bool RecurrenceDescriptor::getSourceExtensionKind(
Instruction *Start, Instruction *Exit, Type *RT, bool &IsSigned,
SmallPtrSetImpl<Instruction *> &Visited,
SmallPtrSetImpl<Instruction *> &CI) {
SmallVector<Instruction *, 8> Worklist;
bool FoundOneOperand = false;
unsigned DstSize = RT->getPrimitiveSizeInBits();
Worklist.push_back(Exit);
// Traverse the instructions in the reduction expression, beginning with the
// exit value.
while (!Worklist.empty()) {
Instruction *I = Worklist.pop_back_val();
for (Use &U : I->operands()) {
// Terminate the traversal if the operand is not an instruction, or we
// reach the starting value.
Instruction *J = dyn_cast<Instruction>(U.get());
if (!J || J == Start)
continue;
// Otherwise, investigate the operation if it is also in the expression.
if (Visited.count(J)) {
Worklist.push_back(J);
continue;
}
// If the operand is not in Visited, it is not a reduction operation, but
// it does feed into one. Make sure it is either a single-use sign- or
// zero-extend instruction.
CastInst *Cast = dyn_cast<CastInst>(J);
bool IsSExtInst = isa<SExtInst>(J);
if (!Cast || !Cast->hasOneUse() || !(isa<ZExtInst>(J) || IsSExtInst))
return false;
// Ensure the source type of the extend is no larger than the reduction
// type. It is not necessary for the types to be identical.
unsigned SrcSize = Cast->getSrcTy()->getPrimitiveSizeInBits();
if (SrcSize > DstSize)
return false;
// Furthermore, ensure that all such extends are of the same kind.
if (FoundOneOperand) {
if (IsSigned != IsSExtInst)
return false;
} else {
FoundOneOperand = true;
IsSigned = IsSExtInst;
}
// Lastly, if the source type of the extend matches the reduction type,
// add the extend to CI so that we can avoid accounting for it in the
// cost model.
if (SrcSize == DstSize)
CI.insert(Cast);
}
}
return true;
}
bool IRTranslator::translateCast(unsigned Opcode, const CastInst &CI) {
unsigned Op = getOrCreateVReg(*CI.getOperand(0));
unsigned Res = getOrCreateVReg(CI);
MIRBuilder.buildInstr(Opcode, {LLT{*CI.getDestTy()}, LLT{*CI.getSrcTy()}})
.addDef(Res)
.addUse(Op);
return true;
}
bool IRTranslator::translateBitCast(const CastInst &CI) {
if (LLT{*CI.getDestTy()} == LLT{*CI.getSrcTy()}) {
MIRBuilder.buildCopy(getOrCreateVReg(CI),
getOrCreateVReg(*CI.getOperand(0)));
return true;
}
return translateCast(TargetOpcode::G_BITCAST, CI);
}
void DSWP::insertProduce(Instruction *u, Instruction *v, DType dtype, int channel, int uthread, int vthread) {
Function *fun = module->getFunction("sync_produce");
vector<Value *> args;
Instruction *insPos = u->getNextNode();
if (insPos == NULL) {
error("here cannot be null");
}
//if (isa<BranchInst>(u)) {
// error("I don't know how do deal with it");
// return;
//}
if (dtype == REG) { // register dep
// cast the value to something that the communication guy likes
CastInst *cast;
if (u->getType()->isIntegerTy()) {
cast = new ZExtInst(u, eleType, u->getName().str() + "_64");
} else if (u->getType()->isFloatingPointTy()) {
if (u->getType()->isFloatTy()) {
cout << "WARNING: float sucks?";
}
cast = new BitCastInst(u, eleType, u->getName().str() + "_64");
} else if (u->getType()->isPointerTy()){
cast = new PtrToIntInst(u, eleType, u->getName().str() + "_64");
} else {
error("what's the hell type");
}
cast->insertBefore(insPos);
args.push_back(cast);
/* TODO: for true memory dependences, need to send anything or just sync?
} else if (dtype == DTRUE) { // true dep
error("check mem dep!!");
StoreInst *store = dyn_cast<StoreInst>(u);
if (store == NULL) {
error("not true dependency!");
}
BitCastInst *cast = new BitCastInst(store->getOperand(0), Type::getInt8PtrTy(*context), u->getName().str() + "_ptr");
cast->insertBefore(insPos);
args.push_back(cast);
*/
} else { // others
// just send a dummy value for synchronization
args.push_back(Constant::getNullValue(Type::getInt64Ty(*context)));
}
// channel ID
args.push_back(ConstantInt::get(Type::getInt32Ty(*context), channel));
// make the actual call
CallInst *call = CallInst::Create(fun, args, "", insPos);
}
void Transformer4Trace::transformLoadInst(Module* module, LoadInst* inst, AliasAnalysis& AA) {
Value * val = inst->getOperand(0);
int svIdx = this->getValueIndex(module, val, AA);
if (svIdx == -1) return;
CastInst* c = CastInst::CreatePointerCast(val, Type::getIntNPtrTy(module->getContext(),POINTER_BIT_SIZE));
c->insertBefore(inst);
Value* lnval = getOtInsertLineNumberValue(module, inst);
this->insertCallInstBefore(inst, F_preload, c, lnval, getOrInsertSrcFileNameValue(module, inst), NULL);
this->insertCallInstAfter(inst, F_load, c, lnval, getOrInsertSrcFileNameValue(module, inst), NULL);
}
void Transformer4Trace::transformMemSet(Module* module, CallInst* call, AliasAnalysis& AA) {
Value* lnval = getOtInsertLineNumberValue(module, call);
Value * val = call->getArgOperand(0);
int svIdx = this->getValueIndex(module, val, AA);
if (svIdx == -1) return;
CastInst* c = CastInst::CreatePointerCast(val, Type::getIntNPtrTy(module->getContext(),POINTER_BIT_SIZE));
c->insertBefore(call);
insertCallInstBefore(call, F_prestore, c, lnval, getOrInsertSrcFileNameValue(module, call), NULL);
insertCallInstAfter(call, F_store, c, lnval, getOrInsertSrcFileNameValue(module, call), NULL);
}
/// If a value has only one user that is a CastInst, return it.
Value *llvm::getUniqueCastUse(Value *Ptr, Loop *Lp, Type *Ty) {
Value *UniqueCast = nullptr;
for (User *U : Ptr->users()) {
CastInst *CI = dyn_cast<CastInst>(U);
if (CI && CI->getType() == Ty) {
if (!UniqueCast)
UniqueCast = CI;
else
return nullptr;
}
}
return UniqueCast;
}
bool CallAnalyzer::visitCastInst(CastInst &I) {
// Propagate constants through ptrtoint.
if (Constant *COp = dyn_cast<Constant>(I.getOperand(0)))
if (Constant *C = ConstantExpr::getCast(I.getOpcode(), COp, I.getType())) {
SimplifiedValues[&I] = C;
return true;
}
// Disable SROA in the face of arbitrary casts we don't whitelist elsewhere.
disableSROA(I.getOperand(0));
return isInstructionFree(&I, TD);
}
/// Try to simplify cast instruction.
bool UnrolledInstAnalyzer::visitCastInst(CastInst &I) {
// Propagate constants through casts.
Constant *COp = dyn_cast<Constant>(I.getOperand(0));
if (!COp)
COp = SimplifiedValues.lookup(I.getOperand(0));
if (COp)
if (Constant *C =
ConstantExpr::getCast(I.getOpcode(), COp, I.getType())) {
SimplifiedValues[&I] = C;
return true;
}
return Base::visitCastInst(I);
}
bool CallAnalyzer::visitCastInst(CastInst &I) {
// Propagate constants through ptrtoint.
Constant *COp = dyn_cast<Constant>(I.getOperand(0));
if (!COp)
COp = SimplifiedValues.lookup(I.getOperand(0));
if (COp)
if (Constant *C = ConstantExpr::getCast(I.getOpcode(), COp, I.getType())) {
SimplifiedValues[&I] = C;
return true;
}
// Disable SROA in the face of arbitrary casts we don't whitelist elsewhere.
disableSROA(I.getOperand(0));
return TargetTransformInfo::TCC_Free == TTI.getUserCost(&I);
}
bool Scalarizer::visitCastInst(CastInst &CI) {
VectorType *VT = dyn_cast<VectorType>(CI.getDestTy());
if (!VT)
return false;
unsigned NumElems = VT->getNumElements();
IRBuilder<> Builder(CI.getParent(), &CI);
Scatterer Op0 = scatter(&CI, CI.getOperand(0));
assert(Op0.size() == NumElems && "Mismatched cast");
ValueVector Res;
Res.resize(NumElems);
for (unsigned I = 0; I < NumElems; ++I)
Res[I] = Builder.CreateCast(CI.getOpcode(), Op0[I], VT->getElementType(),
CI.getName() + ".i" + Twine(I));
gather(&CI, Res);
return true;
}
bool CallAnalyzer::visitCastInst(CastInst &I) {
// Propagate constants through ptrtoint.
if (Constant *COp = dyn_cast<Constant>(I.getOperand(0)))
if (Constant *C = ConstantExpr::getCast(I.getOpcode(), COp, I.getType())) {
SimplifiedValues[&I] = C;
return true;
}
// Disable SROA in the face of arbitrary casts we don't whitelist elsewhere.
