bool SpDefUseInstrumenter::runOnModule(Module &M) {
cerr << "instrument: --- Def-Use pair Spectrum ---\n";
Function *Main = M.getFunction("main");
LLVMContext &C = M.getContext();
if (Main == 0) {
cerr << "WARNING: cannot insert def-use instrumentation into a module"
<< " with no main function!\n";
return false; // No main, no instrumentation!
}
// Add library function prototype
Constant *SpFn = M.getOrInsertFunction("_updateSpectrum",
Type::getVoidTy(C),
Type::getInt32Ty(C), // spectrum index
Type::getInt32Ty(C), // component index
NULL);
unsigned spectrumIndex = IndexManager::getSpectrumIndex();
unsigned nDefs = 0;
unsigned nUses = 0;
// Loop through all functions within module
for (Module::iterator F = M.begin(), ME = M.end(); F != ME; ++F) {
// skip function declarations
if(F->isDeclaration())
continue;
// skip the _registerAll function
if(F->getName()=="_registerAll")
continue;
// Loop through all basic blocks within function
for (Function::iterator B = F->begin(), FE = F->end(); B != FE; ++B) {
//skip dead blocks
//is this really safe??
BasicBlock *bb = B;
if (B!=F->begin() && (pred_begin(bb)==pred_end(bb))) continue; //skip dead blocks
// Loop through all instructions within basic block
for (BasicBlock::iterator I = B->begin(), BE = B->end(); I != BE; I++) {
if(isa<DbgDeclareInst>(*I)) {
// extract source file information from debug intrinsic
DbgDeclareInst &DDI = cast<DbgDeclareInst>(*I);
std::string file, dir;
std::string name;
GlobalVariable *gv = cast<GlobalVariable>(DDI.getVariable());
if(!gv->hasInitializer()) continue;
ConstantStruct *cs = cast<ConstantStruct>(gv->getInitializer());
llvm::GetConstantStringInfo(cs->getOperand(2), name);
unsigned int line = unsigned(cast<ConstantInt>(cs->getOperand(4))->getZExtValue());
Value *V = cast<Value>(cs->getOperand(3));
GlobalVariable *gv2 = cast<GlobalVariable>(cast<ConstantExpr>(V)->getOperand(0));
if(!gv2->hasInitializer()) continue;
ConstantStruct *cs2 = cast<ConstantStruct>(gv2->getInitializer());
llvm::GetConstantStringInfo(cs2->getOperand(3), file);
llvm::GetConstantStringInfo(cs2->getOperand(4), dir);
// get the allocation instruction of the variable definition
AllocaInst *AI;
if(isa<AllocaInst>(DDI.getAddress())) {
AI = cast<AllocaInst>(DDI.getAddress());
} else if (isa<BitCastInst>(DDI.getAddress())) {
AI = cast<AllocaInst>(cast<BitCastInst>(DDI.getAddress())->getOperand(0));
} else {
continue;
}
nDefs++;
// add calls to lib function for each use of the variable
int currUses = 0;
for(AllocaInst::use_iterator U = AI->use_begin(), UE = AI->use_end(); U != UE; ++U) {
if(isa<Instruction>(*U)) {
User *user = *U;
Instruction *insertInst = (Instruction*)user;
// find most likely context location of use
int useline = line;
std::string usefile = file, usedir = dir;
BasicBlock *parent = insertInst->getParent();
BasicBlock::iterator inst = parent->begin();
while(((Instruction *)inst) != insertInst && inst != parent->end()) {
/*TODO: solve DbgStopPointInst problem*/
/*if(isa<DbgStopPointInst>(*inst)) {
DbgStopPointInst &DSPI = cast<DbgStopPointInst>(*inst);
llvm::GetConstantStringInfo(DSPI.getDirectory(), usedir);
llvm::GetConstantStringInfo(DSPI.getFileName(), usefile);
useline = DSPI.getLine();
}*/
inst++;
}
std::stringstream usename;
usename << name << "(use_" << currUses << ")";
// add source context of this invariant to context file
ContextManager::addSpectrumContext(
spectrumIndex, // spectrumIndex
nUses, // componentIndex
usedir, // path
usefile, // file
useline, // line
usename.str()); // name
currUses++;
std::vector<Value*> Args(2);
Args[0] = ConstantInt::get(Type::getInt32Ty(C), spectrumIndex);
Args[1] = ConstantInt::get(Type::getInt32Ty(C), nUses++);
CallInst::Create(SpFn, Args.begin(), Args.end(), "", insertInst);
}
}
}
}
}
}
// add the registration of the instrumented spectrum points in the _registerAll() function
addSpectrumRegistration(M, spectrumIndex, nUses, "Def-Use_Pairs");
std::cerr << "instrument: " << nDefs << " defines with a total number of " << nUses << " uses instrumented\n";
// notify change of program
return true;
}
bool SjLjEHPass::insertSjLjEHSupport(Function &F) {
SmallVector<ReturnInst*,16> Returns;
SmallVector<UnwindInst*,16> Unwinds;
SmallVector<InvokeInst*,16> Invokes;
// Look through the terminators of the basic blocks to find invokes, returns
// and unwinds.
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
if (ReturnInst *RI = dyn_cast<ReturnInst>(BB->getTerminator())) {
// Remember all return instructions in case we insert an invoke into this
// function.
Returns.push_back(RI);
} else if (InvokeInst *II = dyn_cast<InvokeInst>(BB->getTerminator())) {
Invokes.push_back(II);
} else if (UnwindInst *UI = dyn_cast<UnwindInst>(BB->getTerminator())) {
Unwinds.push_back(UI);
}
}
NumInvokes += Invokes.size();
NumUnwinds += Unwinds.size();
// If we don't have any invokes, there's nothing to do.
if (Invokes.empty()) return false;
// Find the eh.selector.*, eh.exception and alloca calls.
//
// Remember any allocas() that aren't in the entry block, as the
// jmpbuf saved SP will need to be updated for them.
//
// We'll use the first eh.selector to determine the right personality
// function to use. For SJLJ, we always use the same personality for the
// whole function, not on a per-selector basis.
// FIXME: That's a bit ugly. Better way?
SmallVector<CallInst*,16> EH_Selectors;
SmallVector<CallInst*,16> EH_Exceptions;
SmallVector<Instruction*,16> JmpbufUpdatePoints;
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
// Note: Skip the entry block since there's nothing there that interests
// us. eh.selector and eh.exception shouldn't ever be there, and we
// want to disregard any allocas that are there.
//
// FIXME: This is awkward. The new EH scheme won't need to skip the entry
// block.
if (BB == F.begin()) {
if (InvokeInst *II = dyn_cast<InvokeInst>(F.begin()->getTerminator())) {
// FIXME: This will be always non-NULL in the new EH.
if (LandingPadInst *LPI = II->getUnwindDest()->getLandingPadInst())
if (!PersonalityFn) PersonalityFn = LPI->getPersonalityFn();
}
continue;
}
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
if (CallInst *CI = dyn_cast<CallInst>(I)) {
if (CI->getCalledFunction() == SelectorFn) {
if (!PersonalityFn) PersonalityFn = CI->getArgOperand(1);
EH_Selectors.push_back(CI);
} else if (CI->getCalledFunction() == ExceptionFn) {
EH_Exceptions.push_back(CI);
} else if (CI->getCalledFunction() == StackRestoreFn) {
JmpbufUpdatePoints.push_back(CI);
}
} else if (AllocaInst *AI = dyn_cast<AllocaInst>(I)) {
JmpbufUpdatePoints.push_back(AI);
} else if (InvokeInst *II = dyn_cast<InvokeInst>(I)) {
// FIXME: This will be always non-NULL in the new EH.
if (LandingPadInst *LPI = II->getUnwindDest()->getLandingPadInst())
if (!PersonalityFn) PersonalityFn = LPI->getPersonalityFn();
}
}
}
// If we don't have any eh.selector calls, we can't determine the personality
// function. Without a personality function, we can't process exceptions.
if (!PersonalityFn) return false;
// We have invokes, so we need to add register/unregister calls to get this
// function onto the global unwind stack.
//
// First thing we need to do is scan the whole function for values that are
// live across unwind edges. Each value that is live across an unwind edge we
// spill into a stack location, guaranteeing that there is nothing live across
// the unwind edge. This process also splits all critical edges coming out of
// invoke's.
splitLiveRangesAcrossInvokes(Invokes);
SmallVector<LandingPadInst*, 16> LandingPads;
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
if (InvokeInst *II = dyn_cast<InvokeInst>(BB->getTerminator()))
// FIXME: This will be always non-NULL in the new EH.
if (LandingPadInst *LPI = II->getUnwindDest()->getLandingPadInst())
LandingPads.push_back(LPI);
}
BasicBlock *EntryBB = F.begin();
// Create an alloca for the incoming jump buffer ptr and the new jump buffer
// that needs to be restored on all exits from the function. This is an
// alloca because the value needs to be added to the global context list.
unsigned Align = 4; // FIXME: Should be a TLI check?
AllocaInst *FunctionContext =
new AllocaInst(FunctionContextTy, 0, Align,
"fcn_context", F.begin()->begin());
Value *Idxs[2];
Type *Int32Ty = Type::getInt32Ty(F.getContext());
Value *Zero = ConstantInt::get(Int32Ty, 0);
// We need to also keep around a reference to the call_site field
Idxs[0] = Zero;
Idxs[1] = ConstantInt::get(Int32Ty, 1);
CallSite = GetElementPtrInst::Create(FunctionContext, Idxs, "call_site",
EntryBB->getTerminator());
// The exception selector comes back in context->data[1]
Idxs[1] = ConstantInt::get(Int32Ty, 2);
Value *FCData = GetElementPtrInst::Create(FunctionContext, Idxs, "fc_data",
EntryBB->getTerminator());
Idxs[1] = ConstantInt::get(Int32Ty, 1);
Value *SelectorAddr = GetElementPtrInst::Create(FCData, Idxs,
"exc_selector_gep",
EntryBB->getTerminator());
// The exception value comes back in context->data[0]
Idxs[1] = Zero;
Value *ExceptionAddr = GetElementPtrInst::Create(FCData, Idxs,
"exception_gep",
EntryBB->getTerminator());
// The result of the eh.selector call will be replaced with a a reference to
// the selector value returned in the function context. We leave the selector
// itself so the EH analysis later can use it.
for (int i = 0, e = EH_Selectors.size(); i < e; ++i) {
CallInst *I = EH_Selectors[i];
Value *SelectorVal = new LoadInst(SelectorAddr, "select_val", true, I);
I->replaceAllUsesWith(SelectorVal);
}
// eh.exception calls are replaced with references to the proper location in
// the context. Unlike eh.selector, the eh.exception calls are removed
// entirely.
for (int i = 0, e = EH_Exceptions.size(); i < e; ++i) {
CallInst *I = EH_Exceptions[i];
// Possible for there to be duplicates, so check to make sure the
// instruction hasn't already been removed.
if (!I->getParent()) continue;
Value *Val = new LoadInst(ExceptionAddr, "exception", true, I);
Type *Ty = Type::getInt8PtrTy(F.getContext());
Val = CastInst::Create(Instruction::IntToPtr, Val, Ty, "", I);
I->replaceAllUsesWith(Val);
I->eraseFromParent();
}
for (unsigned i = 0, e = LandingPads.size(); i != e; ++i)
ReplaceLandingPadVal(F, LandingPads[i], ExceptionAddr, SelectorAddr);
// The entry block changes to have the eh.sjlj.setjmp, with a conditional
// branch to a dispatch block for non-zero returns. If we return normally,
// we're not handling an exception and just register the function context and
// continue.
// Create the dispatch block. The dispatch block is basically a big switch
// statement that goes to all of the invoke landing pads.
BasicBlock *DispatchBlock =
BasicBlock::Create(F.getContext(), "eh.sjlj.setjmp.catch", &F);
// Insert a load of the callsite in the dispatch block, and a switch on its
// value. By default, we issue a trap statement.
BasicBlock *TrapBlock =
BasicBlock::Create(F.getContext(), "trapbb", &F);
CallInst::Create(Intrinsic::getDeclaration(F.getParent(), Intrinsic::trap),
"", TrapBlock);
new UnreachableInst(F.getContext(), TrapBlock);
Value *DispatchLoad = new LoadInst(CallSite, "invoke.num", true,
DispatchBlock);
SwitchInst *DispatchSwitch =
SwitchInst::Create(DispatchLoad, TrapBlock, Invokes.size(),
DispatchBlock);
// Split the entry block to insert the conditional branch for the setjmp.
BasicBlock *ContBlock = EntryBB->splitBasicBlock(EntryBB->getTerminator(),
"eh.sjlj.setjmp.cont");
// Populate the Function Context
// 1. LSDA address
// 2. Personality function address
// 3. jmpbuf (save SP, FP and call eh.sjlj.setjmp)
// LSDA address
Idxs[0] = Zero;
Idxs[1] = ConstantInt::get(Int32Ty, 4);
Value *LSDAFieldPtr =
GetElementPtrInst::Create(FunctionContext, Idxs, "lsda_gep",
EntryBB->getTerminator());
Value *LSDA = CallInst::Create(LSDAAddrFn, "lsda_addr",
EntryBB->getTerminator());
new StoreInst(LSDA, LSDAFieldPtr, true, EntryBB->getTerminator());
Idxs[1] = ConstantInt::get(Int32Ty, 3);
Value *PersonalityFieldPtr =
GetElementPtrInst::Create(FunctionContext, Idxs, "lsda_gep",
EntryBB->getTerminator());
new StoreInst(PersonalityFn, PersonalityFieldPtr, true,
EntryBB->getTerminator());
// Save the frame pointer.
Idxs[1] = ConstantInt::get(Int32Ty, 5);
Value *JBufPtr
= GetElementPtrInst::Create(FunctionContext, Idxs, "jbuf_gep",
EntryBB->getTerminator());
Idxs[1] = ConstantInt::get(Int32Ty, 0);
Value *FramePtr =
GetElementPtrInst::Create(JBufPtr, Idxs, "jbuf_fp_gep",
EntryBB->getTerminator());
Value *Val = CallInst::Create(FrameAddrFn,
ConstantInt::get(Int32Ty, 0),
"fp",
EntryBB->getTerminator());
new StoreInst(Val, FramePtr, true, EntryBB->getTerminator());
// Save the stack pointer.
Idxs[1] = ConstantInt::get(Int32Ty, 2);
Value *StackPtr =
GetElementPtrInst::Create(JBufPtr, Idxs, "jbuf_sp_gep",
EntryBB->getTerminator());
Val = CallInst::Create(StackAddrFn, "sp", EntryBB->getTerminator());
new StoreInst(Val, StackPtr, true, EntryBB->getTerminator());
// Call the setjmp instrinsic. It fills in the rest of the jmpbuf.
Value *SetjmpArg =
CastInst::Create(Instruction::BitCast, JBufPtr,
Type::getInt8PtrTy(F.getContext()), "",
EntryBB->getTerminator());
Value *DispatchVal = CallInst::Create(BuiltinSetjmpFn, SetjmpArg,
"dispatch",
EntryBB->getTerminator());
// Add a call to dispatch_setup after the setjmp call. This is expanded to any
// target-specific setup that needs to be done.
CallInst::Create(DispatchSetupFn, DispatchVal, "", EntryBB->getTerminator());
// check the return value of the setjmp. non-zero goes to dispatcher.
Value *IsNormal = new ICmpInst(EntryBB->getTerminator(),
ICmpInst::ICMP_EQ, DispatchVal, Zero,
"notunwind");
// Nuke the uncond branch.
EntryBB->getTerminator()->eraseFromParent();
// Put in a new condbranch in its place.
BranchInst::Create(ContBlock, DispatchBlock, IsNormal, EntryBB);
// Register the function context and make sure it's known to not throw
CallInst *Register =
CallInst::Create(RegisterFn, FunctionContext, "",
ContBlock->getTerminator());
Register->setDoesNotThrow();
// At this point, we are all set up, update the invoke instructions to mark
// their call_site values, and fill in the dispatch switch accordingly.
for (unsigned i = 0, e = Invokes.size(); i != e; ++i)
markInvokeCallSite(Invokes[i], i+1, CallSite, DispatchSwitch);
// Mark call instructions that aren't nounwind as no-action (call_site ==
// -1). Skip the entry block, as prior to then, no function context has been
// created for this function and any unexpected exceptions thrown will go
// directly to the caller's context, which is what we want anyway, so no need
// to do anything here.
for (Function::iterator BB = F.begin(), E = F.end(); ++BB != E;) {
for (BasicBlock::iterator I = BB->begin(), end = BB->end(); I != end; ++I)
if (CallInst *CI = dyn_cast<CallInst>(I)) {
// Ignore calls to the EH builtins (eh.selector, eh.exception)
Constant *Callee = CI->getCalledFunction();
if (Callee != SelectorFn && Callee != ExceptionFn
&& !CI->doesNotThrow())
insertCallSiteStore(CI, -1, CallSite);
} else if (ResumeInst *RI = dyn_cast<ResumeInst>(I)) {
insertCallSiteStore(RI, -1, CallSite);
}
}
// Replace all unwinds with a branch to the unwind handler.
