/// Evaluate a call to function F, returning true if successful, false if we
/// can't evaluate it. ActualArgs contains the formal arguments for the
/// function.
bool Evaluator::EvaluateFunction(Function *F, Constant *&RetVal,
const SmallVectorImpl<Constant*> &ActualArgs) {
// Check to see if this function is already executing (recursion). If so,
// bail out. TODO: we might want to accept limited recursion.
if (is_contained(CallStack, F))
return false;
CallStack.push_back(F);
// Initialize arguments to the incoming values specified.
unsigned ArgNo = 0;
for (Function::arg_iterator AI = F->arg_begin(), E = F->arg_end(); AI != E;
++AI, ++ArgNo)
setVal(&*AI, ActualArgs[ArgNo]);
// ExecutedBlocks - We only handle non-looping, non-recursive code. As such,
// we can only evaluate any one basic block at most once. This set keeps
// track of what we have executed so we can detect recursive cases etc.
SmallPtrSet<BasicBlock*, 32> ExecutedBlocks;
// CurBB - The current basic block we're evaluating.
BasicBlock *CurBB = &F->front();
BasicBlock::iterator CurInst = CurBB->begin();
while (1) {
BasicBlock *NextBB = nullptr; // Initialized to avoid compiler warnings.
DEBUG(dbgs() << "Trying to evaluate BB: " << *CurBB << "\n");
if (!EvaluateBlock(CurInst, NextBB))
return false;
if (!NextBB) {
// Successfully running until there's no next block means that we found
// the return. Fill it the return value and pop the call stack.
ReturnInst *RI = cast<ReturnInst>(CurBB->getTerminator());
if (RI->getNumOperands())
RetVal = getVal(RI->getOperand(0));
CallStack.pop_back();
return true;
}
// Okay, we succeeded in evaluating this control flow. See if we have
// executed the new block before. If so, we have a looping function,
// which we cannot evaluate in reasonable time.
if (!ExecutedBlocks.insert(NextBB).second)
return false; // looped!
// Okay, we have never been in this block before. Check to see if there
// are any PHI nodes. If so, evaluate them with information about where
// we came from.
PHINode *PN = nullptr;
for (CurInst = NextBB->begin();
(PN = dyn_cast<PHINode>(CurInst)); ++CurInst)
setVal(PN, getVal(PN->getIncomingValueForBlock(CurBB)));
// Advance to the next block.
CurBB = NextBB;
}
}
void InstructionReplace::insertStores(llvm::Module& M)
{
for(llvm::Module::iterator F = M.begin(), ME = M.end(); F != ME; ++F) {
//if(!F->getFnAttributes().hasAttribute(Attributes::AttrVal::MaskedCopy)) { continue; }
llvm::Function* Fun = maskedfn[F];
vector<Value*> outputshares;
ReturnInst* tbd = NULL;
auto arg = Fun->arg_begin();
for(unsigned int i = 0; i <= MaskingOrder; i++) { ++arg; }
for(unsigned int i = 0; i <= MaskingOrder; i++) { outputshares.push_back(arg++); }
for(llvm::Function::iterator BB = Fun->begin(),
FE = Fun->end();
BB != FE;
++BB) {
for( llvm::BasicBlock::iterator i = BB->begin(); i != BB->end(); i++) {
if(tbd != NULL) {tbd->eraseFromParent(); tbd = NULL;}
if(!isa<ReturnInst>(i)) {continue;}
ReturnInst* ri = cast<ReturnInst>(i);
IRBuilder<> ib = llvm::IRBuilder<>(BB->getContext());
ib.SetInsertPoint(i);
vector<Value*> shares = MaskValue(ri->getReturnValue(), ri);
for(unsigned int i = 0; i <= MaskingOrder; i++) { ib.CreateStore(shares[i], outputshares[i]); }
ib.CreateRetVoid();
tbd = ri;
}
}
if(tbd != NULL) {tbd->eraseFromParent(); tbd = NULL;}
}
}
CallInst *BasicBlock::getTerminatingMustTailCall() {
if (InstList.empty())
return nullptr;
ReturnInst *RI = dyn_cast<ReturnInst>(&InstList.back());
if (!RI || RI == &InstList.front())
return nullptr;
Instruction *Prev = RI->getPrevNode();
if (!Prev)
return nullptr;
if (Value *RV = RI->getReturnValue()) {
if (RV != Prev)
return nullptr;
// Look through the optional bitcast.
if (auto *BI = dyn_cast<BitCastInst>(Prev)) {
RV = BI->getOperand(0);
Prev = BI->getPrevNode();
if (!Prev || RV != Prev)
return nullptr;
}
}
if (auto *CI = dyn_cast<CallInst>(Prev)) {
if (CI->isMustTailCall())
return CI;
}
return nullptr;
}
void GCInvariantVerifier::visitReturnInst(ReturnInst &RI) {
if (!RI.getReturnValue())
return;
Type *RTy = RI.getReturnValue()->getType();
if (!RTy->isPointerTy())
return;
unsigned AS = cast<PointerType>(RTy)->getAddressSpace();
Check(!isSpecialAS(AS) || AS == AddressSpace::Tracked,
"Only gc tracked values may be directly returned", &RI);
}
void Lint::visitReturnInst(ReturnInst &I) {
Function *F = I.getParent()->getParent();
Assert(!F->doesNotReturn(),
"Unusual: Return statement in function with noreturn attribute", &I);
if (Value *V = I.getReturnValue()) {
Value *Obj = findValue(V, /*OffsetOk=*/true);
Assert(!isa<AllocaInst>(Obj), "Unusual: Returning alloca value", &I);
}
}
void Interpreter::visitReturnInst(ReturnInst &I) {
ExecutionContext &SF = ECStack.back();
const Type *RetTy = Type::VoidTy;
GenericValue Result;
// Save away the return value... (if we are not 'ret void')
if (I.getNumOperands()) {
RetTy = I.getReturnValue()->getType();
Result = getOperandValue(I.getReturnValue(), SF);
}
popStackAndReturnValueToCaller(RetTy, Result);
}
virtual bool isRegofInstFITarget(Value *reg, Instruction *inst) {
if (isa<CallInst>(inst)) {
CallInst* CI = dyn_cast<CallInst>(inst);
Function* called_func = CI->getCalledFunction();
if (called_func == NULL) {
return false;
}
return reg == CI; // selects dst register
} else if (isa<ReturnInst>(inst)) {
ReturnInst* RI = dyn_cast<ReturnInst>(inst);
return reg == RI->getReturnValue();
} else {
return false;
}
}
bool IRTranslator::translateReturn(const ReturnInst &RI) {
const Value *Ret = RI.getReturnValue();
// The target may mess up with the insertion point, but
// this is not important as a return is the last instruction
// of the block anyway.
return CLI->lowerReturn(MIRBuilder, Ret, !Ret ? 0 : getOrCreateVReg(*Ret));
}
/// Check to see if the function containing the specified tail call consistently
/// returns the same runtime-constant value at all exit points except for
/// IgnoreRI. If so, return the returned value.
static Value *getCommonReturnValue(ReturnInst *IgnoreRI, CallInst *CI) {
Function *F = CI->getParent()->getParent();
Value *ReturnedValue = nullptr;
for (BasicBlock &BBI : *F) {
ReturnInst *RI = dyn_cast<ReturnInst>(BBI.getTerminator());
if (RI == nullptr || RI == IgnoreRI) continue;
// We can only perform this transformation if the value returned is
// evaluatable at the start of the initial invocation of the function,
// instead of at the end of the evaluation.
//
Value *RetOp = RI->getOperand(0);
if (!isDynamicConstant(RetOp, CI, RI))
return nullptr;
if (ReturnedValue && RetOp != ReturnedValue)
return nullptr; // Cannot transform if differing values are returned.
ReturnedValue = RetOp;
}
return ReturnedValue;
}
// getCommonReturnValue - Check to see if the function containing the specified
// tail call consistently returns the same runtime-constant value at all exit
// points except for IgnoreRI. If so, return the returned value.
//
static Value *getCommonReturnValue(ReturnInst *IgnoreRI, CallInst *CI) {
Function *F = CI->getParent()->getParent();
Value *ReturnedValue = 0;
for (Function::iterator BBI = F->begin(), E = F->end(); BBI != E; ++BBI) {
ReturnInst *RI = dyn_cast<ReturnInst>(BBI->getTerminator());
if (RI == 0 || RI == IgnoreRI) continue;
// We can only perform this transformation if the value returned is
// evaluatable at the start of the initial invocation of the function,
// instead of at the end of the evaluation.
//
Value *RetOp = RI->getOperand(0);
if (!isDynamicConstant(RetOp, CI, RI))
return 0;
if (ReturnedValue && RetOp != ReturnedValue)
return 0; // Cannot transform if differing values are returned.