disableSROA(I.getOperand(0));
// No-op casts don't have any cost.
if (I.isLosslessCast())
return true;
// trunc to a native type is free (assuming the target has compare and
// shift-right of the same width).
if (TD && isa<TruncInst>(I) &&
TD->isLegalInteger(TD->getTypeSizeInBits(I.getType())))
return true;
// Result of a cmp instruction is often extended (to be used by other
// cmp instructions, logical or return instructions). These are usually
// no-ops on most sane targets.
if (isa<CmpInst>(I.getOperand(0)))
return true;
// Assume the rest of the casts require work.
return false;
}
void Transformer4Trace::transformPthreadCondSignal(Module* module, CallInst* call, AliasAnalysis& AA) {
Value * val0 = call->getArgOperand(0);
Value * val1 = call->getArgOperand(1);
int svIdx0 = this->getValueIndex(module, val0, AA);
int svIdx1 = this->getValueIndex(module, val1, AA);
if (svIdx0 == -1 && svIdx1 == -1) return;
CastInst* cond = CastInst::CreatePointerCast(val0, Type::getIntNPtrTy(module->getContext(),POINTER_BIT_SIZE));
cond->insertBefore(call);
CastInst* mut = CastInst::CreatePointerCast(val1, Type::getIntNPtrTy(module->getContext(),POINTER_BIT_SIZE));
mut->insertBefore(call);
Value* lnval = getOtInsertLineNumberValue(module, call);
this->insertCallInstBefore(call, F_prenotify, cond, mut, lnval, getOrInsertSrcFileNameValue(module, call), NULL);
this->insertCallInstAfter(call, F_notify, cond, mut, lnval, getOrInsertSrcFileNameValue(module, call), NULL);
}
/// lowerIncomingArguments - To avoid having to handle incoming arguments
/// specially, we lower each arg to a copy instruction in the entry block. This
/// ensures that the argument value itself cannot be live out of the entry
/// block.
void SjLjEHPass::lowerIncomingArguments(Function &F) {
BasicBlock::iterator AfterAllocaInsPt = F.begin()->begin();
while (isa<AllocaInst>(AfterAllocaInsPt) &&
isa<ConstantInt>(cast<AllocaInst>(AfterAllocaInsPt)->getArraySize()))
++AfterAllocaInsPt;
for (Function::arg_iterator
AI = F.arg_begin(), AE = F.arg_end(); AI != AE; ++AI) {
Type *Ty = AI->getType();
// Aggregate types can't be cast, but are legal argument types, so we have
// to handle them differently. We use an extract/insert pair as a
// lightweight method to achieve the same goal.
if (isa<StructType>(Ty) || isa<ArrayType>(Ty) || isa<VectorType>(Ty)) {
Instruction *EI = ExtractValueInst::Create(AI, 0, "", AfterAllocaInsPt);
Instruction *NI = InsertValueInst::Create(AI, EI, 0);
NI->insertAfter(EI);
AI->replaceAllUsesWith(NI);
// Set the operand of the instructions back to the AllocaInst.
EI->setOperand(0, AI);
NI->setOperand(0, AI);
} else {
// This is always a no-op cast because we're casting AI to AI->getType()
// so src and destination types are identical. BitCast is the only
// possibility.
CastInst *NC =
new BitCastInst(AI, AI->getType(), AI->getName() + ".tmp",
AfterAllocaInsPt);
AI->replaceAllUsesWith(NC);
// Set the operand of the cast instruction back to the AllocaInst.
// Normally it's forbidden to replace a CastInst's operand because it
// could cause the opcode to reflect an illegal conversion. However, we're
// replacing it here with the same value it was constructed with. We do
// this because the above replaceAllUsesWith() clobbered the operand, but
// we want this one to remain.
NC->setOperand(0, AI);
}
}
}
void Transformer4Trace::transformOtherFunctionCalls(Module* module, CallInst* call, AliasAnalysis& AA) {
Value* lnval = getOtInsertLineNumberValue(module, call);
for (unsigned i = 0; i < call->getNumArgOperands(); i++) {
Value * arg = call->getArgOperand(i);
CastInst* c = NULL;
if (arg->getType()->isPointerTy()) {
c = CastInst::CreatePointerCast(arg, Type::getIntNPtrTy(module->getContext(),POINTER_BIT_SIZE));
c->insertBefore(call);
} else {
continue;
}
int svIdx = this->getValueIndex(module, arg, AA);
if (svIdx != -1) {
insertCallInstBefore(call, F_prestore, c, lnval, getOrInsertSrcFileNameValue(module, call), NULL);
insertCallInstBefore(call, F_store, c, lnval, getOrInsertSrcFileNameValue(module, call), NULL);
}
}
}
/// lowerIncomingArguments - To avoid having to handle incoming arguments
/// specially, we lower each arg to a copy instruction in the entry block. This
/// ensures that the argument value itself cannot be live out of the entry
/// block.
void SjLjEHPrepare::lowerIncomingArguments(Function &F) {
BasicBlock::iterator AfterAllocaInsPt = F.begin()->begin();
while (isa<AllocaInst>(AfterAllocaInsPt) &&
isa<ConstantInt>(cast<AllocaInst>(AfterAllocaInsPt)->getArraySize()))
++AfterAllocaInsPt;
for (Function::arg_iterator AI = F.arg_begin(), AE = F.arg_end(); AI != AE;
++AI) {
Type *Ty = AI->getType();
if (isa<StructType>(Ty) || isa<ArrayType>(Ty)) {
// Aggregate types can't be cast, but are legal argument types,
// so we have to handle them differently. We use
// select i8 true, %arg, undef to achieve the same goal
Value *TrueValue = ConstantInt::getTrue(F.getContext());
Value *UndefValue = UndefValue::get(Ty);
Instruction *SI = SelectInst::Create(TrueValue, AI, UndefValue,
AI->getName() + ".tmp",
AfterAllocaInsPt);
AI->replaceAllUsesWith(SI);
SI->setOperand(1, AI);
} else {
// This is always a no-op cast because we're casting AI to AI->getType()
// so src and destination types are identical. BitCast is the only
// possibility.
CastInst *NC = new BitCastInst(AI, AI->getType(), AI->getName() + ".tmp",
AfterAllocaInsPt);
AI->replaceAllUsesWith(NC);
// Set the operand of the cast instruction back to the AllocaInst.
// Normally it's forbidden to replace a CastInst's operand because it
// could cause the opcode to reflect an illegal conversion. However, we're
// replacing it here with the same value it was constructed with. We do
// this because the above replaceAllUsesWith() clobbered the operand, but
// we want this one to remain.
NC->setOperand(0, AI);
}
}
}
/// store {atomic|volatile} T %val, T* %ptr memory_order, align sizeof(T)
/// becomes:
/// call void @llvm.nacl.atomic.store.i<size>(%val, %ptr, memory_order)
void AtomicVisitor::visitStoreInst(StoreInst &I) {
return; // XXX EMSCRIPTEN
if (I.isSimple())
return;
PointerHelper<StoreInst> PH(*this, I);
const NaCl::AtomicIntrinsics::AtomicIntrinsic *Intrinsic =
findAtomicIntrinsic(I, Intrinsic::nacl_atomic_store, PH.PET);
checkAlignment(I, I.getAlignment(), PH.BitSize / CHAR_BIT);
Value *V = I.getValueOperand();
if (!V->getType()->isIntegerTy()) {
// The store isn't of an integer type. We define atomics in terms of
// integers, so bitcast the value to store to an integer of the
// proper width.
