/// HandleInlinedInvoke - If we inlined an invoke site, we need to convert calls
/// in the body of the inlined function into invokes and turn unwind
/// instructions into branches to the invoke unwind dest.
///
/// II is the invoke instruction being inlined. FirstNewBlock is the first
/// block of the inlined code (the last block is the end of the function),
/// and InlineCodeInfo is information about the code that got inlined.
static void HandleInlinedInvoke(InvokeInst *II, BasicBlock *FirstNewBlock,
ClonedCodeInfo &InlinedCodeInfo) {
BasicBlock *InvokeDest = II->getUnwindDest();
SmallVector<Value*, 8> InvokeDestPHIValues;
// If there are PHI nodes in the unwind destination block, we need to
// keep track of which values came into them from this invoke, then remove
// the entry for this block.
BasicBlock *InvokeBlock = II->getParent();
for (BasicBlock::iterator I = InvokeDest->begin(); isa<PHINode>(I); ++I) {
PHINode *PN = cast<PHINode>(I);
// Save the value to use for this edge.
InvokeDestPHIValues.push_back(PN->getIncomingValueForBlock(InvokeBlock));
}
Function *Caller = FirstNewBlock->getParent();
// The inlined code is currently at the end of the function, scan from the
// start of the inlined code to its end, checking for stuff we need to
// rewrite. If the code doesn't have calls or unwinds, we know there is
// nothing to rewrite.
if (!InlinedCodeInfo.ContainsCalls && !InlinedCodeInfo.ContainsUnwinds) {
// Now that everything is happy, we have one final detail. The PHI nodes in
// the exception destination block still have entries due to the original
// invoke instruction. Eliminate these entries (which might even delete the
// PHI node) now.
InvokeDest->removePredecessor(II->getParent());
return;
}
for (Function::iterator BB = FirstNewBlock, E = Caller->end(); BB != E; ++BB){
if (InlinedCodeInfo.ContainsCalls)
HandleCallsInBlockInlinedThroughInvoke(BB, InvokeDest,
InvokeDestPHIValues);
if (UnwindInst *UI = dyn_cast<UnwindInst>(BB->getTerminator())) {
// An UnwindInst requires special handling when it gets inlined into an
// invoke site. Once this happens, we know that the unwind would cause
// a control transfer to the invoke exception destination, so we can
// transform it into a direct branch to the exception destination.
BranchInst::Create(InvokeDest, UI);
// Delete the unwind instruction!
UI->eraseFromParent();
// Update any PHI nodes in the exceptional block to indicate that
// there is now a new entry in them.
unsigned i = 0;
for (BasicBlock::iterator I = InvokeDest->begin();
isa<PHINode>(I); ++I, ++i) {
PHINode *PN = cast<PHINode>(I);
PN->addIncoming(InvokeDestPHIValues[i], BB);
}
}
}
// Now that everything is happy, we have one final detail. The PHI nodes in
// the exception destination block still have entries due to the original
// invoke instruction. Eliminate these entries (which might even delete the
// PHI node) now.
InvokeDest->removePredecessor(II->getParent());
}
// InlineFunction - This function inlines the called function into the basic
// block of the caller. This returns false if it is not possible to inline this
// call. The program is still in a well defined state if this occurs though.
//
// Note that this only does one level of inlining. For example, if the
// instruction 'call B' is inlined, and 'B' calls 'C', then the call to 'C' now
// exists in the instruction stream. Similiarly this will inline a recursive
// function by one level.
//
bool llvm::InlineFunction(CallSite CS, InlineFunctionInfo &IFI) {
Instruction *TheCall = CS.getInstruction();
LLVMContext &Context = TheCall->getContext();
assert(TheCall->getParent() && TheCall->getParent()->getParent() &&
"Instruction not in function!");
// If IFI has any state in it, zap it before we fill it in.
IFI.reset();
const Function *CalledFunc = CS.getCalledFunction();
if (CalledFunc == 0 || // Can't inline external function or indirect
CalledFunc->isDeclaration() || // call, or call to a vararg function!
CalledFunc->getFunctionType()->isVarArg()) return false;
// If the call to the callee is not a tail call, we must clear the 'tail'
// flags on any calls that we inline.
bool MustClearTailCallFlags =
!(isa<CallInst>(TheCall) && cast<CallInst>(TheCall)->isTailCall());
// If the call to the callee cannot throw, set the 'nounwind' flag on any
// calls that we inline.
bool MarkNoUnwind = CS.doesNotThrow();
BasicBlock *OrigBB = TheCall->getParent();
Function *Caller = OrigBB->getParent();
// GC poses two hazards to inlining, which only occur when the callee has GC:
// 1. If the caller has no GC, then the callee's GC must be propagated to the
// caller.
// 2. If the caller has a differing GC, it is invalid to inline.
if (CalledFunc->hasGC()) {
if (!Caller->hasGC())
Caller->setGC(CalledFunc->getGC());
else if (CalledFunc->getGC() != Caller->getGC())
return false;
}
// Get an iterator to the last basic block in the function, which will have
// the new function inlined after it.
//
Function::iterator LastBlock = &Caller->back();
// Make sure to capture all of the return instructions from the cloned
// function.
SmallVector<ReturnInst*, 8> Returns;
ClonedCodeInfo InlinedFunctionInfo;
Function::iterator FirstNewBlock;
{ // Scope to destroy VMap after cloning.
ValueMap<const Value*, Value*> VMap;
assert(CalledFunc->arg_size() == CS.arg_size() &&
"No varargs calls can be inlined!");
// Calculate the vector of arguments to pass into the function cloner, which
// matches up the formal to the actual argument values.
