#include "llvm/DerivedTypes.h"
#include "llvm/Module.h"
#include "llvm/Instructions.h"
+#include "llvm/IntrinsicInst.h"
#include "llvm/Intrinsics.h"
-#include "llvm/ParameterAttributes.h"
+#include "llvm/Attributes.h"
#include "llvm/Analysis/CallGraph.h"
+#include "llvm/Analysis/DebugInfo.h"
+#include "llvm/Analysis/InstructionSimplify.h"
#include "llvm/Target/TargetData.h"
+#include "llvm/Transforms/Utils/Local.h"
#include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/StringExtras.h"
#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, InlineFunctionInfo &IFI) {
+ return InlineFunction(CallSite(CI), IFI);
}
-bool llvm::InlineFunction(InvokeInst *II, CallGraph *CG, const TargetData *TD) {
- return InlineFunction(CallSite(II), CG, TD);
+bool llvm::InlineFunction(InvokeInst *II, InlineFunctionInfo &IFI) {
+ return InlineFunction(CallSite(II), IFI);
}
+
+/// 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.
+ ImmutableCallSite CS(CI);
+ SmallVector<Value*, 8> InvokeArgs(CS.arg_begin(), CS.arg_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, because it uses a WeakVH.
+ 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.
///
-/// II is the invoke instruction begin inlined. FirstNewBlock is the first
+/// 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();
- 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
}
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 (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 =
- new InvokeInst(CI->getCalledValue(), Split, InvokeDest,
- InvokeArgs.begin(), InvokeArgs.end(),
- CI->getName(), BB->getTerminator());
- II->setCallingConv(CI->getCallingConv());
- II->setParamAttrs(CI->getParamAttrs());
-
- // Make sure that anything using the call now uses the invoke!
- 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) {
- 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.
- new BranchInst(InvokeDest, UI);
-
- // Delete the unwind instruction!
- UI->getParent()->getInstList().pop_back();
-
- // 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);
}
}
}
/// UpdateCallGraphAfterInlining - Once we have cloned code over from a callee
/// into the caller, update the specified callgraph to reflect the changes we
/// made. Note that it's possible that not all code was copied over, so only
-/// some edges of the callgraph will be remain.
-static void UpdateCallGraphAfterInlining(const Function *Caller,
- const Function *Callee,
+/// some edges of the callgraph may remain.
+static void UpdateCallGraphAfterInlining(CallSite CS,
Function::iterator FirstNewBlock,
- DenseMap<const Value*, Value*> &ValueMap,
- CallGraph &CG) {
- // Update the call graph by deleting the edge from Callee to Caller
+ ValueToValueMapTy &VMap,
+ InlineFunctionInfo &IFI) {
+ CallGraph &CG = *IFI.CG;
+ const Function *Caller = CS.getInstruction()->getParent()->getParent();
+ const Function *Callee = CS.getCalledFunction();
CallGraphNode *CalleeNode = CG[Callee];
CallGraphNode *CallerNode = CG[Caller];
- CallerNode->removeCallEdgeTo(CalleeNode);
-
+
// Since we inlined some uninlined call sites in the callee into the caller,
// add edges from the caller to all of the callees of the callee.
- for (CallGraphNode::iterator I = CalleeNode->begin(),
- E = CalleeNode->end(); I != E; ++I) {
- const Instruction *OrigCall = I->first.getInstruction();
-
- DenseMap<const Value*, Value*>::iterator VMI = ValueMap.find(OrigCall);
+ CallGraphNode::iterator I = CalleeNode->begin(), E = CalleeNode->end();
+
+ // Consider the case where CalleeNode == CallerNode.
+ CallGraphNode::CalledFunctionsVector CallCache;
+ if (CalleeNode == CallerNode) {
+ CallCache.assign(I, E);
+ I = CallCache.begin();
+ E = CallCache.end();
+ }
+
+ for (; I != E; ++I) {
+ const Value *OrigCall = I->first;
+
+ ValueToValueMapTy::iterator VMI = VMap.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 == VMap.end() || VMI->second == 0)
+ continue;
+
+ // If the call was inlined, but then constant folded, there is no edge to
+ // add. Check for this case.
+ Instruction *NewCall = dyn_cast<Instruction>(VMI->second);
+ if (NewCall == 0) continue;
+
+ // Remember that this call site got inlined for the client of
+ // InlineFunction.
+ IFI.InlinedCalls.push_back(NewCall);
+
+ // It's possible that inlining the callsite will cause it to go from an
+ // indirect to a direct call by resolving a function pointer. If this
+ // happens, set the callee of the new call site to a more precise
+ // destination. This can also happen if the call graph node of the caller
+ // was just unnecessarily imprecise.