CastInst *Cast = createCast(I, V, Type::getIntNTy(C, PH.BitSize),
V->getName() + ".cast");
Cast->setDebugLoc(I.getDebugLoc());
V = Cast;
}
checkSizeMatchesType(I, PH.BitSize, V->getType());
Value *Args[] = {V, PH.P, freezeMemoryOrder(I, I.getOrdering())};
replaceInstructionWithIntrinsicCall(I, Intrinsic, PH.OriginalPET, PH.PET,
Args);
}
void Transformer4Trace::transformMemCpyMov(Module* module, CallInst* call, AliasAnalysis& AA) {
Value* lnval = getOtInsertLineNumberValue(module, call);
Value * dst = call->getArgOperand(0);
Value * src = call->getArgOperand(1);
int svIdx_dst = this->getValueIndex(module, dst, AA);
int svIdx_src = this->getValueIndex(module, src, AA);
if (svIdx_dst == -1 && svIdx_src == -1) {
return;
} else if (svIdx_dst != -1 && svIdx_src != -1) {
CastInst* d = CastInst::CreatePointerCast(dst, Type::getIntNPtrTy(module->getContext(),POINTER_BIT_SIZE));
d->insertBefore(call);
CastInst* s = CastInst::CreatePointerCast(src, Type::getIntNPtrTy(module->getContext(),POINTER_BIT_SIZE));
s->insertBefore(call);
if (svIdx_dst != svIdx_src) {
insertCallInstBefore(call, F_prestore, d, lnval, getOrInsertSrcFileNameValue(module, call), NULL);
insertCallInstAfter(call, F_store, d, lnval, getOrInsertSrcFileNameValue(module, call), NULL);
insertCallInstBefore(call, F_preload, s, lnval, getOrInsertSrcFileNameValue(module, call), NULL);
insertCallInstAfter(call, F_load, s, lnval, getOrInsertSrcFileNameValue(module, call), NULL);
} else {
insertCallInstBefore(call, F_prestore, d, lnval, getOrInsertSrcFileNameValue(module, call), NULL);
insertCallInstAfter(call, F_store, d, lnval, getOrInsertSrcFileNameValue(module, call), NULL);
}
} else if (svIdx_dst != -1) {
CastInst* d = CastInst::CreatePointerCast(dst, Type::getIntNPtrTy(module->getContext(),POINTER_BIT_SIZE));
d->insertBefore(call);
insertCallInstBefore(call, F_prestore, d, lnval, getOrInsertSrcFileNameValue(module, call), NULL);
insertCallInstAfter(call, F_store, d, lnval, getOrInsertSrcFileNameValue(module, call), NULL);
} else {
CastInst* s = CastInst::CreatePointerCast(src, Type::getIntNPtrTy(module->getContext(),POINTER_BIT_SIZE));
s->insertBefore(call);
insertCallInstBefore(call, F_preload, s, lnval, getOrInsertSrcFileNameValue(module, call), NULL);
insertCallInstAfter(call, F_load, s, lnval, getOrInsertSrcFileNameValue(module, call), NULL);
}
}
/// Try to simplify cast instruction.
bool UnrolledInstAnalyzer::visitCastInst(CastInst &I) {
// Propagate constants through casts.
Constant *COp = dyn_cast<Constant>(I.getOperand(0));
if (!COp)
COp = SimplifiedValues.lookup(I.getOperand(0));
// If we know a simplified value for this operand and cast is valid, save the
// result to SimplifiedValues.
// The cast can be invalid, because SimplifiedValues contains results of SCEV
// analysis, which operates on integers (and, e.g., might convert i8* null to
// i32 0).
if (COp && CastInst::castIsValid(I.getOpcode(), COp, I.getType())) {
if (Constant *C =
ConstantExpr::getCast(I.getOpcode(), COp, I.getType())) {
SimplifiedValues[&I] = C;
return true;
}
}
return Base::visitCastInst(I);
}
bool PPCCTRLoops::mightUseCTR(const Triple &TT, BasicBlock *BB) {
for (BasicBlock::iterator J = BB->begin(), JE = BB->end();
J != JE; ++J) {
if (CallInst *CI = dyn_cast<CallInst>(J)) {
if (InlineAsm *IA = dyn_cast<InlineAsm>(CI->getCalledValue())) {
// Inline ASM is okay, unless it clobbers the ctr register.
InlineAsm::ConstraintInfoVector CIV = IA->ParseConstraints();
for (unsigned i = 0, ie = CIV.size(); i < ie; ++i) {
InlineAsm::ConstraintInfo &C = CIV[i];
if (C.Type != InlineAsm::isInput)
for (unsigned j = 0, je = C.Codes.size(); j < je; ++j)
if (StringRef(C.Codes[j]).equals_lower("{ctr}"))
return true;
}
continue;
}
if (!TM)
return true;
const TargetLowering *TLI = TM->getTargetLowering();
if (Function *F = CI->getCalledFunction()) {
// Most intrinsics don't become function calls, but some might.
// sin, cos, exp and log are always calls.
unsigned Opcode;
if (F->getIntrinsicID() != Intrinsic::not_intrinsic) {
switch (F->getIntrinsicID()) {
default: continue;
// VisualStudio defines setjmp as _setjmp
#if defined(_MSC_VER) && defined(setjmp) && \
!defined(setjmp_undefined_for_msvc)
# pragma push_macro("setjmp")
# undef setjmp
# define setjmp_undefined_for_msvc
#endif
case Intrinsic::setjmp:
#if defined(_MSC_VER) && defined(setjmp_undefined_for_msvc)
// let's return it to _setjmp state
# pragma pop_macro("setjmp")
# undef setjmp_undefined_for_msvc
#endif
case Intrinsic::longjmp:
// Exclude eh_sjlj_setjmp; we don't need to exclude eh_sjlj_longjmp
// because, although it does clobber the counter register, the
// control can't then return to inside the loop unless there is also
// an eh_sjlj_setjmp.
case Intrinsic::eh_sjlj_setjmp:
case Intrinsic::memcpy:
case Intrinsic::memmove:
case Intrinsic::memset:
case Intrinsic::powi:
case Intrinsic::log:
case Intrinsic::log2:
case Intrinsic::log10:
case Intrinsic::exp:
case Intrinsic::exp2:
case Intrinsic::pow:
case Intrinsic::sin:
case Intrinsic::cos:
return true;
case Intrinsic::copysign:
if (CI->getArgOperand(0)->getType()->getScalarType()->
isPPC_FP128Ty())
return true;
else
continue; // ISD::FCOPYSIGN is never a library call.
case Intrinsic::sqrt: Opcode = ISD::FSQRT; break;
case Intrinsic::floor: Opcode = ISD::FFLOOR; break;
case Intrinsic::ceil: Opcode = ISD::FCEIL; break;
case Intrinsic::trunc: Opcode = ISD::FTRUNC; break;
case Intrinsic::rint: Opcode = ISD::FRINT; break;
case Intrinsic::nearbyint: Opcode = ISD::FNEARBYINT; break;
case Intrinsic::round: Opcode = ISD::FROUND; break;
}
}
// PowerPC does not use [US]DIVREM or other library calls for
// operations on regular types which are not otherwise library calls
// (i.e. soft float or atomics). If adapting for targets that do,
// additional care is required here.
LibFunc::Func Func;
if (!F->hasLocalLinkage() && F->hasName() && LibInfo &&
LibInfo->getLibFunc(F->getName(), Func) &&
LibInfo->hasOptimizedCodeGen(Func)) {
// Non-read-only functions are never treated as intrinsics.
if (!CI->onlyReadsMemory())
return true;
// Conversion happens only for FP calls.
if (!CI->getArgOperand(0)->getType()->isFloatingPointTy())
return true;
switch (Func) {
default: return true;
case LibFunc::copysign:
case LibFunc::copysignf:
continue; // ISD::FCOPYSIGN is never a library call.
case LibFunc::copysignl:
return true;
case LibFunc::fabs:
case LibFunc::fabsf:
case LibFunc::fabsl:
continue; // ISD::FABS is never a library call.
case LibFunc::sqrt:
case LibFunc::sqrtf:
case LibFunc::sqrtl:
Opcode = ISD::FSQRT; break;
case LibFunc::floor:
case LibFunc::floorf:
case LibFunc::floorl:
Opcode = ISD::FFLOOR; break;
case LibFunc::nearbyint:
case LibFunc::nearbyintf:
case LibFunc::nearbyintl:
Opcode = ISD::FNEARBYINT; break;
case LibFunc::ceil:
case LibFunc::ceilf:
case LibFunc::ceill:
Opcode = ISD::FCEIL; break;
case LibFunc::rint:
case LibFunc::rintf:
case LibFunc::rintl:
Opcode = ISD::FRINT; break;
case LibFunc::round:
case LibFunc::roundf:
case LibFunc::roundl:
Opcode = ISD::FROUND; break;
case LibFunc::trunc:
case LibFunc::truncf:
case LibFunc::truncl:
Opcode = ISD::FTRUNC; break;
}
MVT VTy =
TLI->getSimpleValueType(CI->getArgOperand(0)->getType(), true);
if (VTy == MVT::Other)
return true;
if (TLI->isOperationLegalOrCustom(Opcode, VTy))
continue;
else if (VTy.isVector() &&
TLI->isOperationLegalOrCustom(Opcode, VTy.getScalarType()))
continue;
return true;
}
}
return true;
} else if (isa<BinaryOperator>(J) &&
J->getType()->getScalarType()->isPPC_FP128Ty()) {
// Most operations on ppc_f128 values become calls.
return true;
} else if (isa<UIToFPInst>(J) || isa<SIToFPInst>(J) ||
isa<FPToUIInst>(J) || isa<FPToSIInst>(J)) {
CastInst *CI = cast<CastInst>(J);
if (CI->getSrcTy()->getScalarType()->isPPC_FP128Ty() ||
CI->getDestTy()->getScalarType()->isPPC_FP128Ty() ||
(TT.isArch32Bit() &&
(CI->getSrcTy()->getScalarType()->isIntegerTy(64) ||
CI->getDestTy()->getScalarType()->isIntegerTy(64))
))
return true;
} else if (TT.isArch32Bit() &&
J->getType()->getScalarType()->isIntegerTy(64) &&
(J->getOpcode() == Instruction::UDiv ||
J->getOpcode() == Instruction::SDiv ||
J->getOpcode() == Instruction::URem ||
J->getOpcode() == Instruction::SRem)) {
return true;
} else if (isa<IndirectBrInst>(J) || isa<InvokeInst>(J)) {
// On PowerPC, indirect jumps use the counter register.
return true;
} else if (SwitchInst *SI = dyn_cast<SwitchInst>(J)) {
if (!TM)
return true;
const TargetLowering *TLI = TM->getTargetLowering();
if (TLI->supportJumpTables() &&
SI->getNumCases()+1 >= (unsigned) TLI->getMinimumJumpTableEntries())
return true;
}
}
return false;
}
/// splitLiveRangesAcrossInvokes - Each value that is live across an unwind edge
/// we spill into a stack location, guaranteeing that there is nothing live
/// across the unwind edge. This process also splits all critical edges
/// coming out of invoke's.
