/// CloneFunction - Return a copy of the specified function, but without
/// embedding the function into another module. Also, any references specified
/// in the VMap are changed to refer to their mapped value instead of the
/// original one. If any of the arguments to the function are in the VMap,
/// the arguments are deleted from the resultant function. The VMap is
/// updated to include mappings from all of the instructions and basicblocks in
/// the function from their old to new values.
///
Function *llvm::CloneFunction(const Function *F, ValueToValueMapTy &VMap,
bool ModuleLevelChanges,
ClonedCodeInfo *CodeInfo) {
std::vector<Type*> ArgTypes;
// The user might be deleting arguments to the function by specifying them in
// the VMap. If so, we need to not add the arguments to the arg ty vector
//
for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end();
I != E; ++I)
if (VMap.count(I) == 0) // Haven't mapped the argument to anything yet?
ArgTypes.push_back(I->getType());
// Create a new function type...
FunctionType *FTy = FunctionType::get(F->getFunctionType()->getReturnType(),
ArgTypes, F->getFunctionType()->isVarArg());
// Create the new function...
Function *NewF = Function::Create(FTy, F->getLinkage(), F->getName());
// Loop over the arguments, copying the names of the mapped arguments over...
Function::arg_iterator DestI = NewF->arg_begin();
for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end();
I != E; ++I)
if (VMap.count(I) == 0) { // Is this argument preserved?
DestI->setName(I->getName()); // Copy the name over...
VMap[I] = DestI++; // Add mapping to VMap
}
SmallVector<ReturnInst*, 8> Returns; // Ignore returns cloned.
CloneFunctionInto(NewF, F, VMap, ModuleLevelChanges, Returns, "", CodeInfo);
return NewF;
}
static X86MachineFunctionInfo calculateFunctionInfo(const Function *F,
const TargetData *TD) {
X86MachineFunctionInfo Info;
uint64_t Size = 0;
switch (F->getCallingConv()) {
case CallingConv::X86_StdCall:
Info.setDecorationStyle(StdCall);
break;
case CallingConv::X86_FastCall:
Info.setDecorationStyle(FastCall);
break;
default:
return Info;
}
unsigned argNum = 1;
for (Function::const_arg_iterator AI = F->arg_begin(), AE = F->arg_end();
AI != AE; ++AI, ++argNum) {
const Type* Ty = AI->getType();
// 'Dereference' type in case of byval parameter attribute
if (F->paramHasAttr(argNum, Attribute::ByVal))
Ty = cast<PointerType>(Ty)->getElementType();
// Size should be aligned to DWORD boundary
Size += ((TD->getTypePaddedSize(Ty) + 3)/4)*4;
}
// We're not supporting tooooo huge arguments :)
Info.setBytesToPopOnReturn((unsigned int)Size);
return Info;
}
void Andersen::collectConstraintsForGlobals(Module& M)
{
// Create a pointer and an object for each global variable
for (auto const& globalVal: M.globals())
{
NodeIndex gVal = nodeFactory.createValueNode(&globalVal);
NodeIndex gObj = nodeFactory.createObjectNode(&globalVal);
constraints.emplace_back(AndersConstraint::ADDR_OF, gVal, gObj);
}
// Functions and function pointers are also considered global
for (auto const& f: M)
{
// If f is an addr-taken function, create a pointer and an object for it
if (f.hasAddressTaken())
{
NodeIndex fVal = nodeFactory.createValueNode(&f);
NodeIndex fObj = nodeFactory.createObjectNode(&f);
constraints.emplace_back(AndersConstraint::ADDR_OF, fVal, fObj);
}
if (f.isDeclaration() || f.isIntrinsic())
continue;
// Create return node
if (f.getFunctionType()->getReturnType()->isPointerTy())
{
nodeFactory.createReturnNode(&f);
}
// Create vararg node
if (f.getFunctionType()->isVarArg())
nodeFactory.createVarargNode(&f);
// Add nodes for all formal arguments.
for (Function::const_arg_iterator itr = f.arg_begin(), ite = f.arg_end(); itr != ite; ++itr)
{
if (isa<PointerType>(itr->getType()))
nodeFactory.createValueNode(itr);
}
}
// Init globals here since an initializer may refer to a global var/func below it
for (auto const& globalVal: M.globals())
{
NodeIndex gObj = nodeFactory.getObjectNodeFor(&globalVal);
assert(gObj != AndersNodeFactory::InvalidIndex && "Cannot find global object!");
if (globalVal.hasDefinitiveInitializer())
{
addGlobalInitializerConstraints(gObj, globalVal.getInitializer());
}
else
{
// If it doesn't have an initializer (i.e. it's defined in another translation unit), it points to the universal set.
constraints.emplace_back(AndersConstraint::COPY,
gObj, nodeFactory.getUniversalObjNode());
}
}
}
const Type* getArgumentType(const Function* f, const unsigned arg_index) {
assert (f);
assert (arg_index < f->arg_size());
Function::const_arg_iterator A = f->arg_begin();
for (unsigned i=0; i<arg_index; ++i) ++A; //is there a better way? :P
return A->getType();
}
bool TriCoreCallingConvHook::isRegVali64Type (MachineFunction& _mf) {
Function::const_arg_iterator FI;
FI = _mf.getFunction()->arg_begin();
std::advance(FI,curArg);
outs() << "size: " << FI->getType()->getScalarSizeInBits() << "\n";
return (FI->getType()->getScalarSizeInBits() == 64) ? true : false;
}
// Clone OldFunc into NewFunc, transforming the old arguments into references to
// VMap values.
//
void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc,
ValueToValueMapTy &VMap,
bool ModuleLevelChanges,
SmallVectorImpl<ReturnInst*> &Returns,
const char *NameSuffix, ClonedCodeInfo *CodeInfo) {
assert(NameSuffix && "NameSuffix cannot be null!");
#ifndef NDEBUG
for (Function::const_arg_iterator I = OldFunc->arg_begin(),
E = OldFunc->arg_end(); I != E; ++I)
assert(VMap.count(I) && "No mapping from source argument specified!");
#endif
// Clone any attributes.
if (NewFunc->arg_size() == OldFunc->arg_size())
NewFunc->copyAttributesFrom(OldFunc);
else {
//Some arguments were deleted with the VMap. Copy arguments one by one
for (Function::const_arg_iterator I = OldFunc->arg_begin(),
E = OldFunc->arg_end(); I != E; ++I)
if (Argument* Anew = dyn_cast<Argument>(VMap[I]))
Anew->addAttr( OldFunc->getAttributes()
.getParamAttributes(I->getArgNo() + 1));
NewFunc->setAttributes(NewFunc->getAttributes()
.addAttr(0, OldFunc->getAttributes()
.getRetAttributes()));
NewFunc->setAttributes(NewFunc->getAttributes()
.addAttr(~0, OldFunc->getAttributes()
.getFnAttributes()));
}
// Loop over all of the basic blocks in the function, cloning them as
// appropriate. Note that we save BE this way in order to handle cloning of
// recursive functions into themselves.
//
for (Function::const_iterator BI = OldFunc->begin(), BE = OldFunc->end();
BI != BE; ++BI) {
const BasicBlock &BB = *BI;
// Create a new basic block and copy instructions into it!
