/// AddNoCaptureAttrs - Deduce nocapture attributes for the SCC.
bool FunctionAttrs::AddNoCaptureAttrs(const CallGraphSCC &SCC) {
bool Changed = false;
// Check each function in turn, determining which pointer arguments are not
// captured.
for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I) {
Function *F = (*I)->getFunction();
if (F == 0)
// External node - skip it;
continue;
// Definitions with weak linkage may be overridden at linktime with
// something that writes memory, so treat them like declarations.
if (F->isDeclaration() || F->mayBeOverridden())
continue;
for (Function::arg_iterator A = F->arg_begin(), E = F->arg_end(); A!=E; ++A)
if (A->getType()->isPointerTy() && !A->hasNoCaptureAttr() &&
!PointerMayBeCaptured(A, true, /*StoreCaptures=*/false)) {
A->addAttr(Attribute::NoCapture);
++NumNoCapture;
Changed = true;
}
}
return Changed;
}
/// Deduce nocapture attributes for the SCC.
static bool addArgumentAttrs(const SCCNodeSet &SCCNodes) {
bool Changed = false;
ArgumentGraph AG;
// Check each function in turn, determining which pointer arguments are not
// captured.
for (Function *F : SCCNodes) {
// We can infer and propagate function attributes only when we know that the
// definition we'll get at link time is *exactly* the definition we see now.
// For more details, see GlobalValue::mayBeDerefined.
if (!F->hasExactDefinition())
continue;
Changed |= addArgumentAttrsFromCallsites(*F);
// Functions that are readonly (or readnone) and nounwind and don't return
// a value can't capture arguments. Don't analyze them.
if (F->onlyReadsMemory() && F->doesNotThrow() &&
F->getReturnType()->isVoidTy()) {
for (Function::arg_iterator A = F->arg_begin(), E = F->arg_end(); A != E;
++A) {
if (A->getType()->isPointerTy() && !A->hasNoCaptureAttr()) {
A->addAttr(Attribute::NoCapture);
++NumNoCapture;
Changed = true;
}
}
continue;
}
for (Function::arg_iterator A = F->arg_begin(), E = F->arg_end(); A != E;
++A) {
if (!A->getType()->isPointerTy())
continue;
bool HasNonLocalUses = false;
if (!A->hasNoCaptureAttr()) {
ArgumentUsesTracker Tracker(SCCNodes);
PointerMayBeCaptured(&*A, &Tracker);
if (!Tracker.Captured) {
if (Tracker.Uses.empty()) {
// If it's trivially not captured, mark it nocapture now.
A->addAttr(Attribute::NoCapture);
++NumNoCapture;
Changed = true;
} else {
// If it's not trivially captured and not trivially not captured,
// then it must be calling into another function in our SCC. Save
// its particulars for Argument-SCC analysis later.
ArgumentGraphNode *Node = AG[&*A];
for (Argument *Use : Tracker.Uses) {
Node->Uses.push_back(AG[Use]);
if (Use != &*A)
HasNonLocalUses = true;
}
}
}
// Otherwise, it's captured. Don't bother doing SCC analysis on it.
}
if (!HasNonLocalUses && !A->onlyReadsMemory()) {
// Can we determine that it's readonly/readnone without doing an SCC?
// Note that we don't allow any calls at all here, or else our result
// will be dependent on the iteration order through the functions in the
// SCC.
SmallPtrSet<Argument *, 8> Self;
Self.insert(&*A);
Attribute::AttrKind R = determinePointerReadAttrs(&*A, Self);
if (R != Attribute::None) {
A->addAttr(R);
Changed = true;
R == Attribute::ReadOnly ? ++NumReadOnlyArg : ++NumReadNoneArg;
}
}
}
}
// The graph we've collected is partial because we stopped scanning for
// argument uses once we solved the argument trivially. These partial nodes
// show up as ArgumentGraphNode objects with an empty Uses list, and for
// these nodes the final decision about whether they capture has already been
// made. If the definition doesn't have a 'nocapture' attribute by now, it
// captures.