/// FIXME: Move this function to a common utility file (Local.cpp?) so
/// both SjLj and LowerInvoke can use it.
void SjLjEHPass::
splitLiveRangesAcrossInvokes(SmallVector<InvokeInst*,16> &Invokes) {
// First step, split all critical edges from invoke instructions.
for (unsigned i = 0, e = Invokes.size(); i != e; ++i) {
InvokeInst *II = Invokes[i];
SplitCriticalEdge(II, 0, this);
// FIXME: New EH - This if-condition will be always true in the new scheme.
if (II->getUnwindDest()->isLandingPad()) {
SmallVector<BasicBlock*, 2> NewBBs;
SplitLandingPadPredecessors(II->getUnwindDest(), II->getParent(),
".1", ".2", this, NewBBs);
LPadSuccMap[II] = *succ_begin(NewBBs[0]);
} else {
SplitCriticalEdge(II, 1, this);
}
assert(!isa<PHINode>(II->getNormalDest()) &&
!isa<PHINode>(II->getUnwindDest()) &&
"Critical edge splitting left single entry phi nodes?");
}
Function *F = Invokes.back()->getParent()->getParent();
// To avoid having to handle incoming arguments specially, we lower each arg
// to a copy instruction in the entry block. This ensures that the argument
// value itself cannot be live across the entry block.
BasicBlock::iterator AfterAllocaInsertPt = F->begin()->begin();
while (isa<AllocaInst>(AfterAllocaInsertPt) &&
isa<ConstantInt>(cast<AllocaInst>(AfterAllocaInsertPt)->getArraySize()))
++AfterAllocaInsertPt;
for (Function::arg_iterator AI = F->arg_begin(), E = F->arg_end();
AI != E; ++AI) {
Type *Ty = AI->getType();
// Aggregate types can't be cast, but are legal argument types, so we have
// to handle them differently. We use an extract/insert pair as a
// lightweight method to achieve the same goal.
if (isa<StructType>(Ty) || isa<ArrayType>(Ty) || isa<VectorType>(Ty)) {
Instruction *EI = ExtractValueInst::Create(AI, 0, "",AfterAllocaInsertPt);
Instruction *NI = InsertValueInst::Create(AI, EI, 0);
NI->insertAfter(EI);
AI->replaceAllUsesWith(NI);
// Set the operand of the instructions back to the AllocaInst.
EI->setOperand(0, AI);
NI->setOperand(0, AI);
} else {
// This is always a no-op cast because we're casting AI to AI->getType()
// so src and destination types are identical. BitCast is the only
// possibility.
CastInst *NC = new BitCastInst(
AI, AI->getType(), AI->getName()+".tmp", AfterAllocaInsertPt);
AI->replaceAllUsesWith(NC);
// Set the operand of the cast instruction back to the AllocaInst.
// Normally it's forbidden to replace a CastInst's operand because it
// could cause the opcode to reflect an illegal conversion. However,
// we're replacing it here with the same value it was constructed with.
// We do this because the above replaceAllUsesWith() clobbered the
// operand, but we want this one to remain.
NC->setOperand(0, AI);
}
}
// Finally, scan the code looking for instructions with bad live ranges.
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
// Ignore obvious cases we don't have to handle. In particular, most
// instructions either have no uses or only have a single use inside the
// current block. Ignore them quickly.
Instruction *Inst = II;
if (Inst->use_empty()) continue;
if (Inst->hasOneUse() &&
cast<Instruction>(Inst->use_back())->getParent() == BB &&
!isa<PHINode>(Inst->use_back())) continue;
// If this is an alloca in the entry block, it's not a real register
// value.
if (AllocaInst *AI = dyn_cast<AllocaInst>(Inst))
if (isa<ConstantInt>(AI->getArraySize()) && BB == F->begin())
continue;
// Avoid iterator invalidation by copying users to a temporary vector.
SmallVector<Instruction*,16> Users;
for (Value::use_iterator UI = Inst->use_begin(), E = Inst->use_end();
UI != E; ++UI) {
Instruction *User = cast<Instruction>(*UI);
if (User->getParent() != BB || isa<PHINode>(User))
Users.push_back(User);
}
// Find all of the blocks that this value is live in.
std::set<BasicBlock*> LiveBBs;
LiveBBs.insert(Inst->getParent());
while (!Users.empty()) {
Instruction *U = Users.back();
Users.pop_back();
if (!isa<PHINode>(U)) {
MarkBlocksLiveIn(U->getParent(), LiveBBs);
} else {
// Uses for a PHI node occur in their predecessor block.
PHINode *PN = cast<PHINode>(U);
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
if (PN->getIncomingValue(i) == Inst)
MarkBlocksLiveIn(PN->getIncomingBlock(i), LiveBBs);
}
}
// Now that we know all of the blocks that this thing is live in, see if
// it includes any of the unwind locations.
bool NeedsSpill = false;
for (unsigned i = 0, e = Invokes.size(); i != e; ++i) {
BasicBlock *UnwindBlock = Invokes[i]->getUnwindDest();
if (UnwindBlock != BB && LiveBBs.count(UnwindBlock)) {
NeedsSpill = true;
}
}
// If we decided we need a spill, do it.
// FIXME: Spilling this way is overkill, as it forces all uses of
// the value to be reloaded from the stack slot, even those that aren't
// in the unwind blocks. We should be more selective.
if (NeedsSpill) {
++NumSpilled;
DemoteRegToStack(*Inst, true);
}
}
}
bool PPCCTRLoops::mightUseCTR(BasicBlock *BB) {
for (BasicBlock::iterator J = BB->begin(), JE = BB->end();
J != JE; ++J) {
if (CallInst *CI = dyn_cast<CallInst>(J)) {
// Inline ASM is okay, unless it clobbers the ctr register.
if (InlineAsm *IA = dyn_cast<InlineAsm>(CI->getCalledValue())) {
if (asmClobbersCTR(IA))
return true;
continue;
}
if (Function *F = CI->getCalledFunction()) {
// Most intrinsics don't become function calls, but some might.
// sin, cos, exp and log are always calls.
unsigned Opcode = 0;
if (F->getIntrinsicID() != Intrinsic::not_intrinsic) {
switch (F->getIntrinsicID()) {
default: continue;
// If we have a call to ppc_is_decremented_ctr_nonzero, or ppc_mtctr
// we're definitely using CTR.
case Intrinsic::ppc_is_decremented_ctr_nonzero:
case Intrinsic::ppc_mtctr:
return true;
// VisualStudio defines setjmp as _setjmp
#if defined(_MSC_VER) && defined(setjmp) && \
!defined(setjmp_undefined_for_msvc)
# pragma push_macro("setjmp")
# undef setjmp
# define setjmp_undefined_for_msvc
#endif
case Intrinsic::setjmp:
#if defined(_MSC_VER) && defined(setjmp_undefined_for_msvc)
// let's return it to _setjmp state
# pragma pop_macro("setjmp")
# undef setjmp_undefined_for_msvc
#endif
case Intrinsic::longjmp:
// Exclude eh_sjlj_setjmp; we don't need to exclude eh_sjlj_longjmp
// because, although it does clobber the counter register, the
// control can't then return to inside the loop unless there is also
// an eh_sjlj_setjmp.
case Intrinsic::eh_sjlj_setjmp:
case Intrinsic::memcpy:
case Intrinsic::memmove:
case Intrinsic::memset:
case Intrinsic::powi:
case Intrinsic::log:
case Intrinsic::log2:
case Intrinsic::log10:
case Intrinsic::exp:
case Intrinsic::exp2:
case Intrinsic::pow:
case Intrinsic::sin:
case Intrinsic::cos:
return true;
case Intrinsic::copysign:
if (CI->getArgOperand(0)->getType()->getScalarType()->
isPPC_FP128Ty())
return true;
else
continue; // ISD::FCOPYSIGN is never a library call.
case Intrinsic::sqrt: Opcode = ISD::FSQRT; break;
case Intrinsic::floor: Opcode = ISD::FFLOOR; break;
case Intrinsic::ceil: Opcode = ISD::FCEIL; break;
case Intrinsic::trunc: Opcode = ISD::FTRUNC; break;
case Intrinsic::rint: Opcode = ISD::FRINT; break;
case Intrinsic::nearbyint: Opcode = ISD::FNEARBYINT; break;
case Intrinsic::round: Opcode = ISD::FROUND; break;
case Intrinsic::minnum: Opcode = ISD::FMINNUM; break;
case Intrinsic::maxnum: Opcode = ISD::FMAXNUM; break;
case Intrinsic::umul_with_overflow: Opcode = ISD::UMULO; break;
case Intrinsic::smul_with_overflow: Opcode = ISD::SMULO; break;
}
}
// PowerPC does not use [US]DIVREM or other library calls for
// operations on regular types which are not otherwise library calls
// (i.e. soft float or atomics). If adapting for targets that do,
// additional care is required here.