BasicBlock *CBB = CloneBasicBlock(&BB, VMap, NameSuffix, NewFunc, CodeInfo);
VMap[&BB] = CBB; // Add basic block mapping.
if (ReturnInst *RI = dyn_cast<ReturnInst>(CBB->getTerminator()))
Returns.push_back(RI);
}
// Loop over all of the instructions in the function, fixing up operand
// references as we go. This uses VMap to do all the hard work.
for (Function::iterator BB = cast<BasicBlock>(VMap[OldFunc->begin()]),
BE = NewFunc->end(); BB != BE; ++BB)
// Loop over all instructions, fixing each one as we find it...
for (BasicBlock::iterator II = BB->begin(); II != BB->end(); ++II)
RemapInstruction(II, VMap,
ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges);
}
void ProgramCFG::setFuncVariable(const Function *F,string func, CFG* cfg, bool initial){
for (Function::const_arg_iterator it = F->arg_begin(), E = F->arg_end();it != E; ++it) {
Type *Ty = it->getType();
if(initial){
string varNum = it->getName();
string varName = func+"_"+varNum;
if(Ty->isPointerTy()){
Type *ETy = Ty->getPointerElementType();
int ID = cfg->counter_variable++;
Variable var(varName, ID, PTR);
cfg->variableList.push_back(var);
InstParser::setVariable(cfg, NULL, ETy, varName, true);
}
else{
VarType type;
if(Ty->isIntegerTy())
type = INT;
else if(Ty->isFloatingPointTy())
type = FP;
else
errs()<<"0:programCFG.type error\n";
int ID = cfg->counter_variable++;
Variable var(varName, ID, type);
cfg->variableList.push_back(var);
cfg->mainInput.push_back(ID);
}
}
else{
int ID = cfg->counter_variable++;
string varNum = it->getName();
string varName = func+"_"+varNum;
VarType type;
if(Ty->isPointerTy())
type = PTR;
else if(Ty->isIntegerTy())
type = INT;
else if(Ty->isFloatingPointTy())
type = FP;
else
errs()<<"1:programCFG.type error\n";
if(!cfg->hasVariable(varName)){
Variable var(varName, ID, type);
cfg->variableList.push_back(var);
}
else
errs()<<"1:setFuncVariable error 10086!!\t"<<varName<<"\n";
}
}
}
bool AArch64CallLowering::LowerFormalArguments(
MachineIRBuilder &MIRBuilder, const Function::ArgumentListType &Args,
const SmallVectorImpl<unsigned> &VRegs) const {
if (!EMIT_IMPLEMENTATION)
return false;
MachineFunction &MF = MIRBuilder.getMF();
const Function &F = *MF.getFunction();
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(F.getCallingConv(), F.isVarArg(), MF, ArgLocs, F.getContext());
unsigned NumArgs = Args.size();
Function::const_arg_iterator CurOrigArg = Args.begin();
const AArch64TargetLowering &TLI = *getTLI<AArch64TargetLowering>();
for (unsigned i = 0; i != NumArgs; ++i, ++CurOrigArg) {
MVT ValVT = MVT::getVT(CurOrigArg->getType());
CCAssignFn *AssignFn =
TLI.CCAssignFnForCall(F.getCallingConv(), /*IsVarArg=*/false);
bool Res =
AssignFn(i, ValVT, ValVT, CCValAssign::Full, ISD::ArgFlagsTy(), CCInfo);
assert(!Res && "Call operand has unhandled type");
(void)Res;
}
assert(ArgLocs.size() == Args.size() &&
"We have a different number of location and args?!");
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
CCValAssign &VA = ArgLocs[i];
assert(VA.isRegLoc() && "Not yet implemented");
// Transform the arguments in physical registers into virtual ones.
MIRBuilder.getMBB().addLiveIn(VA.getLocReg());
MIRBuilder.buildInstr(TargetOpcode::COPY, VRegs[i], VA.getLocReg());
switch (VA.getLocInfo()) {
default:
llvm_unreachable("Unknown loc info!");
case CCValAssign::Full:
break;
case CCValAssign::BCvt:
// We don't care about bitcast.
break;
case CCValAssign::AExt:
case CCValAssign::SExt:
case CCValAssign::ZExt:
// Zero/Sign extend the register.
assert(0 && "Not yet implemented");
break;
}
}
return true;
}
void llvm::copyFunctionBody(Function &New, const Function &Orig,
ValueToValueMapTy &VMap) {
if (!Orig.isDeclaration()) {
Function::arg_iterator DestI = New.arg_begin();
for (Function::const_arg_iterator J = Orig.arg_begin(); J != Orig.arg_end();
++J) {
DestI->setName(J->getName());
VMap[J] = DestI++;
}
SmallVector<ReturnInst *, 8> Returns; // Ignore returns cloned.
CloneFunctionInto(&New, &Orig, VMap, /*ModuleLevelChanges=*/true, Returns);
}
}
/// AddFastCallStdCallSuffix - Microsoft fastcall and stdcall functions require
/// a suffix on their name indicating the number of words of arguments they
/// take.
static void AddFastCallStdCallSuffix(SmallVectorImpl<char> &OutName,
const Function *F, const DataLayout &TD) {
// Calculate arguments size total.
unsigned ArgWords = 0;
for (Function::const_arg_iterator AI = F->arg_begin(), AE = F->arg_end();
AI != AE; ++AI) {
Type *Ty = AI->getType();
// 'Dereference' type in case of byval parameter attribute
if (AI->hasByValAttr())
Ty = cast<PointerType>(Ty)->getElementType();
// Size should be aligned to DWORD boundary
ArgWords += ((TD.getTypeAllocSize(Ty) + 3)/4)*4;
}
raw_svector_ostream(OutName) << '@' << ArgWords;
}
void PIC16AsmPrinter::EmitFunctionFrame(MachineFunction &MF) {
const Function *F = MF.getFunction();
const TargetData *TD = TM.getTargetData();
// Emit the data section name.
O << "\n";
PIC16Section *fPDataSection =
const_cast<PIC16Section *>(getObjFileLowering().
SectionForFrame(CurrentFnSym->getName()));
fPDataSection->setColor(getFunctionColor(F));
OutStreamer.SwitchSection(fPDataSection);
// Emit function frame label
O << PAN::getFrameLabel(CurrentFnSym->getName()) << ":\n";
const Type *RetType = F->getReturnType();
unsigned RetSize = 0;
if (RetType->getTypeID() != Type::VoidTyID)
RetSize = TD->getTypeAllocSize(RetType);
//Emit function return value space
// FIXME: Do not emit RetvalLable when retsize is zero. To do this
// we will need to avoid printing a global directive for Retval label
// in emitExternandGloblas.
if(RetSize > 0)
O << PAN::getRetvalLabel(CurrentFnSym->getName())
<< " RES " << RetSize << "\n";
else
O << PAN::getRetvalLabel(CurrentFnSym->getName()) << ": \n";
// Emit variable to hold the space for function arguments
unsigned ArgSize = 0;
for (Function::const_arg_iterator argi = F->arg_begin(),
arge = F->arg_end(); argi != arge ; ++argi) {
const Type *Ty = argi->getType();
ArgSize += TD->getTypeAllocSize(Ty);
}
O << PAN::getArgsLabel(CurrentFnSym->getName()) << " RES " << ArgSize << "\n";
// Emit temporary space
int TempSize = PTLI->GetTmpSize();
if (TempSize > 0)
O << PAN::getTempdataLabel(CurrentFnSym->getName()) << " RES "
<< TempSize << '\n';
}
/// Microsoft fastcall and stdcall functions require a suffix on their name
/// indicating the number of words of arguments they take.
static void addByteCountSuffix(raw_ostream &OS, const Function *F,
const DataLayout &DL) {
// Calculate arguments size total.
unsigned ArgWords = 0;
for (Function::const_arg_iterator AI = F->arg_begin(), AE = F->arg_end();
AI != AE; ++AI) {
Type *Ty = AI->getType();
// 'Dereference' type in case of byval or inalloca parameter attribute.
if (AI->hasByValOrInAllocaAttr())
Ty = cast<PointerType>(Ty)->getElementType();
// Size should be aligned to pointer size.
unsigned PtrSize = DL.getPointerSize();
ArgWords += RoundUpToAlignment(DL.getTypeAllocSize(Ty), PtrSize);
}
OS << '@' << ArgWords;
}
void PIC16AsmPrinter::EmitFunctionFrame(MachineFunction &MF) {
const Function *F = MF.getFunction();
std::string FuncName = Mang->getValueName(F);
const TargetData *TD = TM.getTargetData();
// Emit the data section name.
