#include "llvm/Support/CallSite.h"
using namespace llvm;
-bool llvm::InlineFunction(CallInst *CI, CallGraph *CG, const TargetData *TD) {
- return InlineFunction(CallSite(CI), CG, TD);
+bool llvm::InlineFunction(CallInst *CI, CallGraph *CG, const TargetData *TD,
+ SmallVectorImpl<AllocaInst*> *StaticAllocas) {
+ return InlineFunction(CallSite(CI), CG, TD, StaticAllocas);
}
-bool llvm::InlineFunction(InvokeInst *II, CallGraph *CG, const TargetData *TD) {
- return InlineFunction(CallSite(II), CG, TD);
+bool llvm::InlineFunction(InvokeInst *II, CallGraph *CG, const TargetData *TD,
+ SmallVectorImpl<AllocaInst*> *StaticAllocas) {
+ return InlineFunction(CallSite(II), CG, TD, StaticAllocas);
}
+
+/// HandleCallsInBlockInlinedThroughInvoke - When we inline a basic block into
+/// an invoke, we have to turn all of the calls that can throw into
+/// invokes. This function analyze BB to see if there are any calls, and if so,
+/// it rewrites them to be invokes that jump to InvokeDest and fills in the PHI
+/// nodes in that block with the values specified in InvokeDestPHIValues.
+///
+static void HandleCallsInBlockInlinedThroughInvoke(BasicBlock *BB,
+ BasicBlock *InvokeDest,
+ const SmallVectorImpl<Value*> &InvokeDestPHIValues) {
+ for (BasicBlock::iterator BBI = BB->begin(), E = BB->end(); BBI != E; ) {
+ Instruction *I = BBI++;
+
+ // We only need to check for function calls: inlined invoke
+ // instructions require no special handling.
+ CallInst *CI = dyn_cast<CallInst>(I);
+ if (CI == 0) continue;
+
+ // If this call cannot unwind, don't convert it to an invoke.
+ if (CI->doesNotThrow())
+ continue;
+
+ // Convert this function call into an invoke instruction.
+ // First, split the basic block.
+ BasicBlock *Split = BB->splitBasicBlock(CI, CI->getName()+".noexc");
+
+ // Next, create the new invoke instruction, inserting it at the end
+ // of the old basic block.
+ SmallVector<Value*, 8> InvokeArgs(CI->op_begin()+1, CI->op_end());
+ InvokeInst *II =
+ InvokeInst::Create(CI->getCalledValue(), Split, InvokeDest,
+ InvokeArgs.begin(), InvokeArgs.end(),
+ CI->getName(), BB->getTerminator());
+ II->setCallingConv(CI->getCallingConv());
+ II->setAttributes(CI->getAttributes());
+
+ // Make sure that anything using the call now uses the invoke! This also
+ // updates the CallGraph if present.
+ CI->replaceAllUsesWith(II);
+
+ // Delete the unconditional branch inserted by splitBasicBlock
+ BB->getInstList().pop_back();
+ Split->getInstList().pop_front(); // Delete the original call
+
+ // 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)
+ cast<PHINode>(I)->addIncoming(InvokeDestPHIValues[i], BB);
+
+ // This basic block is now complete, the caller will continue scanning the
+ // next one.
+ return;
+ }
+}
+
+
/// 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.
/// 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,
- CallGraph *CG) {
+ ClonedCodeInfo &InlinedCodeInfo) {
BasicBlock *InvokeDest = II->getUnwindDest();
- std::vector<Value*> InvokeDestPHIValues;
+ 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 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 (InlinedCodeInfo.ContainsCalls || InlinedCodeInfo.ContainsUnwinds) {
- for (Function::iterator BB = FirstNewBlock, E = Caller->end();
- BB != E; ++BB) {
- if (InlinedCodeInfo.ContainsCalls) {
- for (BasicBlock::iterator BBI = BB->begin(), E = BB->end(); BBI != E; ){
- Instruction *I = BBI++;
-
- // We only need to check for function calls: inlined invoke
- // instructions require no special handling.
- if (!isa<CallInst>(I)) continue;
- CallInst *CI = cast<CallInst>(I);
-
- // If this call cannot unwind, don't convert it to an invoke.
- if (CI->doesNotThrow())
- continue;
-
- // Convert this function call into an invoke instruction.
- // First, split the basic block.
- BasicBlock *Split = BB->splitBasicBlock(CI, CI->getName()+".noexc");
-
- // Next, create the new invoke instruction, inserting it at the end
- // of the old basic block.
- SmallVector<Value*, 8> InvokeArgs(CI->op_begin()+1, CI->op_end());
- InvokeInst *II =
- InvokeInst::Create(CI->getCalledValue(), Split, InvokeDest,
- InvokeArgs.begin(), InvokeArgs.end(),
- CI->getName(), BB->getTerminator());
- II->setCallingConv(CI->getCallingConv());
- II->setAttributes(CI->getAttributes());
-
- // Make sure that anything using the call now uses the invoke!