CallSite::arg_iterator AI = CS.arg_begin();
unsigned ArgNo = 0;
for (Function::const_arg_iterator I = CalledFunc->arg_begin(),
E = CalledFunc->arg_end(); I != E; ++I, ++AI, ++ArgNo) {
Value *ActualArg = *AI;
// When byval arguments actually inlined, we need to make the copy implied
// by them explicit. However, we don't do this if the callee is readonly
// or readnone, because the copy would be unneeded: the callee doesn't
// modify the struct.
if (CalledFunc->paramHasAttr(ArgNo+1, Attribute::ByVal) &&
!CalledFunc->onlyReadsMemory()) {
const Type *AggTy = cast<PointerType>(I->getType())->getElementType();
const Type *VoidPtrTy =
Type::getInt8PtrTy(Context);
// Create the alloca. If we have TargetData, use nice alignment.
unsigned Align = 1;
if (IFI.TD) Align = IFI.TD->getPrefTypeAlignment(AggTy);
Value *NewAlloca = new AllocaInst(AggTy, 0, Align,
I->getName(),
&*Caller->begin()->begin());
// Emit a memcpy.
const Type *Tys[3] = {VoidPtrTy, VoidPtrTy, Type::getInt64Ty(Context)};
Function *MemCpyFn = Intrinsic::getDeclaration(Caller->getParent(),
Intrinsic::memcpy,
Tys, 3);
Value *DestCast = new BitCastInst(NewAlloca, VoidPtrTy, "tmp", TheCall);
Value *SrcCast = new BitCastInst(*AI, VoidPtrTy, "tmp", TheCall);
Value *Size;
if (IFI.TD == 0)
Size = ConstantExpr::getSizeOf(AggTy);
else
Size = ConstantInt::get(Type::getInt64Ty(Context),
IFI.TD->getTypeStoreSize(AggTy));
// Always generate a memcpy of alignment 1 here because we don't know
// the alignment of the src pointer. Other optimizations can infer
// better alignment.
Value *CallArgs[] = {
DestCast, SrcCast, Size,
ConstantInt::get(Type::getInt32Ty(Context), 1),
ConstantInt::get(Type::getInt1Ty(Context), 0)
};
CallInst *TheMemCpy =
CallInst::Create(MemCpyFn, CallArgs, CallArgs+5, "", TheCall);
// If we have a call graph, update it.
if (CallGraph *CG = IFI.CG) {
CallGraphNode *MemCpyCGN = CG->getOrInsertFunction(MemCpyFn);
CallGraphNode *CallerNode = (*CG)[Caller];
CallerNode->addCalledFunction(TheMemCpy, MemCpyCGN);
}
// Uses of the argument in the function should use our new alloca
// instead.
ActualArg = NewAlloca;
// Calls that we inline may use the new alloca, so we need to clear
// their 'tail' flags.
MustClearTailCallFlags = true;
}
VMap[I] = ActualArg;
}
// We want the inliner to prune the code as it copies. We would LOVE to
// have no dead or constant instructions leftover after inlining occurs
// (which can happen, e.g., because an argument was constant), but we'll be
// happy with whatever the cloner can do.
CloneAndPruneFunctionInto(Caller, CalledFunc, VMap, Returns, ".i",
&InlinedFunctionInfo, IFI.TD, TheCall);
// Remember the first block that is newly cloned over.
FirstNewBlock = LastBlock; ++FirstNewBlock;
// Update the callgraph if requested.
if (IFI.CG)
UpdateCallGraphAfterInlining(CS, FirstNewBlock, VMap, IFI);
}
// If there are any alloca instructions in the block that used to be the entry
// block for the callee, move them to the entry block of the caller. First
// calculate which instruction they should be inserted before. We insert the
// instructions at the end of the current alloca list.
//
{
BasicBlock::iterator InsertPoint = Caller->begin()->begin();
for (BasicBlock::iterator I = FirstNewBlock->begin(),
E = FirstNewBlock->end(); I != E; ) {
AllocaInst *AI = dyn_cast<AllocaInst>(I++);
if (AI == 0) continue;
// If the alloca is now dead, remove it. This often occurs due to code
// specialization.
if (AI->use_empty()) {
AI->eraseFromParent();
continue;
}
if (!isa<Constant>(AI->getArraySize()))
continue;
// Keep track of the static allocas that we inline into the caller if the
// StaticAllocas pointer is non-null.
IFI.StaticAllocas.push_back(AI);
// Scan for the block of allocas that we can move over, and move them
// all at once.
while (isa<AllocaInst>(I) &&
isa<Constant>(cast<AllocaInst>(I)->getArraySize())) {
IFI.StaticAllocas.push_back(cast<AllocaInst>(I));
++I;
}
// Transfer all of the allocas over in a block. Using splice means
// that the instructions aren't removed from the symbol table, then
// reinserted.
Caller->getEntryBlock().getInstList().splice(InsertPoint,
FirstNewBlock->getInstList(),
AI, I);
}
}
// If the inlined code contained dynamic alloca instructions, wrap the inlined
// code with llvm.stacksave/llvm.stackrestore intrinsics.
if (InlinedFunctionInfo.ContainsDynamicAllocas) {
Module *M = Caller->getParent();
// Get the two intrinsics we care about.
Function *StackSave = Intrinsic::getDeclaration(M, Intrinsic::stacksave);
Function *StackRestore=Intrinsic::getDeclaration(M,Intrinsic::stackrestore);
// If we are preserving the callgraph, add edges to the stacksave/restore
// functions for the calls we insert.
CallGraphNode *StackSaveCGN = 0, *StackRestoreCGN = 0, *CallerNode = 0;
if (CallGraph *CG = IFI.CG) {
StackSaveCGN = CG->getOrInsertFunction(StackSave);
StackRestoreCGN = CG->getOrInsertFunction(StackRestore);
CallerNode = (*CG)[Caller];
}
// Insert the llvm.stacksave.