O << "\n";
const char *SectionName = PAN::getFrameSectionName(CurrentFnName).c_str();
const Section *fPDataSection = TAI->getNamedSection(SectionName,
SectionFlags::Writeable);
SwitchToSection(fPDataSection);
// Emit function frame label
O << PAN::getFrameLabel(CurrentFnName) << ":\n";
const Type *RetType = F->getReturnType();
unsigned RetSize = 0;
if (RetType->getTypeID() != Type::VoidTyID)
RetSize = TD->getTypeAllocSize(RetType);
//Emit function return value space
// FIXME: Do not emit RetvalLable when retsize is zero. To do this
// we will need to avoid printing a global directive for Retval label
// in emitExternandGloblas.
if(RetSize > 0)
O << PAN::getRetvalLabel(CurrentFnName) << " RES " << RetSize << "\n";
else
O << PAN::getRetvalLabel(CurrentFnName) << ": \n";
// Emit variable to hold the space for function arguments
unsigned ArgSize = 0;
for (Function::const_arg_iterator argi = F->arg_begin(),
arge = F->arg_end(); argi != arge ; ++argi) {
const Type *Ty = argi->getType();
ArgSize += TD->getTypeAllocSize(Ty);
}
O << PAN::getArgsLabel(CurrentFnName) << " RES " << ArgSize << "\n";
// Emit temporary space
int TempSize = PTLI->GetTmpSize();
if (TempSize > 0 )
O << PAN::getTempdataLabel(CurrentFnName) << " RES " << TempSize <<"\n";
}
std::vector<Type*> lowerJuliaArrayArguments(Function *OldFunc) {
Module* M = OldFunc->getParent();
LLVMContext &context = M->getContext();
NamedMDNode* JuliaArgs = M->getOrInsertNamedMetadata("julia.args");
MDNode *node = JuliaArgs->getOperand(0);
int operand = 0;
std::vector<Type*> ArgTypes;
for (Function::const_arg_iterator I = OldFunc->arg_begin(), E = OldFunc->arg_end(); I != E; ++I) {
Type* argType = I->getType();
if (is_jl_array_type(argType)) {
// Gets the type from custom metadata
Value *value = node->getOperand(operand);
if (MDString* mdstring = dyn_cast<MDString>(value)) {
if (Type* type = extractType(context, mdstring->getString())) {
ArgTypes.push_back(type);
} else {
errs() << "Could not extract type: ";
mdstring->print(errs());
errs() << "\n";
exit(1);
}
} else {
errs() << "Could not extract type: ";
value->print(errs());
errs() << "\n";
exit(1);
}
} else {
ArgTypes.push_back(I->getType());
}
operand++;
}
return ArgTypes;
}
/// Identify lowered values that originated from f128 arguments and record
/// this.
void MipsCCState::PreAnalyzeFormalArgumentsForF128(
const SmallVectorImpl<ISD::InputArg> &Ins) {
const MachineFunction &MF = getMachineFunction();
for (unsigned i = 0; i < Ins.size(); ++i) {
Function::const_arg_iterator FuncArg = MF.getFunction()->arg_begin();
// SRet arguments cannot originate from f128 or {f128} returns so we just
// push false. We have to handle this specially since SRet arguments
// aren't mapped to an original argument.
if (Ins[i].Flags.isSRet()) {
OriginalArgWasF128.push_back(false);
OriginalArgWasFloat.push_back(false);
continue;
}
assert(Ins[i].getOrigArgIndex() < MF.getFunction()->arg_size());
std::advance(FuncArg, Ins[i].getOrigArgIndex());
OriginalArgWasF128.push_back(
originalTypeIsF128(FuncArg->getType(), nullptr));
OriginalArgWasFloat.push_back(FuncArg->getType()->isFloatingPointTy());
}
}
std::unique_ptr<Module> llvm::CloneModule(
const Module *M, ValueToValueMapTy &VMap,
std::function<bool(const GlobalValue *)> ShouldCloneDefinition) {
// First off, we need to create the new module.
std::unique_ptr<Module> New =
llvm::make_unique<Module>(M->getModuleIdentifier(), M->getContext());
New->setDataLayout(M->getDataLayout());
New->setTargetTriple(M->getTargetTriple());
New->setModuleInlineAsm(M->getModuleInlineAsm());
// Loop over all of the global variables, making corresponding globals in the
// new module. Here we add them to the VMap and to the new Module. We
// don't worry about attributes or initializers, they will come later.
//
for (Module::const_global_iterator I = M->global_begin(), E = M->global_end();
I != E; ++I) {
GlobalVariable *GV = new GlobalVariable(*New,
I->getValueType(),
I->isConstant(), I->getLinkage(),
(Constant*) nullptr, I->getName(),
(GlobalVariable*) nullptr,
I->getThreadLocalMode(),
I->getType()->getAddressSpace());
GV->copyAttributesFrom(&*I);
VMap[&*I] = GV;
}
// Loop over the functions in the module, making external functions as before
for (Module::const_iterator I = M->begin(), E = M->end(); I != E; ++I) {
Function *NF =
Function::Create(cast<FunctionType>(I->getValueType()),
I->getLinkage(), I->getName(), New.get());
NF->copyAttributesFrom(&*I);
VMap[&*I] = NF;
}
// Loop over the aliases in the module
for (Module::const_alias_iterator I = M->alias_begin(), E = M->alias_end();
I != E; ++I) {
if (!ShouldCloneDefinition(&*I)) {
// An alias cannot act as an external reference, so we need to create
// either a function or a global variable depending on the value type.
// FIXME: Once pointee types are gone we can probably pick one or the
// other.
GlobalValue *GV;
if (I->getValueType()->isFunctionTy())
GV = Function::Create(cast<FunctionType>(I->getValueType()),
GlobalValue::ExternalLinkage, I->getName(),
New.get());
else
GV = new GlobalVariable(
*New, I->getValueType(), false, GlobalValue::ExternalLinkage,
(Constant *)nullptr, I->getName(), (GlobalVariable *)nullptr,
I->getThreadLocalMode(), I->getType()->getAddressSpace());
VMap[&*I] = GV;
// We do not copy attributes (mainly because copying between different
// kinds of globals is forbidden), but this is generally not required for
// correctness.
continue;
}
auto *GA = GlobalAlias::create(I->getValueType(),
I->getType()->getPointerAddressSpace(),
I->getLinkage(), I->getName(), New.get());
GA->copyAttributesFrom(&*I);
VMap[&*I] = GA;
}
// Now that all of the things that global variable initializer can refer to
// have been created, loop through and copy the global variable referrers
// over... We also set the attributes on the global now.
//
for (Module::const_global_iterator I = M->global_begin(), E = M->global_end();
I != E; ++I) {
GlobalVariable *GV = cast<GlobalVariable>(VMap[&*I]);
if (!ShouldCloneDefinition(&*I)) {
// Skip after setting the correct linkage for an external reference.
GV->setLinkage(GlobalValue::ExternalLinkage);
continue;
}
if (I->hasInitializer())
GV->setInitializer(MapValue(I->getInitializer(), VMap));
}
// Similarly, copy over function bodies now...
//
for (Module::const_iterator I = M->begin(), E = M->end(); I != E; ++I) {
Function *F = cast<Function>(VMap[&*I]);
if (!ShouldCloneDefinition(&*I)) {
// Skip after setting the correct linkage for an external reference.
F->setLinkage(GlobalValue::ExternalLinkage);
// Personality function is not valid on a declaration.
F->setPersonalityFn(nullptr);
continue;
}
if (!I->isDeclaration()) {
Function::arg_iterator DestI = F->arg_begin();
for (Function::const_arg_iterator J = I->arg_begin(); J != I->arg_end();
++J) {
DestI->setName(J->getName());
VMap[&*J] = &*DestI++;
}
SmallVector<ReturnInst*, 8> Returns; // Ignore returns cloned.