// ??? Should this ever happen with sjlj exceptions?
for (unsigned i = 0, e = Unwinds.size(); i != e; ++i) {
BranchInst::Create(TrapBlock, Unwinds[i]);
Unwinds[i]->eraseFromParent();
}
// Following any allocas not in the entry block, update the saved SP in the
// jmpbuf to the new value.
for (unsigned i = 0, e = JmpbufUpdatePoints.size(); i != e; ++i) {
Instruction *AI = JmpbufUpdatePoints[i];
Instruction *StackAddr = CallInst::Create(StackAddrFn, "sp");
StackAddr->insertAfter(AI);
Instruction *StoreStackAddr = new StoreInst(StackAddr, StackPtr, true);
StoreStackAddr->insertAfter(StackAddr);
}
// Finally, for any returns from this function, if this function contains an
// invoke, add a call to unregister the function context.
for (unsigned i = 0, e = Returns.size(); i != e; ++i)
CallInst::Create(UnregisterFn, FunctionContext, "", Returns[i]);
return true;
}
/*
* Rewrite OpenMP call sites and their associated kernel functions -- the folloiwng pattern
call void @GOMP_parallel_start(void (i8*)* @_Z20initialize_variablesiPfS_.omp_fn.4, i8* %.omp_data_o.5571, i32 0) nounwind
call void @_Z20initialize_variablesiPfS_.omp_fn.4(i8* %.omp_data_o.5571) nounwind
call void @GOMP_parallel_end() nounwind
*/
void HeteroOMPTransform::rewrite_omp_call_sites(Module &M) {
SmallVector<Instruction *, 16> toDelete;
DenseMap<Value *, Value *> ValueMap;
for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I){
if (!I->isDeclaration()) {
for (Function::iterator BBI = I->begin(), BBE = I->end(); BBI != BBE; ++BBI) {
bool match = false;
for (BasicBlock::iterator INSNI = BBI->begin(), INSNE = BBI->end(); INSNI != INSNE; ++INSNI) {
if (isa<CallInst>(INSNI)) {
CallSite CI(cast<Instruction>(INSNI));
if (CI.getCalledFunction() != NULL){
string called_func_name = CI.getCalledFunction()->getName();
if (called_func_name == OMP_PARALLEL_START_NAME && CI.arg_size() == 3) {
// change alloc to malloc_shared
// %5 = call i8* @_Z13malloc_sharedm(i64 20) ; <i8*> [#uses=5]
// %6 = bitcast i8* %5 to float* ; <float*> [#uses=2]
AllocaInst *AllocCall;
Value *arg_0 = CI.getArgument(0); // function
Value *arg_1 = CI.getArgument(1); // context
Value *loop_ub = NULL;
Function *function;
BitCastInst* BCI;
Function *kernel_function;
BasicBlock::iterator iI(*INSNI);
//BasicBlock::iterator iJ = iI+1;
iI++; iI++;
//BasicBlock::iterator iK = iI;
CallInst /**next,*/ *next_next;
if (arg_0 != NULL && arg_1 != NULL /*&& (next = dyn_cast<CallInst>(*iJ))*/
&& (next_next = dyn_cast<CallInst>(iI)) && (next_next->getCalledFunction() != NULL)
&& (next_next->getCalledFunction()->getName() == OMP_PARALLEL_END_NAME)
&& (BCI = dyn_cast<BitCastInst>(arg_1)) && (AllocCall = dyn_cast<AllocaInst>(BCI->getOperand(0)))
&& (function = dyn_cast<Function>(arg_0)) && (loop_ub = find_loop_upper_bound (AllocCall))
&& (kernel_function=convert_to_kernel_function (M, function))){
SmallVector<Value*, 16> Args;
Args.push_back(AllocCall->getArraySize());
Instruction *MallocCall = CallInst::Create(mallocFnTy, Args, "", AllocCall);
CastInst *MallocCast = CastInst::Create(Instruction::BitCast, MallocCall, AllocCall->getType(), "", AllocCall);
ValueMap[AllocCall] = MallocCast;
//AllocCall->replaceAllUsesWith(MallocCall);
// Add offload function
Args.clear();
Args.push_back(loop_ub);
Args.push_back(BCI);
Args.push_back(kernel_function);
if (offloadFnTy == NULL) {
init_offload_type(M, kernel_function);
}
Instruction *call = CallInst::Create(offloadFnTy, Args, "", INSNI);
if (find(toDelete.begin(), toDelete.end(), AllocCall) == toDelete.end()){
toDelete.push_back(AllocCall);
}
toDelete.push_back(&(*INSNI));
match = true;
}
}
else if (called_func_name == OMP_PARALLEL_END_NAME && CI.arg_size() == 0 && match) {
toDelete.push_back(&(*INSNI));
match = false;
}
else if (match) {
toDelete.push_back(&(*INSNI));
}
}
}
}
}
}
}
/* Replace AllocCalls by MallocCalls */
for(DenseMap<Value *, Value *>::iterator I = ValueMap.begin(), E = ValueMap.end(); I != E; I++) {
I->first->replaceAllUsesWith(I->second);
}
/* delete the instructions for get_omp_num_thread and get_omp_thread_num */
while(!toDelete.empty()) {
Instruction *g = toDelete.back();
toDelete.pop_back();
g->eraseFromParent();
}
}
bool PruneEH::runOnSCC(CallGraphSCC &SCC) {
SmallPtrSet<CallGraphNode *, 8> SCCNodes;
CallGraph &CG = getAnalysis<CallGraph>();
bool MadeChange = false;
// 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)
SCCNodes.insert(*I);
// First pass, scan all of the functions in the SCC, simplifying them
// according to what we know.
for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I)
if (Function *F = (*I)->getFunction())
MadeChange |= SimplifyFunction(F);
// Next, check to see if any callees might throw or if there are any external
// functions in this SCC: if so, we cannot prune any functions in this SCC.
// Definitions that are weak and not declared non-throwing might be
// overridden at linktime with something that throws, so assume that.
// If this SCC includes the unwind instruction, we KNOW it throws, so
// obviously the SCC might throw.
//
bool SCCMightUnwind = false, SCCMightReturn = false;
for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end();
(!SCCMightUnwind || !SCCMightReturn) && I != E; ++I) {
Function *F = (*I)->getFunction();
if (F == 0) {
SCCMightUnwind = true;
SCCMightReturn = true;
} else if (F->isDeclaration() || F->mayBeOverridden()) {
SCCMightUnwind |= !F->doesNotThrow();
SCCMightReturn |= !F->doesNotReturn();
} else {
bool CheckUnwind = !SCCMightUnwind && !F->doesNotThrow();
bool CheckReturn = !SCCMightReturn && !F->doesNotReturn();
if (!CheckUnwind && !CheckReturn)
continue;
// Check to see if this function performs an unwind or calls an
// unwinding function.
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
if (CheckUnwind && isa<UnwindInst>(BB->getTerminator())) {
// Uses unwind!
SCCMightUnwind = true;
} else if (CheckReturn && isa<ReturnInst>(BB->getTerminator())) {
SCCMightReturn = true;
}
// Invoke instructions don't allow unwinding to continue, so we are
// only interested in call instructions.
if (CheckUnwind && !SCCMightUnwind)
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
if (CallInst *CI = dyn_cast<CallInst>(I)) {
if (CI->doesNotThrow()) {
// This call cannot throw.
} else if (Function *Callee = CI->getCalledFunction()) {
CallGraphNode *CalleeNode = CG[Callee];
// If the callee is outside our current SCC then we may
// throw because it might.
if (!SCCNodes.count(CalleeNode)) {
SCCMightUnwind = true;
break;
}
} else {
// Indirect call, it might throw.
SCCMightUnwind = true;
break;
}
}
if (SCCMightUnwind && SCCMightReturn) break;
}
}
}
// If the SCC doesn't unwind or doesn't throw, note this fact.
if (!SCCMightUnwind || !SCCMightReturn)
for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I) {
Attributes NewAttributes = Attribute::None;
if (!SCCMightUnwind)
NewAttributes |= Attribute::NoUnwind;
if (!SCCMightReturn)
NewAttributes |= Attribute::NoReturn;
Function *F = (*I)->getFunction();
const AttrListPtr &PAL = F->getAttributes();
const AttrListPtr &NPAL = PAL.addAttr(~0, NewAttributes);
if (PAL != NPAL) {
MadeChange = true;
F->setAttributes(NPAL);
}
}
for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I) {
// Convert any invoke instructions to non-throwing functions in this node
// into call instructions with a branch. This makes the exception blocks
// dead.
if (Function *F = (*I)->getFunction())
MadeChange |= SimplifyFunction(F);
}
return MadeChange;
}
/// InputFilename is a LLVM bitcode file. Read it using bitcode reader.
/// Collect global functions and symbol names in symbols vector.
/// Collect external references in references vector.
/// Return LTO_READ_SUCCESS if there is no error.
enum LTOStatus
LTO::readLLVMObjectFile(const std::string &InputFilename,
NameToSymbolMap &symbols,
std::set<std::string> &references)
{
Module *m = getModule(InputFilename);
if (!m)
return LTO_READ_FAILURE;
// Collect Target info
getTarget(m);
if (!Target)
return LTO_READ_FAILURE;
// Use mangler to add GlobalPrefix to names to match linker names.
// FIXME : Instead of hard coding "-" use GlobalPrefix.
Mangler mangler(*m, Target->getTargetAsmInfo()->getGlobalPrefix());
modules.push_back(m);
for (Module::iterator f = m->begin(), e = m->end(); f != e; ++f) {
LTOLinkageTypes lt = getLTOLinkageType(f);
LTOVisibilityTypes vis = getLTOVisibilityType(f);
if (!f->isDeclaration() && lt != LTOInternalLinkage
&& strncmp (f->getName().c_str(), "llvm.", 5)) {
int alignment = ( 16 > f->getAlignment() ? 16 : f->getAlignment());
LLVMSymbol *newSymbol = new LLVMSymbol(lt, vis, f, f->getName(),
mangler.getValueName(f),
Log2_32(alignment));
symbols[newSymbol->getMangledName()] = newSymbol;
allSymbols[newSymbol->getMangledName()] = newSymbol;
}
// Collect external symbols referenced by this function.
for (Function::iterator b = f->begin(), fe = f->end(); b != fe; ++b)
for (BasicBlock::iterator i = b->begin(), be = b->end();
i != be; ++i) {
for (unsigned count = 0, total = i->getNumOperands();
count != total; ++count)
findExternalRefs(i->getOperand(count), references, mangler);
}
}
for (Module::global_iterator v = m->global_begin(), e = m->global_end();
v != e; ++v) {
LTOLinkageTypes lt = getLTOLinkageType(v);
LTOVisibilityTypes vis = getLTOVisibilityType(v);
if (!v->isDeclaration() && lt != LTOInternalLinkage
&& strncmp (v->getName().c_str(), "llvm.", 5)) {
const TargetData *TD = Target->getTargetData();
LLVMSymbol *newSymbol = new LLVMSymbol(lt, vis, v, v->getName(),
mangler.getValueName(v),
TD->getPreferredAlignmentLog(v));
symbols[newSymbol->getMangledName()] = newSymbol;
allSymbols[newSymbol->getMangledName()] = newSymbol;
for (unsigned count = 0, total = v->getNumOperands();
count != total; ++count)
findExternalRefs(v->getOperand(count), references, mangler);
}
}
return LTO_READ_SUCCESS;
}
bool LowerInvoke::insertExpensiveEHSupport(Function &F) {
SmallVector<ReturnInst*,16> Returns;
SmallVector<InvokeInst*,16> Invokes;
UnreachableInst* UnreachablePlaceholder = 0;
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB)
if (ReturnInst *RI = dyn_cast<ReturnInst>(BB->getTerminator())) {
// Remember all return instructions in case we insert an invoke into this
// function.
Returns.push_back(RI);
} else if (InvokeInst *II = dyn_cast<InvokeInst>(BB->getTerminator())) {
Invokes.push_back(II);
}
if (Invokes.empty()) return false;
NumInvokes += Invokes.size();
// TODO: This is not an optimal way to do this. In particular, this always
// inserts setjmp calls into the entries of functions with invoke instructions
// even though there are possibly paths through the function that do not
// execute any invokes. In particular, for functions with early exits, e.g.
// the 'addMove' method in hexxagon, it would be nice to not have to do the
// setjmp stuff on the early exit path. This requires a bit of dataflow, but
// would not be too hard to do.
// If we have an invoke instruction, insert a setjmp that dominates all
// invokes. After the setjmp, use a cond branch that goes to the original
// code path on zero, and to a designated 'catch' block of nonzero.
Value *OldJmpBufPtr = 0;
if (!Invokes.empty()) {
// First thing we need to do is scan the whole function for values that are
// live across unwind edges. Each value that is live across an unwind edge
// we spill into a stack location, guaranteeing that there is nothing live
// across the unwind edge. This process also splits all critical edges
// coming out of invoke's.
splitLiveRangesLiveAcrossInvokes(Invokes);
BasicBlock *EntryBB = F.begin();
// Create an alloca for the incoming jump buffer ptr and the new jump buffer
// that needs to be restored on all exits from the function. This is an
// alloca because the value needs to be live across invokes.
unsigned Align = TLI ? TLI->getJumpBufAlignment() : 0;
AllocaInst *JmpBuf =
new AllocaInst(JBLinkTy, 0, Align,
"jblink", F.begin()->begin());
Value *Idx[] = { Constant::getNullValue(Type::getInt32Ty(F.getContext())),
ConstantInt::get(Type::getInt32Ty(F.getContext()), 1) };
OldJmpBufPtr = GetElementPtrInst::Create(JmpBuf, Idx, "OldBuf",
EntryBB->getTerminator());
// Copy the JBListHead to the alloca.
Value *OldBuf = new LoadInst(JBListHead, "oldjmpbufptr", true,
EntryBB->getTerminator());
new StoreInst(OldBuf, OldJmpBufPtr, true, EntryBB->getTerminator());
// Add the new jumpbuf to the list.
new StoreInst(JmpBuf, JBListHead, true, EntryBB->getTerminator());
// Create the catch block. The catch block is basically a big switch
// statement that goes to all of the invoke catch blocks.
BasicBlock *CatchBB =
BasicBlock::Create(F.getContext(), "setjmp.catch", &F);
// Create an alloca which keeps track of the stack pointer before every
// invoke, this allows us to properly restore the stack pointer after
// long jumping.
AllocaInst *StackPtr = new AllocaInst(Type::getInt8PtrTy(F.getContext()), 0,
"stackptr", EntryBB->begin());
// Create an alloca which keeps track of which invoke is currently
// executing. For normal calls it contains zero.
AllocaInst *InvokeNum = new AllocaInst(Type::getInt32Ty(F.getContext()), 0,
"invokenum",EntryBB->begin());
new StoreInst(ConstantInt::get(Type::getInt32Ty(F.getContext()), 0),
InvokeNum, true, EntryBB->getTerminator());
// Insert a load in the Catch block, and a switch on its value. By default,
// we go to a block that just does an unwind (which is the correct action
// for a standard call). We insert an unreachable instruction here and
// modify the block to jump to the correct unwinding pad later.
BasicBlock *UnwindBB = BasicBlock::Create(F.getContext(), "unwindbb", &F);
UnreachablePlaceholder = new UnreachableInst(F.getContext(), UnwindBB);
Value *CatchLoad = new LoadInst(InvokeNum, "invoke.num", true, CatchBB);
SwitchInst *CatchSwitch =
SwitchInst::Create(CatchLoad, UnwindBB, Invokes.size(), CatchBB);
// Now that things are set up, insert the setjmp call itself.
// Split the entry block to insert the conditional branch for the setjmp.