for (scc_iterator<ArgumentGraph *> I = scc_begin(&AG); !I.isAtEnd(); ++I) {
const std::vector<ArgumentGraphNode *> &ArgumentSCC = *I;
if (ArgumentSCC.size() == 1) {
if (!ArgumentSCC[0]->Definition)
continue; // synthetic root node
// eg. "void f(int* x) { if (...) f(x); }"
if (ArgumentSCC[0]->Uses.size() == 1 &&
ArgumentSCC[0]->Uses[0] == ArgumentSCC[0]) {
Argument *A = ArgumentSCC[0]->Definition;
A->addAttr(Attribute::NoCapture);
++NumNoCapture;
Changed = true;
}
continue;
}
bool SCCCaptured = false;
for (auto I = ArgumentSCC.begin(), E = ArgumentSCC.end();
I != E && !SCCCaptured; ++I) {
ArgumentGraphNode *Node = *I;
if (Node->Uses.empty()) {
if (!Node->Definition->hasNoCaptureAttr())
SCCCaptured = true;
}
}
if (SCCCaptured)
continue;
SmallPtrSet<Argument *, 8> ArgumentSCCNodes;
// Fill ArgumentSCCNodes with the elements of the ArgumentSCC. Used for
// quickly looking up whether a given Argument is in this ArgumentSCC.
for (ArgumentGraphNode *I : ArgumentSCC) {
ArgumentSCCNodes.insert(I->Definition);
}
for (auto I = ArgumentSCC.begin(), E = ArgumentSCC.end();
I != E && !SCCCaptured; ++I) {
ArgumentGraphNode *N = *I;
for (ArgumentGraphNode *Use : N->Uses) {
Argument *A = Use->Definition;
if (A->hasNoCaptureAttr() || ArgumentSCCNodes.count(A))
continue;
SCCCaptured = true;
break;
}
}
if (SCCCaptured)
continue;
for (unsigned i = 0, e = ArgumentSCC.size(); i != e; ++i) {
Argument *A = ArgumentSCC[i]->Definition;
A->addAttr(Attribute::NoCapture);
++NumNoCapture;
Changed = true;
}
// We also want to compute readonly/readnone. With a small number of false
// negatives, we can assume that any pointer which is captured isn't going
// to be provably readonly or readnone, since by definition we can't
// analyze all uses of a captured pointer.
//
// The false negatives happen when the pointer is captured by a function
// that promises readonly/readnone behaviour on the pointer, then the
// pointer's lifetime ends before anything that writes to arbitrary memory.
// Also, a readonly/readnone pointer may be returned, but returning a
// pointer is capturing it.
Attribute::AttrKind ReadAttr = Attribute::ReadNone;
for (unsigned i = 0, e = ArgumentSCC.size(); i != e; ++i) {
Argument *A = ArgumentSCC[i]->Definition;
Attribute::AttrKind K = determinePointerReadAttrs(A, ArgumentSCCNodes);
if (K == Attribute::ReadNone)
continue;
if (K == Attribute::ReadOnly) {
ReadAttr = Attribute::ReadOnly;
continue;
}
ReadAttr = K;
break;
}
if (ReadAttr != Attribute::None) {
for (unsigned i = 0, e = ArgumentSCC.size(); i != e; ++i) {
Argument *A = ArgumentSCC[i]->Definition;
// Clear out existing readonly/readnone attributes
A->removeAttr(Attribute::ReadOnly);
A->removeAttr(Attribute::ReadNone);
A->addAttr(ReadAttr);
ReadAttr == Attribute::ReadOnly ? ++NumReadOnlyArg : ++NumReadNoneArg;
Changed = true;
}
}
}
return Changed;
}
/// AddNoCaptureAttrs - Deduce nocapture attributes for the SCC.
bool FunctionAttrs::AddNoCaptureAttrs(const CallGraphSCC &SCC) {
bool Changed = false;
SmallPtrSet<Function*, 8> SCCNodes;
// Fill SCCNodes with the elements of the SCC. Used for quickly
// looking up whether a given CallGraphNode is in this SCC.
for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I) {
Function *F = (*I)->getFunction();
if (F && !F->isDeclaration() && !F->mayBeOverridden())
SCCNodes.insert(F);
}
ArgumentGraph AG;
// Check each function in turn, determining which pointer arguments are not
// captured.
for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I) {
Function *F = (*I)->getFunction();
if (F == 0)
// External node - only a problem for arguments that we pass to it.
continue;
// Definitions with weak linkage may be overridden at linktime with
// something that captures pointers, so treat them like declarations.
if (F->isDeclaration() || F->mayBeOverridden())
continue;
// Functions that are readonly (or readnone) and nounwind and don't return
// a value can't capture arguments. Don't analyze them.
if (F->onlyReadsMemory() && F->doesNotThrow() &&
F->getReturnType()->isVoidTy()) {
for (Function::arg_iterator A = F->arg_begin(), E = F->arg_end();
A != E; ++A) {
if (A->getType()->isPointerTy() && !A->hasNoCaptureAttr()) {
A->addAttr(Attribute::NoCapture);
++NumNoCapture;
Changed = true;
}
}
continue;
}
for (Function::arg_iterator A = F->arg_begin(), E = F->arg_end(); A!=E; ++A)
if (A->getType()->isPointerTy() && !A->hasNoCaptureAttr()) {
ArgumentUsesTracker Tracker(SCCNodes);
PointerMayBeCaptured(A, &Tracker);
if (!Tracker.Captured) {
if (Tracker.Uses.empty()) {
// If it's trivially not captured, mark it nocapture now.