LibFunc Func;
if (!F->hasLocalLinkage() && F->hasName() && LibInfo &&
LibInfo->getLibFunc(F->getName(), Func) &&
LibInfo->hasOptimizedCodeGen(Func)) {
// Non-read-only functions are never treated as intrinsics.
if (!CI->onlyReadsMemory())
return true;
// Conversion happens only for FP calls.
if (!CI->getArgOperand(0)->getType()->isFloatingPointTy())
return true;
switch (Func) {
default: return true;
case LibFunc_copysign:
case LibFunc_copysignf:
continue; // ISD::FCOPYSIGN is never a library call.
case LibFunc_copysignl:
return true;
case LibFunc_fabs:
case LibFunc_fabsf:
case LibFunc_fabsl:
continue; // ISD::FABS is never a library call.
case LibFunc_sqrt:
case LibFunc_sqrtf:
case LibFunc_sqrtl:
Opcode = ISD::FSQRT; break;
case LibFunc_floor:
case LibFunc_floorf:
case LibFunc_floorl:
Opcode = ISD::FFLOOR; break;
case LibFunc_nearbyint:
case LibFunc_nearbyintf:
case LibFunc_nearbyintl:
Opcode = ISD::FNEARBYINT; break;
case LibFunc_ceil:
case LibFunc_ceilf:
case LibFunc_ceill:
Opcode = ISD::FCEIL; break;
case LibFunc_rint:
case LibFunc_rintf:
case LibFunc_rintl:
Opcode = ISD::FRINT; break;
case LibFunc_round:
case LibFunc_roundf:
case LibFunc_roundl:
Opcode = ISD::FROUND; break;
case LibFunc_trunc:
case LibFunc_truncf:
case LibFunc_truncl:
Opcode = ISD::FTRUNC; break;
case LibFunc_fmin:
case LibFunc_fminf:
case LibFunc_fminl:
Opcode = ISD::FMINNUM; break;
case LibFunc_fmax:
case LibFunc_fmaxf:
case LibFunc_fmaxl:
Opcode = ISD::FMAXNUM; break;
}
}
if (Opcode) {
EVT EVTy =
TLI->getValueType(*DL, CI->getArgOperand(0)->getType(), true);
if (EVTy == MVT::Other)
return true;
if (TLI->isOperationLegalOrCustom(Opcode, EVTy))
continue;
else if (EVTy.isVector() &&
TLI->isOperationLegalOrCustom(Opcode, EVTy.getScalarType()))
continue;
return true;
}
}
return true;
} else if (isa<BinaryOperator>(J) &&
J->getType()->getScalarType()->isPPC_FP128Ty()) {
// Most operations on ppc_f128 values become calls.
return true;
} else if (isa<UIToFPInst>(J) || isa<SIToFPInst>(J) ||
isa<FPToUIInst>(J) || isa<FPToSIInst>(J)) {
CastInst *CI = cast<CastInst>(J);
if (CI->getSrcTy()->getScalarType()->isPPC_FP128Ty() ||
CI->getDestTy()->getScalarType()->isPPC_FP128Ty() ||
isLargeIntegerTy(!TM->isPPC64(), CI->getSrcTy()->getScalarType()) ||
isLargeIntegerTy(!TM->isPPC64(), CI->getDestTy()->getScalarType()))
return true;
} else if (isLargeIntegerTy(!TM->isPPC64(),
J->getType()->getScalarType()) &&
(J->getOpcode() == Instruction::UDiv ||
J->getOpcode() == Instruction::SDiv ||
J->getOpcode() == Instruction::URem ||
J->getOpcode() == Instruction::SRem)) {
return true;
} else if (!TM->isPPC64() &&
isLargeIntegerTy(false, J->getType()->getScalarType()) &&
(J->getOpcode() == Instruction::Shl ||
J->getOpcode() == Instruction::AShr ||
J->getOpcode() == Instruction::LShr)) {
// Only on PPC32, for 128-bit integers (specifically not 64-bit
// integers), these might be runtime calls.
return true;
} else if (isa<IndirectBrInst>(J) || isa<InvokeInst>(J)) {
// On PowerPC, indirect jumps use the counter register.
return true;
} else if (SwitchInst *SI = dyn_cast<SwitchInst>(J)) {
if (SI->getNumCases() + 1 >= (unsigned)TLI->getMinimumJumpTableEntries())
return true;
}
// FREM is always a call.
if (J->getOpcode() == Instruction::FRem)
return true;
if (STI->useSoftFloat()) {
switch(J->getOpcode()) {
case Instruction::FAdd:
case Instruction::FSub:
case Instruction::FMul:
case Instruction::FDiv:
case Instruction::FPTrunc:
case Instruction::FPExt:
case Instruction::FPToUI:
case Instruction::FPToSI:
case Instruction::UIToFP:
case Instruction::SIToFP:
case Instruction::FCmp:
return true;
}
}
for (Value *Operand : J->operands())
if (memAddrUsesCTR(*TM, Operand))
return true;
}
return false;
}
// First thing we need to do is scan the whole function for values that are
// live across unwind edges. Each value that is live across an unwind edge
// we spill into a stack location, guaranteeing that there is nothing live
// across the unwind edge. This process also splits all critical edges
// coming out of invoke's.
void LowerInvoke::
splitLiveRangesLiveAcrossInvokes(std::vector<InvokeInst*> &Invokes) {
// First step, split all critical edges from invoke instructions.
for (unsigned i = 0, e = Invokes.size(); i != e; ++i) {
InvokeInst *II = Invokes[i];
SplitCriticalEdge(II, 0, this);
SplitCriticalEdge(II, 1, this);
assert(!isa<PHINode>(II->getNormalDest()) &&
!isa<PHINode>(II->getUnwindDest()) &&
"critical edge splitting left single entry phi nodes?");
}
Function *F = Invokes.back()->getParent()->getParent();
// To avoid having to handle incoming arguments specially, we lower each arg
// to a copy instruction in the entry block. This ensures that the argument
// value itself cannot be live across the entry block.
BasicBlock::iterator AfterAllocaInsertPt = F->begin()->begin();
while (isa<AllocaInst>(AfterAllocaInsertPt) &&
isa<ConstantInt>(cast<AllocaInst>(AfterAllocaInsertPt)->getArraySize()))
++AfterAllocaInsertPt;
for (Function::arg_iterator AI = F->arg_begin(), E = F->arg_end();
AI != E; ++AI) {
// This is always a no-op cast because we're casting AI to AI->getType() so
// src and destination types are identical. BitCast is the only possibility.
CastInst *NC = new BitCastInst(
AI, AI->getType(), AI->getName()+".tmp", AfterAllocaInsertPt);
AI->replaceAllUsesWith(NC);
// Normally its is forbidden to replace a CastInst's operand because it
// could cause the opcode to reflect an illegal conversion. However, we're
// replacing it here with the same value it was constructed with to simply
// make NC its user.
NC->setOperand(0, AI);
}
// Finally, scan the code looking for instructions with bad live ranges.
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
// Ignore obvious cases we don't have to handle. In particular, most
// instructions either have no uses or only have a single use inside the
// current block. Ignore them quickly.
Instruction *Inst = II;
if (Inst->use_empty()) continue;
if (Inst->hasOneUse() &&
cast<Instruction>(Inst->use_back())->getParent() == BB &&
!isa<PHINode>(Inst->use_back())) continue;
// If this is an alloca in the entry block, it's not a real register
// value.
if (AllocaInst *AI = dyn_cast<AllocaInst>(Inst))
if (isa<ConstantInt>(AI->getArraySize()) && BB == F->begin())
continue;
// Avoid iterator invalidation by copying users to a temporary vector.
std::vector<Instruction*> Users;
for (Value::use_iterator UI = Inst->use_begin(), E = Inst->use_end();
UI != E; ++UI) {
Instruction *User = cast<Instruction>(*UI);
if (User->getParent() != BB || isa<PHINode>(User))
Users.push_back(User);
}
// Scan all of the uses and see if the live range is live across an unwind
// edge. If we find a use live across an invoke edge, create an alloca
// and spill the value.
std::set<InvokeInst*> InvokesWithStoreInserted;
// Find all of the blocks that this value is live in.
std::set<BasicBlock*> LiveBBs;
LiveBBs.insert(Inst->getParent());
while (!Users.empty()) {
Instruction *U = Users.back();
Users.pop_back();
if (!isa<PHINode>(U)) {
MarkBlocksLiveIn(U->getParent(), LiveBBs);
} else {
// Uses for a PHI node occur in their predecessor block.