BasicBlock *ContBlock = EntryBB->splitBasicBlock(EntryBB->getTerminator(),
"setjmp.cont");
Idx[1] = ConstantInt::get(Type::getInt32Ty(F.getContext()), 0);
Value *JmpBufPtr = GetElementPtrInst::Create(JmpBuf, Idx, "TheJmpBuf",
EntryBB->getTerminator());
JmpBufPtr = new BitCastInst(JmpBufPtr,
Type::getInt8PtrTy(F.getContext()),
"tmp", EntryBB->getTerminator());
Value *SJRet = CallInst::Create(SetJmpFn, JmpBufPtr, "sjret",
EntryBB->getTerminator());
// Compare the return value to zero.
Value *IsNormal = new ICmpInst(EntryBB->getTerminator(),
ICmpInst::ICMP_EQ, SJRet,
Constant::getNullValue(SJRet->getType()),
"notunwind");
// Nuke the uncond branch.
EntryBB->getTerminator()->eraseFromParent();
// Put in a new condbranch in its place.
BranchInst::Create(ContBlock, CatchBB, IsNormal, EntryBB);
// At this point, we are all set up, rewrite each invoke instruction.
for (unsigned i = 0, e = Invokes.size(); i != e; ++i)
rewriteExpensiveInvoke(Invokes[i], i+1, InvokeNum, StackPtr, CatchSwitch);
}
// We know that there is at least one unwind.
// Create three new blocks, the block to load the jmpbuf ptr and compare
// against null, the block to do the longjmp, and the error block for if it
// is null. Add them at the end of the function because they are not hot.
BasicBlock *UnwindHandler = BasicBlock::Create(F.getContext(),
"dounwind", &F);
BasicBlock *UnwindBlock = BasicBlock::Create(F.getContext(), "unwind", &F);
BasicBlock *TermBlock = BasicBlock::Create(F.getContext(), "unwinderror", &F);
// If this function contains an invoke, restore the old jumpbuf ptr.
Value *BufPtr;
if (OldJmpBufPtr) {
// Before the return, insert a copy from the saved value to the new value.
BufPtr = new LoadInst(OldJmpBufPtr, "oldjmpbufptr", UnwindHandler);
new StoreInst(BufPtr, JBListHead, UnwindHandler);
} else {
BufPtr = new LoadInst(JBListHead, "ehlist", UnwindHandler);
}
// Load the JBList, if it's null, then there was no catch!
Value *NotNull = new ICmpInst(*UnwindHandler, ICmpInst::ICMP_NE, BufPtr,
Constant::getNullValue(BufPtr->getType()),
"notnull");
BranchInst::Create(UnwindBlock, TermBlock, NotNull, UnwindHandler);
// Create the block to do the longjmp.
// Get a pointer to the jmpbuf and longjmp.
Value *Idx[] = { Constant::getNullValue(Type::getInt32Ty(F.getContext())),
ConstantInt::get(Type::getInt32Ty(F.getContext()), 0) };
Idx[0] = GetElementPtrInst::Create(BufPtr, Idx, "JmpBuf", UnwindBlock);
Idx[0] = new BitCastInst(Idx[0],
Type::getInt8PtrTy(F.getContext()),
"tmp", UnwindBlock);
Idx[1] = ConstantInt::get(Type::getInt32Ty(F.getContext()), 1);
CallInst::Create(LongJmpFn, Idx, "", UnwindBlock);
new UnreachableInst(F.getContext(), UnwindBlock);
// Set up the term block ("throw without a catch").
new UnreachableInst(F.getContext(), TermBlock);
// Insert a call to abort()
CallInst::Create(AbortFn, "",
TermBlock->getTerminator())->setTailCall();
// Replace the inserted unreachable with a branch to the unwind handler.
if (UnreachablePlaceholder) {
BranchInst::Create(UnwindHandler, UnreachablePlaceholder);
UnreachablePlaceholder->eraseFromParent();
}
// Finally, for any returns from this function, if this function contains an
// invoke, restore the old jmpbuf pointer to its input value.
if (OldJmpBufPtr) {
for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
ReturnInst *R = Returns[i];
// Before the return, insert a copy from the saved value to the new value.
Value *OldBuf = new LoadInst(OldJmpBufPtr, "oldjmpbufptr", true, R);
new StoreInst(OldBuf, JBListHead, true, R);
}
}
return true;
}
// UnifyAllExitNodes - Unify all exit nodes of the CFG by creating a new
// BasicBlock, and converting all returns to unconditional branches to this
// new basic block. The singular exit node is returned.
//
// If there are no return stmts in the Function, a null pointer is returned.
//
bool UnifyFunctionExitNodes::runOnFunction(Function &F) {
// Loop over all of the blocks in a function, tracking all of the blocks that
// return.
//
std::vector<BasicBlock*> ReturningBlocks;
std::vector<BasicBlock*> UnwindingBlocks;
std::vector<BasicBlock*> UnreachableBlocks;
for(Function::iterator I = F.begin(), E = F.end(); I != E; ++I)
if (isa<ReturnInst>(I->getTerminator()))
ReturningBlocks.push_back(I);
else if (isa<UnwindInst>(I->getTerminator()))
UnwindingBlocks.push_back(I);
else if (isa<UnreachableInst>(I->getTerminator()))
UnreachableBlocks.push_back(I);
// Handle unwinding blocks first.
if (UnwindingBlocks.empty()) {
UnwindBlock = 0;
} else if (UnwindingBlocks.size() == 1) {
UnwindBlock = UnwindingBlocks.front();
} else {
UnwindBlock = BasicBlock::Create("UnifiedUnwindBlock", &F);
new UnwindInst(UnwindBlock);
for (std::vector<BasicBlock*>::iterator I = UnwindingBlocks.begin(),
E = UnwindingBlocks.end(); I != E; ++I) {
BasicBlock *BB = *I;
BB->getInstList().pop_back(); // Remove the unwind insn
BranchInst::Create(UnwindBlock, BB);
}
}
// Then unreachable blocks.
if (UnreachableBlocks.empty()) {
UnreachableBlock = 0;
} else if (UnreachableBlocks.size() == 1) {
UnreachableBlock = UnreachableBlocks.front();
} else {
UnreachableBlock = BasicBlock::Create("UnifiedUnreachableBlock", &F);
new UnreachableInst(UnreachableBlock);
for (std::vector<BasicBlock*>::iterator I = UnreachableBlocks.begin(),
E = UnreachableBlocks.end(); I != E; ++I) {
BasicBlock *BB = *I;
BB->getInstList().pop_back(); // Remove the unreachable inst.
BranchInst::Create(UnreachableBlock, BB);
}
}
// Now handle return blocks.
if (ReturningBlocks.empty()) {
ReturnBlock = 0;
return false; // No blocks return
} else if (ReturningBlocks.size() == 1) {
ReturnBlock = ReturningBlocks.front(); // Already has a single return block
return false;
}
// Otherwise, we need to insert a new basic block into the function, add a PHI
// nodes (if the function returns values), and convert all of the return
// instructions into unconditional branches.
//
BasicBlock *NewRetBlock = BasicBlock::Create("UnifiedReturnBlock", &F);
SmallVector<Value *, 4> Phis;
unsigned NumRetVals = ReturningBlocks[0]->getTerminator()->getNumOperands();
if (NumRetVals == 0)
ReturnInst::Create(NULL, NewRetBlock);
else if (const StructType *STy = dyn_cast<StructType>(F.getReturnType())) {
Instruction *InsertPt = NULL;
if (NumRetVals == 0)
InsertPt = NewRetBlock->getFirstNonPHI();
PHINode *PN = NULL;
for (unsigned i = 0; i < NumRetVals; ++i) {
if (InsertPt)
PN = PHINode::Create(STy->getElementType(i), "UnifiedRetVal."
+ utostr(i), InsertPt);
else
PN = PHINode::Create(STy->getElementType(i), "UnifiedRetVal."
+ utostr(i), NewRetBlock);
Phis.push_back(PN);
InsertPt = PN;
}
ReturnInst::Create(&Phis[0], NumRetVals, NewRetBlock);
}
else {
// If the function doesn't return void... add a PHI node to the block...
PHINode *PN = PHINode::Create(F.getReturnType(), "UnifiedRetVal");
NewRetBlock->getInstList().push_back(PN);
Phis.push_back(PN);
ReturnInst::Create(PN, NewRetBlock);
}
// Loop over all of the blocks, replacing the return instruction with an
// unconditional branch.
//
for (std::vector<BasicBlock*>::iterator I = ReturningBlocks.begin(),
E = ReturningBlocks.end(); I != E; ++I) {
BasicBlock *BB = *I;
// Add an incoming element to the PHI node for every return instruction that
// is merging into this new block...
if (!Phis.empty()) {
for (unsigned i = 0; i < NumRetVals; ++i)
cast<PHINode>(Phis[i])->addIncoming(BB->getTerminator()->getOperand(i),
BB);
}
BB->getInstList().pop_back(); // Remove the return insn
BranchInst::Create(NewRetBlock, BB);
}
ReturnBlock = NewRetBlock;
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;
}
bool TailCallElim::runOnFunction(Function &F) {
// If this function is a varargs function, we won't be able to PHI the args
// right, so don't even try to convert it...
if (F.getFunctionType()->isVarArg()) return false;
BasicBlock *OldEntry = 0;
bool TailCallsAreMarkedTail = false;
SmallVector<PHINode*, 8> ArgumentPHIs;
bool MadeChange = false;
bool FunctionContainsEscapingAllocas = false;
// CannotTCETailMarkedCall - If true, we cannot perform TCE on tail calls
// marked with the 'tail' attribute, because doing so would cause the stack
// size to increase (real TCE would deallocate variable sized allocas, TCE
// doesn't).
bool CannotTCETailMarkedCall = false;
// Loop over the function, looking for any returning blocks, and keeping track
// of whether this function has any non-trivially used allocas.
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
if (FunctionContainsEscapingAllocas && CannotTCETailMarkedCall)
break;
FunctionContainsEscapingAllocas |=
CheckForEscapingAllocas(BB, CannotTCETailMarkedCall);
}
/// FIXME: The code generator produces really bad code when an 'escaping
/// alloca' is changed from being a static alloca to being a dynamic alloca.
/// Until this is resolved, disable this transformation if that would ever
/// happen. This bug is PR962.
if (FunctionContainsEscapingAllocas)
return false;
// Second pass, change any tail calls to loops.
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
if (ReturnInst *Ret = dyn_cast<ReturnInst>(BB->getTerminator())) {
bool Change = ProcessReturningBlock(Ret, OldEntry, TailCallsAreMarkedTail,
ArgumentPHIs,CannotTCETailMarkedCall);
if (!Change && BB->getFirstNonPHIOrDbg() == Ret)
Change = FoldReturnAndProcessPred(BB, Ret, OldEntry,
TailCallsAreMarkedTail, ArgumentPHIs,
CannotTCETailMarkedCall);
MadeChange |= Change;
}
}
// If we eliminated any tail recursions, it's possible that we inserted some
// silly PHI nodes which just merge an initial value (the incoming operand)
// with themselves. Check to see if we did and clean up our mess if so. This
// occurs when a function passes an argument straight through to its tail
// call.
if (!ArgumentPHIs.empty()) {
for (unsigned i = 0, e = ArgumentPHIs.size(); i != e; ++i) {
PHINode *PN = ArgumentPHIs[i];
// If the PHI Node is a dynamic constant, replace it with the value it is.
if (Value *PNV = SimplifyInstruction(PN)) {
PN->replaceAllUsesWith(PNV);
PN->eraseFromParent();
}
}
}
// Finally, if this function contains no non-escaping allocas, or calls
// setjmp, mark all calls in the function as eligible for tail calls
//(there is no stack memory for them to access).
if (!FunctionContainsEscapingAllocas && !F.callsFunctionThatReturnsTwice())
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB)
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
if (CallInst *CI = dyn_cast<CallInst>(I)) {
CI->setTailCall();
MadeChange = true;
}
return MadeChange;
}
bool llvm::InputValues::runOnModule(Module& M) {
module = &M;
initializeWhiteList();
collectMainArguments();
int one=0;
errs()<<"\n------------------------------------------------\n";
for(Module::iterator Fit = M.begin(), Fend = M.end(); Fit != Fend; Fit++){
for (Function::iterator BBit = Fit->begin(), BBend = Fit->end(); BBit != BBend; BBit++) {
for (BasicBlock::iterator Iit = BBit->begin(), Iend = BBit->end(); Iit != Iend; Iit++) {
if(one<2)
{
Value* V_in;
//FIXME: Temporary assignment of the tainted source for testing...
if(BBit->getName()=="BB_146")
{
// errs() << "Inside BB_0";
if(LoadInst *LI = dyn_cast<LoadInst>(Iit))
{
// LI->dump();
V_in = LI->getPointerOperand();
errs() <<"\ntaint source " << LI->getName() << " ";
errs() << V_in;
insertInInputDepValues(V_in);
//NumInputValues++;
one++;
}
}
}
if (CallInst *CI = dyn_cast<CallInst>(Iit)) {
if (isMarkedCallInst(CI)){
//Values returned by marked instructions
insertInInputDepValues(CI);
for(unsigned int i = 0; i < CI->getNumOperands(); i++){
if (CI->getOperand(i)->getType()->isPointerTy()){
//Arguments with pointer type of marked functions
// insertInInputDepValues(CI->getOperand(i));
}
}
}
}
}
}
}
errs()<<"\n------------------------------------------------\n";
NumInputValues = inputValues.size();
//We don't modify anything, so we must return false;
return false;
}
/// splitLiveRangesAcrossInvokes - Each value that is live across an unwind edge
/// we spill into a stack location, guaranteeing that there is nothing live
/// across the unwind edge. This process also splits all critical edges
/// coming out of invoke's.
void SjLjEHPass::
splitLiveRangesLiveAcrossInvokes(SmallVector<InvokeInst*,16> &Invokes) {
// First step, split all critical edges from invoke instructions.
for (unsigned i = 0, e = Invokes.size(); i != e; ++i) {
InvokeInst *II = Invokes[i];
SplitCriticalEdge(II, 0, this);
SplitCriticalEdge(II, 1, this);
assert(!isa<PHINode>(II->getNormalDest()) &&
!isa<PHINode>(II->getUnwindDest()) &&
"critical edge splitting left single entry phi nodes?");
}
Function *F = Invokes.back()->getParent()->getParent();
// To avoid having to handle incoming arguments specially, we lower each arg
// to a copy instruction in the entry block. This ensures that the argument
// value itself cannot be live across the entry block.
BasicBlock::iterator AfterAllocaInsertPt = F->begin()->begin();
while (isa<AllocaInst>(AfterAllocaInsertPt) &&
isa<ConstantInt>(cast<AllocaInst>(AfterAllocaInsertPt)->getArraySize()))
++AfterAllocaInsertPt;
for (Function::arg_iterator AI = F->arg_begin(), E = F->arg_end();
AI != E; ++AI) {
// This is always a no-op cast because we're casting AI to AI->getType() so
// src and destination types are identical. BitCast is the only possibility.
CastInst *NC = new BitCastInst(
AI, AI->getType(), AI->getName()+".tmp", AfterAllocaInsertPt);
AI->replaceAllUsesWith(NC);
// Normally its is forbidden to replace a CastInst's operand because it
// could cause the opcode to reflect an illegal conversion. However, we're
// replacing it here with the same value it was constructed with to simply
// make NC its user.
NC->setOperand(0, AI);
}
// Finally, scan the code looking for instructions with bad live ranges.
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
// Ignore obvious cases we don't have to handle. In particular, most
// instructions either have no uses or only have a single use inside the
// current block. Ignore them quickly.
Instruction *Inst = II;
if (Inst->use_empty()) continue;
if (Inst->hasOneUse() &&
cast<Instruction>(Inst->use_back())->getParent() == BB &&
!isa<PHINode>(Inst->use_back())) continue;
// If this is an alloca in the entry block, it's not a real register
// value.
if (AllocaInst *AI = dyn_cast<AllocaInst>(Inst))
if (isa<ConstantInt>(AI->getArraySize()) && BB == F->begin())
continue;
// Avoid iterator invalidation by copying users to a temporary vector.
