dead function: "
<< Callee->getName() << "\n");
CallGraphNode *CalleeNode = CG[Callee];
// Remove any call graph edges from the callee to its callees.
CalleeNode->removeAllCalledFunctions();
// Removing the node for callee from the call graph and delete it.
delete CG.removeFunctionFromModule(CalleeNode);
++NumDeleted;
}
// Remove this call site from the list. If possible, use
// swap/pop_back for efficiency, but do not use it if doing so would
// move a call site to a function in this SCC before the
// 'FirstCallInSCC' barrier.
if (SCC.isSingular()) {
CallSites[CSi] = CallSites.back();
CallSites.pop_back();
} else {
CallSites.erase(CallSites.begin() + CSi);
}
--CSi;
Changed = true;
LocalChange = true;
}
} while (LocalChange);
return Changed;
}
bool InternalizePass::runOnModule(Module &M) {
CallGraph *CG = getAnalysisIfAvailable<CallGraph>();
CallGraphNode *ExternalNode = CG ? CG->getExternalCallingNode() : 0;
if (ExternalNames.empty()) {
// Return if we're not in 'all but main' mode and have no external api
if (!AllButMain)
return false;
// If no list or file of symbols was specified, check to see if there is a
// "main" symbol defined in the module. If so, use it, otherwise do not
// internalize the module, it must be a library or something.
//
Function *MainFunc = M.getFunction("main");
if (MainFunc == 0 || MainFunc->isDeclaration())
return false; // No main found, must be a library...
// Preserve main, internalize all else.
ExternalNames.insert(MainFunc->getName());
}
bool Changed = false;
// Never internalize functions which code-gen might insert.
ExternalNames.insert("__stack_chk_fail");
// Mark all functions not in the api as internal.
// FIXME: maybe use private linkage?
for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I)
if (!I->isDeclaration() && // Function must be defined here
// Available externally is really just a "declaration with a body".
!I->hasAvailableExternallyLinkage() &&
!I->hasLocalLinkage() && // Can't already have internal linkage
!ExternalNames.count(I->getName())) {// Not marked to keep external?
I->setLinkage(GlobalValue::InternalLinkage);
// Remove a callgraph edge from the external node to this function.
if (ExternalNode) ExternalNode->removeOneAbstractEdgeTo((*CG)[I]);
Changed = true;
++NumFunctions;
DEBUG(dbgs() << "Internalizing func " << I->getName() << "\n");
}
// Never internalize the llvm.used symbol. It is used to implement
// attribute((used)).
// FIXME: Shouldn't this just filter on llvm.metadata section??
ExternalNames.insert("llvm.used");
ExternalNames.insert("llvm.compiler.used");
// Never internalize anchors used by the machine module info, else the info
// won't find them. (see MachineModuleInfo.)
ExternalNames.insert("llvm.global_ctors");
ExternalNames.insert("llvm.global_dtors");
ExternalNames.insert("llvm.global.annotations");
// Never internalize symbols code-gen inserts.
ExternalNames.insert("__stack_chk_guard");
// Mark all global variables with initializers that are not in the api as
// internal as well.
// FIXME: maybe use private linkage?
for (Module::global_iterator I = M.global_begin(), E = M.global_end();
I != E; ++I)
if (!I->isDeclaration() && !I->hasLocalLinkage() &&
// Available externally is really just a "declaration with a body".
!I->hasAvailableExternallyLinkage() &&
!ExternalNames.count(I->getName())) {
I->setLinkage(GlobalValue::InternalLinkage);
Changed = true;
++NumGlobals;
DEBUG(dbgs() << "Internalized gvar " << I->getName() << "\n");
}
// Mark all aliases that are not in the api as internal as well.
for (Module::alias_iterator I = M.alias_begin(), E = M.alias_end();
I != E; ++I)
if (!I->isDeclaration() && !I->hasInternalLinkage() &&
// Available externally is really just a "declaration with a body".
!I->hasAvailableExternallyLinkage() &&
!ExternalNames.count(I->getName())) {
I->setLinkage(GlobalValue::InternalLinkage);
Changed = true;
++NumAliases;
DEBUG(dbgs() << "Internalized alias " << I->getName() << "\n");
}
return Changed;
}
/// Remove dead functions that are not included in DNR (Do Not Remove) list.
bool LegacyInlinerBase::removeDeadFunctions(CallGraph &CG,
bool AlwaysInlineOnly) {
SmallVector<CallGraphNode *, 16> FunctionsToRemove;
SmallVector<Function *, 16> DeadFunctionsInComdats;
auto RemoveCGN = [&](CallGraphNode *CGN) {
// Remove any call graph edges from the function to its callees.
CGN->removeAllCalledFunctions();
// Remove any edges from the external node to the function's call graph
// node. These edges might have been made irrelegant due to
// optimization of the program.
CG.getExternalCallingNode()->removeAnyCallEdgeTo(CGN);
// Removing the node for callee from the call graph and delete it.
FunctionsToRemove.push_back(CGN);
};
// Scan for all of the functions, looking for ones that should now be removed
// from the program. Insert the dead ones in the FunctionsToRemove set.
for (const auto &I : CG) {
CallGraphNode *CGN = I.second.get();
Function *F = CGN->getFunction();
if (!F || F->isDeclaration())
continue;
// Handle the case when this function is called and we only want to care
// about always-inline functions. This is a bit of a hack to share code
// between here and the InlineAlways pass.
if (AlwaysInlineOnly && !F->hasFnAttribute(Attribute::AlwaysInline))
continue;
// If the only remaining users of the function are dead constants, remove
// them.
F->removeDeadConstantUsers();
if (!F->isDefTriviallyDead())
continue;
// It is unsafe to drop a function with discardable linkage from a COMDAT
// without also dropping the other members of the COMDAT.
// The inliner doesn't visit non-function entities which are in COMDAT
// groups so it is unsafe to do so *unless* the linkage is local.
if (!F->hasLocalLinkage()) {
if (F->hasComdat()) {
DeadFunctionsInComdats.push_back(F);
continue;
}
}
RemoveCGN(CGN);
}
if (!DeadFunctionsInComdats.empty()) {
// Filter out the functions whose comdats remain alive.
filterDeadComdatFunctions(CG.getModule(), DeadFunctionsInComdats);
// Remove the rest.
for (Function *F : DeadFunctionsInComdats)
RemoveCGN(CG[F]);
}
if (FunctionsToRemove.empty())
return false;
// Now that we know which functions to delete, do so. We didn't want to do
// this inline, because that would invalidate our CallGraph::iterator
// objects. :(
//
// Note that it doesn't matter that we are iterating over a non-stable order
// here to do this, it doesn't matter which order the functions are deleted
// in.
array_pod_sort(FunctionsToRemove.begin(), FunctionsToRemove.end());
FunctionsToRemove.erase(
std::unique(FunctionsToRemove.begin(), FunctionsToRemove.end()),
FunctionsToRemove.end());
for (CallGraphNode *CGN : FunctionsToRemove) {
delete CG.removeFunctionFromModule(CGN);
++NumDeleted;
}
return true;
}