CloneFunctionInto(F, &*I, VMap, /*ModuleLevelChanges=*/true, Returns);
}
if (I->hasPersonalityFn())
F->setPersonalityFn(MapValue(I->getPersonalityFn(), VMap));
}
// And aliases
for (Module::const_alias_iterator I = M->alias_begin(), E = M->alias_end();
I != E; ++I) {
// We already dealt with undefined aliases above.
if (!ShouldCloneDefinition(&*I))
continue;
GlobalAlias *GA = cast<GlobalAlias>(VMap[&*I]);
if (const Constant *C = I->getAliasee())
GA->setAliasee(MapValue(C, VMap));
}
// And named metadata....
for (Module::const_named_metadata_iterator I = M->named_metadata_begin(),
E = M->named_metadata_end(); I != E; ++I) {
const NamedMDNode &NMD = *I;
NamedMDNode *NewNMD = New->getOrInsertNamedMetadata(NMD.getName());
for (unsigned i = 0, e = NMD.getNumOperands(); i != e; ++i)
NewNMD->addOperand(MapMetadata(NMD.getOperand(i), VMap));
}
return New;
}
static bool ffiInvoke(RawFunc Fn, Function *F, ArrayRef<GenericValue> ArgVals,
const DataLayout &TD, GenericValue &Result) {
ffi_cif cif;
FunctionType *FTy = F->getFunctionType();
const unsigned NumArgs = F->arg_size();
// TODO: We don't have type information about the remaining arguments, because
// this information is never passed into ExecutionEngine::runFunction().
if (ArgVals.size() > NumArgs && F->isVarArg()) {
report_fatal_error("Calling external var arg function '" + F->getName()
+ "' is not supported by the Interpreter.");
}
unsigned ArgBytes = 0;
std::vector<ffi_type*> args(NumArgs);
for (Function::const_arg_iterator A = F->arg_begin(), E = F->arg_end();
A != E; ++A) {
const unsigned ArgNo = A->getArgNo();
Type *ArgTy = FTy->getParamType(ArgNo);
args[ArgNo] = ffiTypeFor(ArgTy);
ArgBytes += TD.getTypeStoreSize(ArgTy);
}
SmallVector<uint8_t, 128> ArgData;
ArgData.resize(ArgBytes);
uint8_t *ArgDataPtr = ArgData.data();
SmallVector<void*, 16> values(NumArgs);
for (Function::const_arg_iterator A = F->arg_begin(), E = F->arg_end();
A != E; ++A) {
const unsigned ArgNo = A->getArgNo();
Type *ArgTy = FTy->getParamType(ArgNo);
values[ArgNo] = ffiValueFor(ArgTy, ArgVals[ArgNo], ArgDataPtr);
ArgDataPtr += TD.getTypeStoreSize(ArgTy);
}
Type *RetTy = FTy->getReturnType();
ffi_type *rtype = ffiTypeFor(RetTy);
if (ffi_prep_cif(&cif, FFI_DEFAULT_ABI, NumArgs, rtype, args.data()) ==
FFI_OK) {
SmallVector<uint8_t, 128> ret;
if (RetTy->getTypeID() != Type::VoidTyID)
ret.resize(TD.getTypeStoreSize(RetTy));
ffi_call(&cif, Fn, ret.data(), values.data());
switch (RetTy->getTypeID()) {
case Type::IntegerTyID:
switch (cast<IntegerType>(RetTy)->getBitWidth()) {
case 8: Result.IntVal = APInt(8 , *(int8_t *) ret.data()); break;
case 16: Result.IntVal = APInt(16, *(int16_t*) ret.data()); break;
case 32: Result.IntVal = APInt(32, *(int32_t*) ret.data()); break;
case 64: Result.IntVal = APInt(64, *(int64_t*) ret.data()); break;
}
break;
case Type::FloatTyID: Result.FloatVal = *(float *) ret.data(); break;
case Type::DoubleTyID: Result.DoubleVal = *(double*) ret.data(); break;
case Type::PointerTyID: Result.PointerVal = *(void **) ret.data(); break;
default: break;
}
return true;
}
return false;
}
/* ************************************************************************** */
bool RangedAddressSanitizer::doInitialization(Module &M)
{
// Link FastAddressSanitizer functions into the target module
LLVMContext & context = M.getContext();
const char * fasanPath = getenv("FASANMODULE");
if (! fasanPath) {
return false;
}
std::stringstream ss;
ss << fasanPath;
SMDiagnostic diag;
Module * fasanModule = ParseIRFile(ss.str(), diag, context);
if (!fasanModule) {
abort();
}
#if 0 /* using LLVM linking facilities */
Linker linker(&M);
std::string linkErr;
if (linker.linkInModule(fasanModule, Linker::DestroySource, &linkErr)) {
errs() << "[FASAN] Error while linking runtime module: " << fasanModule << "(!!)\n";
abort();
}
#else
PointerType * voidPtrTy = PointerType::getInt8PtrTy(context, 0);
IntegerType * boolTy = IntegerType::get(context, 1);
Type * voidTy = Type::getVoidTy(context);
FunctionType * touchFunType = FunctionType::get(voidTy, ArrayRef<Type*>(voidPtrTy), false);
FunctionType * verifyFunType = FunctionType::get(boolTy, ArrayRef<Type*>(voidPtrTy), false);
ValueToValueMapTy reMap;
reMap[fasanModule->getFunction("__fasan_touch")] = M.getOrInsertFunction("__fasan_touch", touchFunType);
reMap[fasanModule->getFunction("__fasan_verify")] = M.getOrInsertFunction("__fasan_verify", verifyFunType);
// migrate check function
{
std::string errMsg;
Function * checkFunc = fasanModule->getFunction("__fasan_check");
if (!checkFunc) {
abort();
}
#if 1
FunctionType * checkFuncType = checkFunc->getFunctionType();
Function * targetFunc = dyn_cast<Function>(M.getOrInsertFunction("__fasan_check", checkFuncType));
assert(targetFunc && "function cast to const by getOrInsertFunc..?");
// Loop over the arguments, copying the names of the mapped arguments over...
Function::arg_iterator DestI = targetFunc->arg_begin();
for (Function::const_arg_iterator I = checkFunc->arg_begin(), E = checkFunc->arg_end();
I != E; ++I)
if (reMap.count(I) == 0) { // Is this argument preserved?
DestI->setName(I->getName()); // Copy the name over...
reMap[I] = DestI++; // Add mapping to VMap
}
SmallVector<ReturnInst*, 8> Returns; // Ignore returns cloned.
CloneFunctionInto(targetFunc, checkFunc, reMap, false, Returns, "", nullptr);
targetFunc->addAttribute(0,Attribute::SanitizeAddress);
#else
Function * clonedCheckFunc = CloneFunction(checkFunc, reMap, false, 0);
assert(!M.getFunction("__fasan_check") && "already exists in module");
M.getFunctionList().push_back(clonedCheckFunc);
ReuseFn_ = clonedCheckFunc;
clonedCheckFunc->setLinkage(GlobalValue::InternalLinkage); // avoid conflicts during linking
// re-map fake use to local copy
for (auto & BB : *clonedCheckFunc) {
for (auto & Inst : BB) {
RemapInstruction(&Inst, reMap, RF_IgnoreMissingEntries, 0, 0);
}
}
#endif
#endif
}
delete fasanModule;
return true;
}
/// NaClValueEnumerator - Enumerate module-level information.
NaClValueEnumerator::NaClValueEnumerator(const Module *M) {
// Create map for counting frequency of types, and set field
// TypeCountMap accordingly. Note: Pointer field TypeCountMap is
// used to deal with the fact that types are added through various
// method calls in this routine. Rather than pass it as an argument,
// we use a field. The field is a pointer so that the memory
// footprint of count_map can be garbage collected when this
// constructor completes.
TypeCountMapType count_map;
TypeCountMap = &count_map;
IntPtrType = IntegerType::get(M->getContext(), PNaClIntPtrTypeBitSize);
// Enumerate the functions. Note: We do this before global
// variables, so that global variable initializations can refer to
// the functions without a forward reference.
for (Module::const_iterator I = M->begin(), E = M->end(); I != E; ++I) {
EnumerateValue(I);
}
// Enumerate the global variables.
FirstGlobalVarID = Values.size();
for (Module::const_global_iterator I = M->global_begin(),
E = M->global_end(); I != E; ++I)
EnumerateValue(I);
NumGlobalVarIDs = Values.size() - FirstGlobalVarID;
// Enumerate the aliases.
for (Module::const_alias_iterator I = M->alias_begin(), E = M->alias_end();
I != E; ++I)
EnumerateValue(I);
// Remember what is the cutoff between globalvalue's and other constants.
unsigned FirstConstant = Values.size();
// Skip global variable initializers since they are handled within
// WriteGlobalVars of file NaClBitcodeWriter.cpp.