ReturnedValue = RetOp;
}
return ReturnedValue;
}
void LowerEmAsyncify::transformAsyncFunction(Function &F, Instructions const& AsyncCalls) {
assert(!AsyncCalls.empty());
// Pass 0
// collect all the return instructions from the original function
// will use later
std::vector<ReturnInst*> OrigReturns;
for (inst_iterator I = inst_begin(&F), E = inst_end(&F); I != E; ++I) {
if (ReturnInst *RI = dyn_cast<ReturnInst>(&*I)) {
OrigReturns.push_back(RI);
}
}
// Pass 1
// Scan each async call and make the basic structure:
// All these will be cloned into the callback functions
// - allocate the async context before calling an async function
// - check async right after calling an async function, save context & return if async, continue if not
// - retrieve the async return value and free the async context if the called function turns out to be sync
std::vector<AsyncCallEntry> AsyncCallEntries;
AsyncCallEntries.reserve(AsyncCalls.size());
for (Instructions::const_iterator I = AsyncCalls.begin(), E = AsyncCalls.end(); I != E; ++I) {
// prepare blocks
Instruction *CurAsyncCall = *I;
// The block containing the async call
BasicBlock *CurBlock = CurAsyncCall->getParent();
// The block should run after the async call
BasicBlock *AfterCallBlock = SplitBlock(CurBlock, CurAsyncCall->getNextNode());
// The block where we store the context and return
BasicBlock *SaveAsyncCtxBlock = BasicBlock::Create(TheModule->getContext(), "SaveAsyncCtx", &F, AfterCallBlock);
// return a dummy value at the end, to make the block valid
new UnreachableInst(TheModule->getContext(), SaveAsyncCtxBlock);
// allocate the context before making the call
// we don't know the size yet, will fix it later
// we cannot insert the instruction later because,
// we need to make sure that all the instructions and blocks are fixed before we can generate DT and find context variables
// In CallHandler.h `sp` will be put as the second parameter
// such that we can take a note of the original sp
CallInst *AllocAsyncCtxInst = CallInst::Create(AllocAsyncCtxFunction, Constant::getNullValue(I32), "AsyncCtx", CurAsyncCall);
// Right after the call
// check async and return if so
// TODO: we can define truly async functions and partial async functions
{
// remove old terminator, which came from SplitBlock
CurBlock->getTerminator()->eraseFromParent();
// go to SaveAsyncCtxBlock if the previous call is async
// otherwise just continue to AfterCallBlock
CallInst *CheckAsync = CallInst::Create(CheckAsyncFunction, "IsAsync", CurBlock);
BranchInst::Create(SaveAsyncCtxBlock, AfterCallBlock, CheckAsync, CurBlock);
}
// take a note of this async call
AsyncCallEntry CurAsyncCallEntry;
CurAsyncCallEntry.AsyncCallInst = CurAsyncCall;
CurAsyncCallEntry.AfterCallBlock = AfterCallBlock;
CurAsyncCallEntry.AllocAsyncCtxInst = AllocAsyncCtxInst;
CurAsyncCallEntry.SaveAsyncCtxBlock = SaveAsyncCtxBlock;
// create an empty function for the callback, which will be constructed later
CurAsyncCallEntry.CallbackFunc = Function::Create(CallbackFunctionType, F.getLinkage(), F.getName() + "__async_cb", TheModule);
AsyncCallEntries.push_back(CurAsyncCallEntry);
}
// Pass 2
// analyze the context variables and construct SaveAsyncCtxBlock for each async call
// also calculate the size of the context and allocate the async context accordingly
for (std::vector<AsyncCallEntry>::iterator EI = AsyncCallEntries.begin(), EE = AsyncCallEntries.end(); EI != EE; ++EI) {
AsyncCallEntry & CurEntry = *EI;
// Collect everything to be saved
FindContextVariables(CurEntry);
// Pack the variables as a struct
{
// TODO: sort them from large memeber to small ones, in order to make the struct compact even when aligned
SmallVector<Type*, 8> Types;
Types.push_back(CallbackFunctionType->getPointerTo());
for (Values::iterator VI = CurEntry.ContextVariables.begin(), VE = CurEntry.ContextVariables.end(); VI != VE; ++VI) {
Types.push_back((*VI)->getType());
}
CurEntry.ContextStructType = StructType::get(TheModule->getContext(), Types);
}
// fix the size of allocation
CurEntry.AllocAsyncCtxInst->setOperand(0,
ConstantInt::get(I32, DL->getTypeStoreSize(CurEntry.ContextStructType)));
// construct SaveAsyncCtxBlock
{
// fill in SaveAsyncCtxBlock
// temporarily remove the terminator for convenience
CurEntry.SaveAsyncCtxBlock->getTerminator()->eraseFromParent();
assert(CurEntry.SaveAsyncCtxBlock->empty());
Type *AsyncCtxAddrTy = CurEntry.ContextStructType->getPointerTo();
BitCastInst *AsyncCtxAddr = new BitCastInst(CurEntry.AllocAsyncCtxInst, AsyncCtxAddrTy, "AsyncCtxAddr", CurEntry.SaveAsyncCtxBlock);
SmallVector<Value*, 2> Indices;
// store the callback
{
Indices.push_back(ConstantInt::get(I32, 0));
Indices.push_back(ConstantInt::get(I32, 0));
GetElementPtrInst *AsyncVarAddr = GetElementPtrInst::Create(AsyncCtxAddrTy, AsyncCtxAddr, Indices, "", CurEntry.SaveAsyncCtxBlock);
new StoreInst(CurEntry.CallbackFunc, AsyncVarAddr, CurEntry.SaveAsyncCtxBlock);
}
// store the context variables
for (size_t i = 0; i < CurEntry.ContextVariables.size(); ++i) {
Indices.clear();
Indices.push_back(ConstantInt::get(I32, 0));
Indices.push_back(ConstantInt::get(I32, i + 1)); // the 0th element is the callback function
GetElementPtrInst *AsyncVarAddr = GetElementPtrInst::Create(AsyncCtxAddrTy, AsyncCtxAddr, Indices, "", CurEntry.SaveAsyncCtxBlock);
new StoreInst(CurEntry.ContextVariables[i], AsyncVarAddr, CurEntry.SaveAsyncCtxBlock);
}
// to exit the block, we want to return without unwinding the stack frame
CallInst::Create(DoNotUnwindFunction, "", CurEntry.SaveAsyncCtxBlock);
ReturnInst::Create(TheModule->getContext(),
(F.getReturnType()->isVoidTy() ? 0 : Constant::getNullValue(F.getReturnType())),
CurEntry.SaveAsyncCtxBlock);
}
}
// Pass 3
// now all the SaveAsyncCtxBlock's have been constructed
// we can clone F and construct callback functions
// we could not construct the callbacks in Pass 2 because we need _all_ those SaveAsyncCtxBlock's appear in _each_ callback
for (std::vector<AsyncCallEntry>::iterator EI = AsyncCallEntries.begin(), EE = AsyncCallEntries.end(); EI != EE; ++EI) {
AsyncCallEntry & CurEntry = *EI;
Function *CurCallbackFunc = CurEntry.CallbackFunc;
ValueToValueMapTy VMap;
// Add the entry block
// load variables from the context
// also update VMap for CloneFunction
BasicBlock *EntryBlock = BasicBlock::Create(TheModule->getContext(), "AsyncCallbackEntry", CurCallbackFunc);
std::vector<LoadInst *> LoadedAsyncVars;
{
Type *AsyncCtxAddrTy = CurEntry.ContextStructType->getPointerTo();
BitCastInst *AsyncCtxAddr = new BitCastInst(CurCallbackFunc->arg_begin(), AsyncCtxAddrTy, "AsyncCtx", EntryBlock);
SmallVector<Value*, 2> Indices;
for (size_t i = 0; i < CurEntry.ContextVariables.size(); ++i) {
Indices.clear();
Indices.push_back(ConstantInt::get(I32, 0));
Indices.push_back(ConstantInt::get(I32, i + 1)); // the 0th element of AsyncCtx is the callback function
GetElementPtrInst *AsyncVarAddr = GetElementPtrInst::Create(AsyncCtxAddrTy, AsyncCtxAddr, Indices, "", EntryBlock);
LoadedAsyncVars.push_back(new LoadInst(AsyncVarAddr, "", EntryBlock));
// we want the argument to be replaced by the loaded value
if (isa<Argument>(CurEntry.ContextVariables[i]))
VMap[CurEntry.ContextVariables[i]] = LoadedAsyncVars.back();
}
}
// we don't need any argument, just leave dummy entries there to cheat CloneFunctionInto
for (Function::const_arg_iterator AI = F.arg_begin(), AE = F.arg_end(); AI != AE; ++AI) {
if (VMap.count(AI) == 0)
VMap[AI] = Constant::getNullValue(AI->getType());
}
// Clone the function
{
SmallVector<ReturnInst*, 8> Returns;
CloneFunctionInto(CurCallbackFunc, &F, VMap, false, Returns);
// return type of the callback functions is always void
// need to fix the return type
if (!F.getReturnType()->isVoidTy()) {
// for those return instructions that are from the original function
// it means we are 'truly' leaving this function
// need to store the return value right before ruturn
for (size_t i = 0; i < OrigReturns.size(); ++i) {
ReturnInst *RI = cast<ReturnInst>(VMap[OrigReturns[i]]);
// Need to store the return value into the global area
CallInst *RawRetValAddr = CallInst::Create(GetAsyncReturnValueAddrFunction, "", RI);
BitCastInst *RetValAddr = new BitCastInst(RawRetValAddr, F.getReturnType()->getPointerTo(), "AsyncRetValAddr", RI);
new StoreInst(RI->getOperand(0), RetValAddr, RI);
}
// we want to unwind the stack back to where it was before the original function as called
// but we don't actually need to do this here
// at this point it must be true that no callback is pended
// so the scheduler will correct the stack pointer and pop the frame
// here we just fix the return type
for (size_t i = 0; i < Returns.size(); ++i) {
ReplaceInstWithInst(Returns[i], ReturnInst::Create(TheModule->getContext()));
}
}
}
// the callback function does not have any return value
// so clear all the attributes for return
{
AttributeSet Attrs = CurCallbackFunc->getAttributes();
CurCallbackFunc->setAttributes(
Attrs.removeAttributes(TheModule->getContext(), AttributeSet::ReturnIndex, Attrs.getRetAttributes())
);
}
// in the callback function, we never allocate a new async frame
// instead we reuse the existing one
for (std::vector<AsyncCallEntry>::iterator EI = AsyncCallEntries.begin(), EE = AsyncCallEntries.end(); EI != EE; ++EI) {
Instruction *I = cast<Instruction>(VMap[EI->AllocAsyncCtxInst]);
ReplaceInstWithInst(I, CallInst::Create(ReallocAsyncCtxFunction, I->getOperand(0), "ReallocAsyncCtx"));
}
// mapped entry point & async call
BasicBlock *ResumeBlock = cast<BasicBlock>(VMap[CurEntry.AfterCallBlock]);
Instruction *MappedAsyncCall = cast<Instruction>(VMap[CurEntry.AsyncCallInst]);
// To save space, for each async call in the callback function, we just ignore the sync case, and leave it to the scheduler
// TODO need an option for this
{
for (std::vector<AsyncCallEntry>::iterator EI = AsyncCallEntries.begin(), EE = AsyncCallEntries.end(); EI != EE; ++EI) {
AsyncCallEntry & CurEntry = *EI;
Instruction *MappedAsyncCallInst = cast<Instruction>(VMap[CurEntry.AsyncCallInst]);
BasicBlock *MappedAsyncCallBlock = MappedAsyncCallInst->getParent();
BasicBlock *MappedAfterCallBlock = cast<BasicBlock>(VMap[CurEntry.AfterCallBlock]);
// for the sync case of the call, go to NewBlock (instead of MappedAfterCallBlock)
BasicBlock *NewBlock = BasicBlock::Create(TheModule->getContext(), "", CurCallbackFunc, MappedAfterCallBlock);
MappedAsyncCallBlock->getTerminator()->setSuccessor(1, NewBlock);
// store the return value
if (!MappedAsyncCallInst->use_empty()) {
CallInst *RawRetValAddr = CallInst::Create(GetAsyncReturnValueAddrFunction, "", NewBlock);
BitCastInst *RetValAddr = new BitCastInst(RawRetValAddr, MappedAsyncCallInst->getType()->getPointerTo(), "AsyncRetValAddr", NewBlock);
new StoreInst(MappedAsyncCallInst, RetValAddr, NewBlock);
}
// tell the scheduler that we want to keep the current async stack frame
CallInst::Create(DoNotUnwindAsyncFunction, "", NewBlock);
// finally we go to the SaveAsyncCtxBlock, to register the callbac, save the local variables and leave
BasicBlock *MappedSaveAsyncCtxBlock = cast<BasicBlock>(VMap[CurEntry.SaveAsyncCtxBlock]);
BranchInst::Create(MappedSaveAsyncCtxBlock, NewBlock);
}
}
std::vector<AllocaInst*> ToPromote;
// applying loaded variables in the entry block
{
BasicBlockSet ReachableBlocks = FindReachableBlocksFrom(ResumeBlock);
for (size_t i = 0; i < CurEntry.ContextVariables.size(); ++i) {
Value *OrigVar = CurEntry.ContextVariables[i];
if (isa<Argument>(OrigVar)) continue; // already processed
Value *CurVar = VMap[OrigVar];
assert(CurVar != MappedAsyncCall);
if (Instruction *Inst = dyn_cast<Instruction>(CurVar)) {
if (ReachableBlocks.count(Inst->getParent())) {
// Inst could be either defined or loaded from the async context
// Do the dirty works in memory
// TODO: might need to check the safety first
// TODO: can we create phi directly?
