/// See comments in Cloning.h.
BasicBlock *llvm::CloneBasicBlock(const BasicBlock *BB,
ValueToValueMapTy &VMap,
const Twine &NameSuffix, Function *F,
ClonedCodeInfo *CodeInfo) {
BasicBlock *NewBB = BasicBlock::Create(BB->getContext(), "", F);
if (BB->hasName()) NewBB->setName(BB->getName()+NameSuffix);
bool hasCalls = false, hasDynamicAllocas = false, hasStaticAllocas = false;
// Loop over all instructions, and copy them over.
for (BasicBlock::const_iterator II = BB->begin(), IE = BB->end();
II != IE; ++II) {
Instruction *NewInst = II->clone();
if (II->hasName())
NewInst->setName(II->getName()+NameSuffix);
NewBB->getInstList().push_back(NewInst);
VMap[&*II] = NewInst; // Add instruction map to value.
hasCalls |= (isa<CallInst>(II) && !isa<DbgInfoIntrinsic>(II));
if (const AllocaInst *AI = dyn_cast<AllocaInst>(II)) {
if (isa<ConstantInt>(AI->getArraySize()))
hasStaticAllocas = true;
else
hasDynamicAllocas = true;
}
}
if (CodeInfo) {
CodeInfo->ContainsCalls |= hasCalls;
CodeInfo->ContainsDynamicAllocas |= hasDynamicAllocas;
CodeInfo->ContainsDynamicAllocas |= hasStaticAllocas &&
BB != &BB->getParent()->getEntryBlock();
}
return NewBB;
}
/// findStopPoint - Find the stoppoint coressponding to this instruction, that
/// is the stoppoint that dominates this instruction.
const DbgStopPointInst *findStopPoint(const Instruction *Inst) {
if (const DbgStopPointInst *DSI = dyn_cast<DbgStopPointInst>(Inst))
return DSI;
const BasicBlock *BB = Inst->getParent();
BasicBlock::const_iterator I = Inst, B;
while (BB) {
B = BB->begin();
// A BB consisting only of a terminator can't have a stoppoint.
while (I != B) {
--I;
if (const DbgStopPointInst *DSI = dyn_cast<DbgStopPointInst>(I))
return DSI;
}
// This BB didn't have a stoppoint: if there is only one predecessor, look
// for a stoppoint there. We could use getIDom(), but that would require
// dominator info.
BB = I->getParent()->getUniquePredecessor();
if (BB)
I = BB->getTerminator();
}
return 0;
}
void ExampleFunctionPrinter(raw_ostream& O, const Function& F) {
for (Function::const_iterator FI = F.begin(), FE = F.end(); FI != FE; ++FI) {
const BasicBlock* block = FI;
O << block->getName() << ":\n";
PrintInstructionOps(O, NULL);
for (BasicBlock::const_iterator BI = block->begin(), BE = block->end();
BI != BE; ++BI) {
BI->print(O);
PrintInstructionOps(O, &(*BI));
}
}
}
void NaClValueEnumerator::incorporateFunction(const Function &F) {
InstructionCount = 0;
NumModuleValues = Values.size();
// Make sure no insertions outside of a function.
assert(FnForwardTypeRefs.empty());
// Adding function arguments to the value table.
for (Function::const_arg_iterator I = F.arg_begin(), E = F.arg_end();
I != E; ++I)
EnumerateValue(I);
FirstFuncConstantID = Values.size();
// Add all function-level constants to the value table.
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) {
if (const SwitchInst *SI = dyn_cast<SwitchInst>(I)) {
// Handle switch instruction specially, so that we don't write
// out unnecessary vector/array constants used to model case selectors.
if (isa<Constant>(SI->getCondition())) {
EnumerateValue(SI->getCondition());
}
} else {
for (User::const_op_iterator OI = I->op_begin(), E = I->op_end();
OI != E; ++OI) {
if ((isa<Constant>(*OI) && !isa<GlobalValue>(*OI)) ||
isa<InlineAsm>(*OI))
EnumerateValue(*OI);
}
}
}
BasicBlocks.push_back(BB);
ValueMap[BB] = BasicBlocks.size();
}
// Optimize the constant layout.
OptimizeConstants(FirstFuncConstantID, Values.size());
FirstInstID = Values.size();
// Add all of the instructions.
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) {
if (!I->getType()->isVoidTy())
EnumerateValue(I);
}
}
}
/**
* Print the given basic block.
*
* @param block the basic block
*/
void JVMWriter::printBasicBlock(const BasicBlock *block) {
printLabel(getLabelName(block));
if (trace) {
if (block->hasName()) {
std::string n = block->getName();
printTrc(n + ":");
}
}
for(BasicBlock::const_iterator i = block->begin(), e = block->end();
i != e; i++) {
instNum++;
if (trace)
printSimpleInstruction(".line", utostr(trcLineNum+1));
else if(debug >= 1)
printSimpleInstruction(".line", utostr(instNum));
if(debug >= 3 || trace) {
// print current instruction as comment
// note that this block of code significantly increases
// code generation time
std::string str;
raw_string_ostream ss(str); ss << *i;
ss.flush();
if (trace)
printTrc(str);
if (debug >= 3) {
std::string::size_type pos = 0;
while((pos = str.find("\n", pos)) != std::string::npos)
str.replace(pos++, 1, "\n;");
out << ';' << str << '\n';
}
}
if(i->getOpcode() == Instruction::PHI)
// don't handle phi instruction in current block
continue;
printInstruction(i);
if(i->getType() != Type::getVoidTy(block->getContext())
&& i->getOpcode() != Instruction::Invoke)
// instruction doesn't return anything, or is an invoke instruction
// which handles storing the return value itself
printValueStore(i);
}
}
// Test whether two basic blocks have equivalent behaviour.
int FunctionComparator::cmpBasicBlocks(const BasicBlock *BBL,
const BasicBlock *BBR) const {
BasicBlock::const_iterator InstL = BBL->begin(), InstLE = BBL->end();
BasicBlock::const_iterator InstR = BBR->begin(), InstRE = BBR->end();
do {
bool needToCmpOperands = true;
if (int Res = cmpOperations(&*InstL, &*InstR, needToCmpOperands))
return Res;
if (needToCmpOperands) {
assert(InstL->getNumOperands() == InstR->getNumOperands());
for (unsigned i = 0, e = InstL->getNumOperands(); i != e; ++i) {
Value *OpL = InstL->getOperand(i);
Value *OpR = InstR->getOperand(i);
if (int Res = cmpValues(OpL, OpR))
return Res;
// cmpValues should ensure this is true.
assert(cmpTypes(OpL->getType(), OpR->getType()) == 0);
}
}
++InstL;
++InstR;
} while (InstL != InstLE && InstR != InstRE);
if (InstL != InstLE && InstR == InstRE)
return 1;
if (InstL == InstLE && InstR != InstRE)
return -1;
return 0;
}
bool Value::isUsedInBasicBlock(const BasicBlock *BB) const {
// This can be computed either by scanning the instructions in BB, or by
// scanning the use list of this Value. Both lists can be very long, but
// usually one is quite short.
//
// Scan both lists simultaneously until one is exhausted. This limits the
// search to the shorter list.