// Enumerate the aliasees.
for (Module::const_alias_iterator I = M->alias_begin(), E = M->alias_end();
I != E; ++I)
EnumerateValue(I->getAliasee());
// Insert constants that are named at module level into the slot
// pool so that the module symbol table can refer to them...
EnumerateValueSymbolTable(M->getValueSymbolTable());
// Enumerate types used by function bodies and argument lists.
for (Module::const_iterator F = M->begin(), E = M->end(); F != E; ++F) {
for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end();
I != E; ++I)
EnumerateType(I->getType());
for (Function::const_iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); I!=E;++I){
// Don't generate types for elided pointer casts!
if (IsElidedCast(I))
continue;
if (const SwitchInst *SI = dyn_cast<SwitchInst>(I)) {
// Handle switch instruction specially, so that we don't
// write out unnecessary vector/array types used to model case
// selectors.
EnumerateOperandType(SI->getCondition());
} else {
for (User::const_op_iterator OI = I->op_begin(), E = I->op_end();
OI != E; ++OI) {
EnumerateOperandType(*OI);
}
}
EnumerateType(I->getType());
}
}
// Optimized type indicies to put "common" expected types in with small
// indices.
OptimizeTypes(M);
TypeCountMap = NULL;
// Optimize constant ordering.
OptimizeConstants(FirstConstant, Values.size());
}
// 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;
}
Module *llvm::CloneModule(const Module *M, ValueToValueMapTy &VMap) {
// First off, we need to create the new module.
Module *New = new Module(M->getModuleIdentifier(), M->getContext());
New->setDataLayout(M->getDataLayout());
New->setTargetTriple(M->getTargetTriple());
New->setModuleInlineAsm(M->getModuleInlineAsm());
// Loop over all of the global variables, making corresponding globals in the
// new module. Here we add them to the VMap and to the new Module. We
// don't worry about attributes or initializers, they will come later.
//
for (Module::const_global_iterator I = M->global_begin(), E = M->global_end();
I != E; ++I) {
GlobalVariable *GV = new GlobalVariable(*New,
I->getType()->getElementType(),
I->isConstant(), I->getLinkage(),
(Constant*) nullptr, I->getName(),
(GlobalVariable*) nullptr,
I->getThreadLocalMode(),
I->getType()->getAddressSpace());
GV->copyAttributesFrom(I);
VMap[I] = GV;
}
// Loop over the functions in the module, making external functions as before
for (Module::const_iterator I = M->begin(), E = M->end(); I != E; ++I) {
Function *NF =
Function::Create(cast<FunctionType>(I->getType()->getElementType()),
I->getLinkage(), I->getName(), New);
NF->copyAttributesFrom(I);
VMap[I] = NF;
}
// Loop over the aliases in the module
for (Module::const_alias_iterator I = M->alias_begin(), E = M->alias_end();
I != E; ++I) {
auto *PTy = cast<PointerType>(I->getType());
auto *GA =
GlobalAlias::create(PTy->getElementType(), PTy->getAddressSpace(),
I->getLinkage(), I->getName(), New);
GA->copyAttributesFrom(I);
VMap[I] = GA;
}
// Now that all of the things that global variable initializer can refer to
// have been created, loop through and copy the global variable referrers
// over... We also set the attributes on the global now.
//
for (Module::const_global_iterator I = M->global_begin(), E = M->global_end();
I != E; ++I) {
GlobalVariable *GV = cast<GlobalVariable>(VMap[I]);
if (I->hasInitializer())
GV->setInitializer(MapValue(I->getInitializer(), VMap));
}
// Similarly, copy over function bodies now...
//
for (Module::const_iterator I = M->begin(), E = M->end(); I != E; ++I) {
Function *F = cast<Function>(VMap[I]);
if (!I->isDeclaration()) {
Function::arg_iterator DestI = F->arg_begin();
for (Function::const_arg_iterator J = I->arg_begin(); J != I->arg_end();
++J) {
DestI->setName(J->getName());
VMap[J] = DestI++;
}
SmallVector<ReturnInst*, 8> Returns; // Ignore returns cloned.
CloneFunctionInto(F, I, VMap, /*ModuleLevelChanges=*/true, Returns);
}
}
// And aliases
for (Module::const_alias_iterator I = M->alias_begin(), E = M->alias_end();
I != E; ++I) {
GlobalAlias *GA = cast<GlobalAlias>(VMap[I]);
if (const Constant *C = I->getAliasee())
GA->setAliasee(cast<GlobalObject>(MapValue(C, VMap)));
}
// And named metadata....
for (Module::const_named_metadata_iterator I = M->named_metadata_begin(),
E = M->named_metadata_end(); I != E; ++I) {
const NamedMDNode &NMD = *I;
NamedMDNode *NewNMD = New->getOrInsertNamedMetadata(NMD.getName());
for (unsigned i = 0, e = NMD.getNumOperands(); i != e; ++i)
NewNMD->addOperand(MapValue(NMD.getOperand(i), VMap));
}
return New;
}
///
/// Method: visitIntrinsic()
///
/// Description:
/// Generate correct DSNodes for calls to LLVM intrinsic functions.
///
/// Inputs:
/// CS - The CallSite representing the call or invoke to the intrinsic.
/// F - A pointer to the function called by the call site.
///
/// Return value:
/// true - This intrinsic is properly handled by this method.
/// false - This intrinsic is not recognized by DSA.
///
bool GraphBuilder::visitIntrinsic(CallSite CS, Function *F) {
++NumIntrinsicCall;
//
// If this is a debug intrinsic, then don't do any special processing.
//
if (isa<DbgInfoIntrinsic>(CS.getInstruction()))
return true;
switch (F->getIntrinsicID()) {
case Intrinsic::vastart: {
visitVAStartInst(CS);
return true;
}
case Intrinsic::vacopy: {
// Simply merge the two arguments to va_copy.
// This results in loss of precision on the temporaries used to manipulate
// the va_list, and so isn't a big deal. In theory we would build a
// separate graph for this (like the one created in visitVAStartNode)
// and only merge the node containing the variable arguments themselves.
DSNodeHandle destNH = getValueDest(CS.getArgument(0));
DSNodeHandle srcNH = getValueDest(CS.getArgument(1));
destNH.mergeWith(srcNH);
return true;
}
case Intrinsic::stacksave: {
DSNode * Node = createNode();
Node->setAllocaMarker()->setIncompleteMarker()->setUnknownMarker();
Node->foldNodeCompletely();
setDestTo (*(CS.getInstruction()), Node);
return true;
}
case Intrinsic::stackrestore:
getValueDest(CS.getInstruction()).getNode()->setAllocaMarker()
->setIncompleteMarker()
->setUnknownMarker()
->foldNodeCompletely();
return true;
case Intrinsic::vaend:
case Intrinsic::memcpy:
case Intrinsic::memmove: {
// Merge the first & second arguments, and mark the memory read and
// modified.
DSNodeHandle RetNH = getValueDest(*CS.arg_begin());
RetNH.mergeWith(getValueDest(*(CS.arg_begin()+1)));
if (DSNode *N = RetNH.getNode())
N->setModifiedMarker()->setReadMarker();
return true;
}
case Intrinsic::memset:
// Mark the memory modified.
if (DSNode *N = getValueDest(*CS.arg_begin()).getNode())
N->setModifiedMarker();
return true;
case Intrinsic::eh_exception: {
DSNode * Node = createNode();
Node->setIncompleteMarker();
Node->foldNodeCompletely();
setDestTo (*(CS.getInstruction()), Node);
return true;
}
case Intrinsic::eh_selector: {
for (CallSite::arg_iterator I = CS.arg_begin(), E = CS.arg_end();
I != E; ++I) {
if (isa<PointerType>((*I)->getType())) {
DSNodeHandle Ptr = getValueDest(*I);
if(Ptr.getNode()) {
Ptr.getNode()->setReadMarker();
Ptr.getNode()->setIncompleteMarker();
}
}
}
return true;
}
case Intrinsic::eh_typeid_for: {
DSNodeHandle Ptr = getValueDest(*CS.arg_begin());
Ptr.getNode()->setReadMarker();
Ptr.getNode()->setIncompleteMarker();
return true;
}
case Intrinsic::prefetch:
return true;
case Intrinsic::objectsize:
return true;
//
// The return address/frame address aliases with the stack,
// is type-unknown, and should
// have the unknown flag set since we don't know where it goes.
