X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FTransforms%2FUtils%2FInlineFunction.cpp;h=2ed335c95c1b1a4beb34ee7b8f8dce59f82736f6;hb=b1dbcd886a4b5597a839f299054b78b33fb2d6df;hp=6bfdda230eb07eb4f9b51f3055418e472c7433b1;hpb=e4d5c441e04bdc00ccf1804744af670655123b07;p=oota-llvm.git diff --git a/lib/Transforms/Utils/InlineFunction.cpp b/lib/Transforms/Utils/InlineFunction.cpp index 6bfdda230eb..2ed335c95c1 100644 --- a/lib/Transforms/Utils/InlineFunction.cpp +++ b/lib/Transforms/Utils/InlineFunction.cpp @@ -1,18 +1,15 @@ //===- InlineFunction.cpp - Code to perform function inlining -------------===// -// +// // 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. +// //===----------------------------------------------------------------------===// // // This file implements inlining of a function into a call site, resolving // parameters and the return value as appropriate. // -// FIXME: This pass should transform alloca instructions in the called function -// into alloca/dealloca pairs! Or perhaps it should refuse to inline them! -// //===----------------------------------------------------------------------===// #include "llvm/Transforms/Utils/Cloning.h" @@ -21,34 +18,203 @@ #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/ADT/StringExtras.h" #include "llvm/Support/CallSite.h" using namespace llvm; -bool llvm::InlineFunction(CallInst *CI) { return InlineFunction(CallSite(CI)); } -bool llvm::InlineFunction(InvokeInst *II) {return InlineFunction(CallSite(II));} +bool llvm::InlineFunction(CallInst *CI, CallGraph *CG, const TargetData *TD) { + return InlineFunction(CallSite(CI), CG, TD); +} +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 +/// 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 +/// 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 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(I); ++I) { + PHINode *PN = cast(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 (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(I)) continue; + CallInst *CI = cast(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 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->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(I); ++I, ++i) { + PHINode *PN = cast(I); + PN->addIncoming(InvokeDestPHIValues[i], BB); + } + + // This basic block is now complete, start scanning the next one. + break; + } + } + + if (UnwindInst *UI = dyn_cast(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(); + + // 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(I); ++I, ++i) { + PHINode *PN = cast(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()); +} + +/// 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, + Function::iterator FirstNewBlock, + DenseMap &ValueMap, + CallGraph &CG) { + // Update the call graph by deleting the edge from Callee to Caller + 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::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(VMI->second)) + CallerNode->addCalledFunction(CallSite::get(NewCall), I->second); + } + } +} + // 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 +// 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) { +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; + + // If the call to the callee is a non-tail call, we must clear the 'tail' + // flags on any calls that we inline. + bool MustClearTailCallFlags = + isa(TheCall) && !cast(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. // @@ -57,139 +223,206 @@ bool llvm::InlineFunction(CallSite CS) { // Make sure to capture all of the return instructions from the cloned // function. std::vector Returns; + ClonedCodeInfo InlinedFunctionInfo; + Function::iterator FirstNewBlock; + { // Scope to destroy ValueMap after cloning. - // Calculate the vector of arguments to pass into the function cloner... - std::map ValueMap; - assert(std::distance(CalledFunc->arg_begin(), CalledFunc->arg_end()) == + DenseMap ValueMap; + + 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; - - // Clone the entire body of the callee into the caller. - CloneFunctionInto(Caller, CalledFunc, ValueMap, Returns, ".i"); - } - - // Remember the first block that is newly cloned over. - Function::iterator FirstNewBlock = LastBlock; ++FirstNewBlock; + 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(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 = + 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; + } + // 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); + + // 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 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. // - if (isa(FirstNewBlock->begin())) { + { BasicBlock::iterator InsertPoint = Caller->begin()->begin(); for (BasicBlock::iterator I = FirstNewBlock->begin(), E = FirstNewBlock->end(); I != E; ) - if (AllocaInst *AI = dyn_cast(I++)) + if (AllocaInst *AI = dyn_cast(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(AI->getArraySize())) { - // Scan for the block of allocas that we can move over. + // Scan for the block of allocas that we can move over, and move them + // all at once. while (isa(I) && isa(cast(I)->getArraySize())) ++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 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(TheCall)) { - BasicBlock *InvokeDest = II->getUnwindDest(); - std::vector InvokeDestPHIValues; - - // If there are PHI nodes in the exceptional destination block, we need to - // keep track