PHINode *PN = cast<PHINode>(U);
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
if (PN->getIncomingValue(i) == Inst)
MarkBlocksLiveIn(PN->getIncomingBlock(i), LiveBBs);
}
}
// Now that we know all of the blocks that this thing is live in, see if
// it includes any of the unwind locations.
bool NeedsSpill = false;
for (unsigned i = 0, e = Invokes.size(); i != e; ++i) {
BasicBlock *UnwindBlock = Invokes[i]->getUnwindDest();
if (UnwindBlock != BB && LiveBBs.count(UnwindBlock)) {
NeedsSpill = true;
}
}
// If we decided we need a spill, do it.
if (NeedsSpill) {
++NumSpilled;
DemoteRegToStack(*Inst, true);
}
}
}
void Interpreter::visitCastInst(CastInst &I) {
ExecutionContext &SF = ECStack.back();
SetValue(&I, executeCastOperation(I.getOperand(0), I.getType(), SF), SF);
}
void DSWP::insertConsume(Instruction *u, Instruction *v, DType dtype, int channel, int uthread, int vthread) {
Instruction *oldu = dyn_cast<Instruction>(newToOld[u]);
Instruction *insPos = placeEquivalents[vthread][oldu];
if (insPos == NULL) {
insPos = dyn_cast<Instruction>(instMap[vthread][oldu]);
if (insPos == NULL) {
error("can't insert nowhere");
}
}
// call sync_consume(channel)
Function *fun = module->getFunction("sync_consume");
vector<Value *> args;
args.push_back(ConstantInt::get(Type::getInt32Ty(*context), channel));
CallInst *call = CallInst::Create(fun, args, "c" + itoa(channel), insPos);
if (dtype == REG) {
CastInst *cast;
string name = call->getName().str() + "_val";
if (u->getType()->isIntegerTy()) {
cast = new TruncInst(call, u->getType(), name);
}
else if (u->getType()->isFloatingPointTy()) {
if (u->getType()->isFloatTy())
error("cannot deal with double");
cast = new BitCastInst(call, u->getType(), name);
}
else if (u->getType()->isPointerTy()){
cast = new IntToPtrInst(call, u->getType(), name);
} else {
error("what's the hell type");
}
cast->insertBefore(insPos);
// replace the uses
for (Instruction::use_iterator ui = oldu->use_begin(), ue = oldu->use_end(); ui != ue; ++ui) {
Instruction *user = dyn_cast<Instruction>(*ui);
if (user == NULL) {
error("used by a non-instruction?");
}
// make sure it's in the same function...
if (user->getParent()->getParent() != v->getParent()->getParent()) {
continue;
}
// call replaceUses so that it handles phi nodes
map<Value *, Value *> reps;
reps[oldu] = cast;
replaceUses(user, reps);
}
} /* TODO: need to handle true memory dependences more than just syncing?
else if (dtype == DTRUE) { //READ after WRITE
error("check mem dep!!");
if (!isa<LoadInst>(v)) {
error("not true dependency");
}
BitCastInst *cast = new BitCastInst(call, v->getType(), call->getName().str() + "_ptr");
cast->insertBefore(v);
// replace the v with 'cast' in v's thread:
// (other thread with be dealed using dependence)
for (Instruction::use_iterator ui = v->use_begin(), ue = v->use_end(); ui != ue; ui++) {
Instruction *user = dyn_cast<Instruction>(*ui);
if (user == NULL) {
error("how could it be NULL");
}
// int userthread = this->getNewInstAssigned(user);
if (user->getParent()->getParent() != v->getParent()->getParent()) {
continue;
}
for (unsigned i = 0; i < user->getNumOperands(); i++) {
Value * op = user->getOperand(i);
if (op == v) {
user->setOperand(i, cast);
}
}
}
} */ else {
// nothing to do
}
}
/**
* Inserts an argument.
* @param rw_newInstr rewrite rule - new instruction
* @param I instruction
* @param CalleeF function to be called
* @param variables map of found parameters from config
* @param where position of the placement of the new instruction
* @return a vector of arguments for the call that is to be inserted
* and a pointer to the instruction after/before the new call
* is going to be inserted (it is either I or some newly added
* argument)
*/
tuple<vector<Value *>, Instruction*> InsertArgument(InstrumentInstruction rw_newInstr, Instruction *I,
Function* CalleeF, const map <string, Value*>& variables, InstrumentPlacement where) {
std::vector<Value *> args;
unsigned i = 0;
Instruction* nI = I;
for (const string& arg : rw_newInstr.parameters) {
if (i == rw_newInstr.parameters.size() - 1) {
break;
}
auto var = variables.find(arg);
if (var == variables.end()) {
// NOTE: in future think also about other types than ConstantInt
int argInt;
try {
argInt = stoi(arg);
LLVMContext &Context = getGlobalContext();
Value *intValue = ConstantInt::get(Type::getInt32Ty(Context), argInt);
args.push_back(intValue);
} catch (invalid_argument) {
logger.write_error("Problem with instruction arguments: invalid argument.");
} catch (out_of_range) {
logger.write_error("Problem with instruction arguments: out of range.");
}
} else {
unsigned argIndex = 0;
for (Function::ArgumentListType::iterator sit=CalleeF->getArgumentList().begin(); sit != CalleeF->getArgumentList().end(); ++sit) {
Value *argV = &*sit;
if (i == argIndex) {
if (argV->getType() != var->second->getType()) {
//TODO other types?
if (!var->second->getType()->isPtrOrPtrVectorTy() && !var->second->getType()->isIntegerTy()) {
args.push_back(var->second);
} else {
CastInst *CastI;
if (var->second->getType()->isPtrOrPtrVectorTy()) {
CastI = CastInst::CreatePointerCast(var->second, argV->getType());
} else {
CastI = CastInst::CreateIntegerCast(var->second, argV->getType(), true); //TODO do something about signed argument
}
if (Instruction *Inst = dyn_cast<Instruction>(var->second))
CloneMetadata(Inst, CastI);
if (where == InstrumentPlacement::BEFORE) {
// we want to insert before I, that is:
// %c = cast ...
// newInstr
// I
//
// NOTE that we do not set nI in this case,
// so that the new instruction that we will insert
// is inserted before I (and after all arguments
// we added here)
CastI->insertBefore(I);
} else {
// we want to insert after I, that is:
// I
// %c = cast ...
// newInstr
//
// --> we must update the nI, so that the new
// instruction is inserted after the arguments
CastI->insertAfter(I);
nI = CastI;
}
args.push_back(CastI);
}
} else{
args.push_back(var->second);
}
break;
}
argIndex++;
}
}
i++;
}
return make_tuple(args,nI);
}
void vSSA::createSigmasIfNeeded(BasicBlock *BB)
{
TerminatorInst *ti = BB->getTerminator();
// If the condition used in the terminator instruction is a Comparison instruction:
//for each operand of the CmpInst, create sigmas, depending on some conditions
/*
if(isa<BranchInst>(ti)){
BranchInst * bc = cast<BranchInst>(ti);
if(bc->isConditional()){
Value * cond = bc->getCondition();
CmpInst *comparison = dyn_cast<CmpInst>(cond);
for (User::const_op_iterator it = comparison->op_begin(), e = comparison->op_end(); it != e; ++it) {
Value *operand = *it;
if (isa<Instruction>(operand) || isa<Argument>(operand)) {
insertSigmas(ti, operand);
}
}
}
}
*/
// CASE 1: Branch Instruction
BranchInst *bi = NULL;
SwitchInst *si = NULL;
if ((bi = dyn_cast<BranchInst>(ti))) {
if (bi->isConditional()) {
Value *condition = bi->getCondition();
ICmpInst *comparison = dyn_cast<ICmpInst>(condition);
if (comparison) {
// Create sigmas for ICmp operands
for (User::const_op_iterator opit = comparison->op_begin(), opend = comparison->op_end(); opit != opend; ++opit) {
Value *operand = *opit;
if (isa<Instruction>(operand) || isa<Argument>(operand)) {
insertSigmas(ti, operand);
// If the operand is a result of a indirect instruction (e.g. ZExt, SExt, Trunc),
// Create sigmas for the operands of the operands too
CastInst *cinst = NULL;
if ((cinst = dyn_cast<CastInst>(operand))) {
insertSigmas(ti, cinst->getOperand(0));
}
}
}
}
}
}
// CASE 2: Switch Instruction
else if ((si = dyn_cast<SwitchInst>(ti))) {
Value *condition = si->getCondition();