SmallVector<Instruction*,16> Users;
for (Value::use_iterator UI = Inst->use_begin(), E = Inst->use_end();
UI != E; ++UI) {
Instruction *User = cast<Instruction>(*UI);
if (User->getParent() != BB || isa<PHINode>(User))
Users.push_back(User);
}
// Find all of the blocks that this value is live in.
std::set<BasicBlock*> LiveBBs;
LiveBBs.insert(Inst->getParent());
while (!Users.empty()) {
Instruction *U = Users.back();
Users.pop_back();
if (!isa<PHINode>(U)) {
MarkBlocksLiveIn(U->getParent(), LiveBBs);
} else {
// Uses for a PHI node occur in their predecessor block.
PHINode *PN = cast<PHINode>(U);
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
if (PN->getIncomingValue(i) == Inst)
MarkBlocksLiveIn(PN->getIncomingBlock(i), LiveBBs);
}
}
// Now that we know all of the blocks that this thing is live in, see if
// it includes any of the unwind locations.
bool NeedsSpill = false;
for (unsigned i = 0, e = Invokes.size(); i != e; ++i) {
BasicBlock *UnwindBlock = Invokes[i]->getUnwindDest();
if (UnwindBlock != BB && LiveBBs.count(UnwindBlock)) {
NeedsSpill = true;
}
}
// If we decided we need a spill, do it.
if (NeedsSpill) {
++NumSpilled;
DemoteRegToStack(*Inst, true);
}
}
}
// RemoveDeadStuffFromFunction - Remove any arguments and return values from F
// that are not in LiveValues. Transform the function and all of the callees of
// the function to not have these arguments and return values.
//
bool DAE::RemoveDeadStuffFromFunction(Function *F) {
// Don't modify fully live functions
if (LiveFunctions.count(F))
return false;
// Start by computing a new prototype for the function, which is the same as
// the old function, but has fewer arguments and a different return type.
FunctionType *FTy = F->getFunctionType();
std::vector<Type*> Params;
// Keep track of if we have a live 'returned' argument
bool HasLiveReturnedArg = false;
// Set up to build a new list of parameter attributes.
SmallVector<AttributeSet, 8> AttributesVec;
const AttributeSet &PAL = F->getAttributes();
// Remember which arguments are still alive.
SmallVector<bool, 10> ArgAlive(FTy->getNumParams(), false);
// Construct the new parameter list from non-dead arguments. Also construct
// a new set of parameter attributes to correspond. Skip the first parameter
// attribute, since that belongs to the return value.
unsigned i = 0;
for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end();
I != E; ++I, ++i) {
RetOrArg Arg = CreateArg(F, i);
if (LiveValues.erase(Arg)) {
Params.push_back(I->getType());
ArgAlive[i] = true;
// Get the original parameter attributes (skipping the first one, that is
// for the return value.
if (PAL.hasAttributes(i + 1)) {
AttrBuilder B(PAL, i + 1);
if (B.contains(Attribute::Returned))
HasLiveReturnedArg = true;
AttributesVec.
push_back(AttributeSet::get(F->getContext(), Params.size(), B));
}
} else {
++NumArgumentsEliminated;
DEBUG(dbgs() << "DAE - Removing argument " << i << " (" << I->getName()
<< ") from " << F->getName() << "\n");
}
}
// Find out the new return value.
Type *RetTy = FTy->getReturnType();
Type *NRetTy = NULL;
unsigned RetCount = NumRetVals(F);
// -1 means unused, other numbers are the new index
SmallVector<int, 5> NewRetIdxs(RetCount, -1);
std::vector<Type*> RetTypes;
// If there is a function with a live 'returned' argument but a dead return
// value, then there are two possible actions:
// 1) Eliminate the return value and take off the 'returned' attribute on the
// argument.
// 2) Retain the 'returned' attribute and treat the return value (but not the
// entire function) as live so that it is not eliminated.
//
// It's not clear in the general case which option is more profitable because,
// even in the absence of explicit uses of the return value, code generation
// is free to use the 'returned' attribute to do things like eliding
// save/restores of registers across calls. Whether or not this happens is
// target and ABI-specific as well as depending on the amount of register
// pressure, so there's no good way for an IR-level pass to figure this out.
//
// Fortunately, the only places where 'returned' is currently generated by
// the FE are places where 'returned' is basically free and almost always a
// performance win, so the second option can just be used always for now.
//
// This should be revisited if 'returned' is ever applied more liberally.
if (RetTy->isVoidTy() || HasLiveReturnedArg) {
NRetTy = RetTy;
} else {
StructType *STy = dyn_cast<StructType>(RetTy);
if (STy)
// Look at each of the original return values individually.
for (unsigned i = 0; i != RetCount; ++i) {
RetOrArg Ret = CreateRet(F, i);
if (LiveValues.erase(Ret)) {
RetTypes.push_back(STy->getElementType(i));
NewRetIdxs[i] = RetTypes.size() - 1;
} else {
++NumRetValsEliminated;
DEBUG(dbgs() << "DAE - Removing return value " << i << " from "
<< F->getName() << "\n");
}
}
else
// We used to return a single value.
if (LiveValues.erase(CreateRet(F, 0))) {
RetTypes.push_back(RetTy);
NewRetIdxs[0] = 0;
} else {
DEBUG(dbgs() << "DAE - Removing return value from " << F->getName()
<< "\n");
++NumRetValsEliminated;
}
if (RetTypes.size() > 1)
// More than one return type? Return a struct with them. Also, if we used
// to return a struct and didn't change the number of return values,
// return a struct again. This prevents changing {something} into
// something and {} into void.
// Make the new struct packed if we used to return a packed struct
// already.
NRetTy = StructType::get(STy->getContext(), RetTypes, STy->isPacked());
else if (RetTypes.size() == 1)
// One return type? Just a simple value then, but only if we didn't use to
// return a struct with that simple value before.
NRetTy = RetTypes.front();
else if (RetTypes.size() == 0)
// No return types? Make it void, but only if we didn't use to return {}.
NRetTy = Type::getVoidTy(F->getContext());
}
assert(NRetTy && "No new return type found?");
// The existing function return attributes.
AttributeSet RAttrs = PAL.getRetAttributes();
// Remove any incompatible attributes, but only if we removed all return
// values. Otherwise, ensure that we don't have any conflicting attributes
// here. Currently, this should not be possible, but special handling might be
// required when new return value attributes are added.
if (NRetTy->isVoidTy())
RAttrs =
AttributeSet::get(NRetTy->getContext(), AttributeSet::ReturnIndex,
AttrBuilder(RAttrs, AttributeSet::ReturnIndex).
removeAttributes(AttributeFuncs::
typeIncompatible(NRetTy, AttributeSet::ReturnIndex),
AttributeSet::ReturnIndex));
else
assert(!AttrBuilder(RAttrs, AttributeSet::ReturnIndex).
hasAttributes(AttributeFuncs::
typeIncompatible(NRetTy, AttributeSet::ReturnIndex),
AttributeSet::ReturnIndex) &&
"Return attributes no longer compatible?");
if (RAttrs.hasAttributes(AttributeSet::ReturnIndex))
AttributesVec.push_back(AttributeSet::get(NRetTy->getContext(), RAttrs));
if (PAL.hasAttributes(AttributeSet::FunctionIndex))
AttributesVec.push_back(AttributeSet::get(F->getContext(),
PAL.getFnAttributes()));
// Reconstruct the AttributesList based on the vector we constructed.
AttributeSet NewPAL = AttributeSet::get(F->getContext(), AttributesVec);
// Create the new function type based on the recomputed parameters.
FunctionType *NFTy = FunctionType::get(NRetTy, Params, FTy->isVarArg());
// No change?
if (NFTy == FTy)
return false;
// Create the new function body and insert it into the module...
Function *NF = Function::Create(NFTy, F->getLinkage());
NF->copyAttributesFrom(F);
NF->setAttributes(NewPAL);
// Insert the new function before the old function, so we won't be processing
// it again.
F->getParent()->getFunctionList().insert(F, NF);
NF->takeName(F);
// Loop over all of the callers of the function, transforming the call sites
// to pass in a smaller number of arguments into the new function.
//
std::vector<Value*> Args;
while (!F->use_empty()) {
CallSite CS(F->use_back());
Instruction *Call = CS.getInstruction();
AttributesVec.clear();
const AttributeSet &CallPAL = CS.getAttributes();
// The call return attributes.
AttributeSet RAttrs = CallPAL.getRetAttributes();
// Adjust in case the function was changed to return void.
RAttrs =
AttributeSet::get(NF->getContext(), AttributeSet::ReturnIndex,
AttrBuilder(RAttrs, AttributeSet::ReturnIndex).
removeAttributes(AttributeFuncs::
typeIncompatible(NF->getReturnType(),
AttributeSet::ReturnIndex),
AttributeSet::ReturnIndex));
if (RAttrs.hasAttributes(AttributeSet::ReturnIndex))
AttributesVec.push_back(AttributeSet::get(NF->getContext(), RAttrs));
// Declare these outside of the loops, so we can reuse them for the second
// loop, which loops the varargs.
CallSite::arg_iterator I = CS.arg_begin();
unsigned i = 0;
// Loop over those operands, corresponding to the normal arguments to the
// original function, and add those that are still alive.
for (unsigned e = FTy->getNumParams(); i != e; ++I, ++i)
if (ArgAlive[i]) {
Args.push_back(*I);
// Get original parameter attributes, but skip return attributes.
if (CallPAL.hasAttributes(i + 1)) {
AttrBuilder B(CallPAL, i + 1);
// If the return type has changed, then get rid of 'returned' on the
// call site. The alternative is to make all 'returned' attributes on
// call sites keep the return value alive just like 'returned'
// attributes on function declaration but it's less clearly a win
// and this is not an expected case anyway
if (NRetTy != RetTy && B.contains(Attribute::Returned))
B.removeAttribute(Attribute::Returned);
AttributesVec.
push_back(AttributeSet::get(F->getContext(), Args.size(), B));
}
}
// Push any varargs arguments on the list. Don't forget their attributes.
for (CallSite::arg_iterator E = CS.arg_end(); I != E; ++I, ++i) {
Args.push_back(*I);
if (CallPAL.hasAttributes(i + 1)) {
AttrBuilder B(CallPAL, i + 1);
AttributesVec.
push_back(AttributeSet::get(F->getContext(), Args.size(), B));
}
}
if (CallPAL.hasAttributes(AttributeSet::FunctionIndex))
AttributesVec.push_back(AttributeSet::get(Call->getContext(),
CallPAL.getFnAttributes()));
// Reconstruct the AttributesList based on the vector we constructed.
AttributeSet NewCallPAL = AttributeSet::get(F->getContext(), AttributesVec);
Instruction *New;
if (InvokeInst *II = dyn_cast<InvokeInst>(Call)) {
New = InvokeInst::Create(NF, II->getNormalDest(), II->getUnwindDest(),
Args, "", Call);
cast<InvokeInst>(New)->setCallingConv(CS.getCallingConv());
cast<InvokeInst>(New)->setAttributes(NewCallPAL);
} else {
New = CallInst::Create(NF, Args, "", Call);
cast<CallInst>(New)->setCallingConv(CS.getCallingConv());
cast<CallInst>(New)->setAttributes(NewCallPAL);
if (cast<CallInst>(Call)->isTailCall())
cast<CallInst>(New)->setTailCall();
}
New->setDebugLoc(Call->getDebugLoc());
Args.clear();
if (!Call->use_empty()) {
if (New->getType() == Call->getType()) {
// Return type not changed? Just replace users then.
Call->replaceAllUsesWith(New);
New->takeName(Call);
} else if (New->getType()->isVoidTy()) {
// Our return value has uses, but they will get removed later on.
// Replace by null for now.
if (!Call->getType()->isX86_MMXTy())
Call->replaceAllUsesWith(Constant::getNullValue(Call->getType()));
} else {
assert(RetTy->isStructTy() &&
"Return type changed, but not into a void. The old return type"
" must have been a struct!");
Instruction *InsertPt = Call;
if (InvokeInst *II = dyn_cast<InvokeInst>(Call)) {
BasicBlock::iterator IP = II->getNormalDest()->begin();
while (isa<PHINode>(IP)) ++IP;
InsertPt = IP;
}
// We used to return a struct. Instead of doing smart stuff with all the
// uses of this struct, we will just rebuild it using
// extract/insertvalue chaining and let instcombine clean that up.
//
// Start out building up our return value from undef
Value *RetVal = UndefValue::get(RetTy);
for (unsigned i = 0; i != RetCount; ++i)
if (NewRetIdxs[i] != -1) {
Value *V;
if (RetTypes.size() > 1)
// We are still returning a struct, so extract the value from our
// return value
V = ExtractValueInst::Create(New, NewRetIdxs[i], "newret",
InsertPt);
else
// We are now returning a single element, so just insert that
V = New;
// Insert the value at the old position
RetVal = InsertValueInst::Create(RetVal, V, i, "oldret", InsertPt);
}
// Now, replace all uses of the old call instruction with the return
// struct we built
Call->replaceAllUsesWith(RetVal);
New->takeName(Call);
}
}
// Finally, remove the old call from the program, reducing the use-count of
// F.
Call->eraseFromParent();
}
// Since we have now created the new function, splice the body of the old
// function right into the new function, leaving the old rotting hulk of the
// function empty.
NF->getBasicBlockList().splice(NF->begin(), F->getBasicBlockList());
// Loop over the argument list, transferring uses of the old arguments over to
// the new arguments, also transferring over the names as well.
i = 0;
for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(),
I2 = NF->arg_begin(); I != E; ++I, ++i)
if (ArgAlive[i]) {
// If this is a live argument, move the name and users over to the new
// version.
I->replaceAllUsesWith(I2);
I2->takeName(I);
++I2;
} else {
// If this argument is dead, replace any uses of it with null constants
// (these are guaranteed to become unused later on).
if (!I->getType()->isX86_MMXTy())
I->replaceAllUsesWith(Constant::getNullValue(I->getType()));
}
// If we change the return value of the function we must rewrite any return
// instructions. Check this now.
if (F->getReturnType() != NF->getReturnType())
for (Function::iterator BB = NF->begin(), E = NF->end(); BB != E; ++BB)
if (ReturnInst *RI = dyn_cast<ReturnInst>(BB->getTerminator())) {
Value *RetVal;
if (NFTy->getReturnType()->isVoidTy()) {
RetVal = 0;
} else {
assert (RetTy->isStructTy());
// The original return value was a struct, insert
// extractvalue/insertvalue chains to extract only the values we need
// to return and insert them into our new result.
// This does generate messy code, but we'll let it to instcombine to
// clean that up.
Value *OldRet = RI->getOperand(0);
// Start out building up our return value from undef
RetVal = UndefValue::get(NRetTy);
for (unsigned i = 0; i != RetCount; ++i)
if (NewRetIdxs[i] != -1) {
ExtractValueInst *EV = ExtractValueInst::Create(OldRet, i,
"oldret", RI);
if (RetTypes.size() > 1) {
// We're still returning a struct, so reinsert the value into
// our new return value at the new index
RetVal = InsertValueInst::Create(RetVal, EV, NewRetIdxs[i],
"newret", RI);
} else {
// We are now only returning a simple value, so just return the
// extracted value.
RetVal = EV;
}
}
}
// Replace the return instruction with one returning the new return
// value (possibly 0 if we became void).
ReturnInst::Create(F->getContext(), RetVal, RI);
BB->getInstList().erase(RI);
}
// Patch the pointer to LLVM function in debug info descriptor.
FunctionDIMap::iterator DI = FunctionDIs.find(F);
if (DI != FunctionDIs.end())
DI->second.replaceFunction(NF);
// Now that the old function is dead, delete it.
F->eraseFromParent();
return true;
}
/// DeleteDeadVarargs - If this is an function that takes a ... list, and if
/// llvm.vastart is never called, the varargs list is dead for the function.
bool DAE::DeleteDeadVarargs(Function &Fn) {
assert(Fn.getFunctionType()->isVarArg() && "Function isn't varargs!");
if (Fn.isDeclaration() || !Fn.hasLocalLinkage()) return false;
// Ensure that the function is only directly called.
if (Fn.hasAddressTaken())
return false;
// Okay, we know we can transform this function if safe. Scan its body
// looking for calls to llvm.vastart.
for (Function::iterator BB = Fn.begin(), E = Fn.end(); BB != E; ++BB) {
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
if (II->getIntrinsicID() == Intrinsic::vastart)
return false;
}
}
}
// If we get here, there are no calls to llvm.vastart in the function body,
// remove the "..." and adjust all the calls.
// Start by computing a new prototype for the function, which is the same as
// the old function, but doesn't have isVarArg set.