//
case Intrinsic::returnaddress:
case Intrinsic::frameaddress: {
DSNode * Node = createNode();
Node->setAllocaMarker()->setIncompleteMarker()->setUnknownMarker();
Node->foldNodeCompletely();
setDestTo (*(CS.getInstruction()), Node);
return true;
}
// Process lifetime intrinsics
case Intrinsic::lifetime_start:
case Intrinsic::lifetime_end:
case Intrinsic::invariant_start:
case Intrinsic::invariant_end:
return true;
default: {
//ignore pointer free intrinsics
if (!isa<PointerType>(F->getReturnType())) {
bool hasPtr = false;
for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end();
I != E && !hasPtr; ++I)
if (isa<PointerType>(I->getType()))
hasPtr = true;
if (!hasPtr)
return true;
}
DEBUG(errs() << "[dsa:local] Unhandled intrinsic: " << F->getName() << "\n");
assert(0 && "Unhandled intrinsic");
return false;
}
}
}
/// ValueEnumerator - Enumerate module-level information.
ValueEnumerator::ValueEnumerator(const Module *M) {
// Enumerate the global variables.
for (Module::const_global_iterator I = M->global_begin(),
E = M->global_end(); I != E; ++I)
EnumerateValue(I);
// Enumerate the functions.
for (Module::const_iterator I = M->begin(), E = M->end(); I != E; ++I) {
EnumerateValue(I);
EnumerateAttributes(cast<Function>(I)->getAttributes());
}
// Enumerate the aliases.
for (Module::const_alias_iterator I = M->alias_begin(), E = M->alias_end();
I != E; ++I)
EnumerateValue(I);
// Remember what is the cutoff between globalvalue's and other constants.
unsigned FirstConstant = Values.size();
// Enumerate the global variable initializers.
for (Module::const_global_iterator I = M->global_begin(),
E = M->global_end(); I != E; ++I)
if (I->hasInitializer())
EnumerateValue(I->getInitializer());
// Enumerate the aliasees.
for (Module::const_alias_iterator I = M->alias_begin(), E = M->alias_end();
I != E; ++I)
EnumerateValue(I->getAliasee());
// Insert constants and metadata that are named at module level into the slot
// pool so that the module symbol table can refer to them...
EnumerateValueSymbolTable(M->getValueSymbolTable());
EnumerateNamedMetadata(M);
SmallVector<std::pair<unsigned, MDNode*>, 8> MDs;
// Enumerate types used by function bodies and argument lists.
for (Module::const_iterator F = M->begin(), E = M->end(); F != E; ++F) {
for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end();
I != E; ++I)
EnumerateType(I->getType());
for (Function::const_iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); I!=E;++I){
for (User::const_op_iterator OI = I->op_begin(), E = I->op_end();
OI != E; ++OI) {
if (MDNode *MD = dyn_cast<MDNode>(*OI))
if (MD->isFunctionLocal() && MD->getFunction())
// These will get enumerated during function-incorporation.
continue;
EnumerateOperandType(*OI);
}
EnumerateType(I->getType());
if (const CallInst *CI = dyn_cast<CallInst>(I))
EnumerateAttributes(CI->getAttributes());
else if (const InvokeInst *II = dyn_cast<InvokeInst>(I))
EnumerateAttributes(II->getAttributes());
// Enumerate metadata attached with this instruction.
MDs.clear();
I->getAllMetadataOtherThanDebugLoc(MDs);
for (unsigned i = 0, e = MDs.size(); i != e; ++i)
EnumerateMetadata(MDs[i].second);
if (!I->getDebugLoc().isUnknown()) {
MDNode *Scope, *IA;
I->getDebugLoc().getScopeAndInlinedAt(Scope, IA, I->getContext());
if (Scope) EnumerateMetadata(Scope);
if (IA) EnumerateMetadata(IA);
}
}
}
// Optimize constant ordering.
OptimizeConstants(FirstConstant, Values.size());
}
/// NaClValueEnumerator - Enumerate module-level information.
NaClValueEnumerator::NaClValueEnumerator(const Module *M) {
// Create map for counting frequency of types, and set field
// TypeCountMap accordingly. Note: Pointer field TypeCountMap is
// used to deal with the fact that types are added through various
// method calls in this routine. Rather than pass it as an argument,
// we use a field. The field is a pointer so that the memory
// footprint of count_map can be garbage collected when this
// constructor completes.
TypeCountMapType count_map;
TypeCountMap = &count_map;
// Enumerate the global variables.
for (Module::const_global_iterator I = M->global_begin(),
E = M->global_end(); I != E; ++I)
EnumerateValue(I);
// Enumerate the functions.
for (Module::const_iterator I = M->begin(), E = M->end(); I != E; ++I) {
EnumerateValue(I);
EnumerateAttributes(cast<Function>(I)->getAttributes());
}
// Enumerate the aliases.
for (Module::const_alias_iterator I = M->alias_begin(), E = M->alias_end();
I != E; ++I)
EnumerateValue(I);
// Remember what is the cutoff between globalvalue's and other constants.
unsigned FirstConstant = Values.size();
// Enumerate the global variable initializers.
for (Module::const_global_iterator I = M->global_begin(),
E = M->global_end(); I != E; ++I)
if (I->hasInitializer())
EnumerateValue(I->getInitializer());
// Enumerate the aliasees.
for (Module::const_alias_iterator I = M->alias_begin(), E = M->alias_end();
I != E; ++I)
EnumerateValue(I->getAliasee());
// Insert constants and metadata that are named at module level into the slot
// pool so that the module symbol table can refer to them...
EnumerateValueSymbolTable(M->getValueSymbolTable());
EnumerateNamedMetadata(M);
SmallVector<std::pair<unsigned, MDNode*>, 8> MDs;
// Enumerate types used by function bodies and argument lists.
for (Module::const_iterator F = M->begin(), E = M->end(); F != E; ++F) {
for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end();
I != E; ++I)
EnumerateType(I->getType());
for (Function::const_iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); I!=E;++I){
for (User::const_op_iterator OI = I->op_begin(), E = I->op_end();
OI != E; ++OI) {
if (MDNode *MD = dyn_cast<MDNode>(*OI))
if (MD->isFunctionLocal() && MD->getFunction())
// These will get enumerated during function-incorporation.
continue;
EnumerateOperandType(*OI);
}
EnumerateType(I->getType());
if (const CallInst *CI = dyn_cast<CallInst>(I))
EnumerateAttributes(CI->getAttributes());
else if (const InvokeInst *II = dyn_cast<InvokeInst>(I))
EnumerateAttributes(II->getAttributes());
// Enumerate metadata attached with this instruction.
MDs.clear();
I->getAllMetadataOtherThanDebugLoc(MDs);
for (unsigned i = 0, e = MDs.size(); i != e; ++i)
EnumerateMetadata(MDs[i].second);
if (!I->getDebugLoc().isUnknown()) {
MDNode *Scope, *IA;
I->getDebugLoc().getScopeAndInlinedAt(Scope, IA, I->getContext());
if (Scope) EnumerateMetadata(Scope);
if (IA) EnumerateMetadata(IA);
}
}
}
// Optimized type indicies to put "common" expected types in with small
// indices.
OptimizeTypes(M);
TypeCountMap = NULL;
// Optimize constant ordering.
OptimizeConstants(FirstConstant, Values.size());
}
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());
}
}
/// ValueEnumerator - Enumerate module-level information.