A->addAttr(Attribute::NoCapture);
++NumNoCapture;
Changed = true;
} else {
// If it's not trivially captured and not trivially not captured,
// then it must be calling into another function in our SCC. Save
// its particulars for Argument-SCC analysis later.
ArgumentGraphNode *Node = AG[A];
for (SmallVectorImpl<Argument*>::iterator UI = Tracker.Uses.begin(),
UE = Tracker.Uses.end(); UI != UE; ++UI)
Node->Uses.push_back(AG[*UI]);
}
}
// Otherwise, it's captured. Don't bother doing SCC analysis on it.
}
}
// The graph we've collected is partial because we stopped scanning for
// argument uses once we solved the argument trivially. These partial nodes
// show up as ArgumentGraphNode objects with an empty Uses list, and for
// these nodes the final decision about whether they capture has already been
// made. If the definition doesn't have a 'nocapture' attribute by now, it
// captures.
for (scc_iterator<ArgumentGraph*> I = scc_begin(&AG), E = scc_end(&AG);
I != E; ++I) {
std::vector<ArgumentGraphNode*> &ArgumentSCC = *I;
if (ArgumentSCC.size() == 1) {
if (!ArgumentSCC[0]->Definition) continue; // synthetic root node
// eg. "void f(int* x) { if (...) f(x); }"
if (ArgumentSCC[0]->Uses.size() == 1 &&
ArgumentSCC[0]->Uses[0] == ArgumentSCC[0]) {
ArgumentSCC[0]->Definition->addAttr(Attribute::NoCapture);
++NumNoCapture;
Changed = true;
}
continue;
}
bool SCCCaptured = false;
for (std::vector<ArgumentGraphNode*>::iterator I = ArgumentSCC.begin(),
E = ArgumentSCC.end(); I != E && !SCCCaptured; ++I) {
ArgumentGraphNode *Node = *I;
if (Node->Uses.empty()) {
if (!Node->Definition->hasNoCaptureAttr())
SCCCaptured = true;
}
}
if (SCCCaptured) continue;
SmallPtrSet<Argument*, 8> ArgumentSCCNodes;
// Fill ArgumentSCCNodes with the elements of the ArgumentSCC. Used for
// quickly looking up whether a given Argument is in this ArgumentSCC.
for (std::vector<ArgumentGraphNode*>::iterator I = ArgumentSCC.begin(),
E = ArgumentSCC.end(); I != E; ++I) {
ArgumentSCCNodes.insert((*I)->Definition);
}
for (std::vector<ArgumentGraphNode*>::iterator I = ArgumentSCC.begin(),
E = ArgumentSCC.end(); I != E && !SCCCaptured; ++I) {
ArgumentGraphNode *N = *I;
for (SmallVectorImpl<ArgumentGraphNode*>::iterator UI = N->Uses.begin(),
UE = N->Uses.end(); UI != UE; ++UI) {
Argument *A = (*UI)->Definition;
if (A->hasNoCaptureAttr() || ArgumentSCCNodes.count(A))
continue;
SCCCaptured = true;
break;
}
}
if (SCCCaptured) continue;
for (unsigned i = 0, e = ArgumentSCC.size(); i != e; ++i) {
Argument *A = ArgumentSCC[i]->Definition;
A->addAttr(Attribute::NoCapture);
++NumNoCapture;
Changed = true;
}
}
return Changed;
}
//
// Method: runOnModule()
//
// Description:
// Entry point for this LLVM pass.
// Clone functions that take GEPs as arguments
//
// Inputs:
// M - A reference to the LLVM module to transform
//
// Outputs:
// M - The transformed LLVM module.
//
// Return value:
// true - The module was modified.
// false - The module was not modified.
//
bool GEPExprArgs::runOnModule(Module& M) {
bool changed;
do {
changed = false;
for (Module::iterator F = M.begin(); F != M.end(); ++F){
for (Function::iterator B = F->begin(), FE = F->end(); B != FE; ++B) {
for (BasicBlock::iterator I = B->begin(), BE = B->end(); I != BE;) {
CallInst *CI = dyn_cast<CallInst>(I++);
if(!CI)
continue;
if(CI->hasByValArgument())
continue;
// if the GEP calls a function, that is externally defined,
// or might be changed, ignore this call site.