AllocaInst *Addr = DemoteRegToStack(*Inst, false);
new StoreInst(LoadedAsyncVars[i], Addr, EntryBlock);
ToPromote.push_back(Addr);
} else {
// The parent block is not reachable, which means there is no confliction
// it's safe to replace Inst with the loaded value
assert(Inst != LoadedAsyncVars[i]); // this should only happen when OrigVar is an Argument
Inst->replaceAllUsesWith(LoadedAsyncVars[i]);
}
}
}
}
// resolve the return value of the previous async function
// it could be the value just loaded from the global area
// or directly returned by the function (in its sync case)
if (!CurEntry.AsyncCallInst->use_empty()) {
// load the async return value
CallInst *RawRetValAddr = CallInst::Create(GetAsyncReturnValueAddrFunction, "", EntryBlock);
BitCastInst *RetValAddr = new BitCastInst(RawRetValAddr, MappedAsyncCall->getType()->getPointerTo(), "AsyncRetValAddr", EntryBlock);
LoadInst *RetVal = new LoadInst(RetValAddr, "AsyncRetVal", EntryBlock);
AllocaInst *Addr = DemoteRegToStack(*MappedAsyncCall, false);
new StoreInst(RetVal, Addr, EntryBlock);
ToPromote.push_back(Addr);
}
// TODO remove unreachable blocks before creating phi
// We go right to ResumeBlock from the EntryBlock
BranchInst::Create(ResumeBlock, EntryBlock);
/*
* Creating phi's
* Normal stack frames and async stack frames are interleaving with each other.
* In a callback function, if we call an async function, we might need to realloc the async ctx.
* at this point we don't want anything stored after the ctx,
* such that we can free and extend the ctx by simply update STACKTOP.
* Therefore we don't want any alloca's in callback functions.
*
*/
if (!ToPromote.empty()) {
DominatorTreeWrapperPass DTW;
DTW.runOnFunction(*CurCallbackFunc);
PromoteMemToReg(ToPromote, DTW.getDomTree());
}
removeUnreachableBlocks(*CurCallbackFunc);
}
// Pass 4
// Here are modifications to the original function, which we won't want to be cloned into the callback functions
for (std::vector<AsyncCallEntry>::iterator EI = AsyncCallEntries.begin(), EE = AsyncCallEntries.end(); EI != EE; ++EI) {
AsyncCallEntry & CurEntry = *EI;
// remove the frame if no async functinon has been called
CallInst::Create(FreeAsyncCtxFunction, CurEntry.AllocAsyncCtxInst, "", CurEntry.AfterCallBlock->getFirstNonPHI());
}
}
void moduleDepGraph::matchParametersAndReturnValues(Function &F) {
// Only do the matching if F has any use
if (F.isVarArg() || !F.hasNUsesOrMore(1)) {
return;
}
// Data structure which contains the matches between formal and real parameters
// First: formal parameter
// Second: real parameter
SmallVector<std::pair<GraphNode*, GraphNode*>, 4> Parameters(F.arg_size());
// Fetch the function arguments (formal parameters) into the data structure
Function::arg_iterator argptr;
Function::arg_iterator e;
unsigned i;
//Create the PHI nodes for the formal parameters
for (i = 0, argptr = F.arg_begin(), e = F.arg_end(); argptr != e; ++i, ++argptr) {
OpNode* argPHI = new OpNode(Instruction::PHI);
GraphNode* argNode = NULL;
argNode = depGraph->addInst(argptr);
if (argNode != NULL)
depGraph->addEdge(argPHI, argNode);
Parameters[i].first = argPHI;
}
// Check if the function returns a supported value type. If not, no return value matching is done
bool noReturn = F.getReturnType()->isVoidTy();
// Creates the data structure which receives the return values of the function, if there is any
SmallPtrSet<llvm::Value*, 8> ReturnValues;
if (!noReturn) {
// Iterate over the basic blocks to fetch all possible return values
for (Function::iterator bb = F.begin(), bbend = F.end(); bb != bbend; ++bb) {
// Get the terminator instruction of the basic block and check if it's
// a return instruction: if it's not, continue to next basic block
Instruction *terminator = bb->getTerminator();
ReturnInst *RI = dyn_cast<ReturnInst> (terminator);
if (!RI)
continue;
// Get the return value and insert in the data structure
ReturnValues.insert(RI->getReturnValue());
}
}
for (Value::use_iterator UI = F.use_begin(), E = F.use_end(); UI != E; ++UI) {
User *U = *UI;
// Ignore blockaddress uses
if (isa<BlockAddress> (U))
continue;
// Used by a non-instruction, or not the callee of a function, do not
// match.
if (!isa<CallInst> (U) && !isa<InvokeInst> (U))
continue;
Instruction *caller = cast<Instruction> (U);
CallSite CS(caller);
if (!CS.isCallee(UI))
continue;
// Iterate over the real parameters and put them in the data structure
CallSite::arg_iterator AI;
CallSite::arg_iterator EI;
for (i = 0, AI = CS.arg_begin(), EI = CS.arg_end(); AI != EI; ++i, ++AI) {
Parameters[i].second = depGraph->addInst(*AI);
}
// Match formal and real parameters
for (i = 0; i < Parameters.size(); ++i) {
depGraph->addEdge(Parameters[i].second, Parameters[i].first);
}
// Match return values
if (!noReturn) {
OpNode* retPHI = new OpNode(Instruction::PHI);
GraphNode* callerNode = depGraph->addInst(caller);
depGraph->addEdge(retPHI, callerNode);
for (SmallPtrSetIterator<llvm::Value*> ri = ReturnValues.begin(),
re = ReturnValues.end(); ri != re; ++ri) {
GraphNode* retNode = depGraph->addInst(*ri);
depGraph->addEdge(retNode, retPHI);
}
}
// Real parameters are cleaned before moving to the next use (for safety's sake)
for (i = 0; i < Parameters.size(); ++i)
Parameters[i].second = NULL;
}
depGraph->deleteCallNodes(&F);
}
/// InsertStackProtectors - Insert code into the prologue and epilogue of the
/// function.
///
/// - The prologue code loads and stores the stack guard onto the stack.
/// - The epilogue checks the value stored in the prologue against the original
/// value. It calls __stack_chk_fail if they differ.
bool StackProtector::InsertStackProtectors() {
bool SupportsSelectionDAGSP =
EnableSelectionDAGSP && !TM->Options.EnableFastISel;
AllocaInst *AI = nullptr; // Place on stack that stores the stack guard.
for (Function::iterator I = F->begin(), E = F->end(); I != E;) {
BasicBlock *BB = &*I++;
ReturnInst *RI = dyn_cast<ReturnInst>(BB->getTerminator());
if (!RI)
continue;
// Generate prologue instrumentation if not already generated.
if (!HasPrologue) {
HasPrologue = true;
SupportsSelectionDAGSP &= CreatePrologue(F, M, RI, TLI, AI);
}
// SelectionDAG based code generation. Nothing else needs to be done here.
// The epilogue instrumentation is postponed to SelectionDAG.
if (SupportsSelectionDAGSP)
break;
// Set HasIRCheck to true, so that SelectionDAG will not generate its own
// version. SelectionDAG called 'shouldEmitSDCheck' to check whether
// instrumentation has already been generated.