CallInst *SavedPtr = CallInst::Create(StackSave, "savedstack",
FirstNewBlock->begin());
if (IFI.CG) CallerNode->addCalledFunction(SavedPtr, StackSaveCGN);
// Insert a call to llvm.stackrestore before any return instructions in the
// inlined function.
for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
CallInst *CI = CallInst::Create(StackRestore, SavedPtr, "", Returns[i]);
if (IFI.CG) CallerNode->addCalledFunction(CI, StackRestoreCGN);
}
// Count the number of StackRestore calls we insert.
unsigned NumStackRestores = Returns.size();
// If we are inlining an invoke instruction, insert restores before each
// unwind. These unwinds will be rewritten into branches later.
if (InlinedFunctionInfo.ContainsUnwinds && isa<InvokeInst>(TheCall)) {
for (Function::iterator BB = FirstNewBlock, E = Caller->end();
BB != E; ++BB)
if (UnwindInst *UI = dyn_cast<UnwindInst>(BB->getTerminator())) {
CallInst *CI = CallInst::Create(StackRestore, SavedPtr, "", UI);
if (IFI.CG) CallerNode->addCalledFunction(CI, StackRestoreCGN);
++NumStackRestores;
}
}
}
// If we are inlining tail call instruction through a call site that isn't
// marked 'tail', we must remove the tail marker for any calls in the inlined
// code. Also, calls inlined through a 'nounwind' call site should be marked
// 'nounwind'.
if (InlinedFunctionInfo.ContainsCalls &&
(MustClearTailCallFlags || MarkNoUnwind)) {
for (Function::iterator BB = FirstNewBlock, E = Caller->end();
BB != E; ++BB)
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
if (CallInst *CI = dyn_cast<CallInst>(I)) {
if (MustClearTailCallFlags)
CI->setTailCall(false);
if (MarkNoUnwind)
CI->setDoesNotThrow();
}
}
// If we are inlining through a 'nounwind' call site then any inlined 'unwind'
// instructions are unreachable.
if (InlinedFunctionInfo.ContainsUnwinds && MarkNoUnwind)
for (Function::iterator BB = FirstNewBlock, E = Caller->end();
BB != E; ++BB) {
TerminatorInst *Term = BB->getTerminator();
if (isa<UnwindInst>(Term)) {
new UnreachableInst(Context, Term);
BB->getInstList().erase(Term);
}
}
// If we are inlining for an invoke instruction, we must make sure to rewrite
// any inlined 'unwind' instructions into branches to the invoke exception
// destination, and call instructions into invoke instructions.
if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall))
HandleInlinedInvoke(II, FirstNewBlock, InlinedFunctionInfo);
// If we cloned in _exactly one_ basic block, and if that block ends in a
// return instruction, we splice the body of the inlined callee directly into
// the calling basic block.
if (Returns.size() == 1 && std::distance(FirstNewBlock, Caller->end()) == 1) {
// Move all of the instructions right before the call.
OrigBB->getInstList().splice(TheCall, FirstNewBlock->getInstList(),
FirstNewBlock->begin(), FirstNewBlock->end());
// Remove the cloned basic block.
Caller->getBasicBlockList().pop_back();
// If the call site was an invoke instruction, add a branch to the normal
// destination.
if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall))
BranchInst::Create(II->getNormalDest(), TheCall);
// If the return instruction returned a value, replace uses of the call with
// uses of the returned value.
if (!TheCall->use_empty()) {
ReturnInst *R = Returns[0];
if (TheCall == R->getReturnValue())
TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
else
TheCall->replaceAllUsesWith(R->getReturnValue());
}
// Since we are now done with the Call/Invoke, we can delete it.
TheCall->eraseFromParent();
// Since we are now done with the return instruction, delete it also.
Returns[0]->eraseFromParent();
// We are now done with the inlining.
return true;
}
// Otherwise, we have the normal case, of more than one block to inline or
// multiple return sites.
// We want to clone the entire callee function into the hole between the
// "starter" and "ender" blocks. How we accomplish this depends on whether
// this is an invoke instruction or a call instruction.
BasicBlock *AfterCallBB;
if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall)) {
// Add an unconditional branch to make this look like the CallInst case...
BranchInst *NewBr = BranchInst::Create(II->getNormalDest(), TheCall);
// Split the basic block. This guarantees that no PHI nodes will have to be
// updated due to new incoming edges, and make the invoke case more
// symmetric to the call case.
AfterCallBB = OrigBB->splitBasicBlock(NewBr,
CalledFunc->getName()+".exit");
} else { // It's a call
// If this is a call instruction, we need to split the basic block that
// the call lives in.
//
AfterCallBB = OrigBB->splitBasicBlock(TheCall,
CalledFunc->getName()+".exit");
}
// Change the branch that used to go to AfterCallBB to branch to the first
// basic block of the inlined function.
//
TerminatorInst *Br = OrigBB->getTerminator();
assert(Br && Br->getOpcode() == Instruction::Br &&
"splitBasicBlock broken!");
Br->setOperand(0, FirstNewBlock);
// Now that the function is correct, make it a little bit nicer. In
// particular, move the basic blocks inserted from the end of the function
// into the space made by splitting the source basic block.
Caller->getBasicBlockList().splice(AfterCallBB, Caller->getBasicBlockList(),
FirstNewBlock, Caller->end());
// Handle all of the return instructions that we just cloned in, and eliminate
// any users of the original call/invoke instruction.
const Type *RTy = CalledFunc->getReturnType();
if (Returns.size() > 1) {
// The PHI node should go at the front of the new basic block to merge all
// possible incoming values.
PHINode *PHI = 0;
if (!TheCall->use_empty()) {
PHI = PHINode::Create(RTy, TheCall->getName(),
AfterCallBB->begin());
// Anything that used the result of the function call should now use the
// PHI node as their operand.
TheCall->replaceAllUsesWith(PHI);
}
// Loop over all of the return instructions adding entries to the PHI node
// as appropriate.
if (PHI) {
for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
ReturnInst *RI = Returns[i];
assert(RI->getReturnValue()->getType() == PHI->getType() &&
"Ret value not consistent in function!");
PHI->addIncoming(RI->getReturnValue(), RI->getParent());
}
// Now that we inserted the PHI, check to see if it has a single value
// (e.g. all the entries are the same or undef). If so, remove the PHI so
// it doesn't block other optimizations.
if (Value *V = PHI->hasConstantValue()) {
PHI->replaceAllUsesWith(V);
PHI->eraseFromParent();
}
}
// Add a branch to the merge points and remove return instructions.
for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
ReturnInst *RI = Returns[i];
BranchInst::Create(AfterCallBB, RI);
RI->eraseFromParent();
}
} else if (!Returns.empty()) {
// Otherwise, if there is exactly one return value, just replace anything
// using the return value of the call with the computed value.
if (!TheCall->use_empty()) {
if (TheCall == Returns[0]->getReturnValue())
TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
else
TheCall->replaceAllUsesWith(Returns[0]->getReturnValue());
}
// Splice the code from the return block into the block that it will return
// to, which contains the code that was after the call.