Function *F = CI->getCalledFunction();
if (!F || (F->isDeclaration() || F->mayBeOverridden()))
continue;
if(F->hasStructRetAttr())
continue;
if(F->isVarArg())
continue;
// find the argument we must replace
Function::arg_iterator ai = F->arg_begin(), ae = F->arg_end();
unsigned argNum = 1;
for(; argNum < CI->getNumOperands();argNum++, ++ai) {
if(ai->use_empty())
continue;
if (isa<GEPOperator>(CI->getOperand(argNum)))
break;
}
// if no argument was a GEP operator to be changed
if(ai == ae)
continue;
GEPOperator *GEP = dyn_cast<GEPOperator>(CI->getOperand(argNum));
if(!GEP->hasAllConstantIndices())
continue;
// Construct the new Type
// Appends the struct Type at the beginning
std::vector<Type*>TP;
TP.push_back(GEP->getPointerOperand()->getType());
for(unsigned c = 1; c < CI->getNumOperands();c++) {
TP.push_back(CI->getOperand(c)->getType());
}
//return type is same as that of original instruction
FunctionType *NewFTy = FunctionType::get(CI->getType(), TP, false);
Function *NewF;
numSimplified++;
if(numSimplified > 800)
return true;
NewF = Function::Create(NewFTy,
GlobalValue::InternalLinkage,
F->getName().str() + ".TEST",
&M);
Function::arg_iterator NI = NewF->arg_begin();
NI->setName("GEParg");
++NI;
ValueToValueMapTy ValueMap;
for (Function::arg_iterator II = F->arg_begin(); NI != NewF->arg_end(); ++II, ++NI) {
ValueMap[II] = NI;
NI->setName(II->getName());
NI->addAttr(F->getAttributes().getParamAttributes(II->getArgNo() + 1));
}
NewF->setAttributes(NewF->getAttributes().addAttr(
0, F->getAttributes().getRetAttributes()));
// Perform the cloning.
SmallVector<ReturnInst*,100> Returns;
CloneFunctionInto(NewF, F, ValueMap, false, Returns);
std::vector<Value*> fargs;
for(Function::arg_iterator ai = NewF->arg_begin(),
ae= NewF->arg_end(); ai != ae; ++ai) {
fargs.push_back(ai);
}
NewF->setAttributes(NewF->getAttributes().addAttr(
~0, F->getAttributes().getFnAttributes()));
//Get the point to insert the GEP instr.
SmallVector<Value*, 8> Ops(CI->op_begin()+1, CI->op_end());
Instruction *InsertPoint;
for (BasicBlock::iterator insrt = NewF->front().begin();
isa<AllocaInst>(InsertPoint = insrt); ++insrt) {;}
NI = NewF->arg_begin();
SmallVector<Value*, 8> Indices;
Indices.append(GEP->op_begin()+1, GEP->op_end());
GetElementPtrInst *GEP_new = GetElementPtrInst::Create(cast<Value>(NI),
Indices,
"", InsertPoint);
fargs.at(argNum)->replaceAllUsesWith(GEP_new);
unsigned j = argNum + 1;
for(; j < CI->getNumOperands();j++) {
if(CI->getOperand(j) == GEP)
fargs.at(j)->replaceAllUsesWith(GEP_new);
}
SmallVector<AttributeWithIndex, 8> AttributesVec;
// Get the initial attributes of the call
AttrListPtr CallPAL = CI->getAttributes();
Attributes RAttrs = CallPAL.getRetAttributes();
Attributes FnAttrs = CallPAL.getFnAttributes();
if (RAttrs)
AttributesVec.push_back(AttributeWithIndex::get(0, RAttrs));
SmallVector<Value*, 8> Args;
Args.push_back(GEP->getPointerOperand());
for(unsigned j =1;j<CI->getNumOperands();j++) {
Args.push_back(CI->getOperand(j));
// position in the AttributesVec
if (Attributes Attrs = CallPAL.getParamAttributes(j))
AttributesVec.push_back(AttributeWithIndex::get(Args.size(), Attrs));
}
// Create the new attributes vec.