+ if (I->second->getFunction() == 0)
+ if (Function *F = CallSite(NewCall).getCalledFunction()) {
+ // Indirect call site resolved to direct call.
+ CallerNode->addCalledFunction(CallSite(NewCall), CG[F]);
+
+ continue;
+ }
+
+ CallerNode->addCalledFunction(CallSite(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);
}
+/// HandleByValArgument - When inlining a call site that has a byval argument,
+/// we have to make the implicit memcpy explicit by adding it.
+static Value *HandleByValArgument(Value *Arg, Instruction *TheCall,
+ const Function *CalledFunc,
+ InlineFunctionInfo &IFI,
+ unsigned ByValAlignment) {
+ const Type *AggTy = cast<PointerType>(Arg->getType())->getElementType();
+
+ // If the called function is readonly, then it could not mutate the caller's
+ // copy of the byval'd memory. In this case, it is safe to elide the copy and
+ // temporary.
+ if (CalledFunc->onlyReadsMemory()) {
+ // If the byval argument has a specified alignment that is greater than the
+ // passed in pointer, then we either have to round up the input pointer or
+ // give up on this transformation.
+ if (ByValAlignment <= 1) // 0 = unspecified, 1 = no particular alignment.
+ return Arg;
+
+ // If the pointer is already known to be sufficiently aligned, or if we can
+ // round it up to a larger alignment, then we don't need a temporary.
+ if (getOrEnforceKnownAlignment(Arg, ByValAlignment,
+ IFI.TD) >= ByValAlignment)
+ return Arg;
+
+ // Otherwise, we have to make a memcpy to get a safe alignment. This is bad
+ // for code quality, but rarely happens and is required for correctness.
+ }
+
+ LLVMContext &Context = Arg->getContext();
+
+ 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);
+
+ // If the byval had an alignment specified, we *must* use at least that
+ // alignment, as it is required by the byval argument (and uses of the
+ // pointer inside the callee).
+ Align = std::max(Align, ByValAlignment);
+
+ Function *Caller = TheCall->getParent()->getParent();
+
+ Value *NewAlloca = new AllocaInst(AggTy, 0, Align, Arg->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(Arg, 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::getFalse(Context) // isVolatile
+ };
+ 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.
+ return NewAlloca;
+}
// 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
// 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, 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 a non-tail call, we must clear the 'tail'
+ // 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();
+ !(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.
// 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->hasCollector()) {
- if (!Caller->hasCollector())
- Caller->setCollector(CalledFunc->getCollector());
- else if (CalledFunc->getCollector() != Caller->getCollector())
+ 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.
//
// 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;
- { // Scope to destroy ValueMap after cloning.
- DenseMap<const Value*, Value*> ValueMap;
+ { // Scope to destroy VMap after cloning.
+ ValueToValueMapTy VMap;
- assert(std::distance(CalledFunc->arg_begin(), CalledFunc->arg_end()) ==
- std::distance(CS.arg_begin(), CS.arg_end()) &&
+ 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();
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, ParamAttr::ByVal) &&
- !CalledFunc->onlyReadsMemory()) {
- const Type *AggTy = cast<PointerType>(I->getType())->getElementType();
- const Type *VoidPtrTy = PointerType::getUnqual(Type::Int8Ty);
-
- // Create the alloca. If we have TargetData, use nice alignment.
- unsigned Align = 1;
- if (TD) Align = TD->getPrefTypeAlignment(AggTy);
- Value *NewAlloca = new AllocaInst(AggTy, 0, Align, I->getName(),
- Caller->begin()->begin());
- // Emit a memcpy.
- Function *MemCpyFn = Intrinsic::getDeclaration(Caller->getParent(),
- Intrinsic::memcpy_i64);
- Value *DestCast = new BitCastInst(NewAlloca, VoidPtrTy, "tmp", TheCall);
- Value *SrcCast = new BitCastInst(*AI, VoidPtrTy, "tmp", TheCall);
-
- Value *Size;
- if (TD == 0)
- Size = ConstantExpr::getSizeOf(AggTy);
- else
- Size = ConstantInt::get(Type::Int64Ty, 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)
- };
- CallInst *TheMemCpy =
- new CallInst(MemCpyFn, CallArgs, CallArgs+4, "", TheCall);
-
- // If we have a call graph, update it.
- if (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;
+ if (CalledFunc->paramHasAttr(ArgNo+1, Attribute::ByVal)) {
+ ActualArg = HandleByValArgument(ActualArg, TheCall, CalledFunc, IFI,
+ CalledFunc->getParamAlignment(ArgNo+1));
+
+ // Calls that we inline may use the new alloca, so we need to clear
+ // their 'tail' flags if HandleByValArgument introduced a new alloca and
+ // the callee has calls.