BasicBlock::const_iterator BI = BB->begin(), BE = BB->end();
const_user_iterator UI = user_begin(), UE = user_end();
for (; BI != BE && UI != UE; ++BI, ++UI) {
// Scan basic block: Check if this Value is used by the instruction at BI.
if (is_contained(BI->operands(), this))
return true;
// Scan use list: Check if the use at UI is in BB.
const auto *User = dyn_cast<Instruction>(*UI);
if (User && User->getParent() == BB)
return true;
}
return false;
}
void HeterotbbTransform::copy_function (Function* NF, Function* F) {
DenseMap<const Value*, Value *> ValueMap;
// Get the names of the parameters for old function
for(Function::arg_iterator FI = F->arg_begin(), FE=F->arg_end(), DI=NF->arg_begin(); FE!=FI; ++FI,++DI) {
DI->setName(FI->getName());
ValueMap[FI]=DI;
}
for (Function::const_iterator BI=F->begin(),BE = F->end(); BI != BE; ++BI) {
const BasicBlock &FBB = *BI;
BasicBlock *NFBB = BasicBlock::Create(FBB.getContext(), "", NF);
ValueMap[&FBB] = NFBB;
if (FBB.hasName()) {
NFBB->setName(FBB.getName());
//DEBUG(dbgs()<<NFBB->getName()<<"\n");
}
for (BasicBlock::const_iterator II = FBB.begin(), IE = FBB.end(); II != IE; ++II) {
Instruction *NFInst = II->clone(/*F->getContext()*/);
if (II->hasName()) NFInst->setName(II->getName());
const Instruction *FInst = &(*II);
rewrite_instruction((Instruction *)FInst, NFInst, ValueMap);
NFBB->getInstList().push_back(NFInst);
ValueMap[II] = NFInst;
}
}
// Remap the instructions again to take care of forward jumps
for (Function::iterator BB = NF->begin(), BE=NF->end(); BB != BE; ++ BB) {
for (BasicBlock::iterator II = BB->begin(); II != BB->end(); ++II) {
int opIdx = 0;
//DEBUG(dbgs()<<*II<<"\n");
for (User::op_iterator i = II->op_begin(), e = II->op_end(); i != e; ++i, opIdx++) {
Value *V = *i;
if (ValueMap[V] != NULL) {
II->setOperand(opIdx, ValueMap[V]);
}
}
}
}
//NF->dump();
}
/// Test if the given instruction is in a position to be optimized
/// with a tail-call. This roughly means that it's in a block with
/// a return and there's nothing that needs to be scheduled
/// between it and the return.
///
/// This function only tests target-independent requirements.
bool llvm::isInTailCallPosition(ImmutableCallSite CS, const TargetMachine &TM) {
const Instruction *I = CS.getInstruction();
const BasicBlock *ExitBB = I->getParent();
const Instruction *Term = ExitBB->getTerminator();
const ReturnInst *Ret = dyn_cast<ReturnInst>(Term);
// The block must end in a return statement or unreachable.
//
// FIXME: Decline tailcall if it's not guaranteed and if the block ends in
// an unreachable, for now. The way tailcall optimization is currently
// implemented means it will add an epilogue followed by a jump. That is
// not profitable. Also, if the callee is a special function (e.g.
// longjmp on x86), it can end up causing miscompilation that has not
// been fully understood.
if (!Ret &&
(!TM.Options.GuaranteedTailCallOpt || !isa<UnreachableInst>(Term)))
return false;
// If I will have a chain, make sure no other instruction that will have a
// chain interposes between I and the return.
if (I->mayHaveSideEffects() || I->mayReadFromMemory() ||
!isSafeToSpeculativelyExecute(I))
for (BasicBlock::const_iterator BBI = std::prev(ExitBB->end(), 2);; --BBI) {
if (&*BBI == I)
break;
// Debug info intrinsics do not get in the way of tail call optimization.
if (isa<DbgInfoIntrinsic>(BBI))
continue;
// A lifetime end intrinsic should not stop tail call optimization.
if (const IntrinsicInst *II = dyn_cast<IntrinsicInst>(BBI))
if (II->getIntrinsicID() == Intrinsic::lifetime_end)
continue;
if (BBI->mayHaveSideEffects() || BBI->mayReadFromMemory() ||
!isSafeToSpeculativelyExecute(&*BBI))
return false;
}
const Function *F = ExitBB->getParent();
return returnTypeIsEligibleForTailCall(
F, I, Ret, *TM.getSubtargetImpl(*F)->getTargetLowering());
}
/// isUsedInBasicBlock - Return true if this value is used in the specified
/// basic block.
bool Value::isUsedInBasicBlock(const BasicBlock *BB) const {
// Start by scanning over the instructions looking for a use before we start
// the expensive use iteration.
unsigned MaxBlockSize = 3;
for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
if (std::find(I->op_begin(), I->op_end(), this) != I->op_end())
return true;
if (MaxBlockSize-- == 0) // If the block is larger fall back to use_iterator
break;
}
if (MaxBlockSize != 0) // We scanned the entire block and found no use.
return false;
for (const_use_iterator I = use_begin(), E = use_end(); I != E; ++I) {
const Instruction *User = dyn_cast<Instruction>(*I);
if (User && User->getParent() == BB)
return true;
}
return false;
}
void ValueEnumerator::incorporateFunction(const Function &F) {
NumModuleValues = Values.size();
// Adding function arguments to the value table.
for(Function::const_arg_iterator I = F.arg_begin(), E = F.arg_end();
I != E; ++I)
EnumerateValue(I);
FirstFuncConstantID = Values.size();
// Add all function-level constants to the value table.
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 ((isa<Constant>(*OI) && !isa<GlobalValue>(*OI)) ||
isa<InlineAsm>(*OI))
EnumerateValue(*OI);
}
BasicBlocks.push_back(BB);
ValueMap[BB] = BasicBlocks.size();
}
// Optimize the constant layout.
OptimizeConstants(FirstFuncConstantID, Values.size());
// Add the function's parameter attributes so they are available for use in
// the function's instruction.
EnumerateAttributes(F.getAttributes());
FirstInstID = Values.size();
// Add all of the instructions.
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) {
if (I->getType() != Type::getVoidTy(F.getContext()))
EnumerateValue(I);
}
}
}
void ValueEnumerator::incorporateFunction(const Function &F) {
InstructionCount = 0;
NumModuleValues = Values.size();
NumModuleMDs = MDs.size();
// Adding function arguments to the value table.
for (Function::const_arg_iterator I = F.arg_begin(), E = F.arg_end();
I != E; ++I)
EnumerateValue(I);
FirstFuncConstantID = Values.size();
// Add all function-level constants to the value table.
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 ((isa<Constant>(*OI) && !isa<GlobalValue>(*OI)) ||
isa<InlineAsm>(*OI))
EnumerateValue(*OI);
}
BasicBlocks.push_back(BB);
ValueMap[BB] = BasicBlocks.size();
}
// Optimize the constant layout.
OptimizeConstants(FirstFuncConstantID, Values.size());
// Add the function's parameter attributes so they are available for use in
// the function's instruction.
EnumerateAttributes(F.getAttributes());
FirstInstID = Values.size();
SmallVector<LocalAsMetadata *, 8> FnLocalMDVector;
// Add all of the instructions.
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 (auto *MD = dyn_cast<MetadataAsValue>(&*OI))
if (auto *Local = dyn_cast<LocalAsMetadata>(MD->getMetadata()))
// Enumerate metadata after the instructions they might refer to.
FnLocalMDVector.push_back(Local);
}
if (!I->getType()->isVoidTy())
EnumerateValue(I);
}
}
// Add all of the function-local metadata.
for (unsigned i = 0, e = FnLocalMDVector.size(); i != e; ++i)
EnumerateFunctionLocalMetadata(FnLocalMDVector[i]);
}
// Print instruction location, falls back to printing function location,
// (and LLVM instruction if specified).
void printLocation(const llvm::Instruction *I, bool fallback) {
const BasicBlock *BB = I->getParent();
bool approx = false;
BasicBlock::const_iterator It = I;
do {
BasicBlock::const_iterator ItB = BB->begin();
while (It != ItB) {
if (MDNode *N = It->getMetadata("dbg")) {
DILocation Loc(N);
errs() << /*Loc.getDirectory() << "/" <<*/ Loc.getFilename()
<< ":" << Loc.getLineNumber();
if (unsigned Col = Loc.getColumnNumber()) {
errs() << ":" << Col;
}
if (approx)
errs() << "(?)";
errs() << ": ";
DIScope Scope = Loc.getScope();
while (Scope.isLexicalBlock()) {
DILexicalBlock LB(Scope.getNode());
Scope = LB.getContext();
}
if (Scope.isSubprogram()) {
DISubprogram SP(Scope.getNode());
errs() << "in function '" << SP.getDisplayName() << "': ";
}
return;
}
approx = true;
--It;
}
BB = BB->getUniquePredecessor();
if (BB)
It = BB->end();
} while (BB);
printLocation(I->getParent()->getParent());
if (fallback)
errs() << *I << ":\n";
}
/// WriteFunction - Emit a function body to the module stream.
static void WriteFunction(const Function &F, NaClValueEnumerator &VE,
NaClBitstreamWriter &Stream) {
Stream.EnterSubblock(naclbitc::FUNCTION_BLOCK_ID);
VE.incorporateFunction(F);
SmallVector<unsigned, 64> Vals;
// Emit the number of basic blocks, so the reader can create them ahead of
// time.