HasIRCheck = true;
// Generate epilogue instrumentation. The epilogue intrumentation can be
// function-based or inlined depending on which mechanism the target is
// providing.
if (Value* GuardCheck = TLI->getSSPStackGuardCheck(*M)) {
// Generate the function-based epilogue instrumentation.
// The target provides a guard check function, generate a call to it.
IRBuilder<> B(RI);
LoadInst *Guard = B.CreateLoad(AI, true, "Guard");
CallInst *Call = B.CreateCall(GuardCheck, {Guard});
llvm::Function *Function = cast<llvm::Function>(GuardCheck);
Call->setAttributes(Function->getAttributes());
Call->setCallingConv(Function->getCallingConv());
} else {
// Generate the epilogue with inline instrumentation.
// If we do not support SelectionDAG based tail calls, generate IR level
// tail calls.
//
// For each block with a return instruction, convert this:
//
// return:
// ...
// ret ...
//
// into this:
//
// return:
// ...
// %1 = <stack guard>
// %2 = load StackGuardSlot
// %3 = cmp i1 %1, %2
// br i1 %3, label %SP_return, label %CallStackCheckFailBlk
//
// SP_return:
// ret ...
//
// CallStackCheckFailBlk:
// call void @__stack_chk_fail()
// unreachable
// Create the FailBB. We duplicate the BB every time since the MI tail
// merge pass will merge together all of the various BB into one including
// fail BB generated by the stack protector pseudo instruction.
BasicBlock *FailBB = CreateFailBB();
// Split the basic block before the return instruction.
BasicBlock *NewBB = BB->splitBasicBlock(RI->getIterator(), "SP_return");
// Update the dominator tree if we need to.
if (DT && DT->isReachableFromEntry(BB)) {
DT->addNewBlock(NewBB, BB);
DT->addNewBlock(FailBB, BB);
}
// Remove default branch instruction to the new BB.
BB->getTerminator()->eraseFromParent();
// Move the newly created basic block to the point right after the old
// basic block so that it's in the "fall through" position.
NewBB->moveAfter(BB);
// Generate the stack protector instructions in the old basic block.
IRBuilder<> B(BB);
Value *Guard = getStackGuard(TLI, M, B);
LoadInst *LI2 = B.CreateLoad(AI, true);
Value *Cmp = B.CreateICmpEQ(Guard, LI2);
auto SuccessProb =
BranchProbabilityInfo::getBranchProbStackProtector(true);
auto FailureProb =
BranchProbabilityInfo::getBranchProbStackProtector(false);
MDNode *Weights = MDBuilder(F->getContext())
.createBranchWeights(SuccessProb.getNumerator(),
FailureProb.getNumerator());
B.CreateCondBr(Cmp, NewBB, FailBB, Weights);
}
}
// Return if we didn't modify any basic blocks. i.e., there are no return
// statements in the function.
return HasPrologue;
}
// 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;
}
/// mergeEmptyReturnBlocks - If we have more than one empty (other than phi
/// node) return blocks, merge them together to promote recursive block merging.
static bool mergeEmptyReturnBlocks(Function &F) {
bool Changed = false;
BasicBlock *RetBlock = 0;
// Scan all the blocks in the function, looking for empty return blocks.
for (Function::iterator BBI = F.begin(), E = F.end(); BBI != E; ) {
BasicBlock &BB = *BBI++;
// Only look at return blocks.
ReturnInst *Ret = dyn_cast<ReturnInst>(BB.getTerminator());
if (Ret == 0) continue;
// Only look at the block if it is empty or the only other thing in it is a
// single PHI node that is the operand to the return.
if (Ret != &BB.front()) {
// Check for something else in the block.
BasicBlock::iterator I = Ret;
--I;
// Skip over debug info.
while (isa<DbgInfoIntrinsic>(I) && I != BB.begin())
--I;
if (!isa<DbgInfoIntrinsic>(I) &&
(!isa<PHINode>(I) || I != BB.begin() ||
Ret->getNumOperands() == 0 ||
Ret->getOperand(0) != I))
continue;
}
// If this is the first returning block, remember it and keep going.
if (RetBlock == 0) {
RetBlock = &BB;
continue;
}
// Otherwise, we found a duplicate return block. Merge the two.
Changed = true;
// Case when there is no input to the return or when the returned values
// agree is trivial. Note that they can't agree if there are phis in the
// blocks.
if (Ret->getNumOperands() == 0 ||
Ret->getOperand(0) ==
cast<ReturnInst>(RetBlock->getTerminator())->getOperand(0)) {
BB.replaceAllUsesWith(RetBlock);
BB.eraseFromParent();
continue;
}
// If the canonical return block has no PHI node, create one now.
PHINode *RetBlockPHI = dyn_cast<PHINode>(RetBlock->begin());
if (RetBlockPHI == 0) {
Value *InVal = cast<ReturnInst>(RetBlock->getTerminator())->getOperand(0);
pred_iterator PB = pred_begin(RetBlock), PE = pred_end(RetBlock);
RetBlockPHI = PHINode::Create(Ret->getOperand(0)->getType(),
std::distance(PB, PE), "merge",
&RetBlock->front());
for (pred_iterator PI = PB; PI != PE; ++PI)
RetBlockPHI->addIncoming(InVal, *PI);
RetBlock->getTerminator()->setOperand(0, RetBlockPHI);
}
// Turn BB into a block that just unconditionally branches to the return
// block. This handles the case when the two return blocks have a common
// predecessor but that return different things.
RetBlockPHI->addIncoming(Ret->getOperand(0), &BB);
BB.getTerminator()->eraseFromParent();
BranchInst::Create(RetBlock, &BB);
}
return Changed;
}
/// InsertStackProtectors - Insert code into the prologue and epilogue of the
/// function.
///
/// - The prologue code loads and stores the stack guard onto the stack.
/// - The epilogue checks the value stored in the prologue against the original
/// value. It calls __stack_chk_fail if they differ.
bool StackProtector::InsertStackProtectors() {
BasicBlock *FailBB = 0; // The basic block to jump to if check fails.
BasicBlock *FailBBDom = 0; // FailBB's dominator.
AllocaInst *AI = 0; // Place on stack that stores the stack guard.
Value *StackGuardVar = 0; // The stack guard variable.
for (Function::iterator I = F->begin(), E = F->end(); I != E; ) {
BasicBlock *BB = I++;
ReturnInst *RI = dyn_cast<ReturnInst>(BB->getTerminator());
if (!RI) continue;
if (!FailBB) {
// Insert code into the entry block that stores the __stack_chk_guard
// variable onto the stack:
//
// entry:
// StackGuardSlot = alloca i8*
// StackGuard = load __stack_chk_guard
// call void @llvm.stackprotect.create(StackGuard, StackGuardSlot)
//
const PointerType *PtrTy = Type::getInt8PtrTy(RI->getContext());
unsigned AddressSpace, Offset;
if (TLI->getStackCookieLocation(AddressSpace, Offset)) {
Constant *OffsetVal =
ConstantInt::get(Type::getInt32Ty(RI->getContext()), Offset);
StackGuardVar = ConstantExpr::getIntToPtr(OffsetVal,
PointerType::get(PtrTy, AddressSpace));
} else {
StackGuardVar = M->getOrInsertGlobal("__stack_chk_guard", PtrTy);
}
BasicBlock &Entry = F->getEntryBlock();
Instruction *InsPt = &Entry.front();
AI = new AllocaInst(PtrTy, "StackGuardSlot", InsPt);
LoadInst *LI = new LoadInst(StackGuardVar, "StackGuard", false, InsPt);
Value *Args[] = { LI, AI };
CallInst::
Create(Intrinsic::getDeclaration(M, Intrinsic::stackprotector),
&Args[0], array_endof(Args), "", InsPt);
// Create the basic block to jump to when the guard check fails.
FailBB = CreateFailBB();
if (DT)
FailBBDom = DT->isReachableFromEntry(BB) ? BB : 0;
}
// For each block with a return instruction, convert this:
//
// return:
// ...
// ret ...
//
// into this:
//
// return:
// ...
// %1 = load __stack_chk_guard
// %2 = load StackGuardSlot
// %3 = cmp i1 %1, %2
// br i1 %3, label %SP_return, label %CallStackCheckFailBlk
//
// SP_return:
// ret ...
//
// CallStackCheckFailBlk:
// call void @__stack_chk_fail()
// unreachable
// Split the basic block before the return instruction.
BasicBlock *NewBB = BB->splitBasicBlock(RI, "SP_return");
if (DT) {
DT->addNewBlock(NewBB, DT->isReachableFromEntry(BB) ? BB : 0);
FailBBDom = DT->findNearestCommonDominator(FailBBDom, BB);
}
// Remove default branch instruction to the new BB.
BB->getTerminator()->eraseFromParent();
// Move the newly created basic block to the point right after the old basic
// block so that it's in the "fall through" position.
NewBB->moveAfter(BB);
// Generate the stack protector instructions in the old basic block.
LoadInst *LI1 = new LoadInst(StackGuardVar, "", false, BB);
LoadInst *LI2 = new LoadInst(AI, "", true, BB);
ICmpInst *Cmp = new ICmpInst(*BB, CmpInst::ICMP_EQ, LI1, LI2, "");
BranchInst::Create(NewBB, FailBB, Cmp, BB);
}
// Return if we didn't modify any basic blocks. I.e., there are no return
// statements in the function.
if (!FailBB) return false;
if (DT)
DT->addNewBlock(FailBB, FailBBDom);
return true;
}
void GraphBuilder::visitReturnInst(ReturnInst &RI) {
if (RI.getNumOperands() && isa<PointerType>(RI.getOperand(0)->getType()))
G.getOrCreateReturnNodeFor(*FB).mergeWith(getValueDest(RI.getOperand(0)));
}
/// 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. Similarly this will inline a recursive
/// function by one level.
bool llvm::InlineFunction(CallSite CS, InlineFunctionInfo &IFI,
bool InsertLifetime) {
Instruction *TheCall = CS.getInstruction();
assert(TheCall->getParent() && TheCall->getParent()->getParent() &&
"Instruction not in function!");
// If IFI has any state in it, zap it before we fill it in.