of which values came into them from this invoke, then remove - // the entry for this block. - for (BasicBlock::iterator I = InvokeDest->begin(); isa(I); ++I) { - PHINode *PN = cast(I); - // Save the value to use for this edge... - InvokeDestPHIValues.push_back(PN->getIncomingValueForBlock(OrigBB)); + // 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. + Constant *StackSave, *StackRestore; + StackSave = Intrinsic::getDeclaration(M, Intrinsic::stacksave); + 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(StackSave)); + StackRestoreCGN = CG->getOrInsertFunction(cast(StackRestore)); + CallerNode = (*CG)[Caller]; + } + + // Insert the llvm.stacksave. + 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) { + CallInst *CI = CallInst::Create(StackRestore, SavedPtr, "", Returns[i]); + if (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(TheCall)) { + for (Function::iterator BB = FirstNewBlock, E = Caller->end(); + BB != E; ++BB) + if (UnwindInst *UI = dyn_cast(BB->getTerminator())) { + CallInst::Create(StackRestore, SavedPtr, "", UI); + ++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; ) { - // We only need to check for function calls: inlined invoke instructions - // require no special handling... + BB != E; ++BB) + for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) if (CallInst *CI = dyn_cast(I)) { - // Convert this function call into an invoke instruction... if it's - // not an intrinsic function call (which are known to not throw). - if (CI->getCalledFunction() && - CI->getCalledFunction()->getIntrinsicID()) { - ++I; - } else { - // 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. - InvokeInst *II = - new InvokeInst(CI->getCalledValue(), Split, InvokeDest, - std::vector(CI->op_begin()+1, CI->op_end()), - CI->getName(), BB->getTerminator()); - - // 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(I); ++I, ++i) { - PHINode *PN = cast(I); - PN->addIncoming(InvokeDestPHIValues[i], BB); - } - - // This basic block is now complete, start scanning the next one. - break; - } - } else { - ++I; + if (MustClearTailCallFlags) + CI->setTailCall(false); + if (MarkNoUnwind) + CI->setDoesNotThrow(); } - } - - if (UnwindInst *UI = dyn_cast(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(I); ++I, ++i) { - PHINode *PN = cast(I); - PN->addIncoming(InvokeDestPHIValues[i], BB); - } + // 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(Term)) { + new UnreachableInst(Term); + BB->getInstList().erase(Term); } } - // 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()); - } + // 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(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 @@ -200,17 +433,27 @@ bool llvm::InlineFunction(CallSite CS) { 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(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()) - TheCall->replaceAllUsesWith(Returns[0]->getReturnValue()); - + if (!TheCall->use_empty()) { + ReturnInst *R = Returns[0]; + if (isa(TheCall->getType())) { + // Multiple return values. + while (!TheCall->use_empty()) { + GetResultInst *GR = cast(TheCall->use_back()); + 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); @@ -229,16 +472,16 @@ bool llvm::InlineFunction(CallSite CS) { // this is an invoke instruction or a call instruction. BasicBlock *AfterCallBB; if (InvokeInst *II = dyn_cast(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 // 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. @@ -251,7 +494,7 @@ bool llvm::InlineFunction(CallSite CS) { // basic block of the inlined function. // TerminatorInst *Br = OrigBB->getTerminator(); - assert(Br && Br->getOpcode() == Instruction::Br && + assert(Br && Br->getOpcode() == Instruction::Br && "splitBasicBlock broken!"); Br->setOperand(0, FirstNewBlock); @@ -259,62 +502,86 @@ bool llvm::InlineFunction(CallSite CS) { // 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. - if (Returns.size() > 1) { + const Type *RTy = CalledFunc->getReturnType(); + const StructType *STy = dyn_cast(RTy); + if (Returns.size() > 1 || STy) { // The PHI node should go at the front of the new basic block to merge all // possible incoming values. - // - PHINode *PHI = 0; + SmallVector PHIs; if (!TheCall->use_empty()) { - PHI = new PHINode(CalledFunc->getReturnType(), - 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); + 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(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); + } } - - // 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 + + // 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()); + } + } + } + + // Add a branch to the merge points and remove retrun instructions. for (unsigned i = 0, e = Returns.size(); i != e; ++i) { ReturnInst *RI = Returns[i]; - - if (PHI) { - assert(RI->getReturnValue() && "Ret should have value!"); - 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); + 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()); - + // 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(); @@ -323,7 +590,7 @@ bool llvm::InlineFunction(CallSite CS) { // nuke the result. TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType())); } - + // Since we are now done with the Call/Invoke, we can delete it. TheCall->eraseFromParent(); @@ -342,5 +609,6 @@ bool llvm::InlineFunction(CallSite CS) { // Now we can remove the CalleeEntry block, which is now empty. Caller->getBasicBlockList().erase(CalleeEntry); + return true; }