Vals.push_back(VE.getBasicBlocks().size());
Stream.EmitRecord(naclbitc::FUNC_CODE_DECLAREBLOCKS, Vals);
Vals.clear();
// If there are function-local constants, emit them now.
unsigned CstStart, CstEnd;
VE.getFunctionConstantRange(CstStart, CstEnd);
WriteConstants(CstStart, CstEnd, VE, Stream);
// Keep a running idea of what the instruction ID is.
unsigned InstID = CstEnd;
// Finally, emit all the instructions, in order.
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) {
if (WriteInstruction(*I, InstID, VE, Stream, Vals) &&
!I->getType()->isVoidTy())
++InstID;
}
// Emit names for instructions etc.
if (PNaClAllowLocalSymbolTables)
WriteValueSymbolTable(F.getValueSymbolTable(), VE, Stream);
VE.purgeFunction();
Stream.ExitBlock();
}
/// 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());
}
/// Test if the given instruction is in a position to be optimized
/// with a tail-call. This roughly means that it's in a block with
/// a return and there's nothing that needs to be scheduled
/// between it and the return.
///
/// This function only tests target-independent requirements.
bool llvm::isInTailCallPosition(ImmutableCallSite CS, Attribute CalleeRetAttr,
const TargetLowering &TLI) {
const Instruction *I = CS.getInstruction();
const BasicBlock *ExitBB = I->getParent();
const TerminatorInst *Term = ExitBB->getTerminator();
const ReturnInst *Ret = dyn_cast<ReturnInst>(Term);
// The block must end in a return statement or unreachable.
//
// FIXME: Decline tailcall if it's not guaranteed and if the block ends in
// an unreachable, for now. The way tailcall optimization is currently
// implemented means it will add an epilogue followed by a jump. That is
// not profitable. Also, if the callee is a special function (e.g.
// longjmp on x86), it can end up causing miscompilation that has not
// been fully understood.
if (!Ret &&
(!TLI.getTargetMachine().Options.GuaranteedTailCallOpt ||
!isa<UnreachableInst>(Term)))
return false;
// If I will have a chain, make sure no other instruction that will have a
// chain interposes between I and the return.
if (I->mayHaveSideEffects() || I->mayReadFromMemory() ||
!isSafeToSpeculativelyExecute(I))
for (BasicBlock::const_iterator BBI = prior(prior(ExitBB->end())); ;
--BBI) {
if (&*BBI == I)
break;
// Debug info intrinsics do not get in the way of tail call optimization.
if (isa<DbgInfoIntrinsic>(BBI))
continue;
if (BBI->mayHaveSideEffects() || BBI->mayReadFromMemory() ||
!isSafeToSpeculativelyExecute(BBI))
return false;
}
// If the block ends with a void return or unreachable, it doesn't matter
// what the call's return type is.
if (!Ret || Ret->getNumOperands() == 0) return true;
// If the return value is undef, it doesn't matter what the call's
// return type is.
if (isa<UndefValue>(Ret->getOperand(0))) return true;
// Conservatively require the attributes of the call to match those of
// the return. Ignore noalias because it doesn't affect the call sequence.
const Function *F = ExitBB->getParent();
Attribute CallerRetAttr = F->getAttributes().getRetAttributes();
if (AttrBuilder(CalleeRetAttr).removeAttribute(Attribute::NoAlias) !=
AttrBuilder(CallerRetAttr).removeAttribute(Attribute::NoAlias))
return false;
// It's not safe to eliminate the sign / zero extension of the return value.
if (CallerRetAttr.hasAttribute(Attribute::ZExt) ||
CallerRetAttr.hasAttribute(Attribute::SExt))
return false;
// Otherwise, make sure the unmodified return value of I is the return value.
// We handle two cases: multiple return values + scalars.
Value *RetVal = Ret->getOperand(0);
if (!isa<InsertValueInst>(RetVal) || !isa<StructType>(RetVal->getType()))
// Handle scalars first.
return getNoopInput(Ret->getOperand(0), TLI) == I;
// If this is an aggregate return, look through the insert/extract values and
// see if each is transparent.
for (unsigned i = 0, e =cast<StructType>(RetVal->getType())->getNumElements();
i != e; ++i) {
const Value *InScalar = FindInsertedValue(RetVal, i);
if (InScalar == 0) return false;
InScalar = getNoopInput(InScalar, TLI);
// If the scalar value being inserted is an extractvalue of the right index
// from the call, then everything is good.
const ExtractValueInst *EVI = dyn_cast<ExtractValueInst>(InScalar);
if (EVI == 0 || EVI->getOperand(0) != I || EVI->getNumIndices() != 1 ||
EVI->getIndices()[0] != i)
return false;
}
return true;
}
/// analyzeFunction - Fill in the current structure with information gleaned
/// from the specified function.
void InlineCostAnalyzer::FunctionInfo::analyzeFunction(Function *F) {
unsigned NumInsts = 0, NumBlocks = 0, NumVectorInsts = 0;
// Look at the size of the callee. Each basic block counts as 20 units, and
// each instruction counts as 5.
for (Function::const_iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
for (BasicBlock::const_iterator II = BB->begin(), E = BB->end();
II != E; ++II) {
if (isa<PHINode>(II)) continue; // PHI nodes don't count.
// Special handling for calls.
if (isa<CallInst>(II) || isa<InvokeInst>(II)) {
if (isa<DbgInfoIntrinsic>(II))
continue; // Debug intrinsics don't count as size.
CallSite CS = CallSite::get(const_cast<Instruction*>(&*II));
// If this function contains a call to setjmp or _setjmp, never inline
// it. This is a hack because we depend on the user marking their local
// variables as volatile if they are live across a setjmp call, and they
// probably won't do this in callers.
if (Function *F = CS.getCalledFunction())
if (F->isDeclaration() &&
(F->isName("setjmp") || F->isName("_setjmp"))) {
NeverInline = true;
return;
}
// Calls often compile into many machine instructions. Bump up their
// cost to reflect this.
if (!isa<IntrinsicInst>(II))
NumInsts += 5;
}
if (const AllocaInst *AI = dyn_cast<AllocaInst>(II)) {
if (!AI->isStaticAlloca())
this->usesDynamicAlloca = true;
}
if (isa<ExtractElementInst>(II) || isa<VectorType>(II->getType()))
++NumVectorInsts;
// Noop casts, including ptr <-> int, don't count.
if (const CastInst *CI = dyn_cast<CastInst>(II)) {
if (CI->isLosslessCast() || isa<IntToPtrInst>(CI) ||
isa<PtrToIntInst>(CI))
continue;
} else if (const GetElementPtrInst *GEPI =
dyn_cast<GetElementPtrInst>(II)) {
// If a GEP has all constant indices, it will probably be folded with
// a load/store.
bool AllConstant = true;
for (unsigned i = 1, e = GEPI->getNumOperands(); i != e; ++i)
if (!isa<ConstantInt>(GEPI->getOperand(i))) {
AllConstant = false;
break;
}
if (AllConstant) continue;
}
++NumInsts;
}
++NumBlocks;
}
this->NumBlocks = NumBlocks;
this->NumInsts = NumInsts;
this->NumVectorInsts = NumVectorInsts;
// Check out all of the arguments to the function, figuring out how much
// code can be eliminated if one of the arguments is a constant.
for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(); I != E; ++I)
ArgumentWeights.push_back(ArgInfo(CountCodeReductionForConstant(I),
CountCodeReductionForAlloca(I)));
}
/// Test if the given instruction is in a position to be optimized
/// with a tail-call. This roughly means that it's in a block with
/// a return and there's nothing that needs to be scheduled
/// between it and the return.