ValueEnumerator::ValueEnumerator(const Module *M) {
InstructionCount = 0;
// Enumerate the global variables.
for (Module::const_global_iterator I = M->global_begin(),
E = M->global_end(); I != E; ++I)
EnumerateValue(I);
// Enumerate the functions.
for (Module::const_iterator I = M->begin(), E = M->end(); I != E; ++I) {
EnumerateValue(I);
EnumerateAttributes(cast<Function>(I)->getAttributes());
}
// Enumerate the aliases.
for (Module::const_alias_iterator I = M->alias_begin(), E = M->alias_end();
I != E; ++I)
EnumerateValue(I);
// Remember what is the cutoff between globalvalue's and other constants.
unsigned FirstConstant = Values.size();
// Enumerate the global variable initializers.
for (Module::const_global_iterator I = M->global_begin(),
E = M->global_end(); I != E; ++I)
if (I->hasInitializer())
EnumerateValue(I->getInitializer());
// Enumerate the aliasees.
for (Module::const_alias_iterator I = M->alias_begin(), E = M->alias_end();
I != E; ++I)
EnumerateValue(I->getAliasee());
// Enumerate types used by the type symbol table.
EnumerateTypeSymbolTable(M->getTypeSymbolTable());
// Insert constants that are named at module level into the slot pool so that
// the module symbol table can refer to them...
EnumerateValueSymbolTable(M->getValueSymbolTable());
// Enumerate types used by function bodies and argument lists.
for (Module::const_iterator F = M->begin(), E = M->end(); F != E; ++F) {
for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end();
I != E; ++I)
EnumerateType(I->getType());
MetadataContext &TheMetadata = F->getContext().getMetadata();
typedef SmallVector<std::pair<unsigned, TrackingVH<MDNode> >, 2> MDMapTy;
MDMapTy MDs;
for (Function::const_iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); I!=E;++I){
for (User::const_op_iterator OI = I->op_begin(), E = I->op_end();
OI != E; ++OI)
EnumerateOperandType(*OI);
EnumerateType(I->getType());
if (const CallInst *CI = dyn_cast<CallInst>(I))
EnumerateAttributes(CI->getAttributes());
else if (const InvokeInst *II = dyn_cast<InvokeInst>(I))
EnumerateAttributes(II->getAttributes());
// Enumerate metadata attached with this instruction.
MDs.clear();
TheMetadata.getMDs(I, MDs);
for (MDMapTy::const_iterator MI = MDs.begin(), ME = MDs.end(); MI != ME;
++MI)
EnumerateMetadata(MI->second);
}
}
// Optimize constant ordering.
OptimizeConstants(FirstConstant, Values.size());
// Sort the type table by frequency so that most commonly used types are early
// in the table (have low bit-width).
std::stable_sort(Types.begin(), Types.end(), CompareByFrequency);
// Partition the Type ID's so that the single-value types occur before the
// aggregate types. This allows the aggregate types to be dropped from the
// type table after parsing the global variable initializers.
std::partition(Types.begin(), Types.end(), isSingleValueType);
// Now that we rearranged the type table, rebuild TypeMap.
for (unsigned i = 0, e = Types.size(); i != e; ++i)
TypeMap[Types[i].first] = i+1;
}
void CZeroInfo::depthFirstGatherer() {
// Adding the pointer values among the arguments to the alias graph
// We treat them as pointers to global targets.
for (Function::const_arg_iterator I = TheFunction.arg_begin(),
E = TheFunction.arg_end(); I != E; ++I) {
if (I->getType()->getTypeID() == Type::PointerTyID)
PointerAliasGraph.addEdge(I, PointsToTarget::GlobalTarget);
}
df_iterator<const Function*> It = df_begin(&TheFunction), End = df_end(&TheFunction);
for ( ; It != End; It++) {
const BasicBlock *BB = *It;
// Look for store instructions sequentially in the basic block
// updating pointer alias graphs for the other instructions
BasicBlock::const_iterator iterBB;
for (iterBB = BB->begin(); iterBB != BB->end(); ++iterBB) {
const Instruction& I = *iterBB;
// NOTE!!! Removed the if (I) clause here
//
if (I.hasName() &&
I.getType()->getTypeID() == Type::PointerTyID) {
// Each of these cases needs to modify the alias graph appropriately
if (isa<AllocaInst>(I)) {
PointerAliasGraph.addEdge(&I, &I);
}
else if (isa<MallocInst>(I)) {
// TODO: We'll be making this illegal and only allowing
// calls to rmalloc and rfree.
PointerAliasGraph.addEdge(&I, &I);
}
else if (isa<LoadInst>(I)) {
PointerAliasGraph.addEdge(&I, PointsToTarget::DummyTarget);
}
else if (isa<GetElementPtrInst>(I)) {
// Check if the operand is a global value, in which case we
// generate an alias to a generic global value.
if (!isa<ConstantPointerNull>(I.getOperand(0)))
if (isa<GlobalValue>(I.getOperand(0)) ||
isa<Constant>(I.getOperand(0)))
PointerAliasGraph.addEdge(&I, PointsToTarget::GlobalTarget);
else
PointerAliasGraph.addAlias(&I, I.getOperand(0));
else
PointerAliasGraph.addEdge(&I, PointsToTarget::DummyTarget);
}
else if (isa<PHINode>(I)) {
PointerAliasGraph.addEdge(&I, &I);
}
else if (isa<CallInst>(I)) {
PointerAliasGraph.addEdge(&I, PointsToTarget::GlobalTarget);
}
else if (isa<CastInst>(I)) {
PointerAliasGraph.addEdge(&I, PointsToTarget::DummyTarget);
}
}
else if (!I.hasName()) {
if (isa<StoreInst>(I)) {
// We only consider stores of scalar pointers.
if (I.getNumOperands() <= 2 ||
(I.getNumOperands() == 3 &&
I.getOperand(2) != getGlobalContext().getConstantInt(Type::Int32Ty, 0))) {
if (!isa<ConstantPointerNull>(I.getOperand(1))) {
BBPointerLiveInfo[BB][I.getOperand(1)] = true;
df_iterator<const Function*> localIt = df_begin(&TheFunction),
localEnd = df_end(&TheFunction);
for ( ; localIt != localEnd; ++localIt) {
if (DomTree->dominates((BasicBlock *) BB,
(BasicBlock *) *localIt))
BBPointerLiveInfo[*localIt][I.getOperand(1)] = true;
}
} else {
WarningsList += "Stores to null pointers disallowed in CZero\n";
}
}
}
}
}
}
}
bool TriCoreCallingConvHook::isRegValPtrType (MachineFunction& _mf) {
Function::const_arg_iterator FI;
FI = _mf.getFunction()->arg_begin();
std::advance(FI,curArg);
return FI->getType()->isPointerTy()? true : false;
}
SDValue
NVPTXTargetLowering::LowerFormalArguments(SDValue Chain,
CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins,
DebugLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) const {
MachineFunction &MF = DAG.getMachineFunction();
const DataLayout *TD = getDataLayout();
const Function *F = MF.getFunction();
const AttrListPtr &PAL = F->getAttributes();
SDValue Root = DAG.getRoot();
std::vector<SDValue> OutChains;
bool isKernel = llvm::isKernelFunction(*F);
bool isABI = (nvptxSubtarget.getSmVersion() >= 20);
std::vector<Type *> argTypes;
std::vector<const Argument *> theArgs;
for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end();
I != E; ++I) {
theArgs.push_back(I);
argTypes.push_back(I->getType());
}
assert(argTypes.size() == Ins.size() &&
"Ins types and function types did not match");
int idx = 0;
for (unsigned i=0, e=Ins.size(); i!=e; ++i, ++idx) {
Type *Ty = argTypes[i];
EVT ObjectVT = getValueType(Ty);
assert(ObjectVT == Ins[i].VT &&
"Ins type did not match function type");
// If the kernel argument is image*_t or sampler_t, convert it to
// a i32 constant holding the parameter position. This can later
// matched in the AsmPrinter to output the correct mangled name.
if (isImageOrSamplerVal(theArgs[i],
(theArgs[i]->getParent() ?
theArgs[i]->getParent()->getParent() : 0))) {
assert(isKernel && "Only kernels can have image/sampler params");
InVals.push_back(DAG.getConstant(i+1, MVT::i32));
continue;
}
if (theArgs[i]->use_empty()) {
// argument is dead
InVals.push_back(DAG.getNode(ISD::UNDEF, dl, ObjectVT));
continue;
}
// In the following cases, assign a node order of "idx+1"
// to newly created nodes. The SDNOdes for params have to
// appear in the same order as their order of appearance
// in the original function. "idx+1" holds that order.