IFI.reset();
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 the personality function from the callee if it contains a landing pad.
Value *CalleePersonality = 0;
for (Function::const_iterator I = CalledFunc->begin(), E = CalledFunc->end();
I != E; ++I)
if (const InvokeInst *II = dyn_cast<InvokeInst>(I->getTerminator())) {
const BasicBlock *BB = II->getUnwindDest();
const LandingPadInst *LP = BB->getLandingPadInst();
CalleePersonality = LP->getPersonalityFn();
break;
}
// Find the personality function used by the landing pads of the caller. If it
// exists, then check to see that it matches the personality function used in
// the callee.
if (CalleePersonality) {
for (Function::const_iterator I = Caller->begin(), E = Caller->end();
I != E; ++I)
if (const InvokeInst *II = dyn_cast<InvokeInst>(I->getTerminator())) {
const BasicBlock *BB = II->getUnwindDest();
const LandingPadInst *LP = BB->getLandingPadInst();
// If the personality functions match, then we can perform the
// inlining. Otherwise, we can't inline.
// TODO: This isn't 100% true. Some personality functions are proper
// supersets of others and can be used in place of the other.
if (LP->getPersonalityFn() != CalleePersonality)
return false;
break;
}
}
// 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.
SmallVector<ReturnInst*, 8> Returns;
ClonedCodeInfo InlinedFunctionInfo;
Function::iterator FirstNewBlock;
{ // Scope to destroy VMap after cloning.
ValueToValueMapTy VMap;
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;
const Argument *Arg = I;
// 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 (CS.isByValArgument(ArgNo)) {
ActualArg = HandleByValArgument(ActualArg, Arg, TheCall, CalledFunc, IFI,
CalledFunc->getParamAlignment(ArgNo+1));
// Calls that we inline may use the new alloca, so we need to clear
// their 'tail' flags if HandleByValArgument introduced a new alloca and
// the callee has calls.
MustClearTailCallFlags |= ActualArg != *AI;
}
VMap[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, VMap,
/*ModuleLevelChanges=*/false, Returns, ".i",
&InlinedFunctionInfo, IFI.TD, TheCall);
// Remember the first block that is newly cloned over.
FirstNewBlock = LastBlock; ++FirstNewBlock;
// Update the callgraph if requested.
if (IFI.CG)
UpdateCallGraphAfterInlining(CS, FirstNewBlock, VMap, IFI);
// Update inlined instructions' line number information.
fixupLineNumbers(Caller, FirstNewBlock, TheCall);
}
// 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; ) {
AllocaInst *AI = dyn_cast<AllocaInst>(I++);
if (AI == 0) continue;
// 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()))
continue;
// Keep track of the static allocas that we inline into the caller.
IFI.StaticAllocas.push_back(AI);
// 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())) {
IFI.StaticAllocas.push_back(cast<AllocaInst>(I));
++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);
}
}
// Leave lifetime markers for the static alloca's, scoping them to the
// function we just inlined.
if (InsertLifetime && !IFI.StaticAllocas.empty()) {
IRBuilder<> builder(FirstNewBlock->begin());
for (unsigned ai = 0, ae = IFI.StaticAllocas.size(); ai != ae; ++ai) {
AllocaInst *AI = IFI.StaticAllocas[ai];
// If the alloca is already scoped to something smaller than the whole
// function then there's no need to add redundant, less accurate markers.
if (hasLifetimeMarkers(AI))
continue;
// Try to determine the size of the allocation.
ConstantInt *AllocaSize = 0;
if (ConstantInt *AIArraySize =
dyn_cast<ConstantInt>(AI->getArraySize())) {
if (IFI.TD) {
Type *AllocaType = AI->getAllocatedType();
uint64_t AllocaTypeSize = IFI.TD->getTypeAllocSize(AllocaType);
uint64_t AllocaArraySize = AIArraySize->getLimitedValue();
assert(AllocaArraySize > 0 && "array size of AllocaInst is zero");
// Check that array size doesn't saturate uint64_t and doesn't
// overflow when it's multiplied by type size.
if (AllocaArraySize != ~0ULL &&
UINT64_MAX / AllocaArraySize >= AllocaTypeSize) {
AllocaSize = ConstantInt::get(Type::getInt64Ty(AI->getContext()),
AllocaArraySize * AllocaTypeSize);
}
}
}
builder.CreateLifetimeStart(AI, AllocaSize);
for (unsigned ri = 0, re = Returns.size(); ri != re; ++ri) {
IRBuilder<> builder(Returns[ri]);
builder.CreateLifetimeEnd(AI, AllocaSize);
}
}
}
// 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.
Function *StackSave = Intrinsic::getDeclaration(M, Intrinsic::stacksave);
Function *StackRestore=Intrinsic::getDeclaration(M,Intrinsic::stackrestore);
// Insert the llvm.stacksave.
CallInst *SavedPtr = IRBuilder<>(FirstNewBlock, FirstNewBlock->begin())
.CreateCall(StackSave, "savedstack");
// Insert a call to llvm.stackrestore before any return instructions in the
// inlined function.
for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
IRBuilder<>(Returns[i]).CreateCall(StackRestore, SavedPtr);
}
}
// 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 for an invoke instruction, we must make sure to rewrite
// any 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 *NewBr = BranchInst::Create(II->getNormalDest(), TheCall);
NewBr->setDebugLoc(Returns[0]->getDebugLoc());
}
// 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];
if (TheCall == R->getReturnValue())
TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
else
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;
BranchInst *CreatedBranchToNormalDest = NULL;
if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall)) {
// Add an unconditional branch to make this look like the CallInst case...
CreatedBranchToNormalDest = 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(CreatedBranchToNormalDest,
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.
Type *RTy = CalledFunc->getReturnType();
PHINode *PHI = 0;
if (Returns.size() > 1) {
// The PHI node should go at the front of the new basic block to merge all
// possible incoming values.
if (!TheCall->use_empty()) {
PHI = PHINode::Create(RTy, Returns.size(), 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.
DebugLoc Loc;
for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
ReturnInst *RI = Returns[i];
BranchInst* BI = BranchInst::Create(AfterCallBB, RI);
Loc = RI->getDebugLoc();
BI->setDebugLoc(Loc);
RI->eraseFromParent();
}
// We need to set the debug location to *somewhere* inside the
// inlined function. The line number may be nonsensical, but the
// instruction will at least be associated with the right
// function.
if (CreatedBranchToNormalDest)
CreatedBranchToNormalDest->setDebugLoc(Loc);
} 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()) {
if (TheCall == Returns[0]->getReturnValue())
TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
else
TheCall->replaceAllUsesWith(Returns[0]->getReturnValue());
}
// Update PHI nodes that use the ReturnBB to use the AfterCallBB.
BasicBlock *ReturnBB = Returns[0]->getParent();
ReturnBB->replaceAllUsesWith(AfterCallBB);
// Splice the code from the return block into the block that it will return
// to, which contains the code that was after the call.
AfterCallBB->getInstList().splice(AfterCallBB->begin(),
ReturnBB->getInstList());
if (CreatedBranchToNormalDest)
CreatedBranchToNormalDest->setDebugLoc(Returns[0]->getDebugLoc());
// 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.
CalleeEntry->replaceAllUsesWith(OrigBB); // Update PHI nodes
OrigBB->getInstList().splice(Br, CalleeEntry->getInstList());
// Remove the unconditional branch.
OrigBB->getInstList().erase(Br);
// Now we can remove the CalleeEntry block, which is now empty.
Caller->getBasicBlockList().erase(CalleeEntry);
// If we inserted a phi node, check to see if it has a single value (e.g. all
// the entries are the same or undef). If so, remove the PHI so it doesn't
// block other optimizations.
if (PHI) {
if (Value *V = SimplifyInstruction(PHI, IFI.TD)) {
PHI->replaceAllUsesWith(V);
PHI->eraseFromParent();
}
}
return true;
}
/// InsertStackProtectors - Insert code into the prologue and epilogue of the
/// function.
///
/// - The prologue code loads and stores the stack guard onto the stack.
/// - The epilogue checks the value stored in the prologue against the original
/// value. It calls __stack_chk_fail if they differ.
bool StackProtector::InsertStackProtectors() {
bool HasPrologue = false;
bool SupportsSelectionDAGSP =
EnableSelectionDAGSP && !TM->Options.EnableFastISel;
AllocaInst *AI = nullptr; // Place on stack that stores the stack guard.
Value *StackGuardVar = nullptr; // The stack guard variable.
for (Function::iterator I = F->begin(), E = F->end(); I != E;) {
BasicBlock *BB = &*I++;
ReturnInst *RI = dyn_cast<ReturnInst>(BB->getTerminator());
if (!RI)
continue;
if (!HasPrologue) {
HasPrologue = true;
SupportsSelectionDAGSP &=
CreatePrologue(F, M, RI, TLI, Trip, AI, StackGuardVar);
}
if (SupportsSelectionDAGSP) {
// Since we have a potential tail call, insert the special stack check
// intrinsic.
Instruction *InsertionPt = nullptr;
if (CallInst *CI = FindPotentialTailCall(BB, RI, TLI)) {
InsertionPt = CI;
} else {
InsertionPt = RI;
// At this point we know that BB has a return statement so it *DOES*
// have a terminator.
assert(InsertionPt != nullptr &&
"BB must have a terminator instruction at this point.");
}
Function *Intrinsic =
Intrinsic::getDeclaration(M, Intrinsic::stackprotectorcheck);
CallInst::Create(Intrinsic, StackGuardVar, "", InsertionPt);
} else {
// If we do not support SelectionDAG based tail calls, generate IR level
// tail calls.
//
// For each block with a return instruction, convert this:
//
// return:
// ...
// ret ...
//
// into this:
//
// return:
// ...
// %1 = load __stack_chk_guard
// %2 = load StackGuardSlot
// %3 = cmp i1 %1, %2
// br i1 %3, label %SP_return, label %CallStackCheckFailBlk
//
// SP_return:
// ret ...