- CI->replaceAllUsesWith(II);
-
- // Update the callgraph.
- if (CG) {
- // We should be able to do this:
- // (*CG)[Caller]->replaceCallSite(CI, II);
- // but that fails if the old call site isn't in the call graph,
- // which, because of LLVM bug 3601, it sometimes isn't.
- CallGraphNode *CGN = (*CG)[Caller];
- for (CallGraphNode::iterator NI = CGN->begin(), NE = CGN->end();
- NI != NE; ++NI) {
- if (NI->first == CI) {
- NI->first = II;
- break;
- }
- }
- }
-
- // Delete the unconditional branch inserted by splitBasicBlock
- BB->getInstList().pop_back();
- Split->getInstList().pop_front(); // Delete the original call
-
- // 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);
- }
-
- // This basic block is now complete, start scanning the next one.
- break;
- }
- }
-
- 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);
- }
+ // 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);
}
}
}
}
for (; I != E; ++I) {
- const Instruction *OrigCall = I->first.getInstruction();
+ const Value *OrigCall = I->first;
DenseMap<const Value*, Value*>::iterator VMI = ValueMap.find(OrigCall);
// Only copy the edge if the call was inlined!
- if (VMI != ValueMap.end() && VMI->second) {
- // If the call was inlined, but then constant folded, there is no edge to
- // add. Check for this case.
- if (Instruction *NewCall = dyn_cast<Instruction>(VMI->second))
- CallerNode->addCalledFunction(CallSite::get(NewCall), I->second);
- }
+ if (VMI == ValueMap.end() || VMI->second == 0)
+ continue;
+
+ // If the call was inlined, but then constant folded, there is no edge to
+ // add. Check for this case.
+ if (Instruction *NewCall = dyn_cast<Instruction>(VMI->second))
+ CallerNode->addCalledFunction(CallSite::get(NewCall), I->second);
}
+
// Update the call graph by deleting the edge from Callee to Caller. We must
// do this after the loop above in case Caller and Callee are the same.
CallerNode->removeCallEdgeFor(CS);
}
-/// findFnRegionEndMarker - This is a utility routine that is used by
-/// InlineFunction. Return llvm.dbg.region.end intrinsic that corresponds
-/// to the llvm.dbg.func.start of the function F. Otherwise return NULL.
-static const DbgRegionEndInst *findFnRegionEndMarker(const Function *F) {
-
- GlobalVariable *FnStart = NULL;
- const DbgRegionEndInst *FnEnd = NULL;
- for (Function::const_iterator FI = F->begin(), FE =F->end(); FI != FE; ++FI)
- for (BasicBlock::const_iterator BI = FI->begin(), BE = FI->end(); BI != BE;
- ++BI) {
- if (FnStart == NULL) {
- if (const DbgFuncStartInst *FSI = dyn_cast<DbgFuncStartInst>(BI)) {
- DISubprogram SP(cast<GlobalVariable>(FSI->getSubprogram()));
- assert (SP.isNull() == false && "Invalid llvm.dbg.func.start");
- if (SP.describes(F))
- FnStart = SP.getGV();
- }
- } else {
- if (const DbgRegionEndInst *REI = dyn_cast<DbgRegionEndInst>(BI))
- if (REI->getContext() == FnStart)
- FnEnd = REI;
- }
- }
- return FnEnd;
-}
-
// 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.
// exists in the instruction stream. Similiarly this will inline a recursive
// function by one level.
//
-bool llvm::InlineFunction(CallSite CS, CallGraph *CG, const TargetData *TD) {
+bool llvm::InlineFunction(CallSite CS, CallGraph *CG, const TargetData *TD,
+ SmallVectorImpl<AllocaInst*> *StaticAllocas) {
Instruction *TheCall = CS.getInstruction();
LLVMContext &Context = TheCall->getContext();
assert(TheCall->getParent() && TheCall->getParent()->getParent() &&
// Make sure to capture all of the return instructions from the cloned
// function.
- std::vector<ReturnInst*> Returns;
+ SmallVector<ReturnInst*, 8> Returns;
ClonedCodeInfo InlinedFunctionInfo;
Function::iterator FirstNewBlock;
if (CalledFunc->paramHasAttr(ArgNo+1, Attribute::ByVal) &&
!CalledFunc->onlyReadsMemory()) {
const Type *AggTy = cast<PointerType>(I->getType())->getElementType();
- const Type *VoidPtrTy = Context.getPointerTypeUnqual(Type::Int8Ty);
+ const Type *VoidPtrTy =
+ Type::getInt8PtrTy(Context);
// Create the alloca. If we have TargetData, use nice alignment.
unsigned Align = 1;
I->getName(),
&*Caller->begin()->begin());
// Emit a memcpy.