if (FnAttrs != Attribute::None)
AttributesVec.push_back(AttributeWithIndex::get(~0, FnAttrs));
AttrListPtr NewCallPAL = AttrListPtr::get(AttributesVec.begin(),
AttributesVec.end());
CallInst *CallI = CallInst::Create(NewF,Args,"", CI);
CallI->setCallingConv(CI->getCallingConv());
CallI->setAttributes(NewCallPAL);
CI->replaceAllUsesWith(CallI);
CI->eraseFromParent();
changed = true;
}
}
}
} while(changed);
return true;
}
bool CodeGenerator::genFunction(FunctionDefn * fdef) {
// Don't generate undefined functions.
if (fdef->isUndefined() || fdef->isAbstract() || fdef->isInterfaceMethod()) {
return true;
}
DASSERT_OBJ(fdef->isSingular(), fdef);
DASSERT_OBJ(fdef->type(), fdef);
DASSERT_OBJ(fdef->type()->isSingular(), fdef);
// Don't generate intrinsic functions.
if (fdef->isIntrinsic()) {
return true;
}
// Don't generate a function if it has been merged to another function
if (fdef->mergeTo() != NULL || fdef->isUndefined()) {
return true;
}
// Create the function
Function * f = genFunctionValue(fdef);
if (fdef->hasBody() && f->getBasicBlockList().empty()) {
FunctionType * ftype = fdef->functionType();
if (fdef->isSynthetic()) {
f->setLinkage(GlobalValue::LinkOnceODRLinkage);
}
if (gcEnabled_) {
if (SsGC) {
f->setGC("shadow-stack");
} else {
f->setGC("tart-gc");
}
}
if (debug_) {
dbgContext_ = genDISubprogram(fdef);
//dbgContext_ = genLexicalBlock(fdef->location());
dbgInlineContext_ = DIScope();
setDebugLocation(fdef->location());
}
BasicBlock * prologue = BasicBlock::Create(context_, "prologue", f);
// Create the LLVM Basic Blocks corresponding to each high level BB.
// BlockList & blocks = fdef->blocks();
// for (BlockList::iterator b = blocks.begin(); b != blocks.end(); ++b) {
// Block * blk = *b;
// blk->setIRBlock(BasicBlock::Create(context_, blk->label(), f));
// }
builder_.SetInsertPoint(prologue);
// Handle the explicit parameters
unsigned param_index = 0;
Function::arg_iterator it = f->arg_begin();
Value * saveStructRet = structRet_;
if (ftype->isStructReturn()) {
it->addAttr(llvm::Attribute::StructRet);
structRet_ = it;
++it;
}
// Handle the 'self' parameter
if (ftype->selfParam() != NULL) {
ParameterDefn * selfParam = ftype->selfParam();
const Type * selfParamType = selfParam->type().unqualified();
DASSERT_OBJ(fdef->storageClass() == Storage_Instance ||
fdef->storageClass() == Storage_Local, fdef);
DASSERT_OBJ(it != f->arg_end(), ftype);
// Check if the self param is a root.
if (selfParamType->isReferenceType()) {
selfParam->setFlag(ParameterDefn::LValueParam, true);
Value * selfAlloca = builder_.CreateAlloca(
selfParam->type()->irEmbeddedType(), 0, "self.alloca");
builder_.CreateStore(it, selfAlloca);
selfParam->setIRValue(selfAlloca);
markGCRoot(selfAlloca, NULL, "self.alloca");
} else {
// Since selfParam is always a pointer, we don't need to mark the object pointed
// to as a root, since the next call frame up is responsible for tracing it.
ftype->selfParam()->setIRValue(it);
}
it->setName("self");
++it;
}
// If this function needs to make allocations, cache a copy of the
// allocation context pointer for this thread, since it can on some
// platforms be expensive to look up.
if (fdef->flags() & FunctionDefn::MakesAllocs) {
Function * gcGetAllocContext = genFunctionValue(gc_allocContext);
gcAllocContext_ = builder_.CreateCall(gcGetAllocContext, "allocCtx");
}
for (; it != f->arg_end(); ++it, ++param_index) {
// Set the name of the Nth parameter
ParameterDefn * param = ftype->params()[param_index];
DASSERT_OBJ(param != NULL, fdef);
DASSERT_OBJ(param->storageClass() == Storage_Local, param);
QualifiedType paramType = param->internalType();
it->setName(param->name());
Value * paramValue = it;
// If the parameter is a shared reference, then create the shared ref.
if (param->isSharedRef()) {
genLocalVar(param, paramValue);
genGCRoot(param->irValue(), param->sharedRefType(), param->name());
continue;
}
// If the parameter type contains any reference types, then the parameter needs
// to be a root.
bool paramIsRoot = false;
if (paramType->isReferenceType()) {
param->setFlag(ParameterDefn::LValueParam, true);
paramIsRoot = true;
} else if (paramType->containsReferenceType()) {
// TODO: Handle roots of various shapes
//param->setFlag(ParameterDefn::LValueParam, true);
}
// See if we need to make a local copy of the param.