///
/// This function only tests target-independent requirements.
bool llvm::isInTailCallPosition(ImmutableCallSite CS, Attributes CalleeRetAttr,
const TargetLowering &TLI) {
const Instruction *I = CS.getInstruction();
const BasicBlock *ExitBB = I->getParent();
const TerminatorInst *Term = ExitBB->getTerminator();
const ReturnInst *Ret = dyn_cast<ReturnInst>(Term);
// The block must end in a return statement or unreachable.
//
// FIXME: Decline tailcall if it's not guaranteed and if the block ends in
// an unreachable, for now. The way tailcall optimization is currently
// implemented means it will add an epilogue followed by a jump. That is
// not profitable. Also, if the callee is a special function (e.g.
// longjmp on x86), it can end up causing miscompilation that has not
// been fully understood.
if (!Ret &&
(!GuaranteedTailCallOpt || !isa<UnreachableInst>(Term))) return false;
// If I will have a chain, make sure no other instruction that will have a
// chain interposes between I and the return.
if (I->mayHaveSideEffects() || I->mayReadFromMemory() ||
!I->isSafeToSpeculativelyExecute())
for (BasicBlock::const_iterator BBI = prior(prior(ExitBB->end())); ;
--BBI) {
if (&*BBI == I)
break;
// Debug info intrinsics do not get in the way of tail call optimization.
if (isa<DbgInfoIntrinsic>(BBI))
continue;
if (BBI->mayHaveSideEffects() || BBI->mayReadFromMemory() ||
!BBI->isSafeToSpeculativelyExecute())
return false;
}
// If the block ends with a void return or unreachable, it doesn't matter
// what the call's return type is.
if (!Ret || Ret->getNumOperands() == 0) return true;
// If the return value is undef, it doesn't matter what the call's
// return type is.
if (isa<UndefValue>(Ret->getOperand(0))) return true;
// Conservatively require the attributes of the call to match those of
// the return. Ignore noalias because it doesn't affect the call sequence.
const Function *F = ExitBB->getParent();
unsigned CallerRetAttr = F->getAttributes().getRetAttributes();
if ((CalleeRetAttr ^ CallerRetAttr) & ~Attribute::NoAlias)
return false;
// It's not safe to eliminate the sign / zero extension of the return value.
if ((CallerRetAttr & Attribute::ZExt) || (CallerRetAttr & Attribute::SExt))
return false;
// Otherwise, make sure the unmodified return value of I is the return value.
for (const Instruction *U = dyn_cast<Instruction>(Ret->getOperand(0)); ;
U = dyn_cast<Instruction>(U->getOperand(0))) {
if (!U)
return false;
if (!U->hasOneUse())
return false;
if (U == I)
break;
// Check for a truly no-op truncate.
if (isa<TruncInst>(U) &&
TLI.isTruncateFree(U->getOperand(0)->getType(), U->getType()))
continue;
// Check for a truly no-op bitcast.
if (isa<BitCastInst>(U) &&
(U->getOperand(0)->getType() == U->getType() ||
(U->getOperand(0)->getType()->isPointerTy() &&
U->getType()->isPointerTy())))
continue;
// Otherwise it's not a true no-op.
return false;
}
return true;
}
/// CloneBlock - The specified block is found to be reachable, clone it and
/// anything that it can reach.
void PruningFunctionCloner::CloneBlock(const BasicBlock *BB,
std::vector<const BasicBlock*> &ToClone){
TrackingVH<Value> &BBEntry = VMap[BB];
// Have we already cloned this block?
if (BBEntry) return;
// Nope, clone it now.
BasicBlock *NewBB;
BBEntry = NewBB = BasicBlock::Create(BB->getContext());
if (BB->hasName()) NewBB->setName(BB->getName()+NameSuffix);
bool hasCalls = false, hasDynamicAllocas = false, hasStaticAllocas = false;
// Loop over all instructions, and copy them over, DCE'ing as we go. This
// loop doesn't include the terminator.
for (BasicBlock::const_iterator II = BB->begin(), IE = --BB->end();
II != IE; ++II) {
// If this instruction constant folds, don't bother cloning the instruction,
// instead, just add the constant to the value map.
if (Constant *C = ConstantFoldMappedInstruction(II)) {
VMap[II] = C;
continue;
}
Instruction *NewInst = II->clone();
if (II->hasName())
NewInst->setName(II->getName()+NameSuffix);
NewBB->getInstList().push_back(NewInst);
VMap[II] = NewInst; // Add instruction map to value.
hasCalls |= (isa<CallInst>(II) && !isa<DbgInfoIntrinsic>(II));
if (const AllocaInst *AI = dyn_cast<AllocaInst>(II)) {
if (isa<ConstantInt>(AI->getArraySize()))
hasStaticAllocas = true;
else
hasDynamicAllocas = true;
}
}
// Finally, clone over the terminator.
const TerminatorInst *OldTI = BB->getTerminator();
bool TerminatorDone = false;
if (const BranchInst *BI = dyn_cast<BranchInst>(OldTI)) {
if (BI->isConditional()) {
// If the condition was a known constant in the callee...
ConstantInt *Cond = dyn_cast<ConstantInt>(BI->getCondition());
// Or is a known constant in the caller...
if (Cond == 0) {
Value *V = VMap[BI->getCondition()];
Cond = dyn_cast_or_null<ConstantInt>(V);
}
// Constant fold to uncond branch!
if (Cond) {
BasicBlock *Dest = BI->getSuccessor(!Cond->getZExtValue());
VMap[OldTI] = BranchInst::Create(Dest, NewBB);
ToClone.push_back(Dest);
TerminatorDone = true;
}
}
} else if (const SwitchInst *SI = dyn_cast<SwitchInst>(OldTI)) {
// If switching on a value known constant in the caller.
ConstantInt *Cond = dyn_cast<ConstantInt>(SI->getCondition());
if (Cond == 0) { // Or known constant after constant prop in the callee...
Value *V = VMap[SI->getCondition()];
Cond = dyn_cast_or_null<ConstantInt>(V);
}
if (Cond) { // Constant fold to uncond branch!
BasicBlock *Dest = SI->getSuccessor(SI->findCaseValue(Cond));
VMap[OldTI] = BranchInst::Create(Dest, NewBB);
ToClone.push_back(Dest);
TerminatorDone = true;
}
}
if (!TerminatorDone) {
Instruction *NewInst = OldTI->clone();
if (OldTI->hasName())
NewInst->setName(OldTI->getName()+NameSuffix);
NewBB->getInstList().push_back(NewInst);
VMap[OldTI] = NewInst; // Add instruction map to value.
// Recursively clone any reachable successor blocks.
const TerminatorInst *TI = BB->getTerminator();
for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i)
ToClone.push_back(TI->getSuccessor(i));
}
if (CodeInfo) {
CodeInfo->ContainsCalls |= hasCalls;
CodeInfo->ContainsUnwinds |= isa<UnwindInst>(OldTI);
CodeInfo->ContainsDynamicAllocas |= hasDynamicAllocas;
CodeInfo->ContainsDynamicAllocas |= hasStaticAllocas &&
BB != &BB->getParent()->front();
}
if (ReturnInst *RI = dyn_cast<ReturnInst>(NewBB->getTerminator()))
Returns.push_back(RI);
}
static Error ReduceInsts(BugDriver &BD,
bool (*TestFn)(const BugDriver &, Module *)) {
// Attempt to delete instructions using bisection. This should help out nasty
// cases with large basic blocks where the problem is at one end.
if (!BugpointIsInterrupted) {
std::vector<const Instruction *> Insts;
for (const Function &F : *BD.getProgram())
for (const BasicBlock &BB : F)
for (const Instruction &I : BB)
if (!isa<TerminatorInst>(&I))
Insts.push_back(&I);
Expected<bool> Result =
ReduceCrashingInstructions(BD, TestFn).reduceList(Insts);
if (Error E = Result.takeError())
return E;
}
unsigned Simplification = 2;
do {
if (BugpointIsInterrupted)
// TODO: Should we distinguish this with an "interrupted error"?
return Error::success();
--Simplification;
outs() << "\n*** Attempting to reduce testcase by deleting instruc"
<< "tions: Simplification Level #" << Simplification << '\n';
// Now that we have deleted the functions that are unnecessary for the
// program, try to remove instructions that are not necessary to cause the
// crash. To do this, we loop through all of the instructions in the
// remaining functions, deleting them (replacing any values produced with
// nulls), and then running ADCE and SimplifyCFG. If the transformed input
// still triggers failure, keep deleting until we cannot trigger failure
// anymore.