- const Type *Tys[] = { Type::Int64Ty };
+ const Type *Tys[] = { Type::getInt64Ty(Context) };
Function *MemCpyFn = Intrinsic::getDeclaration(Caller->getParent(),
Intrinsic::memcpy,
Tys, 1);
Value *Size;
if (TD == 0)
- Size = Context.getConstantExprSizeOf(AggTy);
+ Size = ConstantExpr::getSizeOf(AggTy);
else
- Size = ConstantInt::get(Type::Int64Ty,
+ Size = ConstantInt::get(Type::getInt64Ty(Context),
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::Int32Ty, 1)
+ DestCast, SrcCast, Size,
+ ConstantInt::get(Type::getInt32Ty(Context), 1)
};
CallInst *TheMemCpy =
CallInst::Create(MemCpyFn, CallArgs, CallArgs+4, "", TheCall);
ValueMap[I] = ActualArg;
}
- // Adjust llvm.dbg.region.end. If the CalledFunc has region end
- // marker then clone that marker after next stop point at the
- // call site. The function body cloner does not clone original
- // region end marker from the CalledFunc. This will ensure that
- // inlined function's scope ends at the right place.
- const DbgRegionEndInst *DREI = findFnRegionEndMarker(CalledFunc);
- if (DREI) {
- for (BasicBlock::iterator BI = TheCall,
- BE = TheCall->getParent()->end(); BI != BE; ++BI) {
- if (DbgStopPointInst *DSPI = dyn_cast<DbgStopPointInst>(BI)) {
- if (DbgRegionEndInst *NewDREI =
- dyn_cast<DbgRegionEndInst>(DREI->clone(Context)))
- NewDREI->insertAfter(DSPI);
- break;
- }
- }
- }
-
// 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, ValueMap, Returns, ".i",
- &InlinedFunctionInfo, TD);
+ &InlinedFunctionInfo, TD, TheCall);
// Remember the first block that is newly cloned over.
FirstNewBlock = LastBlock; ++FirstNewBlock;
{
BasicBlock::iterator InsertPoint = Caller->begin()->begin();
for (BasicBlock::iterator I = FirstNewBlock->begin(),
- E = FirstNewBlock->end(); I != E; )
- if (AllocaInst *AI = dyn_cast<AllocaInst>(I++)) {
- // If the alloca is now dead, remove it. This often occurs due to code
- // specialization.
- if (AI->use_empty()) {
- AI->eraseFromParent();
- continue;
- }
+ 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())) {
- // 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()))
- ++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 (!isa<Constant>(AI->getArraySize()))
+ continue;
+
+ // Keep track of the static allocas that we inline into the caller if the
+ // StaticAllocas pointer is non-null.
+ if (StaticAllocas) 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())) {
+ if (StaticAllocas) 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
if (InlinedFunctionInfo.ContainsDynamicAllocas) {
Module *M = Caller->getParent();
// Get the two intrinsics we care about.
- Constant *StackSave, *StackRestore;
- StackSave = Intrinsic::getDeclaration(M, Intrinsic::stacksave);
- StackRestore = Intrinsic::getDeclaration(M, Intrinsic::stackrestore);
+ 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 (CG) {
- // We know that StackSave/StackRestore are Function*'s, because they are
- // intrinsics which must have the right types.
- StackSaveCGN = CG->getOrInsertFunction(cast<Function>(StackSave));
- StackRestoreCGN = CG->getOrInsertFunction(cast<Function>(StackRestore));
+ StackSaveCGN = CG->getOrInsertFunction(StackSave);
+ StackRestoreCGN = CG->getOrInsertFunction(StackRestore);
CallerNode = (*CG)[Caller];
}
for (Function::iterator BB = FirstNewBlock, E = Caller->end();
BB != E; ++BB)
if (UnwindInst *UI = dyn_cast<UnwindInst>(BB->getTerminator())) {
- CallInst::Create(StackRestore, SavedPtr, "", UI);
+ CallInst *CI = CallInst::Create(StackRestore, SavedPtr, "", UI);
+ if (CG) CallerNode->addCalledFunction(CI, StackRestoreCGN);
++NumStackRestores;
}
}
BB != E; ++BB) {
TerminatorInst *Term = BB->getTerminator();
if (isa<UnwindInst>(Term)) {
- new UnreachableInst(Term);
+ new UnreachableInst(Context, Term);
BB->getInstList().erase(Term);
}
}
// 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, CG);
+ 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
if (!TheCall->use_empty()) {
ReturnInst *R = Returns[0];
if (TheCall == R->getReturnValue())
- TheCall->replaceAllUsesWith(Context.getUndef(TheCall->getType()));
+ TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
else
TheCall->replaceAllUsesWith(R->getReturnValue());
}
"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];
// using the return value of the call with the computed value.
if (!TheCall->use_empty()) {
if (TheCall == Returns[0]->getReturnValue())
- TheCall->replaceAllUsesWith(Context.getUndef(TheCall->getType()));
+ TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
else
TheCall->replaceAllUsesWith(Returns[0]->getReturnValue());
}
} else if (!TheCall->use_empty()) {
// No returns, but something is using the return value of the call. Just
// nuke the result.
- TheCall->replaceAllUsesWith(Context.getUndef(TheCall->getType()));
+ TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType()));
}
// Since we are now done with the Call/Invoke, we can delete it.