BasicBlock *ReturnBB = Returns[0]->getParent();
AfterCallBB->getInstList().splice(AfterCallBB->begin(),
ReturnBB->getInstList());
// Update PHI nodes that use the ReturnBB to use the AfterCallBB.
ReturnBB->replaceAllUsesWith(AfterCallBB);
// Delete the return instruction now and empty ReturnBB now.
Returns[0]->eraseFromParent();
ReturnBB->eraseFromParent();
} else if (!TheCall->use_empty()) {
// No returns, but something is using the return value of the call. Just
// nuke the result.
TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
}
// Since we are now done with the Call/Invoke, we can delete it.
TheCall->eraseFromParent();
// We should always be able to fold the entry block of the function into the
// single predecessor of the block...
assert(cast<BranchInst>(Br)->isUnconditional() && "splitBasicBlock broken!");
BasicBlock *CalleeEntry = cast<BranchInst>(Br)->getSuccessor(0);
// Splice the code entry block into calling block, right before the
// unconditional branch.
OrigBB->getInstList().splice(Br, CalleeEntry->getInstList());
CalleeEntry->replaceAllUsesWith(OrigBB); // Update PHI nodes
// Remove the unconditional branch.
OrigBB->getInstList().erase(Br);
// Now we can remove the CalleeEntry block, which is now empty.
Caller->getBasicBlockList().erase(CalleeEntry);
return true;
}
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<ResumeInst>(BB->getTerminator())) {
// Uses unwind / resume!
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) {
AttrBuilder NewAttributes;
if (!SCCMightUnwind)
NewAttributes.addAttribute(Attribute::NoUnwind);
if (!SCCMightReturn)
NewAttributes.addAttribute(Attribute::NoReturn);
Function *F = (*I)->getFunction();
const AttributeSet &PAL = F->getAttributes().getFnAttributes();
const AttributeSet &NPAL = AttributeSet::get(
F->getContext(), AttributeSet::FunctionIndex, NewAttributes);
if (PAL != NPAL) {
MadeChange = true;
F->addAttributes(AttributeSet::FunctionIndex, 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;
}
bool LoaderPass::runOnModule(Module &M) {
ProfileInfoLoader PIL("profile-loader", Filename, M);
EdgeInformation.clear();
std::vector<unsigned> ECs = PIL.getRawEdgeCounts();
if (ECs.size() > 0) {
unsigned ei = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
if (ei < ECs.size())
EdgeInformation[F][ProfileInfo::getEdge(0, &F->getEntryBlock())] +=
ECs[ei++];
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
// 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 (ei < ECs.size())
EdgeInformation[F][ProfileInfo::getEdge(BB, TI->getSuccessor(s))] +=
ECs[ei++];
}
}
}
if (ei != ECs.size()) {
cerr << "WARNING: profile information is inconsistent with "
<< "the current program!\n";
}
}
BlockInformation.clear();
std::vector<unsigned> BCs = PIL.getRawBlockCounts();
if (BCs.size() > 0) {
unsigned bi = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
if (bi < BCs.size())
BlockInformation[F][BB] = BCs[bi++];
}
if (bi != BCs.size()) {
cerr << "WARNING: profile information is inconsistent with "
<< "the current program!\n";
}
}
FunctionInformation.clear();
std::vector<unsigned> FCs = PIL.getRawFunctionCounts();
if (FCs.size() > 0) {
unsigned fi = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
if (fi < FCs.size())
FunctionInformation[F] = FCs[fi++];
}
if (fi != FCs.size()) {
cerr << "WARNING: profile information is inconsistent with "
<< "the current program!\n";
}
}
return false;
}
// the verifier iterates through each path to gather the total
// number of edge frequencies
bool PathProfileVerifier::runOnModule (Module &M) {
PathProfileInfo& pathProfileInfo = getAnalysis<PathProfileInfo>();
// setup a data structure to map path edges which index an
// array of edge counters
NestedBlockToIndexMap arrayMap;
unsigned i = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
arrayMap[0][F->begin()][0] = i++;
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
TerminatorInst *TI = BB->getTerminator();
unsigned duplicate = 0;
BasicBlock* prev = 0;
for (unsigned s = 0, e = TI->getNumSuccessors(); s != e;
prev = TI->getSuccessor(s), ++s) {
if (prev == TI->getSuccessor(s))
duplicate++;
else duplicate = 0;
arrayMap[BB][TI->getSuccessor(s)][duplicate] = i++;
}
}
}
std::vector<unsigned> edgeArray(i);
// iterate through each path and increment the edge counters as needed
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
pathProfileInfo.setCurrentFunction(F);
DEBUG(dbgs() << "function '" << F->getName() << "' ran "
<< pathProfileInfo.pathsRun()
<< "/" << pathProfileInfo.getPotentialPathCount()
<< " potential paths\n");
for( ProfilePathIterator nextPath = pathProfileInfo.pathBegin(),
endPath = pathProfileInfo.pathEnd();
nextPath != endPath; nextPath++ ) {
ProfilePath* currentPath = nextPath->second;
ProfilePathEdgeVector* pev = currentPath->getPathEdges();
DEBUG(dbgs () << "path #" << currentPath->getNumber() << ": "
<< currentPath->getCount() << "\n");
// setup the entry edge (normally path profiling doesn't care about this)
if (currentPath->getFirstBlockInPath() == &F->getEntryBlock())
edgeArray[arrayMap[0][currentPath->getFirstBlockInPath()][0]]
+= currentPath->getCount();
for( ProfilePathEdgeIterator nextEdge = pev->begin(),
endEdge = pev->end(); nextEdge != endEdge; nextEdge++ ) {