//
unsigned InstructionsToSkipBeforeDeleting = 0;
TryAgain:
// Loop over all of the (non-terminator) instructions remaining in the
// function, attempting to delete them.
unsigned CurInstructionNum = 0;
for (Module::const_iterator FI = BD.getProgram()->begin(),
E = BD.getProgram()->end();
FI != E; ++FI)
if (!FI->isDeclaration())
for (Function::const_iterator BI = FI->begin(), E = FI->end(); BI != E;
++BI)
for (BasicBlock::const_iterator I = BI->begin(), E = --BI->end();
I != E; ++I, ++CurInstructionNum) {
if (InstructionsToSkipBeforeDeleting) {
--InstructionsToSkipBeforeDeleting;
} else {
if (BugpointIsInterrupted)
// TODO: Should this be some kind of interrupted error?
return Error::success();
if (I->isEHPad() || I->getType()->isTokenTy())
continue;
outs() << "Checking instruction: " << *I;
std::unique_ptr<Module> M =
BD.deleteInstructionFromProgram(&*I, Simplification);
// Find out if the pass still crashes on this pass...
if (TestFn(BD, M.get())) {
// Yup, it does, we delete the old module, and continue trying
// to reduce the testcase...
BD.setNewProgram(M.release());
InstructionsToSkipBeforeDeleting = CurInstructionNum;
goto TryAgain; // I wish I had a multi-level break here!
}
}
}
if (InstructionsToSkipBeforeDeleting) {
InstructionsToSkipBeforeDeleting = 0;
goto TryAgain;
}
} while (Simplification);
BD.EmitProgressBitcode(BD.getProgram(), "reduced-instructions");
return Error::success();
}
static bool equals(const BasicBlock *BB1, const BasicBlock *BB2,
DenseMap<const Value *, const Value *> &ValueMap,
DenseMap<const Value *, const Value *> &SpeculationMap) {
// Speculatively add it anyways. If it's false, we'll notice a difference
// later, and this won't matter.
ValueMap[BB1] = BB2;
BasicBlock::const_iterator FI = BB1->begin(), FE = BB1->end();
BasicBlock::const_iterator GI = BB2->begin(), GE = BB2->end();
do {
if (isa<BitCastInst>(FI)) {
++FI;
continue;
}
if (isa<BitCastInst>(GI)) {
++GI;
continue;
}
if (!isEquivalentOperation(FI, GI))
return false;
if (isa<GetElementPtrInst>(FI)) {
const GetElementPtrInst *GEPF = cast<GetElementPtrInst>(FI);
const GetElementPtrInst *GEPG = cast<GetElementPtrInst>(GI);
if (GEPF->hasAllZeroIndices() && GEPG->hasAllZeroIndices()) {
// It's effectively a bitcast.
++FI, ++GI;
continue;
}
// TODO: we only really care about the elements before the index
if (FI->getOperand(0)->getType() != GI->getOperand(0)->getType())
return false;
}
if (ValueMap[FI] == GI) {
++FI, ++GI;
continue;
}
if (ValueMap[FI] != NULL)
return false;
for (unsigned i = 0, e = FI->getNumOperands(); i != e; ++i) {
Value *OpF = IgnoreBitcasts(FI->getOperand(i));
Value *OpG = IgnoreBitcasts(GI->getOperand(i));
if (ValueMap[OpF] == OpG)
continue;
if (ValueMap[OpF] != NULL)
return false;
if (OpF->getValueID() != OpG->getValueID() ||
!isEquivalentType(OpF->getType(), OpG->getType()))
return false;
if (isa<PHINode>(FI)) {
if (SpeculationMap[OpF] == NULL)
SpeculationMap[OpF] = OpG;
else if (SpeculationMap[OpF] != OpG)
return false;
continue;
} else if (isa<BasicBlock>(OpF)) {
assert(isa<TerminatorInst>(FI) &&
"BasicBlock referenced by non-Terminator non-PHI");
// This call changes the ValueMap, hence we can't use
// Value *& = ValueMap[...]
if (!equals(cast<BasicBlock>(OpF), cast<BasicBlock>(OpG), ValueMap,
SpeculationMap))
return false;
} else {
if (!compare(OpF, OpG))
return false;
}
ValueMap[OpF] = OpG;
}
ValueMap[FI] = GI;
++FI, ++GI;
} while (FI != FE && GI != GE);
return FI == FE && GI == GE;
}
/**
* Generate code for
*/
void HeteroOMPTransform::gen_code_per_f (Function* NF, Function* F, Instruction *max_threads){
Function::arg_iterator FI = F->arg_begin();
Argument *ctxname = &*FI;
Function::arg_iterator DestI = NF->arg_begin();
DestI->setName(ctxname->getName());
Argument *ctx_name = &(*DestI);
DestI++;
DestI->setName("tid");
Argument *num_iters = &(*DestI);
#ifdef EXPLICIT_REWRITE
DenseMap<const Value*, Value *> ValueMap;
#else
ValueToValueMapTy ValueMap;
#endif
//get the old basic block and create a new one
Function::const_iterator BI = F->begin();
const BasicBlock &FB = *BI;
BasicBlock *NFBB = BasicBlock::Create(FB.getContext(), "", NF);
if (FB.hasName()){
NFBB->setName(FB.getName());
}
ValueMap[&FB] = NFBB;
//ValueMap[numiters] = num_iters;
ValueMap[ctxname] = ctx_name;
#if EXPLICIT_REWRITE
for (BasicBlock::const_iterator II = FB.begin(), IE = FB.end(); II != IE; ++II) {
Instruction *NFInst = II->clone(/*F->getContext()*/);
// DEBUG(dbgs()<<*II<<"\n");
if (II->hasName()) NFInst->setName(II->getName());
const Instruction *FInst = &(*II);
rewrite_instruction((Instruction *)FInst, NFInst, ValueMap);
NFBB->getInstList().push_back(NFInst);
ValueMap[II] = NFInst;
}
BI++;
for (Function::const_iterator /*BI=F->begin(),*/BE = F->end();BI != BE; ++BI) {
const BasicBlock &FBB = *BI;
BasicBlock *NFBB = BasicBlock::Create(FBB.getContext(), "", NF);
ValueMap[&FBB] = NFBB;
if (FBB.hasName()){
NFBB->setName(FBB.getName());
//DEBUG(dbgs()<<NFBB->getName()<<"\n");
}
for (BasicBlock::const_iterator II = FBB.begin(), IE = FBB.end(); II != IE; ++II) {
Instruction *NFInst = II->clone(/*F->getContext()*/);
if (II->hasName()) NFInst->setName(II->getName());
const Instruction *FInst = &(*II);
rewrite_instruction((Instruction *)FInst, NFInst, ValueMap);
NFBB->getInstList().push_back(NFInst);
ValueMap[II] = NFInst;
}
}
// Remap the instructions again to take care of forward jumps
for (Function::iterator BB = NF->begin(), BE=NF->end(); BB != BE; ++ BB) {
for (BasicBlock::iterator II = BB->begin(); II != BB->end(); ++II){
int opIdx = 0;
//DEBUG(dbgs()<<*II<<"\n");
for (User::op_iterator i = II->op_begin(), e = II->op_end(); i != e; ++i, opIdx++) {
Value *V = *i;
if (ValueMap[V] != NULL) {
II->setOperand(opIdx, ValueMap[V]);
}
}
}
}
#else
SmallVector<ReturnInst*, 8> Returns; // Ignore returns cloned.