if (param->isLValue()) {
Value * paramAlloca = builder_.CreateAlloca(paramType->irEmbeddedType(), 0, param->name());
param->setIRValue(paramAlloca);
if (paramType->typeShape() == Shape_Large_Value) {
paramValue = builder_.CreateLoad(paramValue);
}
builder_.CreateStore(paramValue, paramAlloca);
if (paramIsRoot) {
genGCRoot(paramAlloca, paramType.unqualified(), param->name());
}
} else {
param->setIRValue(paramValue);
}
}
// Generate the body
Function * saveFn = currentFn_;
currentFn_ = f;
#if 0
if (fdef->isGenerator()) {
assert(false);
} else {
#endif
genLocalStorage(fdef->localScopes());
genDISubprogramStart(fdef);
genLocalRoots(fdef->localScopes());
BasicBlock * blkEntry = createBlock("entry");
builder_.SetInsertPoint(blkEntry);
genExpr(fdef->body());
if (!atTerminator()) {
if (fdef->returnType()->isVoidType()) {
builder_.CreateRetVoid();
} else {
// TODO: Use the location from the last statement of the function.
diag.error(fdef) << "Missing return statement at end of non-void function.";
}
}
gcAllocContext_ = NULL;
#if 0
}
#endif
builder_.SetInsertPoint(prologue);
builder_.CreateBr(blkEntry);
currentFn_ = saveFn;
structRet_ = saveStructRet;
if (!diag.inRecovery()) {
if (verifyFunction(*f, PrintMessageAction)) {
f->dump();
DFAIL("Function failed to verify");
}
}
//if (debug_ && !dbgContext_.isNull() && !dbgContext_.Verify()) {
// dbgContext_.Verify();
// DFAIL("BAD DBG");
//}
dbgContext_ = DIScope();
dbgInlineContext_ = DIScope();
builder_.ClearInsertionPoint();
builder_.SetCurrentDebugLocation(llvm::DebugLoc());
}
return true;
}
//
// Method: runOnModule()
//
// Description:
// Entry point for this LLVM pass.
// Clone functions that take LoadInsts as arguments
//
// Inputs:
// M - A reference to the LLVM module to transform
//
// Outputs:
// M - The transformed LLVM module.
//
// Return value:
// true - The module was modified.
// false - The module was not modified.
//
bool LoadArgs::runOnModule(Module& M) {
std::map<std::pair<Function*, const Type * > , Function* > fnCache;
bool changed;
do {
changed = false;
for (Module::iterator Func = M.begin(); Func != M.end(); ++Func) {
for (Function::iterator B = Func->begin(), FE = Func->end(); B != FE; ++B) {
for (BasicBlock::iterator I = B->begin(), BE = B->end(); I != BE;) {
CallInst *CI = dyn_cast<CallInst>(I++);
if(!CI)
continue;
if(CI->hasByValArgument())
continue;
// if the CallInst calls a function, that is externally defined,
// or might be changed, ignore this call site.
Function *F = CI->getCalledFunction();
if (!F || (F->isDeclaration() || F->mayBeOverridden()))
continue;
if(F->hasStructRetAttr())
continue;
if(F->isVarArg())
continue;
// find the argument we must replace
Function::arg_iterator ai = F->arg_begin(), ae = F->arg_end();
unsigned argNum = 0;
for(; argNum < CI->getNumArgOperands();argNum++, ++ai) {
// do not care about dead arguments
if(ai->use_empty())
continue;
if(F->getAttributes().getParamAttributes(argNum).hasAttrSomewhere(Attribute::SExt) ||
F->getAttributes().getParamAttributes(argNum).hasAttrSomewhere(Attribute::ZExt))
continue;
if (isa<LoadInst>(CI->getArgOperand(argNum)))
break;
}
// if no argument was a GEP operator to be changed
if(ai == ae)
continue;
LoadInst *LI = dyn_cast<LoadInst>(CI->getArgOperand(argNum));
Instruction * InsertPt = &(Func->getEntryBlock().front());
AllocaInst *NewVal = new AllocaInst(LI->getType(), "",InsertPt);
StoreInst *Copy = new StoreInst(LI, NewVal);
Copy->insertAfter(LI);
/*if(LI->getParent() != CI->getParent())
continue;
// Also check that there is no store after the load.
// TODO: Check if the load/store do not alias.