if (nextEdge != pev->begin())
DEBUG(dbgs() << " :: ");
BasicBlock* source = nextEdge->getSource();
BasicBlock* target = nextEdge->getTarget();
unsigned duplicateNumber = nextEdge->getDuplicateNumber();
DEBUG(dbgs () << source->getNameStr() << " --{" << duplicateNumber
<< "}--> " << target->getNameStr());
// Ensure all the referenced edges exist
// TODO: make this a separate function
if( !arrayMap.count(source) ) {
errs() << " error [" << F->getNameStr() << "()]: source '"
<< source->getNameStr()
<< "' does not exist in the array map.\n";
} else if( !arrayMap[source].count(target) ) {
errs() << " error [" << F->getNameStr() << "()]: target '"
<< target->getNameStr()
<< "' does not exist in the array map.\n";
} else if( !arrayMap[source][target].count(duplicateNumber) ) {
errs() << " error [" << F->getNameStr() << "()]: edge "
<< source->getNameStr() << " -> " << target->getNameStr()
<< " duplicate number " << duplicateNumber
<< " does not exist in the array map.\n";
} else {
edgeArray[arrayMap[source][target][duplicateNumber]]
+= currentPath->getCount();
}
}
DEBUG(errs() << "\n");
delete pev;
}
}
std::string errorInfo;
std::string filename = EdgeProfileFilename;
// Open a handle to the file
FILE* edgeFile = fopen(filename.c_str(),"wb");
if (!edgeFile) {
errs() << "error: unable to open file '" << filename << "' for output.\n";
return false;
}
errs() << "Generating edge profile '" << filename << "' ...\n";
// write argument info
unsigned type = ArgumentInfo;
unsigned num = pathProfileInfo.argList.size();
int zeros = 0;
fwrite(&type,sizeof(unsigned),1,edgeFile);
fwrite(&num,sizeof(unsigned),1,edgeFile);
fwrite(pathProfileInfo.argList.c_str(),1,num,edgeFile);
if (num&3)
fwrite(&zeros, 1, 4-(num&3), edgeFile);
type = EdgeInfo;
num = edgeArray.size();
fwrite(&type,sizeof(unsigned),1,edgeFile);
fwrite(&num,sizeof(unsigned),1,edgeFile);
// write each edge to the file
for( std::vector<unsigned>::iterator s = edgeArray.begin(),
e = edgeArray.end(); s != e; s++)
fwrite(&*s, sizeof (unsigned), 1, edgeFile);
fclose (edgeFile);
return true;
}
/// NormalizeLandingPads - Normalize and discover landing pads, noting them
/// in the LandingPads set. A landing pad is normal if the only CFG edges
/// that end at it are unwind edges from invoke instructions. If we inlined
/// through an invoke we could have a normal branch from the previous
/// unwind block through to the landing pad for the original invoke.
/// Abnormal landing pads are fixed up by redirecting all unwind edges to
/// a new basic block which falls through to the original.
bool DwarfEHPrepare::NormalizeLandingPads() {
bool Changed = false;
const MCAsmInfo *MAI = TM->getMCAsmInfo();
bool usingSjLjEH = MAI->getExceptionHandlingType() == ExceptionHandling::SjLj;
for (Function::iterator I = F->begin(), E = F->end(); I != E; ++I) {
TerminatorInst *TI = I->getTerminator();
if (!isa<InvokeInst>(TI))
continue;
BasicBlock *LPad = TI->getSuccessor(1);
// Skip landing pads that have already been normalized.
if (LandingPads.count(LPad))
continue;
// Check that only invoke unwind edges end at the landing pad.
bool OnlyUnwoundTo = true;
bool SwitchOK = usingSjLjEH;
for (pred_iterator PI = pred_begin(LPad), PE = pred_end(LPad);
PI != PE; ++PI) {
TerminatorInst *PT = (*PI)->getTerminator();
// The SjLj dispatch block uses a switch instruction. This is effectively
// an unwind edge, so we can disregard it here. There will only ever
// be one dispatch, however, so if there are multiple switches, one
// of them truly is a normal edge, not an unwind edge.
if (SwitchOK && isa<SwitchInst>(PT)) {
SwitchOK = false;
continue;
}
if (!isa<InvokeInst>(PT) || LPad == PT->getSuccessor(0)) {
OnlyUnwoundTo = false;
break;
}
}
if (OnlyUnwoundTo) {
// Only unwind edges lead to the landing pad. Remember the landing pad.
LandingPads.insert(LPad);
continue;
}
// At least one normal edge ends at the landing pad. Redirect the unwind
// edges to a new basic block which falls through into this one.
// Create the new basic block.
BasicBlock *NewBB = BasicBlock::Create(F->getContext(),
LPad->getName() + "_unwind_edge");
// Insert it into the function right before the original landing pad.
LPad->getParent()->getBasicBlockList().insert(LPad, NewBB);
// Redirect unwind edges from the original landing pad to NewBB.
for (pred_iterator PI = pred_begin(LPad), PE = pred_end(LPad); PI != PE; ) {
TerminatorInst *PT = (*PI++)->getTerminator();
if (isa<InvokeInst>(PT) && PT->getSuccessor(1) == LPad)
// Unwind to the new block.
PT->setSuccessor(1, NewBB);
}
// If there are any PHI nodes in LPad, we need to update them so that they
// merge incoming values from NewBB instead.
for (BasicBlock::iterator II = LPad->begin(); isa<PHINode>(II); ++II) {
PHINode *PN = cast<PHINode>(II);
pred_iterator PB = pred_begin(NewBB), PE = pred_end(NewBB);
// Check to see if all of the values coming in via unwind edges are the
// same. If so, we don't need to create a new PHI node.
Value *InVal = PN->getIncomingValueForBlock(*PB);
for (pred_iterator PI = PB; PI != PE; ++PI) {
if (PI != PB && InVal != PN->getIncomingValueForBlock(*PI)) {
InVal = 0;
break;
}
}
if (InVal == 0) {
// Different unwind edges have different values. Create a new PHI node
// in NewBB.
