#include "llvm/Module.h"
#include "llvm/Instructions.h"
#include "llvm/Intrinsics.h"
-#include "llvm/ParameterAttributes.h"
+#include "llvm/Attributes.h"
#include "llvm/Analysis/CallGraph.h"
#include "llvm/Target/TargetData.h"
#include "llvm/ADT/SmallVector.h"
/// 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,
}
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) {
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;
// 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());
InvokeArgs.begin(), InvokeArgs.end(),
CI->getName(), BB->getTerminator());
II->setCallingConv(CI->getCallingConv());
- II->setParamAttrs(CI->getParamAttrs());
-
+ II->setAttributes(CI->getAttributes());
+
// 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;
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->getParent()->getInstList().pop_back();
-
+ UI->eraseFromParent();
+
// Update any PHI nodes in the exceptional block to indicate that
// there is now a new entry in them.
unsigned i = 0;
/// 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
+ 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) {
+ 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 Instruction *OrigCall = I->first.getInstruction();
-
+
DenseMap<const Value*, Value*>::iterator VMI = ValueMap.find(OrigCall);
// Only copy the edge if the call was inlined!
if (VMI != ValueMap.end() && 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);
}
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.
//
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) &&
+ if (CalledFunc->paramHasAttr(ArgNo+1, Attribute::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(),
+ Value *NewAlloca = new AllocaInst(AggTy, 0, Align, I->getName(),
Caller->begin()->begin());
// Emit a memcpy.
+ const Type *Tys[] = { Type::Int64Ty };
Function *MemCpyFn = Intrinsic::getDeclaration(Caller->getParent(),
- Intrinsic::memcpy_i64);
+ Intrinsic::memcpy,
+ Tys, 1);
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.
};
CallInst *TheMemCpy =
CallInst::Create(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;
}
-
+
ValueMap[I] = ActualArg;
}
// happy with whatever the cloner can do.
CloneAndPruneFunctionInto(Caller, CalledFunc, ValueMap, Returns, ".i",
&InlinedFunctionInfo, TD);
-
+
// 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);
+ UpdateCallGraphAfterInlining(CS, FirstNewBlock, ValueMap, *CG);
}
-
+
// 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
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.
StackRestoreCGN = CG->getOrInsertFunction(cast<Function>(StackRestore));
CallerNode = (*CG)[Caller];
}
-
+
// Insert the llvm.stacksave.
- CallInst *SavedPtr = CallInst::Create(StackSave, "savedstack",
+ CallInst *SavedPtr = CallInst::Create(StackSave, "savedstack",
FirstNewBlock->begin());
if (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) {
// 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)) {
}
}
- // 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'.
// uses of the returned value.
if (!TheCall->use_empty()) {
ReturnInst *R = Returns[0];
- if (isa<StructType>(TheCall->getType()) &&
- TheCall->getType() != R->getOperand(0)->getType()) {
- // Multiple-value return statements.
- while (!TheCall->use_empty()) {
- GetResultInst *GR = cast<GetResultInst>(TheCall->use_back());
- Value *RV = R->getOperand(GR->getIndex());
- GR->replaceAllUsesWith(RV);
- GR->eraseFromParent();
- }
- } else
- TheCall->replaceAllUsesWith(R->getReturnValue());
+ 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;
// 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();
- const StructType *STy = dyn_cast<StructType>(RTy);
-
- // We do special handling for multiple-value return statements. If this is
- // a plain aggregate return, don't do the special handling.
- if (!Returns.empty() && Returns[0]->getNumOperands() != 0 &&
- Returns[0]->getOperand(0)->getType() == STy)
- STy = 0;
- if (Returns.size() > 1 || STy) {
+ if (Returns.size() > 1) {
// The PHI node should go at the front of the new basic block to merge all
// possible incoming values.
- SmallVector<PHINode *, 4> PHIs;
+ PHINode *PHI = 0;
if (!TheCall->use_empty()) {
- if (STy) {
- unsigned NumRetVals = STy->getNumElements();
- // Create new phi nodes such that phi node number in the PHIs vector
- // match corresponding return value operand number.
- Instruction *InsertPt = AfterCallBB->begin();
- for (unsigned i = 0; i < NumRetVals; ++i) {
- PHINode *PHI = PHINode::Create(STy->getElementType(i),
- TheCall->getName() + "." + utostr(i),
- InsertPt);
- PHIs.push_back(PHI);
- }
- // TheCall results are used by GetResult instructions.
- while (!TheCall->use_empty()) {
- GetResultInst *GR = cast<GetResultInst>(TheCall->use_back());
- GR->replaceAllUsesWith(PHIs[GR->getIndex()]);
- GR->eraseFromParent();
- }
- } else {
- PHINode *PHI = PHINode::Create(RTy, TheCall->getName(),
- AfterCallBB->begin());
- PHIs.push_back(PHI);
- // Anything that used the result of the function call should now use the
- // PHI node as their operand.
- TheCall->replaceAllUsesWith(PHI);
- }
+ 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 (!PHIs.empty()) {
- // There is atleast one return value.
- unsigned NumRetVals = 1;
- if (STy)
- NumRetVals = STy->getNumElements();
- for (unsigned j = 0; j < NumRetVals; ++j) {
- PHINode *PHI = PHIs[j];
- // Each PHI node will receive one value from each return instruction.
- for(unsigned i = 0, e = Returns.size(); i != e; ++i) {
- ReturnInst *RI = Returns[i];
- assert(RI->getReturnValue(j)->getType() == PHI->getType() &&
- "Ret value not consistent in function!");
- PHI->addIncoming(RI->getReturnValue(j /*PHI number matches operand number*/),
- RI->getParent());
- }
+ // 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 points and remove retrun instructions.
+ // 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);
// using the return value of the call with the computed value.
if (!TheCall->use_empty())
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();
// Now we can remove the CalleeEntry block, which is now empty.
Caller->getBasicBlockList().erase(CalleeEntry);
-
+
return true;
}