if (isa<Instruction>(condition) || isa<Argument>(condition)) {
insertSigmas(ti, condition);
// If the operand is a result of a indirect instruction (e.g. ZExt, SExt, Trunc),
// Create sigmas for the operands of the operands too
CastInst *cinst = NULL;
if ((cinst = dyn_cast<CastInst>(condition))) {
insertSigmas(ti, cinst->getOperand(0));
}
}
}
}
bool TypeChecksOpt::runOnModule(Module &M) {
TS = &getAnalysis<dsa::TypeSafety<TDDataStructures> >();
// Create the necessary prototypes
VoidTy = IntegerType::getVoidTy(M.getContext());
Int8Ty = IntegerType::getInt8Ty(M.getContext());
Int32Ty = IntegerType::getInt32Ty(M.getContext());
Int64Ty = IntegerType::getInt64Ty(M.getContext());
VoidPtrTy = PointerType::getUnqual(Int8Ty);
TypeTagTy = Int8Ty;
TypeTagPtrTy = PointerType::getUnqual(TypeTagTy);
Constant *memsetF = M.getOrInsertFunction ("llvm.memset.i64", VoidTy,
VoidPtrTy,
Int8Ty,
Int64Ty,
Int32Ty,
NULL);
trackGlobal = M.getOrInsertFunction("trackGlobal",
VoidTy,
VoidPtrTy,/*ptr*/
TypeTagTy,/*type*/
Int64Ty,/*size*/
Int32Ty,/*tag*/
NULL);
trackInitInst = M.getOrInsertFunction("trackInitInst",
VoidTy,
VoidPtrTy,/*ptr*/
Int64Ty,/*size*/
Int32Ty,/*tag*/
NULL);
trackUnInitInst = M.getOrInsertFunction("trackUnInitInst",
VoidTy,
VoidPtrTy,/*ptr*/
Int64Ty,/*size*/
Int32Ty,/*tag*/
NULL);
trackStoreInst = M.getOrInsertFunction("trackStoreInst",
VoidTy,
VoidPtrTy,/*ptr*/
TypeTagTy,/*type*/
Int64Ty,/*size*/
Int32Ty,/*tag*/
NULL);
checkTypeInst = M.getOrInsertFunction("checkType",
VoidTy,
TypeTagTy,/*type*/
Int64Ty,/*size*/
TypeTagPtrTy,
VoidPtrTy,/*ptr*/
Int32Ty,/*tag*/
NULL);
copyTypeInfo = M.getOrInsertFunction("copyTypeInfo",
VoidTy,
VoidPtrTy,/*dest ptr*/
VoidPtrTy,/*src ptr*/
Int64Ty,/*size*/
Int32Ty,/*tag*/
NULL);
setTypeInfo = M.getOrInsertFunction("setTypeInfo",
VoidTy,
VoidPtrTy,/*dest ptr*/
TypeTagPtrTy,/*metadata*/
Int64Ty,/*size*/
TypeTagTy,
VoidPtrTy,
Int32Ty,/*tag*/
NULL);
trackStringInput = M.getOrInsertFunction("trackStringInput",
VoidTy,
VoidPtrTy,
Int32Ty,
NULL);
getTypeTag = M.getOrInsertFunction("getTypeTag",
VoidTy,
VoidPtrTy, /*ptr*/
Int64Ty, /*size*/
TypeTagPtrTy, /*dest for type tag*/
Int32Ty, /*tag*/
NULL);
MallocFunc = M.getFunction("malloc");
for(Value::use_iterator User = trackGlobal->use_begin(); User != trackGlobal->use_end(); ++User) {
CallInst *CI = dyn_cast<CallInst>(*User);
assert(CI);
if(TS->isTypeSafe(CI->getOperand(1)->stripPointerCasts(), CI->getParent()->getParent())) {
std::vector<Value*>Args;
Args.push_back(CI->getOperand(1));
Args.push_back(CI->getOperand(3));
Args.push_back(CI->getOperand(4));
CallInst::Create(trackInitInst, Args, "", CI);
toDelete.push_back(CI);
}
}
for(Value::use_iterator User = checkTypeInst->use_begin(); User != checkTypeInst->use_end(); ++User) {
CallInst *CI = dyn_cast<CallInst>(*User);
assert(CI);
if(TS->isTypeSafe(CI->getOperand(4)->stripPointerCasts(), CI->getParent()->getParent())) {
toDelete.push_back(CI);
}
}
for(Value::use_iterator User = trackStoreInst->use_begin(); User != trackStoreInst->use_end(); ++User) {
CallInst *CI = dyn_cast<CallInst>(*User);
assert(CI);
if(TS->isTypeSafe(CI->getOperand(1)->stripPointerCasts(), CI->getParent()->getParent())) {
toDelete.push_back(CI);
}
}
// for alloca's if they are type known
// assume initialized with TOP
for(Value::use_iterator User = trackUnInitInst->use_begin(); User != trackUnInitInst->use_end(); ) {
CallInst *CI = dyn_cast<CallInst>(*(User++));
assert(CI);
// check if operand is an alloca inst.
if(TS->isTypeSafe(CI->getOperand(1)->stripPointerCasts(), CI->getParent()->getParent())) {
CI->setCalledFunction(trackInitInst);
if(AllocaInst *AI = dyn_cast<AllocaInst>(CI->getOperand(1)->stripPointerCasts())) {
// Initialize the allocation to NULL
std::vector<Value *> Args2;
Args2.push_back(CI->getOperand(1));
Args2.push_back(ConstantInt::get(Int8Ty, 0));
Args2.push_back(CI->getOperand(2));
Args2.push_back(ConstantInt::get(Int32Ty, AI->getAlignment()));
CallInst::Create(memsetF, Args2, "", CI);
}
}
}
if(MallocFunc) {
for(Value::use_iterator User = MallocFunc->use_begin(); User != MallocFunc->use_end(); User ++) {
CallInst *CI = dyn_cast<CallInst>(*User);
if(!CI)
continue;
if(TS->isTypeSafe(CI, CI->getParent()->getParent())){
CastInst *BCI = BitCastInst::CreatePointerCast(CI, VoidPtrTy);
CastInst *Size = CastInst::CreateSExtOrBitCast(CI->getOperand(1), Int64Ty);
Size->insertAfter(CI);
BCI->insertAfter(Size);
std::vector<Value *>Args;
Args.push_back(BCI);
Args.push_back(Size);
Args.push_back(ConstantInt::get(Int32Ty, 0));
CallInst *CINew = CallInst::Create(trackInitInst, Args);
CINew->insertAfter(BCI);
}
}
}
// also do for mallocs/calloc/other allocators???
// other allocators??
for(Value::use_iterator User = copyTypeInfo->use_begin(); User != copyTypeInfo->use_end(); ++User) {
CallInst *CI = dyn_cast<CallInst>(*User);
assert(CI);
if(TS->isTypeSafe(CI->getOperand(1)->stripPointerCasts(), CI->getParent()->getParent())) {
std::vector<Value*> Args;
Args.push_back(CI->getOperand(1));
Args.push_back(CI->getOperand(3)); // size
Args.push_back(CI->getOperand(4));
CallInst::Create(trackInitInst, Args, "", CI);
toDelete.push_back(CI);
}
}
for(Value::use_iterator User = setTypeInfo->use_begin(); User != setTypeInfo->use_end(); ++User) {
CallInst *CI = dyn_cast<CallInst>(*User);
assert(CI);
if(TS->isTypeSafe(CI->getOperand(1)->stripPointerCasts(), CI->getParent()->getParent())) {
std::vector<Value*> Args;
Args.push_back(CI->getOperand(1));
Args.push_back(CI->getOperand(3)); // size
Args.push_back(CI->getOperand(6));
CallInst::Create(trackInitInst, Args, "", CI);
toDelete.push_back(CI);
}
}
for(Value::use_iterator User = getTypeTag->use_begin(); User != getTypeTag->use_end(); ++User) {
CallInst *CI = dyn_cast<CallInst>(*User);
assert(CI);
if(TS->isTypeSafe(CI->getOperand(1)->stripPointerCasts(), CI->getParent()->getParent())) {
AllocaInst *AI = dyn_cast<AllocaInst>(CI->getOperand(3)->stripPointerCasts());
assert(AI);
std::vector<Value*>Args;
Args.push_back(CI->getOperand(3));
Args.push_back(ConstantInt::get(Int8Ty, 255));
Args.push_back(CI->getOperand(2));
Args.push_back(ConstantInt::get(Int32Ty, AI->getAlignment()));
CallInst::Create(memsetF, Args, "", CI);
toDelete.push_back(CI);
}
}
numSafe += toDelete.size();
while(!toDelete.empty()) {
Instruction *I = toDelete.back();
toDelete.pop_back();
I->eraseFromParent();
}
return (numSafe > 0);
}
/// InsertPHITranslatedPointer - Insert a computation of the PHI translated
/// version of 'V' for the edge PredBB->CurBB into the end of the PredBB
/// block. All newly created instructions are added to the NewInsts list.
/// This returns null on failure.
///
Value *PHITransAddr::
InsertPHITranslatedSubExpr(Value *InVal, BasicBlock *CurBB,
BasicBlock *PredBB, const DominatorTree &DT,
SmallVectorImpl<Instruction*> &NewInsts) {
// See if we have a version of this value already available and dominating
// PredBB. If so, there is no need to insert a new instance of it.
PHITransAddr Tmp(InVal, DL, AC);
if (!Tmp.PHITranslateValue(CurBB, PredBB, &DT, /*MustDominate=*/true))
return Tmp.getAddr();
// We don't need to PHI translate values which aren't instructions.
auto *Inst = dyn_cast<Instruction>(InVal);
if (!Inst)
return nullptr;
// Handle cast of PHI translatable value.
if (CastInst *Cast = dyn_cast<CastInst>(Inst)) {
if (!isSafeToSpeculativelyExecute(Cast)) return nullptr;
Value *OpVal = InsertPHITranslatedSubExpr(Cast->getOperand(0),
CurBB, PredBB, DT, NewInsts);
if (!OpVal) return nullptr;
// Otherwise insert a cast at the end of PredBB.