//
// CallStackCheckFailBlk:
// call void @__stack_chk_fail()
// unreachable
// Create the FailBB. We duplicate the BB every time since the MI tail
// merge pass will merge together all of the various BB into one including
// fail BB generated by the stack protector pseudo instruction.
BasicBlock *FailBB = CreateFailBB();
// Split the basic block before the return instruction.
BasicBlock *NewBB = BB->splitBasicBlock(RI->getIterator(), "SP_return");
// Update the dominator tree if we need to.
if (DT && DT->isReachableFromEntry(BB)) {
DT->addNewBlock(NewBB, BB);
DT->addNewBlock(FailBB, BB);
}
// Remove default branch instruction to the new BB.
BB->getTerminator()->eraseFromParent();
// Move the newly created basic block to the point right after the old
// basic block so that it's in the "fall through" position.
NewBB->moveAfter(BB);
// Generate the stack protector instructions in the old basic block.
IRBuilder<> B(BB);
LoadInst *LI1 = B.CreateLoad(StackGuardVar);
LoadInst *LI2 = B.CreateLoad(AI);
Value *Cmp = B.CreateICmpEQ(LI1, LI2);
unsigned SuccessWeight =
BranchProbabilityInfo::getBranchWeightStackProtector(true);
unsigned FailureWeight =
BranchProbabilityInfo::getBranchWeightStackProtector(false);
MDNode *Weights = MDBuilder(F->getContext())
.createBranchWeights(SuccessWeight, FailureWeight);
B.CreateCondBr(Cmp, NewBB, FailBB, Weights);
}
}
// Return if we didn't modify any basic blocks. i.e., there are no return
// statements in the function.
return HasPrologue;
}
//
// Method: runOnModule()
//
// Description:
// Entry point for this LLVM pass.
// If a function returns a struct, make it return
// a pointer to the struct.
//
// Inputs:
// M - A reference to the LLVM module to transform
//
// Outputs:
// M - The transformed LLVM module.
//
// Return value:
// true - The module was modified.
// false - The module was not modified.
//
bool StructRet::runOnModule(Module& M) {
const llvm::DataLayout targetData(&M);
std::vector<Function*> worklist;
for (Module::iterator I = M.begin(); I != M.end(); ++I)
if (!I->mayBeOverridden()) {
if(I->hasAddressTaken())
continue;
if(I->getReturnType()->isStructTy()) {
worklist.push_back(I);
}
}
while(!worklist.empty()) {
Function *F = worklist.back();
worklist.pop_back();
Type *NewArgType = F->getReturnType()->getPointerTo();
// Construct the new Type
std::vector<Type*>TP;
TP.push_back(NewArgType);
for (Function::arg_iterator ii = F->arg_begin(), ee = F->arg_end();
ii != ee; ++ii) {
TP.push_back(ii->getType());
}
FunctionType *NFTy = FunctionType::get(F->getReturnType(), TP, F->isVarArg());
// Create the new function body and insert it into the module.
Function *NF = Function::Create(NFTy,
F->getLinkage(),
F->getName(), &M);
ValueToValueMapTy ValueMap;
Function::arg_iterator NI = NF->arg_begin();
NI->setName("ret");
++NI;
for (Function::arg_iterator II = F->arg_begin(); II != F->arg_end(); ++II, ++NI) {
ValueMap[II] = NI;
NI->setName(II->getName());
AttributeSet attrs = F->getAttributes().getParamAttributes(II->getArgNo() + 1);
if (!attrs.isEmpty())
NI->addAttr(attrs);
}
// Perform the cloning.
SmallVector<ReturnInst*,100> Returns;
if (!F->isDeclaration())
CloneFunctionInto(NF, F, ValueMap, false, Returns);
std::vector<Value*> fargs;
for(Function::arg_iterator ai = NF->arg_begin(),
ae= NF->arg_end(); ai != ae; ++ai) {
fargs.push_back(ai);
}
NF->setAttributes(NF->getAttributes().addAttributes(
M.getContext(), 0, F->getAttributes().getRetAttributes()));
NF->setAttributes(NF->getAttributes().addAttributes(
M.getContext(), ~0, F->getAttributes().getFnAttributes()));
for (Function::iterator B = NF->begin(), FE = NF->end(); B != FE; ++B) {
for (BasicBlock::iterator I = B->begin(), BE = B->end(); I != BE;) {
ReturnInst * RI = dyn_cast<ReturnInst>(I++);
if(!RI)
continue;
LoadInst *LI = dyn_cast<LoadInst>(RI->getOperand(0));
assert(LI && "Return should be preceded by a load instruction");
IRBuilder<> Builder(RI);
Builder.CreateMemCpy(fargs.at(0),
LI->getPointerOperand(),
targetData.getTypeStoreSize(LI->getType()),
targetData.getPrefTypeAlignment(LI->getType()));
}
}
for(Value::use_iterator ui = F->use_begin(), ue = F->use_end();
ui != ue; ) {
CallInst *CI = dyn_cast<CallInst>(*ui++);
if(!CI)
continue;
if(CI->getCalledFunction() != F)
continue;
if(CI->hasByValArgument())
continue;
AllocaInst *AllocaNew = new AllocaInst(F->getReturnType(), 0, "", CI);
SmallVector<Value*, 8> Args;
//this should probably be done in a different manner
AttributeSet NewCallPAL=AttributeSet();
// Get the initial attributes of the call
AttributeSet CallPAL = CI->getAttributes();
AttributeSet RAttrs = CallPAL.getRetAttributes();
AttributeSet FnAttrs = CallPAL.getFnAttributes();
if (!RAttrs.isEmpty())
NewCallPAL=NewCallPAL.addAttributes(F->getContext(),0, RAttrs);
Args.push_back(AllocaNew);
for(unsigned j = 0; j < CI->getNumOperands()-1; j++) {
Args.push_back(CI->getOperand(j));
// position in the NewCallPAL
AttributeSet Attrs = CallPAL.getParamAttributes(j);
if (!Attrs.isEmpty())
NewCallPAL=NewCallPAL.addAttributes(F->getContext(),Args.size(), Attrs);
}
// Create the new attributes vec.
if (!FnAttrs.isEmpty())
NewCallPAL=NewCallPAL.addAttributes(F->getContext(),~0, FnAttrs);
CallInst *CallI = CallInst::Create(NF, Args, "", CI);
CallI->setCallingConv(CI->getCallingConv());
CallI->setAttributes(NewCallPAL);
LoadInst *LI = new LoadInst(AllocaNew, "", CI);
CI->replaceAllUsesWith(LI);
CI->eraseFromParent();
}
if(F->use_empty())
F->eraseFromParent();
}
return true;
}
void AAAnalyzer::handle_inst(Instruction *inst, FunctionWrapper * parent_func) {
//outs()<<*inst<<"\n"; outs().flush();
switch (inst->getOpcode()) {
// common/bitwise binary operations
// Terminator instructions
case Instruction::Ret:
{
ReturnInst* retInst = ((ReturnInst*) inst);
if (retInst->getNumOperands() > 0 && !retInst->getOperandUse(0)->getType()->isVoidTy()) {
parent_func->addRet(retInst->getOperandUse(0));
}
}
break;
case Instruction::Resume:
{
Value* resume = ((ResumeInst*) inst)->getOperand(0);
parent_func->addResume(resume);
}
break;
case Instruction::Switch:
case Instruction::Br:
case Instruction::IndirectBr:
case Instruction::Unreachable:
break;
// vector operations
case Instruction::ExtractElement:
{
}
break;
case Instruction::InsertElement:
{
}
break;
case Instruction::ShuffleVector:
{
}
break;
// aggregate operations
case Instruction::ExtractValue:
{
Value * agg = ((ExtractValueInst*) inst)->getAggregateOperand();
DyckVertex* aggV = wrapValue(agg);
Type* aggTy = agg->getType();
ArrayRef<unsigned> indices = ((ExtractValueInst*) inst)->getIndices();
DyckVertex* currentStruct = aggV;
for (unsigned int i = 0; i < indices.size(); i++) {
if (isa<CompositeType>(aggTy) && aggTy->isSized()) {
if (!aggTy->isStructTy()) {
aggTy = ((CompositeType*) aggTy)->getTypeAtIndex(indices[i]);
#ifndef ARRAY_SIMPLIFIED
current = addPtrOffset(current, (int) indices[i] * dl.getTypeAllocSize(aggTy), dgraph);
#endif
if (i == indices.size() - 1) {
this->makeAlias(currentStruct, wrapValue(inst));
}
} else {
aggTy = ((CompositeType*) aggTy)->getTypeAtIndex(indices[i]);
if (i != indices.size() - 1) {
currentStruct = this->addField(currentStruct, -2 - (int) indices[i], NULL);
} else {
currentStruct = this->addField(currentStruct, -2 - (int) indices[i], wrapValue(inst));
}
}
} else {
break;
}
}
}
break;
case Instruction::InsertValue:
{
DyckVertex* resultV = wrapValue(inst);
Value * agg = ((InsertValueInst*) inst)->getAggregateOperand();
if (!isa<UndefValue>(agg)) {
makeAlias(resultV, wrapValue(agg));
}
Value * val = ((InsertValueInst*) inst)->getInsertedValueOperand();
DyckVertex* insertedVal = wrapValue(val);
Type *aggTy = inst->getType();
ArrayRef<unsigned> indices = ((InsertValueInst*) inst)->getIndices();
DyckVertex* currentStruct = resultV;
for (unsigned int i = 0; i < indices.size(); i++) {
if (isa<CompositeType>(aggTy) && aggTy->isSized()) {
if (!aggTy->isStructTy()) {
aggTy = ((CompositeType*) aggTy)->getTypeAtIndex(indices[i]);
#ifndef ARRAY_SIMPLIFIED
current = addPtrOffset(current, (int) indices[i] * dl.getTypeAllocSize(aggTy), dgraph);
#endif
if (i == indices.size() - 1) {
this->makeAlias(currentStruct, insertedVal);
}
} else {
aggTy = ((CompositeType*) aggTy)->getTypeAtIndex(indices[i]);
if (i != indices.size() - 1) {
currentStruct = this->addField(currentStruct, -2 - (int) indices[i], NULL);
} else {
currentStruct = this->addField(currentStruct, -2 - (int) indices[i], insertedVal);
}
}
} else {
break;
}
}
}
break;
// memory accessing and addressing operations
case Instruction::Alloca:
{
}
break;
case Instruction::Fence:
{
}
break;
case Instruction::AtomicCmpXchg:
{
Value * retXchg = inst;
Value * ptrXchg = inst->getOperand(0);
Value * newXchg = inst->getOperand(2);
addPtrTo(wrapValue(ptrXchg), wrapValue(retXchg));
addPtrTo(wrapValue(ptrXchg), wrapValue(newXchg));
}
break;
case Instruction::AtomicRMW:
{
Value * retRmw = inst;
Value * ptrRmw = ((AtomicRMWInst*) inst)->getPointerOperand();
addPtrTo(wrapValue(ptrRmw), wrapValue(retRmw));
switch (((AtomicRMWInst*) inst)->getOperation()) {
case AtomicRMWInst::Max:
case AtomicRMWInst::Min:
case AtomicRMWInst::UMax:
case AtomicRMWInst::UMin:
case AtomicRMWInst::Xchg:
{
Value * newRmw = ((AtomicRMWInst*) inst)->getValOperand();
addPtrTo(wrapValue(ptrRmw), wrapValue(newRmw));
}
break;
default:
//others are binary ops like add/sub/...
