//
// The LLVM Compiler Infrastructure
//
-// This file was developed by the LLVM research group and is distributed under
-// the University of Illinois Open Source License. See LICENSE.TXT for details.
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
#include "llvm/Module.h"
#include "llvm/Instructions.h"
#include "llvm/Intrinsics.h"
+#include "llvm/ParameterAttributes.h"
#include "llvm/Analysis/CallGraph.h"
+#include "llvm/Target/TargetData.h"
+#include "llvm/ADT/SmallVector.h"
#include "llvm/Support/CallSite.h"
using namespace llvm;
-bool llvm::InlineFunction(CallInst *CI, CallGraph *CG) {
- return InlineFunction(CallSite(CI), CG);
+bool llvm::InlineFunction(CallInst *CI, CallGraph *CG, const TargetData *TD) {
+ return InlineFunction(CallSite(CI), CG, TD);
}
-bool llvm::InlineFunction(InvokeInst *II, CallGraph *CG) {
- return InlineFunction(CallSite(II), CG);
+bool llvm::InlineFunction(InvokeInst *II, CallGraph *CG, const TargetData *TD) {
+ return InlineFunction(CallSite(II), CG, TD);
}
/// HandleInlinedInvoke - If we inlined an invoke site, we need to convert calls
if (!isa<CallInst>(I)) continue;
CallInst *CI = cast<CallInst>(I);
- // If this is an intrinsic function call, don't convert it to an
- // invoke.
- if (CI->getCalledFunction() &&
- CI->getCalledFunction()->getIntrinsicID())
+ // 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,
- std::vector<Value*>(CI->op_begin()+1, CI->op_end()),
+ 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);
static void UpdateCallGraphAfterInlining(const Function *Caller,
const Function *Callee,
Function::iterator FirstNewBlock,
- std::map<const Value*, Value*> &ValueMap,
+ DenseMap<const Value*, Value*> &ValueMap,
CallGraph &CG) {
// Update the call graph by deleting the edge from Callee to Caller
CallGraphNode *CalleeNode = CG[Callee];
E = CalleeNode->end(); I != E; ++I) {
const Instruction *OrigCall = I->first.getInstruction();
- std::map<const Value*, Value*>::iterator VMI = ValueMap.find(OrigCall);
+ 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
// exists in the instruction stream. Similiarly this will inline a recursive
// function by one level.
//
-bool llvm::InlineFunction(CallSite CS, CallGraph *CG) {
+bool llvm::InlineFunction(CallSite CS, CallGraph *CG, const TargetData *TD) {
Instruction *TheCall = CS.getInstruction();
assert(TheCall->getParent() && TheCall->getParent()->getParent() &&
"Instruction not in function!");
const Function *CalledFunc = CS.getCalledFunction();
if (CalledFunc == 0 || // Can't inline external function or indirect
- CalledFunc->isExternal() || // call, or call to a vararg function!
+ CalledFunc->isDeclaration() || // call, or call to a vararg function!
CalledFunc->getFunctionType()->isVarArg()) return false;
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->hasCollector()) {
+ if (!Caller->hasCollector())
+ Caller->setCollector(CalledFunc->getCollector());
+ else if (CalledFunc->getCollector() != Caller->getCollector())
+ return false;
+ }
+
// Get an iterator to the last basic block in the function, which will have
// the new function inlined after it.
//
std::vector<ReturnInst*> Returns;
ClonedCodeInfo InlinedFunctionInfo;
Function::iterator FirstNewBlock;
-
+
{ // Scope to destroy ValueMap after cloning.
- std::map<const Value*, Value*> ValueMap;
+ DenseMap<const Value*, Value*> ValueMap;
- // Calculate the vector of arguments to pass into the function cloner, which
- // matches up the formal to the actual argument values.
assert(std::distance(CalledFunc->arg_begin(), CalledFunc->arg_end()) ==
std::distance(CS.arg_begin(), CS.arg_end()) &&
"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)
- ValueMap[I] = *AI;
+ 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;
+ }
+
+ ValueMap[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);
+ &InlinedFunctionInfo, TD);
// 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 (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.
++I;
// Transfer all of the allocas over in a block. Using splice means
- // that they instructions aren't removed from the symbol table, then
+ // that the instructions aren't removed from the symbol table, then
// reinserted.
- Caller->front().getInstList().splice(InsertPoint,
- FirstNewBlock->getInstList(),
- AI, I);
+ 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 *SBytePtr = PointerType::get(Type::SByteTy);
+ const Type *BytePtr = PointerType::getUnqual(Type::Int8Ty);
// Get the two intrinsics we care about.
- Function *StackSave, *StackRestore;
- StackSave = M->getOrInsertFunction("llvm.stacksave", SBytePtr, NULL);
+ Constant *StackSave, *StackRestore;
+ StackSave = M->getOrInsertFunction("llvm.stacksave", BytePtr, NULL);
StackRestore = M->getOrInsertFunction("llvm.stackrestore", Type::VoidTy,
- SBytePtr, NULL);
+ BytePtr, NULL);
// 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) {
- StackSaveCGN = CG->getOrInsertFunction(StackSave);
- StackRestoreCGN = CG->getOrInsertFunction(StackRestore);
+ // 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));
CallerNode = (*CG)[Caller];
}
// 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.
- if (MustClearTailCallFlags && InlinedFunctionInfo.ContainsCalls) {
+ // 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))
- CI->setTailCall(false);
+ 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(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 the return instruction returned a value, replace uses of the call with
// uses of the returned value.
- if (!TheCall->use_empty())
- TheCall->replaceAllUsesWith(Returns[0]->getReturnValue());
-
+ 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
+ TheCall->replaceAllUsesWith(R->getReturnValue());
+ }
// Since we are now done with the Call/Invoke, we can delete it.
TheCall->getParent()->getInstList().erase(TheCall);