CloneFunctionInto(NF, F, ValueMap, false, Returns, "");
#endif
//max_threads->dump();
/* Remap openmp omp_num_threads() and omp_thread_num() */
/*
* define internal void @_Z20initialize_variablesiPfS_.omp_fn.4(i8* nocapture %.omp_data_i) nounwind ssp {
* entry:
* %0 = bitcast i8* %.omp_data_i to i32* ; <i32*> [#uses=1]
* %1 = load i32* %0, align 8 ; <i32> [#uses=2]
* %2 = tail call i32 @omp_get_num_threads() nounwind readnone ; <i32> [#uses=2]
* %3 = tail call i32 @omp_get_thread_num() nounwind readnone ; <i32> [#uses=2]
%4 = sdiv i32 %1, %2
%5 = mul nsw i32 %4, %2
%6 = icmp ne i32 %5, %1
%7 = zext i1 %6 to i32
*/
vector<Instruction *> toDelete;
for (Function::iterator BB = NF->begin(), BE=NF->end(); BB != BE; ++ BB) {
for (BasicBlock::iterator II = BB->begin(); II != BB->end(); ++II){
if (isa<CallInst>(II)) {
CallSite CI(cast<Instruction>(II));
if (CI.getCalledFunction() != NULL){
string called_func_name = CI.getCalledFunction()->getName();
if (called_func_name == OMP_GET_NUM_THREADS_NAME && CI.arg_size() == 0) {
II->replaceAllUsesWith(ValueMap[max_threads]);
toDelete.push_back(II);
}
else if (called_func_name == OMP_GET_THREAD_NUM_NAME && CI.arg_size() == 0) {
II->replaceAllUsesWith(num_iters);
toDelete.push_back(II);
}
}
}
}
}
/* Delete the last branch instruction of the first basic block -- Assuming it is safe */
Function::iterator nfBB = NF->begin();
TerminatorInst *lastI = nfBB->getTerminator();
BranchInst *bI;
BasicBlock *returnBlock;
if ((bI = dyn_cast<BranchInst>(lastI)) && bI->isConditional() &&
(returnBlock = bI->getSuccessor(1)) &&
(returnBlock->getName() == "return")) {
/* modify to a unconditional branch to next basic block and not return */
Instruction *bbI = BranchInst::Create(bI->getSuccessor(0),lastI);
bbI->dump();
toDelete.push_back(lastI);
}
//NF->dump();
while(!toDelete.empty()) {
Instruction *g = toDelete.back();
//g->replaceAllUsesWith(UndefValue::get(g->getType()));
toDelete.pop_back();
g->eraseFromParent();
}
//NF->dump();
}
void HeterotbbTransform::gen_opt_code_per_f (Function* NF, Function* F) {
// Get the names of the parameters for old function
Function::arg_iterator FI = F->arg_begin();
Argument *classname = &*FI;
FI++;
Argument *numiters = &*FI;
// Set the names of the parameters for new function
Function::arg_iterator DestI = NF->arg_begin();
DestI->setName(classname->getName());
Argument *class_name = &(*DestI);
//second argument
DestI++;
DestI->setName(numiters->getName());
Argument *num_iters = &(*DestI);
#ifdef EXPLICIT_REWRITE
DenseMap<const Value*, Value *> ValueMap;
#else
ValueToValueMapTy ValueMap;
#endif
#if EXPLICIT_REWRITE
//get the old basic block and create a new one
Function::const_iterator BI = F->begin();
const BasicBlock &FB = *BI;
BasicBlock *NFBB = BasicBlock::Create(FB.getContext(), "", NF);
if (FB.hasName()) {
NFBB->setName(FB.getName());
//DEBUG(dbgs()<<FB.getName()<<"\n");
}
ValueMap[&FB] = NFBB;
ValueMap[numiters] = num_iters;
//must create a new instruction which casts i32* back to the class name
CastInst *StrucRevCast = CastInst::Create(Instruction::BitCast, class_name,
classname->getType(), classname->getName(), NFBB);
ValueMap[classname] = StrucRevCast;
for (BasicBlock::const_iterator II = FB.begin(), IE = FB.end(); II != IE; ++II) {
Instruction *NFInst = II->clone(/*F->getContext()*/);
// DEBUG(dbgs()<<*II<<"\n");
if (II->hasName()) NFInst->setName(II->getName());
const Instruction *FInst = &(*II);
rewrite_instruction((Instruction *)FInst, NFInst, ValueMap);
NFBB->getInstList().push_back(NFInst);
ValueMap[II] = NFInst;
}
BI++;
for (Function::const_iterator /*BI=F->begin(),*/BE = F->end(); BI != BE; ++BI) {
const BasicBlock &FBB = *BI;
BasicBlock *NFBB = BasicBlock::Create(FBB.getContext(), "", NF);
ValueMap[&FBB] = NFBB;
if (FBB.hasName()) {
NFBB->setName(FBB.getName());
//DEBUG(dbgs()<<NFBB->getName()<<"\n");
}
for (BasicBlock::const_iterator II = FBB.begin(), IE = FBB.end(); II != IE; ++II) {
Instruction *NFInst = II->clone(/*F->getContext()*/);
if (II->hasName()) NFInst->setName(II->getName());
const Instruction *FInst = &(*II);
rewrite_instruction((Instruction *)FInst, NFInst, ValueMap);
NFBB->getInstList().push_back(NFInst);
ValueMap[II] = NFInst;
}
}
// Remap the instructions again to take care of forward jumps
for (Function::iterator BB = NF->begin(), BE=NF->end(); BB != BE; ++ BB) {
for (BasicBlock::iterator II = BB->begin(); II != BB->end(); ++II) {
int opIdx = 0;
//DEBUG(dbgs()<<*II<<"\n");
for (User::op_iterator i = II->op_begin(), e = II->op_end(); i != e; ++i, opIdx++) {
Value *V = *i;
if (ValueMap[V] != NULL) {
II->setOperand(opIdx, ValueMap[V]);
}
}
}
}
#else
Function::const_iterator BI = F->begin();
const BasicBlock &FB = *BI;
BasicBlock *NFBB = BasicBlock::Create(FB.getContext(), "", NF);
if (FB.hasName()) {
NFBB->setName(FB.getName());
}
ValueMap[&FB] = NFBB;
CastInst *StrucRevCast = CastInst::Create(Instruction::BitCast, class_name,
classname->getType(), classname->getName(), NFBB);
ValueMap[classname] = StrucRevCast;
ValueMap[numiters] = num_iters;
CloneFunctionWithExistingBBInto(NF, NFBB, F, ValueMap, "");
#endif
}
/// 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());
}
/// processModule - Process entire module and collect debug info.