PHINode *NewPN = PHINode::Create(PN->getType(), PN->getName()+".unwind",
NewBB);
// Add an entry for each unwind edge, using the value from the old PHI.
for (pred_iterator PI = PB; PI != PE; ++PI)
NewPN->addIncoming(PN->getIncomingValueForBlock(*PI), *PI);
// Now use this new PHI as the common incoming value for NewBB in PN.
InVal = NewPN;
}
// Revector exactly one entry in the PHI node to come from NewBB
// and delete all other entries that come from unwind edges. If
// there are both normal and unwind edges from the same predecessor,
// this leaves an entry for the normal edge.
for (pred_iterator PI = PB; PI != PE; ++PI)
PN->removeIncomingValue(*PI);
PN->addIncoming(InVal, NewBB);
}
// Add a fallthrough from NewBB to the original landing pad.
BranchInst::Create(LPad, NewBB);
// Now update DominatorTree and DominanceFrontier analysis information.
if (DT)
DT->splitBlock(NewBB);
if (DF)
DF->splitBlock(NewBB);
// Remember the newly constructed landing pad. The original landing pad
// LPad is no longer a landing pad now that all unwind edges have been
// revectored to NewBB.
LandingPads.insert(NewBB);
++NumLandingPadsSplit;
Changed = true;
}
return Changed;
}
bool TailCallElim::ProcessReturningBlock(ReturnInst *Ret, BasicBlock *&OldEntry,
bool &TailCallsAreMarkedTail,
SmallVector<PHINode*, 8> &ArgumentPHIs,
bool CannotTailCallElimCallsMarkedTail) {
BasicBlock *BB = Ret->getParent();
Function *F = BB->getParent();
if (&BB->front() == Ret) // Make sure there is something before the ret...
return false;
// If the return is in the entry block, then making this transformation would
// turn infinite recursion into an infinite loop. This transformation is ok
// in theory, but breaks some code like:
// double fabs(double f) { return __builtin_fabs(f); } // a 'fabs' call
// disable this xform in this case, because the code generator will lower the
// call to fabs into inline code.
if (BB == &F->getEntryBlock())
return false;
// Scan backwards from the return, checking to see if there is a tail call in
// this block. If so, set CI to it.
CallInst *CI;
BasicBlock::iterator BBI = Ret;
while (1) {
CI = dyn_cast<CallInst>(BBI);
if (CI && CI->getCalledFunction() == F)
break;
if (BBI == BB->begin())
return false; // Didn't find a potential tail call.
--BBI;
}
// If this call is marked as a tail call, and if there are dynamic allocas in
// the function, we cannot perform this optimization.
if (CI->isTailCall() && CannotTailCallElimCallsMarkedTail)
return false;
// If we are introducing accumulator recursion to eliminate associative
// operations after the call instruction, this variable contains the initial
// value for the accumulator. If this value is set, we actually perform
// accumulator recursion elimination instead of simple tail recursion
// elimination.
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.
for (BBI = CI, ++BBI; &*BBI != Ret; ++BBI)
if (!CanMoveAboveCall(BBI, CI)) {
// If we can't move the instruction above the call, it might be because it
// is an associative operation that could be tranformed 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(Ret, CI))
return false;
// 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(),
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->getNumOperands()-1; i != e; ++i)
ArgumentPHIs[i]->addIncoming(CI->getOperand(i+1), 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.
PHINode *AccPN = PHINode::Create(AccRecInstr->getType(), "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 = pred_begin(OldEntry), PE = pred_end(OldEntry);
PI != PE; ++PI) {
if (*PI == &F->getEntryBlock())
AccPN->addIncoming(AccumulatorRecursionEliminationInitVal, *PI);
else
AccPN->addIncoming(AccPN, *PI);
}
// Add an incoming argument for the current block, which is computed by our
// associative 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);
// 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::Create(OldEntry, Ret);
BB->getInstList().erase(Ret); // Remove return.
BB->getInstList().erase(CI); // Remove call.
++NumEliminated;
return true;
}
bool LoaderPass::runOnModule(Module &M) {
ProfileInfoLoader PIL("profile-loader", Filename);
EdgeInformation.clear();
std::vector<unsigned> Counters = PIL.getRawEdgeCounts();
if (Counters.size() > 0) {
ReadCount = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
DEBUG(dbgs() << "Working on " << F->getName() << "\n");
readEdge(getEdge(0,&F->getEntryBlock()), Counters);
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
TerminatorInst *TI = BB->getTerminator();
for (unsigned s = 0, e = TI->getNumSuccessors(); s != e; ++s) {
readEdge(getEdge(BB,TI->getSuccessor(s)), Counters);
}
}
}
if (ReadCount != Counters.size()) {
errs() << "WARNING: profile information is inconsistent with "
<< "the current program!\n";
}
NumEdgesRead = ReadCount;
}
Counters = PIL.getRawOptimalEdgeCounts();
if (Counters.size() > 0) {
ReadCount = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
DEBUG(dbgs() << "Working on " << F->getName() << "\n");
readEdge(getEdge(0,&F->getEntryBlock()), Counters);
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
TerminatorInst *TI = BB->getTerminator();
if (TI->getNumSuccessors() == 0) {
readEdge(getEdge(BB,0), Counters);
}
for (unsigned s = 0, e = TI->getNumSuccessors(); s != e; ++s) {
readEdge(getEdge(BB,TI->getSuccessor(s)), Counters);
}
}
while (SpanningTree.size() > 0) {
unsigned size = SpanningTree.size();
BBisUnvisited.clear();
for (std::set<Edge>::iterator ei = SpanningTree.begin(),
ee = SpanningTree.end(); ei != ee; ++ei) {
BBisUnvisited.insert(ei->first);
BBisUnvisited.insert(ei->second);
}
while (BBisUnvisited.size() > 0) {
recurseBasicBlock(*BBisUnvisited.begin());
}
if (SpanningTree.size() == size) {
DEBUG(dbgs()<<"{");
for (std::set<Edge>::iterator ei = SpanningTree.begin(),
ee = SpanningTree.end(); ei != ee; ++ei) {
DEBUG(dbgs()<< *ei <<",");
}
assert(0 && "No edge calculated!");
}
}
}
if (ReadCount != Counters.size()) {
errs() << "WARNING: profile information is inconsistent with "
<< "the current program!\n";
}
NumEdgesRead = ReadCount;
}
BlockInformation.clear();
Counters = PIL.getRawBlockCounts();
if (Counters.size() > 0) {
ReadCount = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
if (ReadCount < Counters.size())
// Here the data realm changes from the unsigned of the file to the
// double of the ProfileInfo. This conversion is save because we know
// that everything thats representable in unsinged is also
// representable in double.