+ MustClearTailCallFlags |= ActualArg != *AI;
}
-
- ValueMap[I] = ActualArg;
+
+ 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, ValueMap, Returns, ".i",
- &InlinedFunctionInfo, TD);
-
+ CloneAndPruneFunctionInto(Caller, CalledFunc, VMap,
+ /*ModuleLevelChanges=*/false, Returns, ".i",
+ &InlinedFunctionInfo, IFI.TD, TheCall);
+
// Remember the first block that is newly cloned over.
FirstNewBlock = LastBlock; ++FirstNewBlock;
-
+
// Update the callgraph if requested.
- if (CG)
- UpdateCallGraphAfterInlining(Caller, CalledFunc, FirstNewBlock, ValueMap,
- *CG);
+ 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
{
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;
- }
-
- 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);
- }
+ 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.
+ 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();
- const Type *BytePtr = PointerType::getUnqual(Type::Int8Ty);
// Get the two intrinsics we care about.
- Constant *StackSave, *StackRestore;
- StackSave = M->getOrInsertFunction("llvm.stacksave", BytePtr, NULL);
- StackRestore = M->getOrInsertFunction("llvm.stackrestore", Type::VoidTy,
- BytePtr, NULL);
+ 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));
+ if (CallGraph *CG = IFI.CG) {
+ StackSaveCGN = CG->getOrInsertFunction(StackSave);
+ StackRestoreCGN = CG->getOrInsertFunction(StackRestore);
CallerNode = (*CG)[Caller];
}
-
+
// Insert the llvm.stacksave.
- CallInst *SavedPtr = new CallInst(StackSave, "savedstack",
- FirstNewBlock->begin());
- if (CG) CallerNode->addCalledFunction(SavedPtr, StackSaveCGN);
-
+ 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 = new CallInst(StackRestore, SavedPtr, "", Returns[i]);
- if (CG) CallerNode->addCalledFunction(CI, StackRestoreCGN);
+ 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())) {
- new CallInst(StackRestore, SavedPtr, "", UI);
+ 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
+ // 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'.
BB != E; ++BB) {
TerminatorInst *Term = BB->getTerminator();
if (isa<UnwindInst>(Term)) {
- new UnreachableInst(Term);
+ new UnreachableInst(Context, Term);
BB->getInstList().erase(Term);
}
}
// If the call site was an invoke instruction, add a branch to the normal
// destination.
if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall))
- new BranchInst(II->getNormalDest(), 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 (R->getNumOperands() > 1) {
- // Multiple return values.
- for (Value::use_iterator RUI = TheCall->use_begin(),
- RUE = TheCall->use_end(); RUI != RUE; ) {
- GetResultInst *GR = dyn_cast<GetResultInst>(RUI++);
- assert (GR && "Invalid Call instruction use!");
- Value *RV = R->getOperand(GR->getIndex());
- GR->replaceAllUsesWith(RV);
- GR->eraseFromParent();
- }
- } else
+ 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->getParent()->getInstList().erase(TheCall);
+ TheCall->eraseFromParent();
// Since we are now done with the return instruction, delete it also.
- Returns[0]->getParent()->getInstList().erase(Returns[0]);
+ Returns[0]->eraseFromParent();
// We are now done with the inlining.
return true;
if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall)) {
// Add an unconditional branch to make this look like the CallInst case...
- BranchInst *NewBr = new BranchInst(II->getNormalDest(), TheCall);
+ 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
// 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();
+
+ PHINode *PHI = 0;
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 = new PHINode(CalledFunc->getReturnType(),
- TheCall->getName(), AfterCallBB->begin());
-
+ PHI = PHINode::Create(RTy, Returns.size(), 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, turning them into unconditional
- // branches to the merge point now, and adding entries to the PHI node as
- // appropriate.
- for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
- ReturnInst *RI = Returns[i];
-
- if (PHI) {
- assert(RI->getReturnValue() && "Ret should have value!");
+ // 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());
}
+ }
- // Add a branch to the merge point where the PHI node lives if it exists.
- new BranchInst(AfterCallBB, RI);
- // Delete the return instruction now
- RI->getParent()->getInstList().erase(RI);
+ // 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())
- TheCall->replaceAllUsesWith(Returns[0]->getReturnValue());
+ 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.
// Now we can remove the CalleeEntry block, which is now empty.
Caller->getBasicBlockList().erase(CalleeEntry);
-
+
+ // If we inserted a phi node, 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 (PHI)
+ if (Value *V = SimplifyInstruction(PHI, IFI.TD)) {
+ PHI->replaceAllUsesWith(V);
+ PHI->eraseFromParent();
+ }
+
return true;
}
projects / oota-llvm.git / blobdiff