void DebugInfoFinder::processModule(const Module &M) {
if (NamedMDNode *CU_Nodes = M.getNamedMetadata("llvm.dbg.cu")) {
for (unsigned i = 0, e = CU_Nodes->getNumOperands(); i != e; ++i) {
DICompileUnit CU(CU_Nodes->getOperand(i));
addCompileUnit(CU);
if (CU.getVersion() > LLVMDebugVersion10) {
DIArray GVs = CU.getGlobalVariables();
for (unsigned i = 0, e = GVs.getNumElements(); i != e; ++i) {
DIGlobalVariable DIG(GVs.getElement(i));
if (addGlobalVariable(DIG))
processType(DIG.getType());
}
DIArray SPs = CU.getSubprograms();
for (unsigned i = 0, e = SPs.getNumElements(); i != e; ++i)
processSubprogram(DISubprogram(SPs.getElement(i)));
DIArray EnumTypes = CU.getEnumTypes();
for (unsigned i = 0, e = EnumTypes.getNumElements(); i != e; ++i)
processType(DIType(EnumTypes.getElement(i)));
DIArray RetainedTypes = CU.getRetainedTypes();
for (unsigned i = 0, e = RetainedTypes.getNumElements(); i != e; ++i)
processType(DIType(RetainedTypes.getElement(i)));
return;
}
}
}
for (Module::const_iterator I = M.begin(), E = M.end(); I != E; ++I)
for (Function::const_iterator FI = (*I).begin(), FE = (*I).end();
FI != FE; ++FI)
for (BasicBlock::const_iterator BI = (*FI).begin(), BE = (*FI).end();
BI != BE; ++BI) {
if (const DbgDeclareInst *DDI = dyn_cast<DbgDeclareInst>(BI))
processDeclare(DDI);
DebugLoc Loc = BI->getDebugLoc();
if (Loc.isUnknown())
continue;
LLVMContext &Ctx = BI->getContext();
DIDescriptor Scope(Loc.getScope(Ctx));
if (Scope.isCompileUnit())
addCompileUnit(DICompileUnit(Scope));
else if (Scope.isSubprogram())
processSubprogram(DISubprogram(Scope));
else if (Scope.isLexicalBlockFile()) {
DILexicalBlockFile DBF = DILexicalBlockFile(Scope);
processLexicalBlock(DILexicalBlock(DBF.getScope()));
}
else if (Scope.isLexicalBlock())
processLexicalBlock(DILexicalBlock(Scope));
if (MDNode *IA = Loc.getInlinedAt(Ctx))
processLocation(DILocation(IA));
}
if (NamedMDNode *NMD = M.getNamedMetadata("llvm.dbg.gv")) {
for (unsigned i = 0, e = NMD->getNumOperands(); i != e; ++i) {
DIGlobalVariable DIG(cast<MDNode>(NMD->getOperand(i)));
if (addGlobalVariable(DIG)) {
if (DIG.getVersion() <= LLVMDebugVersion10)
addCompileUnit(DIG.getCompileUnit());
processType(DIG.getType());
}
}
}
if (NamedMDNode *NMD = M.getNamedMetadata("llvm.dbg.sp"))
for (unsigned i = 0, e = NMD->getNumOperands(); i != e; ++i)
processSubprogram(DISubprogram(NMD->getOperand(i)));
}
/// analyzeBasicBlock - Fill in the current structure with information gleaned
/// from the specified block.
void CodeMetrics::analyzeBasicBlock(const BasicBlock *BB,
const DataLayout *TD) {
++NumBlocks;
unsigned NumInstsBeforeThisBB = NumInsts;
for (BasicBlock::const_iterator II = BB->begin(), E = BB->end();
II != E; ++II) {
if (isInstructionFree(II, TD))
continue;
// Special handling for calls.
if (isa<CallInst>(II) || isa<InvokeInst>(II)) {
ImmutableCallSite CS(cast<Instruction>(II));
if (const Function *F = CS.getCalledFunction()) {
// If a function is both internal and has a single use, then it is
// extremely likely to get inlined in the future (it was probably
// exposed by an interleaved devirtualization pass).
if (!CS.isNoInline() && F->hasInternalLinkage() && F->hasOneUse())
++NumInlineCandidates;
// If this call is to function itself, then the function is recursive.
// Inlining it into other functions is a bad idea, because this is
// basically just a form of loop peeling, and our metrics aren't useful
// for that case.
if (F == BB->getParent())
isRecursive = true;
}
if (!callIsSmall(CS)) {
// Each argument to a call takes on average one instruction to set up.
NumInsts += CS.arg_size();
// We don't want inline asm to count as a call - that would prevent loop
// unrolling. The argument setup cost is still real, though.
if (!isa<InlineAsm>(CS.getCalledValue()))
++NumCalls;
}
}
if (const AllocaInst *AI = dyn_cast<AllocaInst>(II)) {
if (!AI->isStaticAlloca())
this->usesDynamicAlloca = true;
}
if (isa<ExtractElementInst>(II) || II->getType()->isVectorTy())
++NumVectorInsts;
if (const CallInst *CI = dyn_cast<CallInst>(II))
if (CI->hasFnAttr(Attribute::NoDuplicate))
notDuplicatable = true;
if (const InvokeInst *InvI = dyn_cast<InvokeInst>(II))
if (InvI->hasFnAttr(Attribute::NoDuplicate))
notDuplicatable = true;
++NumInsts;
}
if (isa<ReturnInst>(BB->getTerminator()))
++NumRets;
// We never want to inline functions that contain an indirectbr. This is
// incorrect because all the blockaddress's (in static global initializers
// for example) would be referring to the original function, and this indirect
// jump would jump from the inlined copy of the function into the original
// function which is extremely undefined behavior.
// FIXME: This logic isn't really right; we can safely inline functions
// with indirectbr's as long as no other function or global references the
// blockaddress of a block within the current function. And as a QOI issue,
// if someone is using a blockaddress without an indirectbr, and that
// reference somehow ends up in another function or global, we probably
// don't want to inline this function.
notDuplicatable |= isa<IndirectBrInst>(BB->getTerminator());
// Remember NumInsts for this BB.
NumBBInsts[BB] = NumInsts - NumInstsBeforeThisBB;
}
/// 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());
}
/// The specified block is found to be reachable, clone it and
/// anything that it can reach.
void PruningFunctionCloner::CloneBlock(const BasicBlock *BB,
BasicBlock::const_iterator StartingInst,
std::vector<const BasicBlock*> &ToClone){
WeakVH &BBEntry = VMap[BB];
// Have we already cloned this block?
if (BBEntry) return;
// Nope, clone it now.
BasicBlock *NewBB;
BBEntry = NewBB = BasicBlock::Create(BB->getContext());
if (BB->hasName()) NewBB->setName(BB->getName()+NameSuffix);
// It is only legal to clone a function if a block address within that
// function is never referenced outside of the function. Given that, we
// want to map block addresses from the old function to block addresses in
// the clone. (This is different from the generic ValueMapper
// implementation, which generates an invalid blockaddress when
// cloning a function.)
//
// Note that we don't need to fix the mapping for unreachable blocks;
// the default mapping there is safe.
if (BB->hasAddressTaken()) {
Constant *OldBBAddr = BlockAddress::get(const_cast<Function*>(OldFunc),
const_cast<BasicBlock*>(BB));
VMap[OldBBAddr] = BlockAddress::get(NewFunc, NewBB);
}
bool hasCalls = false, hasDynamicAllocas = false, hasStaticAllocas = false;
// Loop over all instructions, and copy them over, DCE'ing as we go. This
// loop doesn't include the terminator.
for (BasicBlock::const_iterator II = StartingInst, IE = --BB->end();
II != IE; ++II) {
Instruction *NewInst = II->clone();
// Eagerly remap operands to the newly cloned instruction, except for PHI
// nodes for which we defer processing until we update the CFG.
if (!isa<PHINode>(NewInst)) {
RemapInstruction(NewInst, VMap,
ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges);
// If we can simplify this instruction to some other value, simply add
// a mapping to that value rather than inserting a new instruction into
// the basic block.
if (Value *V =
SimplifyInstruction(NewInst, BB->getModule()->getDataLayout())) {
// On the off-chance that this simplifies to an instruction in the old
// function, map it back into the new function.
if (Value *MappedV = VMap.lookup(V))
V = MappedV;
VMap[&*II] = V;
delete NewInst;
continue;
}
}
if (II->hasName())
NewInst->setName(II->getName()+NameSuffix);
VMap[&*II] = NewInst; // Add instruction map to value.
NewBB->getInstList().push_back(NewInst);
hasCalls |= (isa<CallInst>(II) && !isa<DbgInfoIntrinsic>(II));
if (CodeInfo)
if (auto CS = ImmutableCallSite(&*II))
if (CS.hasOperandBundles())
CodeInfo->OperandBundleCallSites.push_back(NewInst);
if (const AllocaInst *AI = dyn_cast<AllocaInst>(II)) {
if (isa<ConstantInt>(AI->getArraySize()))
hasStaticAllocas = true;
else
hasDynamicAllocas = true;
}
}
// Finally, clone over the terminator.
const TerminatorInst *OldTI = BB->getTerminator();
bool TerminatorDone = false;
if (const BranchInst *BI = dyn_cast<BranchInst>(OldTI)) {
if (BI->isConditional()) {
// If the condition was a known constant in the callee...
