X-Git-Url: http://plrg.eecs.uci.edu/git/?p=oota-llvm.git;a=blobdiff_plain;f=lib%2FTransforms%2FUtils%2FInlineFunction.cpp;h=0a9aa7a8d751fc2fe35590852b950e729ff6412f;hp=9362d26e2b267a575e1e17126b2fd0dedb2fefff;hb=0e13821c96937830ec817f08095c3cef1fdcac8d;hpb=e07007c2bd0dcb775ff3eecabb4c31ce8dc5fab0 diff --git a/lib/Transforms/Utils/InlineFunction.cpp b/lib/Transforms/Utils/InlineFunction.cpp index 9362d26e2b2..0a9aa7a8d75 100644 --- a/lib/Transforms/Utils/InlineFunction.cpp +++ b/lib/Transforms/Utils/InlineFunction.cpp @@ -1,244 +1,529 @@ //===- 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 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 malloc/free pairs! Or perhaps it should refuse to inline them! -// //===----------------------------------------------------------------------===// #include "llvm/Transforms/Utils/Cloning.h" -#include "llvm/Constant.h" +#include "llvm/Constants.h" #include "llvm/DerivedTypes.h" #include "llvm/Module.h" #include "llvm/Instructions.h" #include "llvm/Intrinsics.h" +#include "llvm/Analysis/CallGraph.h" +#include "llvm/ADT/SmallVector.h" #include "llvm/Support/CallSite.h" -#include "llvm/Transforms/Utils/Local.h" +using namespace llvm; + +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 = + 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(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. + 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); + } + } + } + } + + // 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); + } + } +} -bool InlineFunction(CallInst *CI) { return InlineFunction(CallSite(CI)); } -bool InlineFunction(InvokeInst *II) { return InlineFunction(CallSite(II)); } // 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 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; - BasicBlock *OrigBB = TheCall->getParent(); - Function *Caller = OrigBB->getParent(); - - // We want to clone the entire callee function into the whole between the - // "starter" and "ender" blocks. How we accomplish this depends on whether - // this is an invoke instruction or a call instruction. - - BasicBlock *InvokeDest = 0; // Exception handling destination - BasicBlock *AfterCallBB; - if (InvokeInst *II = dyn_cast(TheCall)) { - AfterCallBB = II->getNormalDest(); - InvokeDest = II->getExceptionalDest(); - // Add an unconditional branch to make this look like the CallInst case... - new BranchInst(AfterCallBB, TheCall); + // 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(); - // Remove (unlink) the InvokeInst from the function... - OrigBB->getInstList().remove(TheCall); - } else { // It's a call - // If this is a call instruction, we need to split the basic block that the - // call lives in. - // - AfterCallBB = OrigBB->splitBasicBlock(TheCall, - CalledFunc->getName()+".entry"); - // Remove (unlink) the CallInst from the function... - AfterCallBB->getInstList().remove(TheCall); - } + // If the call to the callee cannot throw, set the 'nounwind' flag on any + // calls that we inline. + bool MarkNoUnwind = CS.doesNotThrow(); - // If we have a return value generated by this call, convert it into a PHI - // node that gets values from each of the old RET instructions in the original - // function. - // - PHINode *PHI = 0; - if (!TheCall->use_empty()) { - // The PHI node should go at the front of the new basic block to merge all - // possible incoming values. - // - PHI = new PHINode(CalledFunc->getReturnType(), TheCall->getName(), - AfterCallBB->begin()); + BasicBlock *OrigBB = TheCall->getParent(); + Function *Caller = OrigBB->getParent(); - // Anything that used the result of the function call should now use the PHI - // node as their operand. - // - TheCall->replaceAllUsesWith(PHI); + + // 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. // Function::iterator LastBlock = &Caller->back(); - // Calculate the vector of arguments to pass into the function cloner... - std::map ValueMap; - assert(std::distance(CalledFunc->abegin(), CalledFunc->aend()) == - std::distance(CS.arg_begin(), CS.arg_end()) && - "No varargs calls can be inlined!"); + // 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. + DenseMap 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!"); + CallSite::arg_iterator AI = CS.arg_begin(); + for (Function::const_arg_iterator I = CalledFunc->arg_begin(), + E = CalledFunc->arg_end(); I != E; ++I, ++AI) + ValueMap[I] = *AI; + + // 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. + // + { + BasicBlock::iterator InsertPoint = Caller->begin()->begin(); + for (BasicBlock::iterator I = FirstNewBlock->begin(), + E = FirstNewBlock->end(); I != E; ) + 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, 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 the instructions aren't removed from the symbol table, then + // reinserted. + Caller->getEntryBlock().getInstList().splice( + InsertPoint, + FirstNewBlock->getInstList(), + AI, I); + } + } + } - CallSite::arg_iterator AI = CS.arg_begin(); - for (Function::const_aiterator I = CalledFunc->abegin(), E=CalledFunc->aend(); - I != E; ++I, ++AI) - ValueMap[I] = *AI; + // 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); + + // 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 = new CallInst(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 = new CallInst(StackRestore, SavedPtr, "", Returns[i]); + if (CG) CallerNode->addCalledFunction(CI, StackRestoreCGN); + } - // Since we are now done with the Call/Invoke, we can delete it. - delete TheCall; + // 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())) { + new CallInst(StackRestore, SavedPtr, "", UI); + ++NumStackRestores; + } + } + } - // Make a vector to capture the return instructions in the cloned function... - std::vector Returns; + // 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; ++I) + if (CallInst *CI = dyn_cast(I)) { + if (MustClearTailCallFlags) + CI->setTailCall(false); + if (MarkNoUnwind) + CI->setDoesNotThrow(); + } + } - // Populate the value map with all of the globals in the program. - // FIXME: This should be the default for CloneFunctionInto! - Module &M = *Caller->getParent(); - for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) - ValueMap[I] = I; - for (Module::giterator I = M.gbegin(), E = M.gend(); I != E; ++I) - ValueMap[I] = I; - - // Do all of the hard part of cloning the callee into the caller... - CloneFunctionInto(Caller, CalledFunc, ValueMap, Returns, ".i"); - - // Loop over all of the return instructions, turning them into unconditional - // branches to the merge point now... - for (unsigned i = 0, e = Returns.size(); i != e; ++i) { - ReturnInst *RI = Returns[i]; - BasicBlock *BB = RI->getParent(); - - // Add a branch to the merge point where the PHI node lives if it exists. - new BranchInst(AfterCallBB, RI); - - if (PHI) { // The PHI node should include this value! - assert(RI->getReturnValue() && "Ret should have value!"); - assert(RI->getReturnValue()->getType() == PHI->getType() && - "Ret value not consistent in function!"); - PHI->addIncoming(RI->getReturnValue(), 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); + } } - // Delete the return instruction now - BB->getInstList().erase(RI); + // 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 + // the calling basic block. + if (Returns.size() == 1 && std::distance(FirstNewBlock, Caller->end()) == 1) { + // Move all of the instructions right before the call. + OrigBB->getInstList().splice(TheCall, FirstNewBlock->getInstList(), + 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); + + // 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()); + + // Since we are now done with the Call/Invoke, we can delete it. + TheCall->getParent()->getInstList().erase(TheCall); + + // Since we are now done with the return instruction, delete it also. + Returns[0]->getParent()->getInstList().erase(Returns[0]); + + // We are now done with the inlining. + return true; } - // Check to see if the PHI node only has one argument. This is a common - // case resulting from there only being a single return instruction in the - // function call. Because this is so common, eliminate the PHI node. - // - if (PHI && PHI->getNumIncomingValues() == 1) { - PHI->replaceAllUsesWith(PHI->getIncomingValue(0)); - PHI->getParent()->getInstList().erase(PHI); + // Otherwise, we have the normal case, of more than one block to inline or + // multiple return sites. + + // We want to clone the entire callee function into the hole between the + // "starter" and "ender" blocks. How we accomplish this depends on whether + // 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); + + // 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. + // + AfterCallBB = OrigBB->splitBasicBlock(TheCall, + CalledFunc->getName()+".exit"); } // Change the branch that used to go to AfterCallBB to branch to the first // basic block of the inlined function. // TerminatorInst *Br = OrigBB->getTerminator(); - assert(Br && Br->getOpcode() == Instruction::Br && - "splitBasicBlock broken!"); - Br->setOperand(0, ++LastBlock); + assert(Br && Br->getOpcode() == Instruction::Br && + "splitBasicBlock broken!"); + Br->setOperand(0, FirstNewBlock); - // 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(LastBlock->begin())) { - BasicBlock::iterator InsertPoint = Caller->begin()->begin(); - while (isa(InsertPoint)) ++InsertPoint; - - for (BasicBlock::iterator I = LastBlock->begin(), E = LastBlock->end(); - I != E; ) - if (AllocaInst *AI = dyn_cast(I)) { - ++I; // Move to the next instruction - LastBlock->getInstList().remove(AI); - Caller->front().getInstList().insert(InsertPoint, AI); - } else { - ++I; - } - } - // If we just inlined a call due to an invoke instruction, scan the inlined - // function checking for function calls that should now be made into invoke - // instructions, and for unwind's which should be turned into branches. - if (InvokeDest) - for (Function::iterator BB = LastBlock, 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... - if (CallInst *CI = dyn_cast(I)) { - // 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. - InvokeInst *II = - new InvokeInst(CI->getCalledValue(), Split, InvokeDest, - std::vector(CI->op_begin()+1, CI->op_end()), - CI->getName(), BB->getTerminator()); + // 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()); - // Make sure that anything using the call now uses the invoke! - CI->replaceAllUsesWith(II); + // 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) { + // 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()); + + // Anything that used the result of the function call should now use the + // PHI node as their operand. + // + TheCall->replaceAllUsesWith(PHI); + } - // Delete the unconditional branch inserted by splitBasicBlock - BB->getInstList().pop_back(); - Split->getInstList().pop_front(); // Delete the original call - - // This basic block is now complete, start scanning the next one. - break; - } else { - ++I; - } + // 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!"); + assert(RI->getReturnValue()->getType() == PHI->getType() && + "Ret value not consistent in function!"); + PHI->addIncoming(RI->getReturnValue(), RI->getParent()); } - 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 *BI = new BranchInst(InvokeDest, UI); + // Add a branch to the merge point where the PHI node lives if it exists. + new BranchInst(AfterCallBB, RI); - // Delete the unwind instruction! - UI->getParent()->getInstList().pop_back(); - } + // Delete the return instruction now + RI->getParent()->getInstList().erase(RI); } - // 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(), - LastBlock, Caller->end()); + } 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(); + } else if (!TheCall->use_empty()) { + // No returns, but something is using the return value of the call. Just + // nuke the result. + TheCall->replaceAllUsesWith(UndefValue::get(TheCall->getType())); + } + + // Since we are now done with the Call/Invoke, we can delete it. + TheCall->eraseFromParent(); // We should always be able to fold the entry block of the function into the // single predecessor of the block... assert(cast(Br)->isUnconditional() && "splitBasicBlock broken!"); BasicBlock *CalleeEntry = cast(Br)->getSuccessor(0); - SimplifyCFG(CalleeEntry); + + // Splice the code entry block into calling block, right before the + // unconditional branch. + OrigBB->getInstList().splice(Br, CalleeEntry->getInstList()); + CalleeEntry->replaceAllUsesWith(OrigBB); // Update PHI nodes + + // Remove the unconditional branch. + OrigBB->getInstList().erase(Br); + + // Now we can remove the CalleeEntry block, which is now empty. + Caller->getBasicBlockList().erase(CalleeEntry); - // Okay, continue the CFG cleanup. It's often the case that there is only a - // single return instruction in the callee function. If this is the case, - // then we have an unconditional branch from the return block to the - // 'AfterCallBB'. Check for this case, and eliminate the branch is possible. - SimplifyCFG(AfterCallBB); return true; }