CastInst *New = CastInst::Create(Cast->getOpcode(), OpVal, InVal->getType(),
InVal->getName() + ".phi.trans.insert",
PredBB->getTerminator());
New->setDebugLoc(Inst->getDebugLoc());
NewInsts.push_back(New);
return New;
}
// Handle getelementptr with at least one PHI operand.
if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Inst)) {
SmallVector<Value*, 8> GEPOps;
BasicBlock *CurBB = GEP->getParent();
for (unsigned i = 0, e = GEP->getNumOperands(); i != e; ++i) {
Value *OpVal = InsertPHITranslatedSubExpr(GEP->getOperand(i),
CurBB, PredBB, DT, NewInsts);
if (!OpVal) return nullptr;
GEPOps.push_back(OpVal);
}
GetElementPtrInst *Result = GetElementPtrInst::Create(
GEP->getSourceElementType(), GEPOps[0], makeArrayRef(GEPOps).slice(1),
InVal->getName() + ".phi.trans.insert", PredBB->getTerminator());
Result->setDebugLoc(Inst->getDebugLoc());
Result->setIsInBounds(GEP->isInBounds());
NewInsts.push_back(Result);
return Result;
}
#if 0
// FIXME: This code works, but it is unclear that we actually want to insert
// a big chain of computation in order to make a value available in a block.
// This needs to be evaluated carefully to consider its cost trade offs.
// Handle add with a constant RHS.
if (Inst->getOpcode() == Instruction::Add &&
isa<ConstantInt>(Inst->getOperand(1))) {
// PHI translate the LHS.
Value *OpVal = InsertPHITranslatedSubExpr(Inst->getOperand(0),
CurBB, PredBB, DT, NewInsts);
if (OpVal == 0) return 0;
BinaryOperator *Res = BinaryOperator::CreateAdd(OpVal, Inst->getOperand(1),
InVal->getName()+".phi.trans.insert",
PredBB->getTerminator());
Res->setHasNoSignedWrap(cast<BinaryOperator>(Inst)->hasNoSignedWrap());
Res->setHasNoUnsignedWrap(cast<BinaryOperator>(Inst)->hasNoUnsignedWrap());
NewInsts.push_back(Res);
return Res;
}
#endif
return nullptr;
}
bool partition::runOnLoop(Loop* L, LPPassManager &LPM) {
errs() << "*************************** Loop encountered: ************************" << '\n' << L->getHeader()->getName() << '\n';
if (function->getName() != "main")
return false;
IntegerType* int32Ty = Type::getInt32Ty(*context);
IntegerType* int64Ty = Type::getInt64Ty(*context);
PointerType* voidPtrTy = Type::getInt8PtrTy(*context);
FunctionType* funcTy = FunctionType::get(int32Ty, false);
Constant* func1_c;
Function* func1;
func1_c = module->getOrInsertFunction("func1", funcTy);
func1 = cast<Function>(func1_c);
Function* pro = module->getFunction("produce");
Function* con = module->getFunction("consume");
BasicBlock* func1EntryBlock = BasicBlock::Create(*context, "entry.func1", func1);
AllocaInst* i_var = new AllocaInst(int32Ty, NULL, 4, "i", func1EntryBlock);
Value* liveIn;
BasicBlock *forCond, *forBody, *forInc;
ValueToValueMapTy VMap;
std::map<BasicBlock *, BasicBlock *> BlockMap;
for (Loop::block_iterator BB = L->block_begin(), BBe = L->block_end(); BB != BBe; ++BB) {
BasicBlock* func1Block = CloneBasicBlock(*BB, VMap, ".func1", func1);
BlockMap[*BB] = func1Block;
if ((*BB)->getName() == "for.cond")
forCond = func1Block;
if ((*BB)->getName() == "for.body")
forBody = func1Block;
if ((*BB)->getName() == "for.inc")
forInc = func1Block;
for (BasicBlock::iterator it = func1Block->begin(), ite = func1Block->end(); it != ite; ++it) {
for (User::op_iterator oit = it->op_begin(), oite = it->op_end(); oit != oite; ++oit) {
if (VMap[*oit] != NULL) {
*oit = VMap[*oit];
} else {
Constant* cons = dyn_cast<Constant>(*oit);
BranchInst* br = dyn_cast<BranchInst>(it);
if (cons == NULL && br == NULL) {
liveIn = *oit;
*oit = i_var;
}
}
}
}
if ((*BB)->getName() == "for.body") {
Instruction* term = (*BB)->getTerminator();
term->removeFromParent();
for (int i = 0; i < 7; i++) {
(*BB)->back().eraseFromParent();
}
term->insertAfter(&(*BB)->back());
(*BB)->front().eraseFromParent();
LoadInst* load = new LoadInst(liveIn, "", false, 4, term);
std::vector<Value *> produce_args;
CastInst* cast = CastInst::CreateIntegerCast(load, int64Ty, true);
cast->insertAfter(load);
produce_args.push_back(cast);
ConstantInt* val = ConstantInt::get(int32Ty, (uint32_t) 3);
produce_args.push_back(val);
CallInst::Create(pro, ArrayRef<Value*>(produce_args), "", term);
produce_args.pop_back();
val = ConstantInt::get(int32Ty, (uint32_t) 2);
produce_args.push_back(val);
CallInst::Create(pro, ArrayRef<Value*>(produce_args), "", term);
}
}
// set branch instructions to restructure the CFG in created function
BasicBlock* func1EndBlock = BasicBlock::Create(*context, "if.end.func1", func1);
BasicBlock* garbageBB = BasicBlock::Create(*context, "garbage", func1);
ConstantInt* retVal_g = ConstantInt::get(int32Ty, (uint32_t) 0);
ReturnInst* ret_g = ReturnInst::Create(*context, retVal_g, garbageBB);
for (Function::iterator fit = func1->begin(), fite = func1->end(); fit != fite; ++fit) {
if (fit->getTerminator() == NULL || fit->getName() == "garbage")
continue;
BranchInst* br = dyn_cast<BranchInst>(fit->getTerminator());
int numSuccessors = br->getNumSuccessors();
for (int i = 0; i < numSuccessors; i++) {
BasicBlock* successor = br->getSuccessor(i);
if (BlockMap[successor] != NULL) {
br->setSuccessor(i, BlockMap[successor]);
}
else {
br->setSuccessor(i, func1EndBlock);
}
}
/*
if (fit->getName() == "for.body.func1") {
for (int i = 0; i < 4; i++) {
BasicBlock::iterator it = fit->begin();
it->moveBefore(ret_g);
}
}
*/
}
garbageBB->eraseFromParent();
BranchInst* br = dyn_cast<BranchInst>(forBody->getTerminator());
br->setSuccessor(0, forCond);
forInc->eraseFromParent();
// Create return instruction for func1EndBlock and set a branch from loop header to func1EndBlock
ConstantInt* retVal = ConstantInt::get(int32Ty, (uint32_t) 0);
ReturnInst* ret1 = ReturnInst::Create(*context, retVal, func1EndBlock);
BasicBlock* loopHeader = BlockMap.at(L->getHeader());
BranchInst* brInst = BranchInst::Create(loopHeader, func1EntryBlock);
// add produce function call
std::vector<Value *> produce_args;
ConstantInt* val = ConstantInt::get(int64Ty, (uint64_t) 0);
produce_args.push_back(val);
val = ConstantInt::get(int32Ty, (uint32_t) 5);
produce_args.push_back(val);
CallInst::Create(pro, ArrayRef<Value*>(produce_args), "", func1EndBlock->getTerminator());
// add consume function call
int q_id = 2;
for (Value::use_iterator uit = i_var->use_begin(), uite = i_var->use_end(); uit != uite; ++uit) {
std::vector<Value *> consume_args;
ConstantInt* val = ConstantInt::get(int32Ty, (uint32_t) q_id);
consume_args.push_back(val);
CallInst* call = CallInst::Create(con, ArrayRef<Value*>(consume_args));
Instruction* inst = dyn_cast<Instruction>(*uit);
call->insertAfter(inst);
CastInst* cast = CastInst::CreateIntegerCast(call, int32Ty, true);
cast->insertAfter(call);
(*uit)->replaceAllUsesWith(cast);
inst->eraseFromParent();
q_id++;
}
i_var->eraseFromParent();
// add produce and consume function calls to main thread
// transmit the function pointer to created function by a produce call
BasicBlock* loopPreheader = L->getLoopPreheader();
produce_args.clear();
CastInst* cast = CastInst::CreatePointerCast(func1, int64Ty);
cast->insertBefore(loopPreheader->getTerminator());
produce_args.push_back(cast);
val = ConstantInt::get(int32Ty, (uint32_t) 0);
produce_args.push_back(val);
CallInst::Create(pro, ArrayRef<Value*>(produce_args), "", loopPreheader->getTerminator());
// transmit induction variable to created function by a produce call
Instruction* load = &L->getHeader()->front();
produce_args.clear();
cast = CastInst::CreateIntegerCast(load, int64Ty, true);
cast->insertAfter(load);
produce_args.push_back(cast);
val = ConstantInt::get(int32Ty, (uint32_t) 4);
produce_args.push_back(val);
CallInst::Create(pro, ArrayRef<Value*>(produce_args))->insertAfter(cast);
return true;
}