BlockInformation[F][BB] = (double)Counters[ReadCount++];
}
if (ReadCount != Counters.size()) {
errs() << "WARNING: profile information is inconsistent with "
<< "the current program!\n";
}
}
FunctionInformation.clear();
Counters = PIL.getRawFunctionCounts();
if (Counters.size() > 0) {
ReadCount = 0;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->isDeclaration()) continue;
if (ReadCount < Counters.size())
// Here the data realm changes from the unsigned of the file to the
// double of the ProfileInfo. This conversion is save because we know
// that everything thats representable in unsinged is also
// representable in double.
FunctionInformation[F] = (double)Counters[ReadCount++];
}
if (ReadCount != Counters.size()) {
errs() << "WARNING: profile information is inconsistent with "
<< "the current program!\n";
}
}
return false;
}
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()))
MadeChange |= ProcessReturningBlock(Ret, OldEntry, TailCallsAreMarkedTail,
ArgumentPHIs,CannotTCETailMarkedCall);
// 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 = PN->hasConstantValue()) {
PN->replaceAllUsesWith(PNV);
PN->eraseFromParent();
}
}
}
// Finally, if this function contains no non-escaping allocas, mark all calls
// in the function as eligible for tail calls (there is no stack memory for
// them to access).
if (!FunctionContainsEscapingAllocas)
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 ReduceCrashingBlocks::TestBlocks(std::vector<const BasicBlock*> &BBs) {
// Clone the program to try hacking it apart...
ValueToValueMapTy VMap;
Module *M = CloneModule(BD.getProgram(), VMap);
// Convert list to set for fast lookup...
SmallPtrSet<BasicBlock*, 8> Blocks;
for (unsigned i = 0, e = BBs.size(); i != e; ++i)
Blocks.insert(cast<BasicBlock>(VMap[BBs[i]]));
outs() << "Checking for crash with only these blocks:";
unsigned NumPrint = Blocks.size();
if (NumPrint > 10) NumPrint = 10;
for (unsigned i = 0, e = NumPrint; i != e; ++i)
outs() << " " << BBs[i]->getName();
if (NumPrint < Blocks.size())
outs() << "... <" << Blocks.size() << " total>";
outs() << ": ";
// Loop over and delete any hack up any blocks that are not listed...
for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
for (Function::iterator BB = I->begin(), E = I->end(); BB != E; ++BB)
if (!Blocks.count(BB) && BB->getTerminator()->getNumSuccessors()) {
// Loop over all of the successors of this block, deleting any PHI nodes
// that might include it.
for (succ_iterator SI = succ_begin(BB), E = succ_end(BB); SI != E; ++SI)
(*SI)->removePredecessor(BB);
TerminatorInst *BBTerm = BB->getTerminator();
if (!BB->getTerminator()->getType()->isVoidTy())
BBTerm->replaceAllUsesWith(Constant::getNullValue(BBTerm->getType()));
// Replace the old terminator instruction.
BB->getInstList().pop_back();
new UnreachableInst(BB->getContext(), BB);
}
// The CFG Simplifier pass may delete one of the basic blocks we are
// interested in. If it does we need to take the block out of the list. Make
// a "persistent mapping" by turning basic blocks into <function, name> pairs.
// This won't work well if blocks are unnamed, but that is just the risk we
// have to take.
std::vector<std::pair<std::string, std::string> > BlockInfo;
for (SmallPtrSet<BasicBlock*, 8>::iterator I = Blocks.begin(),
E = Blocks.end(); I != E; ++I)
BlockInfo.push_back(std::make_pair((*I)->getParent()->getName(),
(*I)->getName()));
// Now run the CFG simplify pass on the function...
std::vector<std::string> Passes;
Passes.push_back("simplifycfg");
Passes.push_back("verify");
Module *New = BD.runPassesOn(M, Passes);
delete M;
if (!New) {
errs() << "simplifycfg failed!\n";
exit(1);
}
M = New;
// Try running on the hacked up program...
if (TestFn(BD, M)) {
BD.setNewProgram(M); // It crashed, keep the trimmed version...
// Make sure to use basic block pointers that point into the now-current
// module, and that they don't include any deleted blocks.
BBs.clear();
const ValueSymbolTable &GST = M->getValueSymbolTable();
for (unsigned i = 0, e = BlockInfo.size(); i != e; ++i) {
Function *F = cast<Function>(GST.lookup(BlockInfo[i].first));
ValueSymbolTable &ST = F->getValueSymbolTable();
Value* V = ST.lookup(BlockInfo[i].second);
if (V && V->getType() == Type::getLabelTy(V->getContext()))
BBs.push_back(cast<BasicBlock>(V));
}
return true;
}
delete M; // It didn't crash, try something else.