///@TODO
break;
}
}
break;
case Instruction::Load:
{
Value *lval = inst;
Value *ladd = inst->getOperand(0);
addPtrTo(wrapValue(ladd), wrapValue(lval));
}
break;
case Instruction::Store:
{
Value * sval = inst->getOperand(0);
Value * sadd = inst->getOperand(1);
addPtrTo(wrapValue(sadd), wrapValue(sval));
}
break;
case Instruction::GetElementPtr:
{
makeAlias(wrapValue(inst), handle_gep((GEPOperator*) inst));
}
break;
// conversion operations
case Instruction::Trunc:
case Instruction::ZExt:
case Instruction::SExt:
case Instruction::FPTrunc:
case Instruction::FPExt:
case Instruction::FPToUI:
case Instruction::FPToSI:
case Instruction::UIToFP:
case Instruction::SIToFP:
case Instruction::BitCast:
case Instruction::PtrToInt:
case Instruction::IntToPtr:
{
Value * itpv = inst->getOperand(0);
makeAlias(wrapValue(inst), wrapValue(itpv));
}
break;
// other operations
case Instruction::Invoke: // invoke is a terminal operation
{
InvokeInst * invoke = (InvokeInst*) inst;
LandingPadInst* lpd = invoke->getLandingPadInst();
parent_func->addLandingPad(invoke, lpd);
Value * cv = invoke->getCalledValue();
vector<Value*> args;
for (unsigned i = 0; i < invoke->getNumArgOperands(); i++) {
args.push_back(invoke->getArgOperand(i));
}
this->handle_invoke_call_inst(invoke, cv, &args, parent_func);
}
break;
case Instruction::Call:
{
CallInst * callinst = (CallInst*) inst;
if (callinst->isInlineAsm()) {
break;
}
Value * cv = callinst->getCalledValue();
vector<Value*> args;
for (unsigned i = 0; i < callinst->getNumArgOperands(); i++) {
args.push_back(callinst->getArgOperand(i));
}
this->handle_invoke_call_inst(callinst, cv, &args, parent_func);
}
break;
case Instruction::PHI:
{
PHINode *phi = (PHINode *) inst;
int nums = phi->getNumIncomingValues();
for (int i = 0; i < nums; i++) {
Value * p = phi->getIncomingValue(i);
makeAlias(wrapValue(inst), wrapValue(p));
}
}
break;
case Instruction::Select:
{
Value *first = ((SelectInst*) inst)->getTrueValue();
Value *second = ((SelectInst*) inst)->getFalseValue();
makeAlias(wrapValue(inst), wrapValue(first));
makeAlias(wrapValue(inst), wrapValue(second));
}
break;
case Instruction::VAArg:
{
parent_func->addVAArg(inst);
DyckVertex* vaarg = wrapValue(inst);
Value * ptrVaarg = inst->getOperand(0);
addPtrTo(wrapValue(ptrVaarg), vaarg);
}
break;
case Instruction::LandingPad: // handled with invoke inst
case Instruction::ICmp:
case Instruction::FCmp:
default:
break;
}
}
// visitReturnInst - We want to destroy the setjmp map upon exit from the
// function.
void LowerSetJmp::visitReturnInst(ReturnInst &RI) {
Function* Func = RI.getParent()->getParent();
CallInst::Create(DestroySJMap, GetSetJmpMap(Func), "", &RI);
}
void Lint::visitReturnInst(ReturnInst &I) {
Function *F = I.getParent()->getParent();
Assert1(!F->doesNotReturn(),
"Unusual: Return statement in function with noreturn attribute",
&I);
}
/// InsertStackProtectors - Insert code into the prologue and epilogue of the
/// function.
///
/// - The prologue code loads and stores the stack guard onto the stack.
/// - The epilogue checks the value stored in the prologue against the original
/// value. It calls __stack_chk_fail if they differ.
bool StackProtector::InsertStackProtectors() {
bool SupportsSelectionDAGSP =
EnableSelectionDAGSP && !TM->Options.EnableFastISel;
AllocaInst *AI = nullptr; // Place on stack that stores the stack guard.
Value *StackGuardVar = nullptr; // The stack guard variable.
for (Function::iterator I = F->begin(), E = F->end(); I != E;) {
BasicBlock *BB = &*I++;
ReturnInst *RI = dyn_cast<ReturnInst>(BB->getTerminator());
if (!RI)
continue;
if (!HasPrologue) {
HasPrologue = true;
SupportsSelectionDAGSP &=
CreatePrologue(F, M, RI, TLI, AI, StackGuardVar);
}
if (!SupportsSelectionDAGSP) {
// If we do not support SelectionDAG based tail calls, generate IR level
// tail calls.
//
// For each block with a return instruction, convert this:
//
// return:
// ...
// ret ...
//
// into this:
//
// return:
// ...
// %1 = load __stack_chk_guard
// %2 = load StackGuardSlot
// %3 = cmp i1 %1, %2
// br i1 %3, label %SP_return, label %CallStackCheckFailBlk
//
// SP_return:
// ret ...
//
// CallStackCheckFailBlk:
// call void @__stack_chk_fail()
// unreachable
// Create the FailBB. We duplicate the BB every time since the MI tail
// merge pass will merge together all of the various BB into one including
// fail BB generated by the stack protector pseudo instruction.
BasicBlock *FailBB = CreateFailBB();
// Set HasIRCheck to true, so that SelectionDAG will not generate its own
// version.
HasIRCheck = true;
// Split the basic block before the return instruction.
BasicBlock *NewBB = BB->splitBasicBlock(RI->getIterator(), "SP_return");
// Update the dominator tree if we need to.
if (DT && DT->isReachableFromEntry(BB)) {
DT->addNewBlock(NewBB, BB);
DT->addNewBlock(FailBB, BB);
}
// Remove default branch instruction to the new BB.
BB->getTerminator()->eraseFromParent();
// Move the newly created basic block to the point right after the old
// basic block so that it's in the "fall through" position.
NewBB->moveAfter(BB);
// Generate the stack protector instructions in the old basic block.
IRBuilder<> B(BB);
LoadInst *LI1 = B.CreateLoad(StackGuardVar);
LoadInst *LI2 = B.CreateLoad(AI);
Value *Cmp = B.CreateICmpEQ(LI1, LI2);
auto SuccessProb =
BranchProbabilityInfo::getBranchProbStackProtector(true);
auto FailureProb =
BranchProbabilityInfo::getBranchProbStackProtector(false);
MDNode *Weights = MDBuilder(F->getContext())
.createBranchWeights(SuccessProb.getNumerator(),
FailureProb.getNumerator());
B.CreateCondBr(Cmp, NewBB, FailBB, Weights);
}
}
// Return if we didn't modify any basic blocks. i.e., there are no return
// statements in the function.
return HasPrologue;
}
// 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 a non-tail call, we must clear the 'tail'
// flags on any calls that we inline.
bool MustClearTailCallFlags =
isa<CallInst>(TheCall) && !cast<CallInst>(TheCall)->isTailCall();
BasicBlock *OrigBB = TheCall->getParent();
Function *Caller = OrigBB->getParent();
// 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;
// Calculate the vector of arguments to pass into the function cloner, which
// matches up the formal to the actual argument values.
assert(std::distance(CalledFunc->arg_begin(), CalledFunc->arg_end()) ==
std::distance(CS.arg_begin(), CS.arg_end()) &&
"No varargs calls can be inlined!");
CallSite::arg_iterator AI = CS.arg_begin();
for (Function::const_arg_iterator I = CalledFunc->arg_begin(),
E = CalledFunc->arg_end(); I != E; ++I, ++AI)
ValueMap[I] = *AI;
// 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(Caller, CalledFunc, 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();
const Type *BytePtr = PointerType::get(Type::Int8Ty);
// Get the two intrinsics we care about.
Constant *StackSave, *StackRestore;
StackSave = M->getOrInsertFunction("llvm.stacksave", BytePtr, NULL);
StackRestore = M->getOrInsertFunction("llvm.stackrestore", Type::VoidTy,
BytePtr, NULL);
// 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 = new CallInst(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 = new CallInst(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())) {
new CallInst(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.
if (MustClearTailCallFlags && InlinedFunctionInfo.ContainsCalls) {
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))
CI->setTailCall(false);
}
// 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))
new BranchInst(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())
TheCall->replaceAllUsesWith(Returns[0]->getReturnValue());
// Since we are now done with the Call/Invoke, we can delete it.