if (PAL.getParamAttributes(i+1).hasAttribute(Attributes::ByVal) == false) {
// A plain scalar.
if (isABI || isKernel) {
// If ABI, load from the param symbol
SDValue Arg = getParamSymbol(DAG, idx);
Value *srcValue = new Argument(PointerType::get(ObjectVT.getTypeForEVT(
F->getContext()),
llvm::ADDRESS_SPACE_PARAM));
SDValue p = DAG.getLoad(ObjectVT, dl, Root, Arg,
MachinePointerInfo(srcValue), false, false,
false,
TD->getABITypeAlignment(ObjectVT.getTypeForEVT(
F->getContext())));
if (p.getNode())
DAG.AssignOrdering(p.getNode(), idx+1);
InVals.push_back(p);
}
else {
// If no ABI, just move the param symbol
SDValue Arg = getParamSymbol(DAG, idx, ObjectVT);
SDValue p = DAG.getNode(NVPTXISD::MoveParam, dl, ObjectVT, Arg);
if (p.getNode())
DAG.AssignOrdering(p.getNode(), idx+1);
InVals.push_back(p);
}
continue;
}
// Param has ByVal attribute
if (isABI || isKernel) {
// Return MoveParam(param symbol).
// Ideally, the param symbol can be returned directly,
// but when SDNode builder decides to use it in a CopyToReg(),
// machine instruction fails because TargetExternalSymbol
// (not lowered) is target dependent, and CopyToReg assumes
// the source is lowered.
SDValue Arg = getParamSymbol(DAG, idx, getPointerTy());
SDValue p = DAG.getNode(NVPTXISD::MoveParam, dl, ObjectVT, Arg);
if (p.getNode())
DAG.AssignOrdering(p.getNode(), idx+1);
if (isKernel)
InVals.push_back(p);
else {
SDValue p2 = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, ObjectVT,
DAG.getConstant(Intrinsic::nvvm_ptr_local_to_gen, MVT::i32),
p);
InVals.push_back(p2);
}
} else {
// Have to move a set of param symbols to registers and
// store them locally and return the local pointer in InVals
const PointerType *elemPtrType = dyn_cast<PointerType>(argTypes[i]);
assert(elemPtrType &&
"Byval parameter should be a pointer type");
Type *elemType = elemPtrType->getElementType();
// Compute the constituent parts
SmallVector<EVT, 16> vtparts;
SmallVector<uint64_t, 16> offsets;
ComputeValueVTs(*this, elemType, vtparts, &offsets, 0);
unsigned totalsize = 0;
for (unsigned j=0, je=vtparts.size(); j!=je; ++j)
totalsize += vtparts[j].getStoreSizeInBits();
SDValue localcopy = DAG.getFrameIndex(MF.getFrameInfo()->
CreateStackObject(totalsize/8, 16, false),
getPointerTy());
unsigned sizesofar = 0;
std::vector<SDValue> theChains;
for (unsigned j=0, je=vtparts.size(); j!=je; ++j) {
unsigned numElems = 1;
if (vtparts[j].isVector()) numElems = vtparts[j].getVectorNumElements();
for (unsigned k=0, ke=numElems; k!=ke; ++k) {
EVT tmpvt = vtparts[j];
if (tmpvt.isVector()) tmpvt = tmpvt.getVectorElementType();
SDValue arg = DAG.getNode(NVPTXISD::MoveParam, dl, tmpvt,
getParamSymbol(DAG, idx, tmpvt));
SDValue addr = DAG.getNode(ISD::ADD, dl, getPointerTy(), localcopy,
DAG.getConstant(sizesofar, getPointerTy()));
theChains.push_back(DAG.getStore(Chain, dl, arg, addr,
MachinePointerInfo(), false, false, 0));
sizesofar += tmpvt.getStoreSizeInBits()/8;
++idx;
}
}
--idx;
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, &theChains[0],
theChains.size());
InVals.push_back(localcopy);
}
}
// Clang will check explicit VarArg and issue error if any. However, Clang
// will let code with
// implicit var arg like f() pass.
// We treat this case as if the arg list is empty.
//if (F.isVarArg()) {
// assert(0 && "VarArg not supported yet!");
//}
if (!OutChains.empty())
DAG.setRoot(DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
&OutChains[0], OutChains.size()));
return Chain;
}
Module *llvm::CloneModule(const Module *M,
DenseMap<const Value*, Value*> &ValueMap) {
// First off, we need to create the new module...
Module *New = new Module(M->getModuleIdentifier());
New->setDataLayout(M->getDataLayout());
New->setTargetTriple(M->getTargetTriple());
New->setModuleInlineAsm(M->getModuleInlineAsm());
// Copy all of the type symbol table entries over.
const TypeSymbolTable &TST = M->getTypeSymbolTable();
for (TypeSymbolTable::const_iterator TI = TST.begin(), TE = TST.end();
TI != TE; ++TI)
New->addTypeName(TI->first, TI->second);
// Copy all of the dependent libraries over.
for (Module::lib_iterator I = M->lib_begin(), E = M->lib_end(); I != E; ++I)
New->addLibrary(*I);
// Loop over all of the global variables, making corresponding globals in the
// new module. Here we add them to the ValueMap and to the new Module. We
// don't worry about attributes or initializers, they will come later.
//
for (Module::const_global_iterator I = M->global_begin(), E = M->global_end();
I != E; ++I) {
GlobalVariable *GV = new GlobalVariable(I->getType()->getElementType(),
false,
GlobalValue::ExternalLinkage, 0,
I->getName(), New);
GV->setAlignment(I->getAlignment());
ValueMap[I] = GV;
}
// Loop over the functions in the module, making external functions as before
for (Module::const_iterator I = M->begin(), E = M->end(); I != E; ++I) {
Function *NF =
Function::Create(cast<FunctionType>(I->getType()->getElementType()),
GlobalValue::ExternalLinkage, I->getName(), New);
NF->copyAttributesFrom(I);
ValueMap[I] = NF;
}
// Loop over the aliases in the module
for (Module::const_alias_iterator I = M->alias_begin(), E = M->alias_end();
I != E; ++I)
ValueMap[I] = new GlobalAlias(I->getType(), GlobalAlias::ExternalLinkage,
I->getName(), NULL, New);
// Now that all of the things that global variable initializer can refer to
// have been created, loop through and copy the global variable referrers
// over... We also set the attributes on the global now.
//
for (Module::const_global_iterator I = M->global_begin(), E = M->global_end();
I != E; ++I) {
GlobalVariable *GV = cast<GlobalVariable>(ValueMap[I]);
if (I->hasInitializer())
GV->setInitializer(cast<Constant>(MapValue(I->getInitializer(),
ValueMap)));
GV->setLinkage(I->getLinkage());
GV->setThreadLocal(I->isThreadLocal());
GV->setConstant(I->isConstant());
}
// Similarly, copy over function bodies now...
//
for (Module::const_iterator I = M->begin(), E = M->end(); I != E; ++I) {
Function *F = cast<Function>(ValueMap[I]);
if (!I->isDeclaration()) {
Function::arg_iterator DestI = F->arg_begin();
for (Function::const_arg_iterator J = I->arg_begin(); J != I->arg_end();
++J) {
DestI->setName(J->getName());
ValueMap[J] = DestI++;
}
std::vector<ReturnInst*> Returns; // Ignore returns cloned...
CloneFunctionInto(F, I, ValueMap, Returns);
}
F->setLinkage(I->getLinkage());
}
// And aliases
for (Module::const_alias_iterator I = M->alias_begin(), E = M->alias_end();
I != E; ++I) {
GlobalAlias *GA = cast<GlobalAlias>(ValueMap[I]);
GA->setLinkage(I->getLinkage());
if (const Constant* C = I->getAliasee())
GA->setAliasee(cast<Constant>(MapValue(C, ValueMap)));
}
return New;
}