return false;
}
void StructFieldReach::InitBeforeAfterSet(Function * pF)
{
for(Function::iterator BB = pF->begin(); BB != pF->end(); BB++)
{
for(BasicBlock::iterator II = BB->begin(); II != BB->end(); II ++)
{
vector<Instruction *> vecPredInst;
this->InstPredInstVecMapping[II] = vecPredInst;
vector<Instruction *> vecSuccInst;
this->InstSuccInstVecMapping[II] = vecSuccInst;
set<MemFootPrint *> beforeSet;
this->InstBeforeSetMapping[II] = beforeSet;
set<MemFootPrint *> afterSet;
this->InstAfterSetMapping[II] = afterSet;
set<MemFootPrint *> beforeExtendSet;
this->InstBeforeExtendSetMapping[II] = beforeExtendSet;
set<MemFootPrint *> afterExtendSet;
this->InstAfterExtendSetMapping[II] = afterExtendSet;
}
}
for(Function::iterator BB = pF->begin(); BB != pF->end(); BB ++)
{
BasicBlock::iterator itFirstInst = BB->begin();
if(itFirstInst == BB->end() )
{
continue;
}
for(pred_iterator pred = pred_begin(BB); pred != pred_end(BB); pred++ )
{
this->InstPredInstVecMapping[itFirstInst].push_back((*pred)->getTerminator());
}
BasicBlock::iterator itLastInst = BB->getTerminator();
for(succ_iterator succ = succ_begin(BB); succ != succ_end(BB); succ++ )
{
if((*succ)->begin() == (*succ)->end())
{
continue;
}
this->InstSuccInstVecMapping[itLastInst].push_back((*succ)->begin());
}
for(BasicBlock::iterator II = BB->begin(); II != BB->end(); II ++)
{
BasicBlock::iterator ITmp;
if(II != BB->begin())
{
ITmp = II;
ITmp--;
this->InstPredInstVecMapping[II].push_back(ITmp);
}
ITmp = II;
ITmp ++;
if(ITmp != BB->end() )
{
this->InstSuccInstVecMapping[II].push_back(ITmp);
}
}
}
}
/// Tests whether a function is "malloc-like".
///
/// A function is "malloc-like" if it returns either null or a pointer that
/// doesn't alias any other pointer visible to the caller.
static bool isFunctionMallocLike(Function *F, const SCCNodeSet &SCCNodes) {
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:
case Instruction::AddrSpaceCast:
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 (Value *IncValue : PN->incoming_values())
FlowsToReturn.insert(IncValue);
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;
}
/// setupEntryBlockAndCallSites - Setup the entry block by creating and filling
/// the function context and marking the call sites with the appropriate
/// values. These values are used by the DWARF EH emitter.
bool SjLjEHPrepare::setupEntryBlockAndCallSites(Function &F) {
SmallVector<ReturnInst *, 16> Returns;
SmallVector<InvokeInst *, 16> Invokes;
SmallSetVector<LandingPadInst *, 16> LPads;
// Look through the terminators of the basic blocks to find invokes.
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB)
if (InvokeInst *II = dyn_cast<InvokeInst>(BB->getTerminator())) {
if (Function *Callee = II->getCalledFunction())
if (Callee->isIntrinsic() &&
Callee->getIntrinsicID() == Intrinsic::donothing) {
// Remove the NOP invoke.
BranchInst::Create(II->getNormalDest(), II);
II->eraseFromParent();
continue;
}
Invokes.push_back(II);
LPads.insert(II->getUnwindDest()->getLandingPadInst());
} else if (ReturnInst *RI = dyn_cast<ReturnInst>(BB->getTerminator())) {
Returns.push_back(RI);
}
if (Invokes.empty())
return false;
NumInvokes += Invokes.size();
lowerIncomingArguments(F);
lowerAcrossUnwindEdges(F, Invokes);
Value *FuncCtx =
setupFunctionContext(F, makeArrayRef(LPads.begin(), LPads.end()));
BasicBlock *EntryBB = F.begin();
IRBuilder<> Builder(EntryBB->getTerminator());
// Get a reference to the jump buffer.
Value *JBufPtr = Builder.CreateConstGEP2_32(FuncCtx, 0, 5, "jbuf_gep");
// Save the frame pointer.
Value *FramePtr = Builder.CreateConstGEP2_32(JBufPtr, 0, 0, "jbuf_fp_gep");
Value *Val = Builder.CreateCall(FrameAddrFn, Builder.getInt32(0), "fp");
Builder.CreateStore(Val, FramePtr, /*isVolatile=*/true);
// Save the stack pointer.
Value *StackPtr = Builder.CreateConstGEP2_32(JBufPtr, 0, 2, "jbuf_sp_gep");
Val = Builder.CreateCall(StackAddrFn, "sp");
Builder.CreateStore(Val, StackPtr, /*isVolatile=*/true);
// Call the setjmp instrinsic. It fills in the rest of the jmpbuf.
Value *SetjmpArg = Builder.CreateBitCast(JBufPtr, Builder.getInt8PtrTy());
Builder.CreateCall(BuiltinSetjmpFn, SetjmpArg);
// Store a pointer to the function context so that the back-end will know
// where to look for it.
Value *FuncCtxArg = Builder.CreateBitCast(FuncCtx, Builder.getInt8PtrTy());
Builder.CreateCall(FuncCtxFn, FuncCtxArg);
// At this point, we are all set up, update the invoke instructions to mark
// their call_site values.
for (unsigned I = 0, E = Invokes.size(); I != E; ++I) {
insertCallSiteStore(Invokes[I], I + 1);
ConstantInt *CallSiteNum =
ConstantInt::get(Type::getInt32Ty(F.getContext()), I + 1);
// Record the call site value for the back end so it stays associated with
// the invoke.
CallInst::Create(CallSiteFn, CallSiteNum, "", Invokes[I]);
}
// 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)) {
if (!CI->doesNotThrow())
insertCallSiteStore(CI, -1);
} else if (ResumeInst *RI = dyn_cast<ResumeInst>(I)) {
insertCallSiteStore(RI, -1);
}
// Register the function context and make sure it's known to not throw
CallInst *Register =
CallInst::Create(RegisterFn, FuncCtx, "", EntryBB->getTerminator());
Register->setDoesNotThrow();
// Following any allocas not in the entry block, update the saved SP in the
// jmpbuf to the new value.
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
if (BB == F.begin())
continue;
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
if (CallInst *CI = dyn_cast<CallInst>(I)) {
if (CI->getCalledFunction() != StackRestoreFn)
continue;
} else if (!isa<AllocaInst>(I)) {
continue;
}
Instruction *StackAddr = CallInst::Create(StackAddrFn, "sp");
StackAddr->insertAfter(I);
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, FuncCtx, "", Returns[I]);
return true;
}