TheCall->getParent()->getInstList().erase(TheCall);
// Since we are now done with the return instruction, delete it also.
Returns[0]->getParent()->getInstList().erase(Returns[0]);
// 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 = new BranchInst(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.
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 = new PHINode(CalledFunc->getReturnType(),
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, turning them into unconditional
// branches to the merge point now, and adding entries to the PHI node as
// appropriate.
for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
ReturnInst *RI = Returns[i];
if (PHI) {
assert(RI->getReturnValue() && "Ret should have value!");
assert(RI->getReturnValue()->getType() == PHI->getType() &&
"Ret value not consistent in function!");
PHI->addIncoming(RI->getReturnValue(), RI->getParent());
}
// Add a branch to the merge point where the PHI node lives if it exists.
new BranchInst(AfterCallBB, RI);
// Delete the return instruction now
RI->getParent()->getInstList().erase(RI);
}
} 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;
}
bool AMDGPURewriteOutArguments::runOnFunction(Function &F) {
if (skipFunction(F))
return false;
// TODO: Could probably handle variadic functions.
if (F.isVarArg() || F.hasStructRetAttr() ||
AMDGPU::isEntryFunctionCC(F.getCallingConv()))
return false;
MDA = &getAnalysis<MemoryDependenceWrapperPass>().getMemDep();
unsigned ReturnNumRegs = 0;
SmallSet<int, 4> OutArgIndexes;
SmallVector<Type *, 4> ReturnTypes;
Type *RetTy = F.getReturnType();
if (!RetTy->isVoidTy()) {
ReturnNumRegs = DL->getTypeStoreSize(RetTy) / 4;
if (ReturnNumRegs >= MaxNumRetRegs)
return false;
ReturnTypes.push_back(RetTy);
}
SmallVector<Argument *, 4> OutArgs;
for (Argument &Arg : F.args()) {
if (isOutArgumentCandidate(Arg)) {
LLVM_DEBUG(dbgs() << "Found possible out argument " << Arg
<< " in function " << F.getName() << '\n');
OutArgs.push_back(&Arg);
}
}
if (OutArgs.empty())
return false;
using ReplacementVec = SmallVector<std::pair<Argument *, Value *>, 4>;
DenseMap<ReturnInst *, ReplacementVec> Replacements;
SmallVector<ReturnInst *, 4> Returns;
for (BasicBlock &BB : F) {
if (ReturnInst *RI = dyn_cast<ReturnInst>(&BB.back()))
Returns.push_back(RI);
}
if (Returns.empty())
return false;
bool Changing;
do {
Changing = false;
// Keep retrying if we are able to successfully eliminate an argument. This
// helps with cases with multiple arguments which may alias, such as in a
// sincos implemntation. If we have 2 stores to arguments, on the first
// attempt the MDA query will succeed for the second store but not the
// first. On the second iteration we've removed that out clobbering argument
// (by effectively moving it into another function) and will find the second
// argument is OK to move.
for (Argument *OutArg : OutArgs) {
bool ThisReplaceable = true;
SmallVector<std::pair<ReturnInst *, StoreInst *>, 4> ReplaceableStores;
Type *ArgTy = OutArg->getType()->getPointerElementType();
// Skip this argument if converting it will push us over the register
// count to return limit.
// TODO: This is an approximation. When legalized this could be more. We
// can ask TLI for exactly how many.
unsigned ArgNumRegs = DL->getTypeStoreSize(ArgTy) / 4;
if (ArgNumRegs + ReturnNumRegs > MaxNumRetRegs)
continue;
// An argument is convertible only if all exit blocks are able to replace
// it.
for (ReturnInst *RI : Returns) {
BasicBlock *BB = RI->getParent();
MemDepResult Q = MDA->getPointerDependencyFrom(MemoryLocation(OutArg),
true, BB->end(), BB, RI);
StoreInst *SI = nullptr;
if (Q.isDef())
SI = dyn_cast<StoreInst>(Q.getInst());
if (SI) {
LLVM_DEBUG(dbgs() << "Found out argument store: " << *SI << '\n');
ReplaceableStores.emplace_back(RI, SI);
} else {
ThisReplaceable = false;
break;
}
}
if (!ThisReplaceable)
continue; // Try the next argument candidate.
for (std::pair<ReturnInst *, StoreInst *> Store : ReplaceableStores) {
Value *ReplVal = Store.second->getValueOperand();
auto &ValVec = Replacements[Store.first];
if (llvm::find_if(ValVec,
[OutArg](const std::pair<Argument *, Value *> &Entry) {
return Entry.first == OutArg;}) != ValVec.end()) {
LLVM_DEBUG(dbgs()
<< "Saw multiple out arg stores" << *OutArg << '\n');
// It is possible to see stores to the same argument multiple times,
// but we expect these would have been optimized out already.
ThisReplaceable = false;
break;
}
ValVec.emplace_back(OutArg, ReplVal);
Store.second->eraseFromParent();
}
if (ThisReplaceable) {
ReturnTypes.push_back(ArgTy);
OutArgIndexes.insert(OutArg->getArgNo());
++NumOutArgumentsReplaced;
Changing = true;
}
}
} while (Changing);
if (Replacements.empty())
return false;
LLVMContext &Ctx = F.getParent()->getContext();
StructType *NewRetTy = StructType::create(Ctx, ReturnTypes, F.getName());
FunctionType *NewFuncTy = FunctionType::get(NewRetTy,
F.getFunctionType()->params(),
F.isVarArg());
LLVM_DEBUG(dbgs() << "Computed new return type: " << *NewRetTy << '\n');
Function *NewFunc = Function::Create(NewFuncTy, Function::PrivateLinkage,
F.getName() + ".body");
F.getParent()->getFunctionList().insert(F.getIterator(), NewFunc);
NewFunc->copyAttributesFrom(&F);
NewFunc->setComdat(F.getComdat());
// We want to preserve the function and param attributes, but need to strip
// off any return attributes, e.g. zeroext doesn't make sense with a struct.
NewFunc->stealArgumentListFrom(F);
AttrBuilder RetAttrs;
RetAttrs.addAttribute(Attribute::SExt);
RetAttrs.addAttribute(Attribute::ZExt);
RetAttrs.addAttribute(Attribute::NoAlias);
NewFunc->removeAttributes(AttributeList::ReturnIndex, RetAttrs);
// TODO: How to preserve metadata?
// Move the body of the function into the new rewritten function, and replace
// this function with a stub.
NewFunc->getBasicBlockList().splice(NewFunc->begin(), F.getBasicBlockList());
for (std::pair<ReturnInst *, ReplacementVec> &Replacement : Replacements) {
ReturnInst *RI = Replacement.first;
IRBuilder<> B(RI);
B.SetCurrentDebugLocation(RI->getDebugLoc());
int RetIdx = 0;
Value *NewRetVal = UndefValue::get(NewRetTy);
Value *RetVal = RI->getReturnValue();
if (RetVal)
NewRetVal = B.CreateInsertValue(NewRetVal, RetVal, RetIdx++);
for (std::pair<Argument *, Value *> ReturnPoint : Replacement.second) {
Argument *Arg = ReturnPoint.first;
Value *Val = ReturnPoint.second;
Type *EltTy = Arg->getType()->getPointerElementType();
if (Val->getType() != EltTy) {
Type *EffectiveEltTy = EltTy;
if (StructType *CT = dyn_cast<StructType>(EltTy)) {
assert(CT->getNumElements() == 1);
EffectiveEltTy = CT->getElementType(0);
}
if (DL->getTypeSizeInBits(EffectiveEltTy) !=
DL->getTypeSizeInBits(Val->getType())) {
assert(isVec3ToVec4Shuffle(EffectiveEltTy, Val->getType()));
Val = B.CreateShuffleVector(Val, UndefValue::get(Val->getType()),
{ 0, 1, 2 });
}
Val = B.CreateBitCast(Val, EffectiveEltTy);
// Re-create single element composite.
if (EltTy != EffectiveEltTy)
Val = B.CreateInsertValue(UndefValue::get(EltTy), Val, 0);
}
NewRetVal = B.CreateInsertValue(NewRetVal, Val, RetIdx++);
}
if (RetVal)
RI->setOperand(0, NewRetVal);
else {
B.CreateRet(NewRetVal);
RI->eraseFromParent();
}
}
SmallVector<Value *, 16> StubCallArgs;
for (Argument &Arg : F.args()) {
if (OutArgIndexes.count(Arg.getArgNo())) {
// It's easier to preserve the type of the argument list. We rely on
// DeadArgumentElimination to take care of these.
StubCallArgs.push_back(UndefValue::get(Arg.getType()));
} else {
StubCallArgs.push_back(&Arg);
}
}
BasicBlock *StubBB = BasicBlock::Create(Ctx, "", &F);
IRBuilder<> B(StubBB);
CallInst *StubCall = B.CreateCall(NewFunc, StubCallArgs);
int RetIdx = RetTy->isVoidTy() ? 0 : 1;
for (Argument &Arg : F.args()) {
if (!OutArgIndexes.count(Arg.getArgNo()))
continue;
PointerType *ArgType = cast<PointerType>(Arg.getType());
auto *EltTy = ArgType->getElementType();
unsigned Align = Arg.getParamAlignment();
if (Align == 0)
Align = DL->getABITypeAlignment(EltTy);
Value *Val = B.CreateExtractValue(StubCall, RetIdx++);
Type *PtrTy = Val->getType()->getPointerTo(ArgType->getAddressSpace());
// We can peek through bitcasts, so the type may not match.
Value *PtrVal = B.CreateBitCast(&Arg, PtrTy);
B.CreateAlignedStore(Val, PtrVal, Align);
}
if (!RetTy->isVoidTy()) {
B.CreateRet(B.CreateExtractValue(StubCall, 0));
} else {
B.CreateRetVoid();
}
// The function is now a stub we want to inline.
F.addFnAttr(Attribute::AlwaysInline);
++NumOutArgumentFunctionsReplaced;
return true;
}