BasicBlock::iterator bii = LI->getParent()->begin();
Instruction *BII = bii;
while(BII != LI) {
++bii;
BII = bii;
}
while(BII != CI) {
if(isa<StoreInst>(BII))
break;
++bii;
BII = bii;
}
if(isa<StoreInst>(bii)){
continue;
}*/
// Construct the new Type
// Appends the struct Type at the beginning
std::vector<Type*>TP;
for(unsigned c = 0; c < CI->getNumArgOperands();c++) {
if(c == argNum)
TP.push_back(LI->getPointerOperand()->getType());
TP.push_back(CI->getArgOperand(c)->getType());
}
//return type is same as that of original instruction
FunctionType *NewFTy = FunctionType::get(CI->getType(), TP, false);
numSimplified++;
//if(numSimplified > 1000)
//return true;
Function *NewF;
std::map<std::pair<Function*, const Type* > , Function* >::iterator Test;
Test = fnCache.find(std::make_pair(F, NewFTy));
if(Test != fnCache.end()) {
NewF = Test->second;
} else {
NewF = Function::Create(NewFTy,
GlobalValue::InternalLinkage,
F->getName().str() + ".TEST",
&M);
fnCache[std::make_pair(F, NewFTy)] = NewF;
Function::arg_iterator NI = NewF->arg_begin();
ValueToValueMapTy ValueMap;
unsigned count = 0;
for (Function::arg_iterator II = F->arg_begin(); NI != NewF->arg_end(); ++count, ++NI) {
if(count == argNum) {
NI->setName("LDarg");
continue;
}
ValueMap[II] = NI;
NI->setName(II->getName());
NI->addAttr(F->getAttributes().getParamAttributes(II->getArgNo() + 1));
++II;
}
// Perform the cloning.
SmallVector<ReturnInst*,100> Returns;
CloneFunctionInto(NewF, F, ValueMap, false, Returns);
std::vector<Value*> fargs;
for(Function::arg_iterator ai = NewF->arg_begin(),
ae= NewF->arg_end(); ai != ae; ++ai) {
fargs.push_back(ai);
}
NewF->setAttributes(NewF->getAttributes().addAttributes(
F->getContext(), 0, F->getAttributes().getRetAttributes()));
NewF->setAttributes(NewF->getAttributes().addAttributes(
F->getContext(), ~0, F->getAttributes().getFnAttributes()));
//Get the point to insert the GEP instr.
Instruction *InsertPoint;
for (BasicBlock::iterator insrt = NewF->front().begin(); isa<AllocaInst>(InsertPoint = insrt); ++insrt) {;}
LoadInst *LI_new = new LoadInst(fargs.at(argNum), "", InsertPoint);
fargs.at(argNum+1)->replaceAllUsesWith(LI_new);
}
//this does not seem to be a good idea
AttributeSet NewCallPAL=AttributeSet();
// Get the initial attributes of the call
AttributeSet CallPAL = CI->getAttributes();
AttributeSet RAttrs = CallPAL.getRetAttributes();
AttributeSet FnAttrs = CallPAL.getFnAttributes();
if (!RAttrs.isEmpty())
NewCallPAL=NewCallPAL.addAttributes(F->getContext(),0, RAttrs);
SmallVector<Value*, 8> Args;
for(unsigned j =0;j<CI->getNumArgOperands();j++) {
if(j == argNum) {
Args.push_back(NewVal);
}
Args.push_back(CI->getArgOperand(j));
// position in the NewCallPAL
AttributeSet Attrs = CallPAL.getParamAttributes(j+1);
if (!Attrs.isEmpty())
NewCallPAL=NewCallPAL.addAttributes(F->getContext(),Args.size(), Attrs);
}
// Create the new attributes vec.
if (!FnAttrs.isEmpty())
NewCallPAL=NewCallPAL.addAttributes(F->getContext(),~0, FnAttrs);
CallInst *CallI = CallInst::Create(NewF,Args,"", CI);
CallI->setCallingConv(CI->getCallingConv());
CallI->setAttributes(NewCallPAL);
CI->replaceAllUsesWith(CallI);
CI->eraseFromParent();
changed = true;
}
}
}
} while(changed);
return true;
}
//
// Method: runOnModule()
//
// Description:
// Entry point for this LLVM pass.
// If a function returns a struct, make it return
// a pointer to the struct.
//
// Inputs:
// M - A reference to the LLVM module to transform
//
// Outputs:
// M - The transformed LLVM module.
//
// Return value:
// true - The module was modified.
// false - The module was not modified.