FunctionType *FTy = Fn.getFunctionType();
std::vector<Type*> Params(FTy->param_begin(), FTy->param_end());
FunctionType *NFTy = FunctionType::get(FTy->getReturnType(),
Params, false);
unsigned NumArgs = Params.size();
// Create the new function body and insert it into the module...
Function *NF = Function::Create(NFTy, Fn.getLinkage());
NF->copyAttributesFrom(&Fn);
Fn.getParent()->getFunctionList().insert(&Fn, NF);
NF->takeName(&Fn);
// Loop over all of the callers of the function, transforming the call sites
// to pass in a smaller number of arguments into the new function.
//
std::vector<Value*> Args;
for (Value::use_iterator I = Fn.use_begin(), E = Fn.use_end(); I != E; ) {
CallSite CS(*I++);
if (!CS)
continue;
Instruction *Call = CS.getInstruction();
// Pass all the same arguments.
Args.assign(CS.arg_begin(), CS.arg_begin() + NumArgs);
// Drop any attributes that were on the vararg arguments.
AttributeSet PAL = CS.getAttributes();
if (!PAL.isEmpty() && PAL.getSlotIndex(PAL.getNumSlots() - 1) > NumArgs) {
SmallVector<AttributeSet, 8> AttributesVec;
for (unsigned i = 0; PAL.getSlotIndex(i) <= NumArgs; ++i)
AttributesVec.push_back(PAL.getSlotAttributes(i));
if (PAL.hasAttributes(AttributeSet::FunctionIndex))
AttributesVec.push_back(AttributeSet::get(Fn.getContext(),
PAL.getFnAttributes()));
PAL = AttributeSet::get(Fn.getContext(), AttributesVec);
}
Instruction *New;
if (InvokeInst *II = dyn_cast<InvokeInst>(Call)) {
New = InvokeInst::Create(NF, II->getNormalDest(), II->getUnwindDest(),
Args, "", Call);
cast<InvokeInst>(New)->setCallingConv(CS.getCallingConv());
cast<InvokeInst>(New)->setAttributes(PAL);
} else {
New = CallInst::Create(NF, Args, "", Call);
cast<CallInst>(New)->setCallingConv(CS.getCallingConv());
cast<CallInst>(New)->setAttributes(PAL);
if (cast<CallInst>(Call)->isTailCall())
cast<CallInst>(New)->setTailCall();
}
New->setDebugLoc(Call->getDebugLoc());
Args.clear();
if (!Call->use_empty())
Call->replaceAllUsesWith(New);
New->takeName(Call);
// Finally, remove the old call from the program, reducing the use-count of
// F.
Call->eraseFromParent();
}
// Since we have now created the new function, splice the body of the old
// function right into the new function, leaving the old rotting hulk of the
// function empty.
NF->getBasicBlockList().splice(NF->begin(), Fn.getBasicBlockList());
// Loop over the argument list, transferring uses of the old arguments over to
// the new arguments, also transferring over the names as well. While we're at
// it, remove the dead arguments from the DeadArguments list.
//
for (Function::arg_iterator I = Fn.arg_begin(), E = Fn.arg_end(),
I2 = NF->arg_begin(); I != E; ++I, ++I2) {
// Move the name and users over to the new version.
I->replaceAllUsesWith(I2);
I2->takeName(I);
}
// Patch the pointer to LLVM function in debug info descriptor.
FunctionDIMap::iterator DI = FunctionDIs.find(&Fn);
if (DI != FunctionDIs.end())
DI->second.replaceFunction(NF);
// Fix up any BlockAddresses that refer to the function.
Fn.replaceAllUsesWith(ConstantExpr::getBitCast(NF, Fn.getType()));
// Delete the bitcast that we just created, so that NF does not
// appear to be address-taken.
NF->removeDeadConstantUsers();
// Finally, nuke the old function.
Fn.eraseFromParent();
return true;
}
bool GCOVProfiler::emitProfileArcs() {
NamedMDNode *CU_Nodes = M->getNamedMetadata("llvm.dbg.cu");
if (!CU_Nodes) return false;
bool Result = false;
bool InsertIndCounterIncrCode = false;
for (unsigned i = 0, e = CU_Nodes->getNumOperands(); i != e; ++i) {
DICompileUnit CU(CU_Nodes->getOperand(i));
DIArray SPs = CU.getSubprograms();
SmallVector<std::pair<GlobalVariable *, MDNode *>, 8> CountersBySP;
for (unsigned i = 0, e = SPs.getNumElements(); i != e; ++i) {
DISubprogram SP(SPs.getElement(i));
if (!SP.Verify()) continue;
Function *F = SP.getFunction();
if (!F) continue;
if (!Result) Result = true;
unsigned Edges = 0;
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
TerminatorInst *TI = BB->getTerminator();
if (isa<ReturnInst>(TI))
++Edges;
else
Edges += TI->getNumSuccessors();
}
ArrayType *CounterTy =
ArrayType::get(Type::getInt64Ty(*Ctx), Edges);
GlobalVariable *Counters =
new GlobalVariable(*M, CounterTy, false,
GlobalValue::InternalLinkage,
Constant::getNullValue(CounterTy),
"__llvm_gcov_ctr");
CountersBySP.push_back(std::make_pair(Counters, (MDNode*)SP));
UniqueVector<BasicBlock *> ComplexEdgePreds;
UniqueVector<BasicBlock *> ComplexEdgeSuccs;
unsigned Edge = 0;
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
TerminatorInst *TI = BB->getTerminator();
int Successors = isa<ReturnInst>(TI) ? 1 : TI->getNumSuccessors();
if (Successors) {
IRBuilder<> Builder(TI);
if (Successors == 1) {
Value *Counter = Builder.CreateConstInBoundsGEP2_64(Counters, 0,
Edge);
Value *Count = Builder.CreateLoad(Counter);
Count = Builder.CreateAdd(Count, Builder.getInt64(1));
Builder.CreateStore(Count, Counter);
} else if (BranchInst *BI = dyn_cast<BranchInst>(TI)) {
Value *Sel = Builder.CreateSelect(BI->getCondition(),
Builder.getInt64(Edge),
Builder.getInt64(Edge + 1));
SmallVector<Value *, 2> Idx;
Idx.push_back(Builder.getInt64(0));
Idx.push_back(Sel);
Value *Counter = Builder.CreateInBoundsGEP(Counters, Idx);
Value *Count = Builder.CreateLoad(Counter);
Count = Builder.CreateAdd(Count, Builder.getInt64(1));
Builder.CreateStore(Count, Counter);
} else {
ComplexEdgePreds.insert(BB);
for (int i = 0; i != Successors; ++i)
ComplexEdgeSuccs.insert(TI->getSuccessor(i));
}
Edge += Successors;
}
}
if (!ComplexEdgePreds.empty()) {
GlobalVariable *EdgeTable =
buildEdgeLookupTable(F, Counters,
ComplexEdgePreds, ComplexEdgeSuccs);
GlobalVariable *EdgeState = getEdgeStateValue();
for (int i = 0, e = ComplexEdgePreds.size(); i != e; ++i) {
IRBuilder<> Builder(ComplexEdgePreds[i+1]->getTerminator());
Builder.CreateStore(Builder.getInt32(i), EdgeState);
}
for (int i = 0, e = ComplexEdgeSuccs.size(); i != e; ++i) {
// call runtime to perform increment
BasicBlock::iterator InsertPt =
ComplexEdgeSuccs[i+1]->getFirstInsertionPt();
IRBuilder<> Builder(InsertPt);
Value *CounterPtrArray =
Builder.CreateConstInBoundsGEP2_64(EdgeTable, 0,
i * ComplexEdgePreds.size());
// Build code to increment the counter.
InsertIndCounterIncrCode = true;
Builder.CreateCall2(getIncrementIndirectCounterFunc(),
EdgeState, CounterPtrArray);
}
}
}
Function *WriteoutF = insertCounterWriteout(CountersBySP);
Function *FlushF = insertFlush(CountersBySP);
// Create a small bit of code that registers the "__llvm_gcov_writeout" to
// be executed at exit and the "__llvm_gcov_flush" function to be executed
// when "__gcov_flush" is called.
FunctionType *FTy = FunctionType::get(Type::getVoidTy(*Ctx), false);
Function *F = Function::Create(FTy, GlobalValue::InternalLinkage,
"__llvm_gcov_init", M);
F->setUnnamedAddr(true);
F->setLinkage(GlobalValue::InternalLinkage);
F->addFnAttr(Attribute::NoInline);
if (Options.NoRedZone)
F->addFnAttr(Attribute::NoRedZone);
BasicBlock *BB = BasicBlock::Create(*Ctx, "entry", F);
IRBuilder<> Builder(BB);
FTy = FunctionType::get(Type::getVoidTy(*Ctx), false);
Type *Params[] = {
PointerType::get(FTy, 0),
PointerType::get(FTy, 0)
};
FTy = FunctionType::get(Builder.getVoidTy(), Params, false);
// Inialize the environment and register the local writeout and flush
// functions.
Constant *GCOVInit = M->getOrInsertFunction("llvm_gcov_init", FTy);
Builder.CreateCall2(GCOVInit, WriteoutF, FlushF);
Builder.CreateRetVoid();
appendToGlobalCtors(*M, F, 0);
}
if (InsertIndCounterIncrCode)
insertIndirectCounterIncrement();
return Result;
}
//
// Method: runOnFunction()
//
// Description:
// Entry point for this LLVM pass.
//
// Return value:
// true - The function was modified.
// false - The function was not modified.
//
bool
BreakConstantGEPs::runOnFunction (Function & F) {
if (!pocl::Workgroup::isKernelToProcess(F)) return false;
bool modified = false;
// Worklist of values to check for constant GEP expressions
std::vector<Instruction *> Worklist;
//
// Initialize the worklist by finding all instructions that have one or more
// operands containing a constant GEP expression.
//
for (Function::iterator BB = F.begin(); BB != F.end(); ++BB) {
for (BasicBlock::iterator i = BB->begin(); i != BB->end(); ++i) {
//
// Scan through the operands of this instruction. If it is a constant
// expression GEP, insert an instruction GEP before the instruction.
//
Instruction * I = &*i;
for (unsigned index = 0; index < I->getNumOperands(); ++index) {
if (hasConstantGEP (I->getOperand(index))) {
Worklist.push_back (I);
}
}
}
}
//
// Determine whether we will modify anything.
//
if (Worklist.size()) modified = true;
//
// While the worklist is not empty, take an item from it, convert the
// operands into instructions if necessary, and determine if the newly
// added instructions need to be processed as well.
//
while (Worklist.size()) {
Instruction * I = Worklist.back();
Worklist.pop_back();
//
// Scan through the operands of this instruction and convert each into an
// instruction. Note that this works a little differently for phi
// instructions because the new instruction must be added to the
// appropriate predecessor block.
//
if (PHINode * PHI = dyn_cast<PHINode>(I)) {
for (unsigned index = 0; index < PHI->getNumIncomingValues(); ++index) {
//
// For PHI Nodes, if an operand is a constant expression with a GEP, we
// want to insert the new instructions in the predecessor basic block.
//
// Note: It seems that it's possible for a phi to have the same
// incoming basic block listed multiple times; this seems okay as long
// the same value is listed for the incoming block.
//
Instruction * InsertPt = PHI->getIncomingBlock(index)->getTerminator();
if (ConstantExpr * CE = hasConstantGEP (PHI->getIncomingValue(index))) {
Instruction * NewInst = convertExpression (CE, InsertPt);
for (unsigned i2 = index; i2 < PHI->getNumIncomingValues(); ++i2) {
if ((PHI->getIncomingBlock (i2)) == PHI->getIncomingBlock (index))
PHI->setIncomingValue (i2, NewInst);
}
Worklist.push_back (NewInst);
}
}
} else {
for (unsigned index = 0; index < I->getNumOperands(); ++index) {
//
// For other instructions, we want to insert instructions replacing
// constant expressions immediently before the instruction using the
// constant expression.
//
if (ConstantExpr * CE = hasConstantGEP (I->getOperand(index))) {
Instruction * NewInst = convertExpression (CE, I);
I->replaceUsesOfWith (CE, NewInst);
Worklist.push_back (NewInst);
}
}
}
}
return modified;
}
bool TailCallElim::EliminateRecursiveTailCall(CallInst *CI, ReturnInst *Ret,
BasicBlock *&OldEntry,
bool &TailCallsAreMarkedTail,
SmallVector<PHINode*, 8> &ArgumentPHIs,
bool CannotTailCallElimCallsMarkedTail) {
// If we are introducing accumulator recursion to eliminate operations after
// the call instruction that are both associative and commutative, the initial
// value for the accumulator is placed in this variable. If this value is set
// then we actually perform accumulator recursion elimination instead of
// simple tail recursion elimination. If the operation is an LLVM instruction
// (eg: "add") then it is recorded in AccumulatorRecursionInstr. If not, then
// we are handling the case when the return instruction returns a constant C
// which is different to the constant returned by other return instructions
// (which is recorded in AccumulatorRecursionEliminationInitVal). This is a
// special case of accumulator recursion, the operation being "return C".
Value *AccumulatorRecursionEliminationInitVal = 0;
Instruction *AccumulatorRecursionInstr = 0;
// Ok, we found a potential tail call. We can currently only transform the
// tail call if all of the instructions between the call and the return are
// movable to above the call itself, leaving the call next to the return.
// Check that this is the case now.
BasicBlock::iterator BBI = CI;
for (++BBI; &*BBI != Ret; ++BBI) {
if (CanMoveAboveCall(BBI, CI)) continue;
// If we can't move the instruction above the call, it might be because it
// is an associative and commutative operation that could be transformed
// using accumulator recursion elimination. Check to see if this is the
// case, and if so, remember the initial accumulator value for later.
if ((AccumulatorRecursionEliminationInitVal =
CanTransformAccumulatorRecursion(BBI, CI))) {
// Yes, this is accumulator recursion. Remember which instruction
// accumulates.
AccumulatorRecursionInstr = BBI;
} else {
return false; // Otherwise, we cannot eliminate the tail recursion!
}
}
// We can only transform call/return pairs that either ignore the return value
// of the call and return void, ignore the value of the call and return a
// constant, return the value returned by the tail call, or that are being
// accumulator recursion variable eliminated.
if (Ret->getNumOperands() == 1 && Ret->getReturnValue() != CI &&
!isa<UndefValue>(Ret->getReturnValue()) &&
AccumulatorRecursionEliminationInitVal == 0 &&
!getCommonReturnValue(0, CI)) {
// One case remains that we are able to handle: the current return
// instruction returns a constant, and all other return instructions
// return a different constant.
if (!isDynamicConstant(Ret->getReturnValue(), CI, Ret))
return false; // Current return instruction does not return a constant.
// Check that all other return instructions return a common constant. If
// so, record it in AccumulatorRecursionEliminationInitVal.
AccumulatorRecursionEliminationInitVal = getCommonReturnValue(Ret, CI);
if (!AccumulatorRecursionEliminationInitVal)
return false;
}
BasicBlock *BB = Ret->getParent();
Function *F = BB->getParent();
// OK! We can transform this tail call. If this is the first one found,
// create the new entry block, allowing us to branch back to the old entry.
if (OldEntry == 0) {
OldEntry = &F->getEntryBlock();
BasicBlock *NewEntry = BasicBlock::Create(F->getContext(), "", F, OldEntry);
NewEntry->takeName(OldEntry);
OldEntry->setName("tailrecurse");
BranchInst::Create(OldEntry, NewEntry);
// If this tail call is marked 'tail' and if there are any allocas in the
// entry block, move them up to the new entry block.
TailCallsAreMarkedTail = CI->isTailCall();
if (TailCallsAreMarkedTail)
// Move all fixed sized allocas from OldEntry to NewEntry.
for (BasicBlock::iterator OEBI = OldEntry->begin(), E = OldEntry->end(),
NEBI = NewEntry->begin(); OEBI != E; )
if (AllocaInst *AI = dyn_cast<AllocaInst>(OEBI++))
if (isa<ConstantInt>(AI->getArraySize()))
AI->moveBefore(NEBI);
// Now that we have created a new block, which jumps to the entry
// block, insert a PHI node for each argument of the function.
// For now, we initialize each PHI to only have the real arguments
// which are passed in.
Instruction *InsertPos = OldEntry->begin();
for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end();
I != E; ++I) {
PHINode *PN = PHINode::Create(I->getType(), 2,
I->getName() + ".tr", InsertPos);
I->replaceAllUsesWith(PN); // Everyone use the PHI node now!