ConstantInt *Cond = dyn_cast<ConstantInt>(BI->getCondition());
// Or is a known constant in the caller...
if (!Cond) {
Value *V = VMap[BI->getCondition()];
Cond = dyn_cast_or_null<ConstantInt>(V);
}
// Constant fold to uncond branch!
if (Cond) {
BasicBlock *Dest = BI->getSuccessor(!Cond->getZExtValue());
VMap[OldTI] = BranchInst::Create(Dest, NewBB);
ToClone.push_back(Dest);
TerminatorDone = true;
}
}
} else if (const SwitchInst *SI = dyn_cast<SwitchInst>(OldTI)) {
// If switching on a value known constant in the caller.
ConstantInt *Cond = dyn_cast<ConstantInt>(SI->getCondition());
if (!Cond) { // Or known constant after constant prop in the callee...
Value *V = VMap[SI->getCondition()];
Cond = dyn_cast_or_null<ConstantInt>(V);
}
if (Cond) { // Constant fold to uncond branch!
SwitchInst::ConstCaseIt Case = SI->findCaseValue(Cond);
BasicBlock *Dest = const_cast<BasicBlock*>(Case.getCaseSuccessor());
VMap[OldTI] = BranchInst::Create(Dest, NewBB);
ToClone.push_back(Dest);
TerminatorDone = true;
}
}
if (!TerminatorDone) {
Instruction *NewInst = OldTI->clone();
if (OldTI->hasName())
NewInst->setName(OldTI->getName()+NameSuffix);
NewBB->getInstList().push_back(NewInst);
VMap[OldTI] = NewInst; // Add instruction map to value.
if (CodeInfo)
if (auto CS = ImmutableCallSite(OldTI))
if (CS.hasOperandBundles())
CodeInfo->OperandBundleCallSites.push_back(NewInst);
// Recursively clone any reachable successor blocks.
const TerminatorInst *TI = BB->getTerminator();
for (const BasicBlock *Succ : TI->successors())
ToClone.push_back(Succ);
}
if (CodeInfo) {
CodeInfo->ContainsCalls |= hasCalls;
CodeInfo->ContainsDynamicAllocas |= hasDynamicAllocas;
CodeInfo->ContainsDynamicAllocas |= hasStaticAllocas &&
BB != &BB->getParent()->front();
}
}
static bool equals(const BasicBlock *BB1, const BasicBlock *BB2,
DenseMap<const Value *, const Value *> &ValueMap,
DenseMap<const Value *, const Value *> &SpeculationMap) {
// Specutively add it anyways. If it's false, we'll notice a difference later, and
// this won't matter.
ValueMap[BB1] = BB2;
BasicBlock::const_iterator FI = BB1->begin(), FE = BB1->end();
BasicBlock::const_iterator GI = BB2->begin(), GE = BB2->end();
do {
if (!FI->isSameOperationAs(const_cast<Instruction *>(&*GI)))
return false;
if (FI->getNumOperands() != GI->getNumOperands())
return false;
if (ValueMap[FI] == GI) {
++FI, ++GI;
continue;
}
if (ValueMap[FI] != NULL)
return false;
for (unsigned i = 0, e = FI->getNumOperands(); i != e; ++i) {
Value *OpF = FI->getOperand(i);
Value *OpG = GI->getOperand(i);
if (ValueMap[OpF] == OpG)
continue;
if (ValueMap[OpF] != NULL)
return false;
assert(OpF->getType() == OpG->getType() &&
"Two of the same operation has operands of different type.");
if (OpF->getValueID() != OpG->getValueID())
return false;
if (isa<PHINode>(FI)) {
if (SpeculationMap[OpF] == NULL)
SpeculationMap[OpF] = OpG;
else if (SpeculationMap[OpF] != OpG)
return false;
continue;
} else if (isa<BasicBlock>(OpF)) {
assert(isa<TerminatorInst>(FI) &&
"BasicBlock referenced by non-Terminator non-PHI");
// This call changes the ValueMap, hence we can't use
// Value *& = ValueMap[...]
if (!equals(cast<BasicBlock>(OpF), cast<BasicBlock>(OpG), ValueMap,
SpeculationMap))
return false;
} else {
if (!compare(OpF, OpG))
return false;
}
ValueMap[OpF] = OpG;
}
ValueMap[FI] = GI;
++FI, ++GI;
} while (FI != FE && GI != GE);
return FI == FE && GI == GE;
}
/// analyzeBasicBlock - Fill in the current structure with information gleaned
/// from the specified block.
void CodeMetrics::analyzeBasicBlock(const BasicBlock *BB) {
++NumBlocks;
unsigned NumInstsBeforeThisBB = NumInsts;
for (BasicBlock::const_iterator II = BB->begin(), E = BB->end();
II != E; ++II) {
if (isa<PHINode>(II)) continue; // PHI nodes don't count.
// Special handling for calls.
if (isa<CallInst>(II) || isa<InvokeInst>(II)) {
if (isa<DbgInfoIntrinsic>(II))
continue; // Debug intrinsics don't count as size.
ImmutableCallSite CS(cast<Instruction>(II));
// If this function contains a call to setjmp or _setjmp, never inline
// it. This is a hack because we depend on the user marking their local
// variables as volatile if they are live across a setjmp call, and they
// probably won't do this in callers.
if (const Function *F = CS.getCalledFunction()) {
// If a function is both internal and has a single use, then it is
// extremely likely to get inlined in the future (it was probably
// exposed by an interleaved devirtualization pass).
if (F->hasInternalLinkage() && F->hasOneUse())
++NumInlineCandidates;
if (F->isDeclaration() &&
(F->getName() == "setjmp" || F->getName() == "_setjmp"))
callsSetJmp = true;
// If this call is to function itself, then the function is recursive.
// Inlining it into other functions is a bad idea, because this is
// basically just a form of loop peeling, and our metrics aren't useful
// for that case.
if (F == BB->getParent())
isRecursive = true;
}
if (!isa<IntrinsicInst>(II) && !callIsSmall(CS.getCalledFunction())) {
// Each argument to a call takes on average one instruction to set up.
NumInsts += CS.arg_size();
// We don't want inline asm to count as a call - that would prevent loop
// unrolling. The argument setup cost is still real, though.
if (!isa<InlineAsm>(CS.getCalledValue()))
++NumCalls;
}
}
if (const AllocaInst *AI = dyn_cast<AllocaInst>(II)) {
if (!AI->isStaticAlloca())
this->usesDynamicAlloca = true;
}
if (isa<ExtractElementInst>(II) || II->getType()->isVectorTy())
++NumVectorInsts;
if (const CastInst *CI = dyn_cast<CastInst>(II)) {
// Noop casts, including ptr <-> int, don't count.
if (CI->isLosslessCast() || isa<IntToPtrInst>(CI) ||
isa<PtrToIntInst>(CI))
continue;
// Result of a cmp instruction is often extended (to be used by other
// cmp instructions, logical or return instructions). These are usually
// nop on most sane targets.
if (isa<CmpInst>(CI->getOperand(0)))
continue;
} else if (const GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(II)){
// If a GEP has all constant indices, it will probably be folded with
// a load/store.
if (GEPI->hasAllConstantIndices())
continue;
}
++NumInsts;
}
if (isa<ReturnInst>(BB->getTerminator()))
++NumRets;
// We never want to inline functions that contain an indirectbr. This is
// incorrect because all the blockaddress's (in static global initializers
// for example) would be referring to the original function, and this indirect
// jump would jump from the inlined copy of the function into the original
// function which is extremely undefined behavior.
if (isa<IndirectBrInst>(BB->getTerminator()))
containsIndirectBr = true;
// Remember NumInsts for this BB.
NumBBInsts[BB] = NumInsts - NumInstsBeforeThisBB;
}