//
bool StructRet::runOnModule(Module& M) {
const llvm::DataLayout targetData(&M);
std::vector<Function*> worklist;
for (Module::iterator I = M.begin(); I != M.end(); ++I)
if (!I->mayBeOverridden()) {
if(I->hasAddressTaken())
continue;
if(I->getReturnType()->isStructTy()) {
worklist.push_back(I);
}
}
while(!worklist.empty()) {
Function *F = worklist.back();
worklist.pop_back();
Type *NewArgType = F->getReturnType()->getPointerTo();
// Construct the new Type
std::vector<Type*>TP;
TP.push_back(NewArgType);
for (Function::arg_iterator ii = F->arg_begin(), ee = F->arg_end();
ii != ee; ++ii) {
TP.push_back(ii->getType());
}
FunctionType *NFTy = FunctionType::get(F->getReturnType(), TP, F->isVarArg());
// Create the new function body and insert it into the module.
Function *NF = Function::Create(NFTy,
F->getLinkage(),
F->getName(), &M);
ValueToValueMapTy ValueMap;
Function::arg_iterator NI = NF->arg_begin();
NI->setName("ret");
++NI;
for (Function::arg_iterator II = F->arg_begin(); II != F->arg_end(); ++II, ++NI) {
ValueMap[II] = NI;
NI->setName(II->getName());
AttributeSet attrs = F->getAttributes().getParamAttributes(II->getArgNo() + 1);
if (!attrs.isEmpty())
NI->addAttr(attrs);
}
// Perform the cloning.
SmallVector<ReturnInst*,100> Returns;
if (!F->isDeclaration())
CloneFunctionInto(NF, F, ValueMap, false, Returns);
std::vector<Value*> fargs;
for(Function::arg_iterator ai = NF->arg_begin(),
ae= NF->arg_end(); ai != ae; ++ai) {
fargs.push_back(ai);
}
NF->setAttributes(NF->getAttributes().addAttributes(
M.getContext(), 0, F->getAttributes().getRetAttributes()));
NF->setAttributes(NF->getAttributes().addAttributes(
M.getContext(), ~0, F->getAttributes().getFnAttributes()));
for (Function::iterator B = NF->begin(), FE = NF->end(); B != FE; ++B) {
for (BasicBlock::iterator I = B->begin(), BE = B->end(); I != BE;) {
ReturnInst * RI = dyn_cast<ReturnInst>(I++);
if(!RI)
continue;
LoadInst *LI = dyn_cast<LoadInst>(RI->getOperand(0));
assert(LI && "Return should be preceded by a load instruction");
IRBuilder<> Builder(RI);
Builder.CreateMemCpy(fargs.at(0),
LI->getPointerOperand(),
targetData.getTypeStoreSize(LI->getType()),
targetData.getPrefTypeAlignment(LI->getType()));
}
}
for(Value::use_iterator ui = F->use_begin(), ue = F->use_end();
ui != ue; ) {
CallInst *CI = dyn_cast<CallInst>(*ui++);
if(!CI)
continue;
if(CI->getCalledFunction() != F)
continue;
if(CI->hasByValArgument())
continue;
AllocaInst *AllocaNew = new AllocaInst(F->getReturnType(), 0, "", CI);
SmallVector<Value*, 8> Args;
//this should probably be done in a different manner
AttributeSet NewCallPAL=AttributeSet();
// Get the initial attributes of the call
AttributeSet CallPAL = CI->getAttributes();
AttributeSet RAttrs = CallPAL.getRetAttributes();
AttributeSet FnAttrs = CallPAL.getFnAttributes();
if (!RAttrs.isEmpty())
NewCallPAL=NewCallPAL.addAttributes(F->getContext(),0, RAttrs);
Args.push_back(AllocaNew);
for(unsigned j = 0; j < CI->getNumOperands()-1; j++) {
Args.push_back(CI->getOperand(j));
// position in the NewCallPAL
AttributeSet Attrs = CallPAL.getParamAttributes(j);
if (!Attrs.isEmpty())
NewCallPAL=NewCallPAL.addAttributes(F->getContext(),Args.size(), Attrs);
}
// Create the new attributes vec.
if (!FnAttrs.isEmpty())
NewCallPAL=NewCallPAL.addAttributes(F->getContext(),~0, FnAttrs);
CallInst *CallI = CallInst::Create(NF, Args, "", CI);
CallI->setCallingConv(CI->getCallingConv());
CallI->setAttributes(NewCallPAL);
LoadInst *LI = new LoadInst(AllocaNew, "", CI);
CI->replaceAllUsesWith(LI);
CI->eraseFromParent();
}
if(F->use_empty())
F->eraseFromParent();
}
return true;
}