PN->addIncoming(I, NewEntry);
ArgumentPHIs.push_back(PN);
}
}
// If this function has self recursive calls in the tail position where some
// are marked tail and some are not, only transform one flavor or another. We
// have to choose whether we move allocas in the entry block to the new entry
// block or not, so we can't make a good choice for both. NOTE: We could do
// slightly better here in the case that the function has no entry block
// allocas.
if (TailCallsAreMarkedTail && !CI->isTailCall())
return false;
// Ok, now that we know we have a pseudo-entry block WITH all of the
// required PHI nodes, add entries into the PHI node for the actual
// parameters passed into the tail-recursive call.
for (unsigned i = 0, e = CI->getNumArgOperands(); i != e; ++i)
ArgumentPHIs[i]->addIncoming(CI->getArgOperand(i), BB);
// If we are introducing an accumulator variable to eliminate the recursion,
// do so now. Note that we _know_ that no subsequent tail recursion
// eliminations will happen on this function because of the way the
// accumulator recursion predicate is set up.
//
if (AccumulatorRecursionEliminationInitVal) {
Instruction *AccRecInstr = AccumulatorRecursionInstr;
// Start by inserting a new PHI node for the accumulator.
pred_iterator PB = pred_begin(OldEntry), PE = pred_end(OldEntry);
PHINode *AccPN =
PHINode::Create(AccumulatorRecursionEliminationInitVal->getType(),
std::distance(PB, PE) + 1,
"accumulator.tr", OldEntry->begin());
// Loop over all of the predecessors of the tail recursion block. For the
// real entry into the function we seed the PHI with the initial value,
// computed earlier. For any other existing branches to this block (due to
// other tail recursions eliminated) the accumulator is not modified.
// Because we haven't added the branch in the current block to OldEntry yet,
// it will not show up as a predecessor.
for (pred_iterator PI = PB; PI != PE; ++PI) {
BasicBlock *P = *PI;
if (P == &F->getEntryBlock())
AccPN->addIncoming(AccumulatorRecursionEliminationInitVal, P);
else
AccPN->addIncoming(AccPN, P);
}
if (AccRecInstr) {
// Add an incoming argument for the current block, which is computed by
// our associative and commutative accumulator instruction.
AccPN->addIncoming(AccRecInstr, BB);
// Next, rewrite the accumulator recursion instruction so that it does not
// use the result of the call anymore, instead, use the PHI node we just
// inserted.
AccRecInstr->setOperand(AccRecInstr->getOperand(0) != CI, AccPN);
} else {
// Add an incoming argument for the current block, which is just the
// constant returned by the current return instruction.
AccPN->addIncoming(Ret->getReturnValue(), BB);
}
// Finally, rewrite any return instructions in the program to return the PHI
// node instead of the "initval" that they do currently. This loop will
// actually rewrite the return value we are destroying, but that's ok.
for (Function::iterator BBI = F->begin(), E = F->end(); BBI != E; ++BBI)
if (ReturnInst *RI = dyn_cast<ReturnInst>(BBI->getTerminator()))
RI->setOperand(0, AccPN);
++NumAccumAdded;
}
// Now that all of the PHI nodes are in place, remove the call and
// ret instructions, replacing them with an unconditional branch.
BranchInst *NewBI = BranchInst::Create(OldEntry, Ret);
NewBI->setDebugLoc(CI->getDebugLoc());
BB->getInstList().erase(Ret); // Remove return.
BB->getInstList().erase(CI); // Remove call.
++NumEliminated;
return true;
}
bool Inliner::runOnSCC(CallGraphSCC &SCC) {
CallGraph &CG = getAnalysis<CallGraph>();
const DataLayout *TD = getAnalysisIfAvailable<DataLayout>();
const TargetLibraryInfo *TLI = getAnalysisIfAvailable<TargetLibraryInfo>();
SmallPtrSet<Function*, 8> SCCFunctions;
DEBUG(dbgs() << "Inliner visiting SCC:");
for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I) {
Function *F = (*I)->getFunction();
if (F) SCCFunctions.insert(F);
DEBUG(dbgs() << " " << (F ? F->getName() : "INDIRECTNODE"));
}
// Scan through and identify all call sites ahead of time so that we only
// inline call sites in the original functions, not call sites that result
// from inlining other functions.
SmallVector<std::pair<CallSite, int>, 16> CallSites;
// When inlining a callee produces new call sites, we want to keep track of
// the fact that they were inlined from the callee. This allows us to avoid
// infinite inlining in some obscure cases. To represent this, we use an
// index into the InlineHistory vector.
SmallVector<std::pair<Function*, int>, 8> InlineHistory;
for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I) {
Function *F = (*I)->getFunction();
if (!F) continue;
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
CallSite CS(cast<Value>(I));
// If this isn't a call, or it is a call to an intrinsic, it can
// never be inlined.
if (!CS || isa<IntrinsicInst>(I))
continue;
// If this is a direct call to an external function, we can never inline
// it. If it is an indirect call, inlining may resolve it to be a
// direct call, so we keep it.
if (CS.getCalledFunction() && CS.getCalledFunction()->isDeclaration())
continue;
CallSites.push_back(std::make_pair(CS, -1));
}
}
DEBUG(dbgs() << ": " << CallSites.size() << " call sites.\n");
// If there are no calls in this function, exit early.
if (CallSites.empty())
return false;
// Now that we have all of the call sites, move the ones to functions in the
// current SCC to the end of the list.
unsigned FirstCallInSCC = CallSites.size();
for (unsigned i = 0; i < FirstCallInSCC; ++i)
if (Function *F = CallSites[i].first.getCalledFunction())
if (SCCFunctions.count(F))
std::swap(CallSites[i--], CallSites[--FirstCallInSCC]);
InlinedArrayAllocasTy InlinedArrayAllocas;
InlineFunctionInfo InlineInfo(&CG, TD);
// Now that we have all of the call sites, loop over them and inline them if
// it looks profitable to do so.
bool Changed = false;
bool LocalChange;
do {
LocalChange = false;
// Iterate over the outer loop because inlining functions can cause indirect
// calls to become direct calls.
for (unsigned CSi = 0; CSi != CallSites.size(); ++CSi) {
CallSite CS = CallSites[CSi].first;
Function *Caller = CS.getCaller();
Function *Callee = CS.getCalledFunction();
// If this call site is dead and it is to a readonly function, we should
// just delete the call instead of trying to inline it, regardless of
// size. This happens because IPSCCP propagates the result out of the
// call and then we're left with the dead call.
if (isInstructionTriviallyDead(CS.getInstruction(), TLI)) {
DEBUG(dbgs() << " -> Deleting dead call: "
<< *CS.getInstruction() << "\n");
// Update the call graph by deleting the edge from Callee to Caller.
CG[Caller]->removeCallEdgeFor(CS);
CS.getInstruction()->eraseFromParent();
++NumCallsDeleted;
} else {
// We can only inline direct calls to non-declarations.
if (Callee == 0 || Callee->isDeclaration()) continue;
// If this call site was obtained by inlining another function, verify
// that the include path for the function did not include the callee
// itself. If so, we'd be recursively inlining the same function,
// which would provide the same callsites, which would cause us to
// infinitely inline.
int InlineHistoryID = CallSites[CSi].second;
if (InlineHistoryID != -1 &&
InlineHistoryIncludes(Callee, InlineHistoryID, InlineHistory))
continue;
// If the policy determines that we should inline this function,
// try to do so.
if (!shouldInline(CS))
continue;
// Attempt to inline the function.
if (!InlineCallIfPossible(CS, InlineInfo, InlinedArrayAllocas,
InlineHistoryID, InsertLifetime))
continue;
++NumInlined;
// If inlining this function gave us any new call sites, throw them
// onto our worklist to process. They are useful inline candidates.
if (!InlineInfo.InlinedCalls.empty()) {
// Create a new inline history entry for this, so that we remember
// that these new callsites came about due to inlining Callee.
int NewHistoryID = InlineHistory.size();
InlineHistory.push_back(std::make_pair(Callee, InlineHistoryID));
for (unsigned i = 0, e = InlineInfo.InlinedCalls.size();
i != e; ++i) {
Value *Ptr = InlineInfo.InlinedCalls[i];
CallSites.push_back(std::make_pair(CallSite(Ptr), NewHistoryID));
}
}
}
// If we inlined or deleted the last possible call site to the function,
// delete the function body now.
if (Callee && Callee->use_empty() && Callee->hasLocalLinkage() &&
// TODO: Can remove if in SCC now.
!SCCFunctions.count(Callee) &&
// The function may be apparently dead, but if there are indirect
// callgraph references to the node, we cannot delete it yet, this
// could invalidate the CGSCC iterator.
CG[Callee]->getNumReferences() == 0) {
DEBUG(dbgs() << " -> Deleting dead function: "
<< Callee->getName() << "\n");
CallGraphNode *CalleeNode = CG[Callee];
// Remove any call graph edges from the callee to its callees.
CalleeNode->removeAllCalledFunctions();
// Removing the node for callee from the call graph and delete it.
delete CG.removeFunctionFromModule(CalleeNode);
++NumDeleted;
}
// Remove this call site from the list. If possible, use
// swap/pop_back for efficiency, but do not use it if doing so would
// move a call site to a function in this SCC before the
// 'FirstCallInSCC' barrier.
if (SCC.isSingular()) {
CallSites[CSi] = CallSites.back();
CallSites.pop_back();
} else {
CallSites.erase(CallSites.begin()+CSi);
}
--CSi;
Changed = true;
LocalChange = true;
}
} while (LocalChange);
return Changed;
}
// insertFastDiv - Substitutes the div/rem instruction with code that checks the
// value of the operands and uses a shorter-faster div/rem instruction when
// possible and the longer-slower div/rem instruction otherwise.
static bool insertFastDiv(Function &F,
Function::iterator &I,
BasicBlock::iterator &J,
IntegerType *BypassType,
bool UseDivOp,
bool UseSignedOp,
DivCacheTy &PerBBDivCache) {
// Get instruction operands
Instruction *Instr = J;
Value *Dividend = Instr->getOperand(0);
Value *Divisor = Instr->getOperand(1);
if (isa<ConstantInt>(Divisor) ||
(isa<ConstantInt>(Dividend) && isa<ConstantInt>(Divisor))) {
// Operations with immediate values should have
// been solved and replaced during compile time.
return false;
}
// Basic Block is split before divide
BasicBlock *MainBB = I;
BasicBlock *SuccessorBB = I->splitBasicBlock(J);
++I; //advance iterator I to successorBB
// Add new basic block for slow divide operation
BasicBlock *SlowBB = BasicBlock::Create(F.getContext(), "",
MainBB->getParent(), SuccessorBB);
SlowBB->moveBefore(SuccessorBB);
IRBuilder<> SlowBuilder(SlowBB, SlowBB->begin());
Value *SlowQuotientV;
Value *SlowRemainderV;
if (UseSignedOp) {
SlowQuotientV = SlowBuilder.CreateSDiv(Dividend, Divisor);
SlowRemainderV = SlowBuilder.CreateSRem(Dividend, Divisor);
} else {
SlowQuotientV = SlowBuilder.CreateUDiv(Dividend, Divisor);
SlowRemainderV = SlowBuilder.CreateURem(Dividend, Divisor);
}
SlowBuilder.CreateBr(SuccessorBB);
// Add new basic block for fast divide operation
BasicBlock *FastBB = BasicBlock::Create(F.getContext(), "",
MainBB->getParent(), SuccessorBB);
FastBB->moveBefore(SlowBB);
IRBuilder<> FastBuilder(FastBB, FastBB->begin());
Value *ShortDivisorV = FastBuilder.CreateCast(Instruction::Trunc, Divisor,
BypassType);
Value *ShortDividendV = FastBuilder.CreateCast(Instruction::Trunc, Dividend,
BypassType);
// udiv/urem because optimization only handles positive numbers
Value *ShortQuotientV = FastBuilder.CreateExactUDiv(ShortDividendV,
ShortDivisorV);
Value *ShortRemainderV = FastBuilder.CreateURem(ShortDividendV,
ShortDivisorV);
Value *FastQuotientV = FastBuilder.CreateCast(Instruction::ZExt,
ShortQuotientV,
Dividend->getType());
Value *FastRemainderV = FastBuilder.CreateCast(Instruction::ZExt,
ShortRemainderV,
Dividend->getType());
FastBuilder.CreateBr(SuccessorBB);
// Phi nodes for result of div and rem
IRBuilder<> SuccessorBuilder(SuccessorBB, SuccessorBB->begin());
PHINode *QuoPhi = SuccessorBuilder.CreatePHI(Instr->getType(), 2);
QuoPhi->addIncoming(SlowQuotientV, SlowBB);
QuoPhi->addIncoming(FastQuotientV, FastBB);
PHINode *RemPhi = SuccessorBuilder.CreatePHI(Instr->getType(), 2);
RemPhi->addIncoming(SlowRemainderV, SlowBB);
RemPhi->addIncoming(FastRemainderV, FastBB);
// Replace Instr with appropriate phi node
if (UseDivOp)
Instr->replaceAllUsesWith(QuoPhi);
else
Instr->replaceAllUsesWith(RemPhi);
Instr->eraseFromParent();
// Combine operands into a single value with OR for value testing below
MainBB->getInstList().back().eraseFromParent();
IRBuilder<> MainBuilder(MainBB, MainBB->end());
Value *OrV = MainBuilder.CreateOr(Dividend, Divisor);
// BitMask is inverted to check if the operands are
// larger than the bypass type
uint64_t BitMask = ~BypassType->getBitMask();
Value *AndV = MainBuilder.CreateAnd(OrV, BitMask);
// Compare operand values and branch
Value *ZeroV = MainBuilder.getInt32(0);
Value *CmpV = MainBuilder.CreateICmpEQ(AndV, ZeroV);
MainBuilder.CreateCondBr(CmpV, FastBB, SlowBB);
// point iterator J at first instruction of successorBB
J = I->begin();
// Cache phi nodes to be used later in place of other instances
// of div or rem with the same sign, dividend, and divisor
DivOpInfo Key(UseSignedOp, Dividend, Divisor);
DivPhiNodes Value(QuoPhi, RemPhi);
PerBBDivCache.insert(std::pair<DivOpInfo, DivPhiNodes>(Key, Value));
return true;
}
/*
* Main args are always input
* Functions currently considered as input functions:
* scanf
* fscanf
* gets
* fgets
* fread
* fgetc
* getc
* getchar
* recv
* recvmsg
* read
* recvfrom
* fread
*/
bool InputDep::runOnModule(Module &M) {
// DEBUG (errs() << "Function " << F.getName() << "\n";);
NumInputValues = 0;
bool inserted;
Function* main = M.getFunction("main");
if (main) {
MDNode *mdn = main->begin()->begin()->getMetadata("dbg");
for (Function::arg_iterator Arg = main->arg_begin(), aEnd =
main->arg_end(); Arg != aEnd; Arg++) {
inputDepValues.insert(Arg);
NumInputValues++;
if (mdn) {
DILocation Loc(mdn);
unsigned Line = Loc.getLineNumber();
lineNo[Arg] = Line-1; //educated guess (can only get line numbers from insts, suppose main is declared one line before 1st inst
}
}
}
for (Module::iterator F = M.begin(), eM = M.end(); F != eM; ++F) {
for (Function::iterator BB = F->begin(), e = F->end(); BB != e; ++BB) {
for (BasicBlock::iterator I = BB->begin(), ee = BB->end(); I != ee; ++I) {
if (CallInst *CI = dyn_cast<CallInst>(I)) {
Function *Callee = CI->getCalledFunction();
if (Callee) {
Value* V;
inserted = false;
StringRef Name = Callee->getName();
if (Name.equals("main")) {
errs() << "main\n";
V = CI->getArgOperand(1); //char* argv[]
inputDepValues.insert(V);
inserted = true;
//errs() << "Input " << *V << "\n";
}
if (Name.equals("__isoc99_scanf") || Name.equals(
"scanf")) {
for (unsigned i = 1, eee = CI->getNumArgOperands(); i
!= eee; ++i) { // skip format string (i=1)
V = CI->getArgOperand(i);
if (V->getType()->isPointerTy()) {
inputDepValues.insert(V);
inserted = true;
//errs() << "Input " << *V << "\n";
}
}
} else if (Name.equals("__isoc99_fscanf")
|| Name.equals("fscanf")) {
for (unsigned i = 2, eee = CI->getNumArgOperands(); i
!= eee; ++i) { // skip file pointer and format string (i=1)
V = CI->getArgOperand(i);
if (V->getType()->isPointerTy()) {
inputDepValues.insert(V);
inserted = true;
//errs() << "Input " << *V << "\n";
}
}
} else if ((Name.equals("gets") || Name.equals("fgets")
|| Name.equals("fread"))
|| Name.equals("getwd")
|| Name.equals("getcwd")) {
V = CI->getArgOperand(0); //the first argument receives the input for these functions
if (V->getType()->isPointerTy()) {
inputDepValues.insert(V);
inserted = true;
//errs() << "Input " << *V << "\n";
}
} else if ((Name.equals("fgetc") || Name.equals("getc")
|| Name.equals("getchar"))) {
inputDepValues.insert(CI);
inserted = true;
//errs() << "Input " << *CI << "\n";
} else if (Name.equals("recv")
|| Name.equals("recvmsg")
|| Name.equals("read")) {
Value* V = CI->getArgOperand(1);
if (V->getType()->isPointerTy()) {
inputDepValues.insert(V);
inserted = true;
//errs() << "Input " << *V << "\n";
}
} else if (Name.equals("recvfrom")) {
V = CI->getArgOperand(1);
if (V->getType()->isPointerTy()) {
inputDepValues.insert(V);
inserted = true;
//errs() << "Input " << *V << "\n";
}
V = CI->getArgOperand(4);
if (V->getType()->isPointerTy()) {
inputDepValues.insert(V);
inserted = true;
//errs() << "Input " << *V << "\n";
}
}
if (inserted) {
if (MDNode *mdn = I->getMetadata("dbg")) {
NumInputValues++;
DILocation Loc(mdn);
unsigned Line = Loc.getLineNumber();
lineNo[V] = Line;
}
}
}
}
}
}
}
DEBUG(printer());
return false;
}
// Check to see if this function returns one or more constants. If so, replace
// all callers that use those return values with the constant value. This will
// leave in the actual return values and instructions, but deadargelim will
// clean that up.
//
// Additionally if a function always returns one of its arguments directly,
// callers will be updated to use the value they pass in directly instead of
// using the return value.
bool IPCP::PropagateConstantReturn(Function &F) {
if (F.getReturnType()->isVoidTy())
return false; // No return value.
// If this function could be overridden later in the link stage, we can't
// propagate information about its results into callers.
if (F.mayBeOverridden())
return false;
// Check to see if this function returns a constant.
SmallVector<Value *,4> RetVals;
StructType *STy = dyn_cast<StructType>(F.getReturnType());
if (STy)
for (unsigned i = 0, e = STy->getNumElements(); i < e; ++i)
RetVals.push_back(UndefValue::get(STy->getElementType(i)));
else
RetVals.push_back(UndefValue::get(F.getReturnType()));
unsigned NumNonConstant = 0;
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB)
if (ReturnInst *RI = dyn_cast<ReturnInst>(BB->getTerminator())) {
for (unsigned i = 0, e = RetVals.size(); i != e; ++i) {
// Already found conflicting return values?
Value *RV = RetVals[i];
if (!RV)
continue;
// Find the returned value
Value *V;
if (!STy)
V = RI->getOperand(0);
else
V = FindInsertedValue(RI->getOperand(0), i);
if (V) {
// Ignore undefs, we can change them into anything
if (isa<UndefValue>(V))
continue;
// Try to see if all the rets return the same constant or argument.
if (isa<Constant>(V) || isa<Argument>(V)) {
if (isa<UndefValue>(RV)) {
// No value found yet? Try the current one.
RetVals[i] = V;
continue;
}
// Returning the same value? Good.
if (RV == V)
continue;
}
}
// Different or no known return value? Don't propagate this return
// value.
RetVals[i] = 0;
// All values non-constant? Stop looking.
if (++NumNonConstant == RetVals.size())
return false;
}
}
// If we got here, the function returns at least one constant value. Loop
// over all users, replacing any uses of the return value with the returned
// constant.
bool MadeChange = false;
for (Value::use_iterator UI = F.use_begin(), E = F.use_end(); UI != E; ++UI) {
CallSite CS(*UI);
Instruction* Call = CS.getInstruction();
// Not a call instruction or a call instruction that's not calling F
// directly?
if (!Call || !CS.isCallee(UI))
continue;
// Call result not used?
if (Call->use_empty())
continue;
MadeChange = true;
if (STy == 0) {
Value* New = RetVals[0];
if (Argument *A = dyn_cast<Argument>(New))
// Was an argument returned? Then find the corresponding argument in
// the call instruction and use that.
New = CS.getArgument(A->getArgNo());
Call->replaceAllUsesWith(New);
continue;
}
for (Value::use_iterator I = Call->use_begin(), E = Call->use_end();
I != E;) {
Instruction *Ins = cast<Instruction>(*I);
// Increment now, so we can remove the use
++I;
// Find the index of the retval to replace with
int index = -1;
if (ExtractValueInst *EV = dyn_cast<ExtractValueInst>(Ins))
if (EV->hasIndices())
index = *EV->idx_begin();
// If this use uses a specific return value, and we have a replacement,
// replace it.
if (index != -1) {
Value *New = RetVals[index];
if (New) {
if (Argument *A = dyn_cast<Argument>(New))
// Was an argument returned? Then find the corresponding argument in
// the call instruction and use that.
New = CS.getArgument(A->getArgNo());
Ins->replaceAllUsesWith(New);
Ins->eraseFromParent();
}
}
}
}
if (MadeChange) ++NumReturnValProped;
return MadeChange;
}
/*
* Clone a given function removing dead stores
*/
Function* DeadStoreEliminationPass::cloneFunctionWithoutDeadStore(Function *Fn,
Instruction* caller, std::string suffix) {
Function *NF = Function::Create(Fn->getFunctionType(), Fn->getLinkage());
NF->copyAttributesFrom(Fn);
// Copy the parameter names, to ease function inspection afterwards.
Function::arg_iterator NFArg = NF->arg_begin();
for (Function::arg_iterator Arg = Fn->arg_begin(), ArgEnd = Fn->arg_end();
Arg != ArgEnd; ++Arg, ++NFArg) {
NFArg->setName(Arg->getName());
}
// To avoid name collision, we should select another name.
NF->setName(Fn->getName() + suffix);
// Fill clone content
ValueToValueMapTy VMap;
SmallVector<ReturnInst*, 8> Returns;
Function::arg_iterator NI = NF->arg_begin();
for (Function::arg_iterator I = Fn->arg_begin();
NI != NF->arg_end(); ++I, ++NI) {
VMap[I] = NI;
}
CloneAndPruneFunctionInto(NF, Fn, VMap, false, Returns);
// Remove dead stores
std::set<Value*> deadArgs = deadArguments[caller];
std::set<Value*> removeStoresTo;
Function::arg_iterator NFArgIter = NF->arg_begin();
for (Function::arg_iterator FnArgIter = Fn->arg_begin(); FnArgIter !=
Fn->arg_end(); ++FnArgIter, ++NFArgIter) {
Value *FnArg = FnArgIter;
if (deadArgs.count(FnArg)) {
removeStoresTo.insert(NFArgIter);
}
}
std::vector<Instruction*> toRemove;
for (Function::iterator BB = NF->begin(); BB != NF->end(); ++BB) {
for (BasicBlock::iterator I = BB->begin(); I != BB->end(); ++I) {
Instruction *inst = I;
if (!isa<StoreInst>(inst)) continue;
StoreInst *SI = dyn_cast<StoreInst>(inst);
Value *ptrOp = SI->getPointerOperand();
if (removeStoresTo.count(ptrOp)) {
DEBUG(errs() << "will remove this store: " << *inst << "\n");
toRemove.push_back(inst);
}
}
}
for (std::vector<Instruction*>::iterator it = toRemove.begin();
it != toRemove.end(); ++it) {
Instruction* inst = *it;
inst->eraseFromParent();
RemovedStores++;
}
// Insert the clone function before the original
Fn->getParent()->getFunctionList().insert(Fn, NF);
return NF;
}
/// IsFunctionMallocLike - A function is malloc-like if it returns either null
/// or a pointer that doesn't alias any other pointer visible to the caller.
bool FunctionAttrs::IsFunctionMallocLike(Function *F,
SmallPtrSet<Function*, 8> &SCCNodes) const {
SmallSetVector<Value *, 8> FlowsToReturn;
for (Function::iterator I = F->begin(), E = F->end(); I != E; ++I)
if (ReturnInst *Ret = dyn_cast<ReturnInst>(I->getTerminator()))
FlowsToReturn.insert(Ret->getReturnValue());
for (unsigned i = 0; i != FlowsToReturn.size(); ++i) {
Value *RetVal = FlowsToReturn[i];
if (Constant *C = dyn_cast<Constant>(RetVal)) {
if (!C->isNullValue() && !isa<UndefValue>(C))
return false;
continue;
}
if (isa<Argument>(RetVal))
return false;
if (Instruction *RVI = dyn_cast<Instruction>(RetVal))
switch (RVI->getOpcode()) {
// Extend the analysis by looking upwards.
case Instruction::BitCast:
case Instruction::GetElementPtr:
FlowsToReturn.insert(RVI->getOperand(0));
continue;
case Instruction::Select: {
SelectInst *SI = cast<SelectInst>(RVI);
FlowsToReturn.insert(SI->getTrueValue());
FlowsToReturn.insert(SI->getFalseValue());
continue;
}
case Instruction::PHI: {
PHINode *PN = cast<PHINode>(RVI);
for (int i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
FlowsToReturn.insert(PN->getIncomingValue(i));
continue;
}
// Check whether the pointer came from an allocation.
case Instruction::Alloca:
break;
case Instruction::Call:
case Instruction::Invoke: {
CallSite CS(RVI);
if (CS.paramHasAttr(0, Attribute::NoAlias))
break;
if (CS.getCalledFunction() &&
SCCNodes.count(CS.getCalledFunction()))
break;
} // fall-through
default:
return false; // Did not come from an allocation.
}
if (PointerMayBeCaptured(RetVal, false, /*StoreCaptures=*/false))
return false;
}
return true;
}
//
// Returns of float, double and complex need to be handled with a helper
// function.
//
static bool fixupFPReturnAndCall
(Function &F, Module *M, const MipsSubtarget &Subtarget) {
bool Modified = false;
LLVMContext &C = M->getContext();
Type *MyVoid = Type::getVoidTy(C);
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB)
for (BasicBlock::iterator I = BB->begin(), E = BB->end();
I != E; ++I) {
Instruction &Inst = *I;
if (const ReturnInst *RI = dyn_cast<ReturnInst>(I)) {
Value *RVal = RI->getReturnValue();
if (!RVal) continue;
//
// If there is a return value and it needs a helper function,
// figure out which one and add a call before the actual
// return to this helper. The purpose of the helper is to move
// floating point values from their soft float return mapping to
// where they would have been mapped to in floating point registers.
//
Type *T = RVal->getType();
FPReturnVariant RV = whichFPReturnVariant(T);
if (RV == NoFPRet) continue;
static const char* Helper[NoFPRet] =
{"__mips16_ret_sf", "__mips16_ret_df", "__mips16_ret_sc",
"__mips16_ret_dc"};
const char *Name = Helper[RV];
AttributeSet A;
Value *Params[] = {RVal};
Modified = true;
//
// These helper functions have a different calling ABI so
// this __Mips16RetHelper indicates that so that later
// during call setup, the proper call lowering to the helper
// functions will take place.
//
A = A.addAttribute(C, AttributeSet::FunctionIndex,
"__Mips16RetHelper");
A = A.addAttribute(C, AttributeSet::FunctionIndex,
Attribute::ReadNone);
A = A.addAttribute(C, AttributeSet::FunctionIndex,
Attribute::NoInline);
Value *F = (M->getOrInsertFunction(Name, A, MyVoid, T, NULL));
CallInst::Create(F, Params, "", &Inst );
} else if (const CallInst *CI = dyn_cast<CallInst>(I)) {
const Value* V = CI->getCalledValue();
const Type* T = nullptr;
if (V) T = V->getType();
const PointerType *PFT=nullptr;
if (T) PFT = dyn_cast<PointerType>(T);
const FunctionType *FT=nullptr;
if (PFT) FT = dyn_cast<FunctionType>(PFT->getElementType());
Function *F_ = CI->getCalledFunction();
if (FT && needsFPReturnHelper(*FT) &&
!(F_ && isIntrinsicInline(F_))) {
Modified=true;
F.addFnAttr("saveS2");
}
if (F_ && !isIntrinsicInline(F_)) {
// pic mode calls are handled by already defined
// helper functions
if (needsFPReturnHelper(*F_)) {
Modified=true;
F.addFnAttr("saveS2");
}
if (Subtarget.getRelocationModel() != Reloc::PIC_ ) {
if (needsFPHelperFromSig(*F_)) {
assureFPCallStub(*F_, M, Subtarget);
Modified=true;
}
}
}
}
}
return Modified;
}
bool EdgeProfiler::runOnModule(Module &M) {
Function *Main = M.getFunction("main");
if (Main == 0) {
errs() << "WARNING: cannot insert edge profiling into a module"
<< " with no main function!\n";
return false; // No main, no instrumentation!
}
std::set<BasicBlock*> BlocksToInstrument;
unsigned NumEdges = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
// Reserve space for (0,entry) edge.
++NumEdges;
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
// Keep track of which blocks need to be instrumented. We don't want to
// instrument blocks that are added as the result of breaking critical
// edges!
BlocksToInstrument.insert(BB);
NumEdges += BB->getTerminator()->getNumSuccessors();
}
}
Type *ATy = ArrayType::get(Type::getInt32Ty(M.getContext()), NumEdges);
GlobalVariable *Counters =
new GlobalVariable(M, ATy, false, GlobalValue::InternalLinkage,
Constant::getNullValue(ATy), "EdgeProfCounters");
NumEdgesInserted = NumEdges;
// Instrument all of the edges...
unsigned i = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
// Create counter for (0,entry) edge.
IncrementCounterInBlock(&F->getEntryBlock(), i++, Counters);
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
if (BlocksToInstrument.count(BB)) { // Don't instrument inserted blocks
// Okay, we have to add a counter of each outgoing edge. If the
// outgoing edge is not critical don't split it, just insert the counter
// in the source or destination of the edge.
TerminatorInst *TI = BB->getTerminator();
for (unsigned s = 0, e = TI->getNumSuccessors(); s != e; ++s) {
// If the edge is critical, split it.
SplitCriticalEdge(TI, s, this);
// Okay, we are guaranteed that the edge is no longer critical. If we
// only have a single successor, insert the counter in this block,
// otherwise insert it in the successor block.
if (TI->getNumSuccessors() == 1) {
// Insert counter at the start of the block
IncrementCounterInBlock(BB, i++, Counters, false);
} else {
// Insert counter at the start of the block
IncrementCounterInBlock(TI->getSuccessor(s), i++, Counters);
}
}
}
}
// Add the initialization call to main.
InsertProfilingInitCall(Main, "llvm_start_edge_profiling", Counters);
return true;
}
/// splitLiveRangesAcrossInvokes - Each value that is live across an unwind edge
/// we spill into a stack location, guaranteeing that there is nothing live
/// across the unwind edge. This process also splits all critical edges
/// coming out of invoke's.
/// FIXME: Move this function to a common utility file (Local.cpp?) so
/// both SjLj and LowerInvoke can use it.
void SjLjEHPass::
splitLiveRangesAcrossInvokes(SmallVector<InvokeInst*,16> &Invokes) {
// First step, split all critical edges from invoke instructions.
for (unsigned i = 0, e = Invokes.size(); i != e; ++i) {
InvokeInst *II = Invokes[i];
SplitCriticalEdge(II, 0, this);
// FIXME: New EH - This if-condition will be always true in the new scheme.
if (II->getUnwindDest()->isLandingPad()) {
SmallVector<BasicBlock*, 2> NewBBs;
SplitLandingPadPredecessors(II->getUnwindDest(), II->getParent(),
".1", ".2", this, NewBBs);
LPadSuccMap[II] = *succ_begin(NewBBs[0]);
} else {
SplitCriticalEdge(II, 1, this);
}
assert(!isa<PHINode>(II->getNormalDest()) &&
!isa<PHINode>(II->getUnwindDest()) &&
"Critical edge splitting left single entry phi nodes?");
}
Function *F = Invokes.back()->getParent()->getParent();
// To avoid having to handle incoming arguments specially, we lower each arg
// to a copy instruction in the entry block. This ensures that the argument
// value itself cannot be live across the entry block.
BasicBlock::iterator AfterAllocaInsertPt = F->begin()->begin();
while (isa<AllocaInst>(AfterAllocaInsertPt) &&
isa<ConstantInt>(cast<AllocaInst>(AfterAllocaInsertPt)->getArraySize()))
++AfterAllocaInsertPt;
for (Function::arg_iterator AI = F->arg_begin(), E = F->arg_end();
AI != E; ++AI) {
Type *Ty = AI->getType();
// Aggregate types can't be cast, but are legal argument types, so we have
// to handle them differently. We use an extract/insert pair as a
// lightweight method to achieve the same goal.
if (isa<StructType>(Ty) || isa<ArrayType>(Ty) || isa<VectorType>(Ty)) {
Instruction *EI = ExtractValueInst::Create(AI, 0, "",AfterAllocaInsertPt);
Instruction *NI = InsertValueInst::Create(AI, EI, 0);
NI->insertAfter(EI);
AI->replaceAllUsesWith(NI);
// Set the operand of the instructions back to the AllocaInst.
EI->setOperand(0, AI);
NI->setOperand(0, AI);
} else {
// This is always a no-op cast because we're casting AI to AI->getType()
// so src and destination types are identical. BitCast is the only
// possibility.
CastInst *NC = new BitCastInst(
AI, AI->getType(), AI->getName()+".tmp", AfterAllocaInsertPt);
AI->replaceAllUsesWith(NC);
// Set the operand of the cast instruction back to the AllocaInst.
// Normally it's forbidden to replace a CastInst's operand because it
// could cause the opcode to reflect an illegal conversion. However,
// we're replacing it here with the same value it was constructed with.
// We do this because the above replaceAllUsesWith() clobbered the
// operand, but we want this one to remain.
NC->setOperand(0, AI);
}
}
// Finally, scan the code looking for instructions with bad live ranges.
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
// Ignore obvious cases we don't have to handle. In particular, most
// instructions either have no uses or only have a single use inside the
// current block. Ignore them quickly.
Instruction *Inst = II;
if (Inst->use_empty()) continue;
if (Inst->hasOneUse() &&
cast<Instruction>(Inst->use_back())->getParent() == BB &&
!isa<PHINode>(Inst->use_back())) continue;
// If this is an alloca in the entry block, it's not a real register
// value.
if (AllocaInst *AI = dyn_cast<AllocaInst>(Inst))
if (isa<ConstantInt>(AI->getArraySize()) && BB == F->begin())
continue;
// Avoid iterator invalidation by copying users to a temporary vector.
SmallVector<Instruction*,16> Users;
for (Value::use_iterator UI = Inst->use_begin(), E = Inst->use_end();
UI != E; ++UI) {
Instruction *User = cast<Instruction>(*UI);
if (User->getParent() != BB || isa<PHINode>(User))
Users.push_back(User);
}
// Find all of the blocks that this value is live in.
std::set<BasicBlock*> LiveBBs;
LiveBBs.insert(Inst->getParent());
while (!Users.empty()) {
Instruction *U = Users.back();
Users.pop_back();
if (!isa<PHINode>(U)) {
MarkBlocksLiveIn(U->getParent(), LiveBBs);
} else {
// Uses for a PHI node occur in their predecessor block.
PHINode *PN = cast<PHINode>(U);
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
if (PN->getIncomingValue(i) == Inst)
MarkBlocksLiveIn(PN->getIncomingBlock(i), LiveBBs);
}
}
// Now that we know all of the blocks that this thing is live in, see if
// it includes any of the unwind locations.
bool NeedsSpill = false;
for (unsigned i = 0, e = Invokes.size(); i != e; ++i) {
BasicBlock *UnwindBlock = Invokes[i]->getUnwindDest();
if (UnwindBlock != BB && LiveBBs.count(UnwindBlock)) {
NeedsSpill = true;
}
}
// If we decided we need a spill, do it.
// FIXME: Spilling this way is overkill, as it forces all uses of
// the value to be reloaded from the stack slot, even those that aren't
// in the unwind blocks. We should be more selective.
if (NeedsSpill) {
++NumSpilled;
DemoteRegToStack(*Inst, true);
}
}
}
bool klee::PhiCleanerPass::runOnFunction(Function &f) {
bool changed = false;
for (Function::iterator b = f.begin(), be = f.end(); b != be; ++b) {
BasicBlock::iterator it = b->begin();
if (it->getOpcode() == Instruction::PHI) {
PHINode *reference = cast<PHINode>(it);
std::set<Value*> phis;
phis.insert(reference);
unsigned numBlocks = reference->getNumIncomingValues();
for (++it; isa<PHINode>(*it); ++it) {
PHINode *pi = cast<PHINode>(it);
assert(numBlocks == pi->getNumIncomingValues());
// see if it is out of order
unsigned i;
for (i=0; i<numBlocks; i++)
if (pi->getIncomingBlock(i) != reference->getIncomingBlock(i))
break;
if (i!=numBlocks) {
std::vector<Value*> values;
values.reserve(numBlocks);
for (unsigned i=0; i<numBlocks; i++)
values[i] = pi->getIncomingValueForBlock(reference->getIncomingBlock(i));
for (unsigned i=0; i<numBlocks; i++) {
pi->setIncomingBlock(i, reference->getIncomingBlock(i));
pi->setIncomingValue(i, values[i]);
}
changed = true;
}
// see if it uses any previously defined phi nodes
for (i=0; i<numBlocks; i++) {
Value *value = pi->getIncomingValue(i);
if (phis.find(value) != phis.end()) {
// fix by making a "move" at the end of the incoming block
// to a new temporary, which is thus known not to be a phi
// result. we could be somewhat more efficient about this
// by sharing temps and by reordering phi instructions so
// this isn't completely necessary, but in the end this is
// just a pathological case which does not occur very
// often.
Instruction *tmp =
new BitCastInst(value,
value->getType(),
value->getName() + ".phiclean",
pi->getIncomingBlock(i)->getTerminator());
pi->setIncomingValue(i, tmp);
}
changed = true;
}
phis.insert(pi);
}
}
}
return changed;
}
bool PruneEH::runOnSCC(const std::vector<CallGraphNode *> &SCC) {
CallGraph &CG = getAnalysis<CallGraph>();
bool MadeChange = false;
// First pass, scan all of the functions in the SCC, simplifying them
// according to what we know.
for (unsigned i = 0, e = SCC.size(); i != e; ++i)
if (Function *F = SCC[i]->getFunction())
MadeChange |= SimplifyFunction(F);
// Next, check to see if any callees might throw or if there are any external
// functions in this SCC: if so, we cannot prune any functions in this SCC.
// If this SCC includes the unwind instruction, we KNOW it throws, so
// obviously the SCC might throw.
//
bool SCCMightUnwind = false, SCCMightReturn = false;
for (unsigned i = 0, e = SCC.size();
(!SCCMightUnwind || !SCCMightReturn) && i != e; ++i) {
Function *F = SCC[i]->getFunction();
if (F == 0 || (F->isDeclaration() && !F->getIntrinsicID())) {
SCCMightUnwind = true;
SCCMightReturn = true;
} else {
if (F->isDeclaration())
SCCMightReturn = true;
// Check to see if this function performs an unwind or calls an
// unwinding function.
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
if (isa<UnwindInst>(BB->getTerminator())) { // Uses unwind!
SCCMightUnwind = true;
} else if (isa<ReturnInst>(BB->getTerminator())) {
SCCMightReturn = true;
}
// Invoke instructions don't allow unwinding to continue, so we are
// only interested in call instructions.
if (!SCCMightUnwind)
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
if (CallInst *CI = dyn_cast<CallInst>(I)) {
if (Function *Callee = CI->getCalledFunction()) {
CallGraphNode *CalleeNode = CG[Callee];
// If the callee is outside our current SCC, or if it is not
// known to throw, then we might throw also.
if (std::find(SCC.begin(), SCC.end(), CalleeNode) == SCC.end()&&
!DoesNotUnwind.count(CalleeNode)) {
SCCMightUnwind = true;
break;
}
} else {
// Indirect call, it might throw.
SCCMightUnwind = true;
break;
}
}
if (SCCMightUnwind && SCCMightReturn) break;
}
}
}
// If the SCC doesn't unwind or doesn't throw, note this fact.
if (!SCCMightUnwind)
for (unsigned i = 0, e = SCC.size(); i != e; ++i)
DoesNotUnwind.insert(SCC[i]);
if (!SCCMightReturn)
for (unsigned i = 0, e = SCC.size(); i != e; ++i)
DoesNotReturn.insert(SCC[i]);
for (unsigned i = 0, e = SCC.size(); i != e; ++i) {
// Convert any invoke instructions to non-throwing functions in this node
// into call instructions with a branch. This makes the exception blocks
// dead.
if (Function *F = SCC[i]->getFunction())
MadeChange |= SimplifyFunction(F);
}
return MadeChange;
}
Function* Decompiler::decompileFunction(unsigned Address) {
// Check that Address is inside the current section.
// TODO: Find a better way to do this check. What we really care about is
// avoiding reads to library calls and areas of memory we can't "see".
const object::SectionRef Sect = Dis->getCurrentSection();
uint64_t SectStart, SectEnd;
Sect.getAddress(SectStart);
Sect.getSize(SectEnd);
SectEnd += SectStart;
if (Address < SectStart || Address > SectEnd) {
errs() << "Address out of bounds for section (is this a library call?): "
<< format("%1" PRIx64, Address) << "\n";
return NULL;
}
MachineFunction *MF = Dis->disassemble(Address);
// Get Function Name
// TODO: Determine Function Type
FunctionType *FType = FunctionType::get(Type::getPrimitiveType(*Context,
Type::VoidTyID), false);
Function *F =
cast<Function>(Mod->getOrInsertFunction(MF->getName(), FType));
if (!F->empty()) {
return F;
}
// Create a basic block to hold entry point (alloca) information
BasicBlock *entry = getOrCreateBasicBlock("entry", F);
// For each basic block
MachineFunction::iterator BI = MF->begin(), BE = MF->end();
while (BI != BE) {
// Add branch from "entry"
if (BI == MF->begin()) {
entry->getInstList().push_back(
BranchInst::Create(getOrCreateBasicBlock(BI->getName(), F)));
} else {
getOrCreateBasicBlock(BI->getName(), F);
}
++BI;
}
BI = MF->begin();
while (BI != BE) {
if (decompileBasicBlock(BI, F) == NULL) {
printError("Unable to decompile basic block!");
}
++BI;
}
// During Decompilation, did any "in-between" basic blocks get created?
// Nothing ever splits the entry block, so we skip it.
for (Function::iterator I = ++F->begin(), E = F->end(); I != E; ++I) {
if (!(I->empty())) {
continue;
}
// Right now, the only way to get the right offset is to parse its name
// it sucks, but it works.
StringRef Name = I->getName();
if (Name == "end" || Name == "entry") continue; // these can be empty
size_t Off = F->getName().size() + 1;
size_t Size = Name.size() - Off;
StringRef BBAddrStr = Name.substr(Off, Size);
unsigned long long BBAddr;
getAsUnsignedInteger(BBAddrStr, 10, BBAddr);
BBAddr += Address;
DEBUG(errs() << "Split Target: " << Name << "\t Address: "
<< BBAddr << "\n");
// split Block at AddrStr
Function::iterator SB; // Split basic block
BasicBlock::iterator SI, SE; // Split instruction
// Note the ++, nothing ever splits the entry block.
for (SB = ++F->begin(); SB != E; ++SB) {
DEBUG(outs() << "SB: " << SB->getName()
<< "\tRange: " << Dis->getDebugOffset(SB->begin()->getDebugLoc())
<< " " << Dis->getDebugOffset(SB->getTerminator()->getDebugLoc())
<< "\n");
if (SB->empty() || BBAddr < getBasicBlockAddress(SB)
|| BBAddr > Dis->getDebugOffset(SB->getTerminator()->getDebugLoc())) {
continue;
}
// Reorder instructions based on Debug Location
sortBasicBlock(SB);
DEBUG(errs() << "Found Split Block: " << SB->getName() << "\n");
// Find iterator to split on.
for (SI = SB->begin(), SE = SB->end(); SI != SE; ++SI) {
// outs() << "SI: " << SI->getDebugLoc().getLine() << "\n";
if (Dis->getDebugOffset(SI->getDebugLoc()) == BBAddr) break;
if (Dis->getDebugOffset(SI->getDebugLoc()) > BBAddr) {
errs() << "Could not find address inside basic block!\n"
<< "SI: " << Dis->getDebugOffset(SI->getDebugLoc()) << "\n"
<< "BBAddr: " << BBAddr << "\n";
break;
}
}
break;
}
if (!SB || SI == SE || SB == E) {
errs() << "Decompiler: Failed to find instruction offset in function!\n";
continue;
}
// outs() << SB->getName() << " " << SI->getName() << "\n";
// outs() << "Creating Block...";
splitBasicBlockIntoBlock(SB, SI, I);
}
// Clean up unnecessary stores and loads
FunctionPassManager FPM(Mod);
// FPM.add(createPromoteMemoryToRegisterPass()); // See Scalar.h for more.
FPM.add(createTypeRecoveryPass());
FPM.run(*F);
return F;
}
/**
* 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 CallTargetFinder<dsa>::findIndTargets(Module &M)
{
dsa* T = &getAnalysis<dsa>();
const DSCallGraph & callgraph = T->getCallGraph();
DSGraph* G = T->getGlobalsGraph();
DSGraph::ScalarMapTy& SM = G->getScalarMap();
for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I)
if (!I->isDeclaration())
for (Function::iterator F = I->begin(), FE = I->end(); F != FE; ++F)
for (BasicBlock::iterator B = F->begin(), BE = F->end(); B != BE; ++B)
if (isa<CallInst>(B) || isa<InvokeInst>(B)) {
CallSite cs(B);
AllSites.push_back(cs);
Function* CF = cs.getCalledFunction();
if (isa<UndefValue>(cs.getCalledValue())) continue;
if (isa<InlineAsm>(cs.getCalledValue())) continue;
//
// If the called function is casted from one function type to
// another, peer into the cast instruction and pull out the actual
// function being called.
//
if (!CF)
CF = dyn_cast<Function>(cs.getCalledValue()->stripPointerCasts());
if (!CF) {
Value * calledValue = cs.getCalledValue()->stripPointerCasts();
if (isa<ConstantPointerNull>(calledValue)) {
++DirCall;
CompleteSites.insert(cs);
} else {
IndCall++;
DSCallGraph::callee_iterator csi = callgraph.callee_begin(cs),
cse = callgraph.callee_end(cs);
while(csi != cse) {
const Function *F = *csi;
DSCallGraph::scc_iterator sccii = callgraph.scc_begin(F),
sccee = callgraph.scc_end(F);
for(;sccii != sccee; ++sccii) {
DSGraph::ScalarMapTy::const_iterator I = SM.find(SM.getLeaderForGlobal(*sccii));
if (I != SM.end()) {
IndMap[cs].push_back (*sccii);
}
}
++csi;
}
const Function *F1 = (cs).getInstruction()->getParent()->getParent();
F1 = callgraph.sccLeader(&*F1);
DSCallGraph::scc_iterator sccii = callgraph.scc_begin(F1),
sccee = callgraph.scc_end(F1);
for(;sccii != sccee; ++sccii) {
DSGraph::ScalarMapTy::const_iterator I = SM.find(SM.getLeaderForGlobal(*sccii));
if (I != SM.end()) {
IndMap[cs].push_back (*sccii);
}
}
DSNode* N = T->getDSGraph(*cs.getCaller())
->getNodeForValue(cs.getCalledValue()).getNode();
assert (N && "CallTarget: findIndTargets: No DSNode!");
if (!N->isIncompleteNode() && !N->isExternalNode() && IndMap[cs].size()) {
CompleteSites.insert(cs);
++CompleteInd;
}
if (!N->isIncompleteNode() && !N->isExternalNode() && !IndMap[cs].size()) {
++CompleteEmpty;
DEBUG(errs() << "Call site empty: '"
<< cs.getInstruction()->getName()
<< "' In '"
<< cs.getInstruction()->getParent()->getParent()->getName()
<< "'\n");
}
}
} else {
++DirCall;
IndMap[cs].push_back(CF);
CompleteSites.insert(cs);
}
}
}