//===-- ArgumentPromotion.cpp - Promote by-reference arguments ------------===// // // The LLVM Compiler Infrastructure // // This file is distributed under the University of Illinois Open Source // License. See LICENSE.TXT for details. // //===----------------------------------------------------------------------===// // // This pass promotes "by reference" arguments to be "by value" arguments. In // practice, this means looking for internal functions that have pointer // arguments. If it can prove, through the use of alias analysis, that an // argument is *only* loaded, then it can pass the value into the function // instead of the address of the value. This can cause recursive simplification // of code and lead to the elimination of allocas (especially in C++ template // code like the STL). // // This pass also handles aggregate arguments that are passed into a function, // scalarizing them if the elements of the aggregate are only loaded. Note that // by default it refuses to scalarize aggregates which would require passing in // more than three operands to the function, because passing thousands of // operands for a large array or structure is unprofitable! This limit can be // configured or disabled, however. // // Note that this transformation could also be done for arguments that are only // stored to (returning the value instead), but does not currently. This case // would be best handled when and if LLVM begins supporting multiple return // values from functions. // //===----------------------------------------------------------------------===// #define DEBUG_TYPE "argpromotion" #include "llvm/Transforms/IPO.h" #include "llvm/Constants.h" #include "llvm/DerivedTypes.h" #include "llvm/Module.h" #include "llvm/CallGraphSCCPass.h" #include "llvm/Instructions.h" #include "llvm/Analysis/AliasAnalysis.h" #include "llvm/Analysis/CallGraph.h" #include "llvm/Target/TargetData.h" #include "llvm/Support/CallSite.h" #include "llvm/Support/CFG.h" #include "llvm/Support/Debug.h" #include "llvm/ADT/DepthFirstIterator.h" #include "llvm/ADT/Statistic.h" #include "llvm/ADT/StringExtras.h" #include "llvm/Support/Compiler.h" #include using namespace llvm; STATISTIC(NumArgumentsPromoted , "Number of pointer arguments promoted"); STATISTIC(NumAggregatesPromoted, "Number of aggregate arguments promoted"); STATISTIC(NumByValArgsPromoted , "Number of byval arguments promoted"); STATISTIC(NumArgumentsDead , "Number of dead pointer args eliminated"); namespace { /// ArgPromotion - The 'by reference' to 'by value' argument promotion pass. /// struct VISIBILITY_HIDDEN ArgPromotion : public CallGraphSCCPass { virtual void getAnalysisUsage(AnalysisUsage &AU) const { AU.addRequired(); AU.addRequired(); CallGraphSCCPass::getAnalysisUsage(AU); } virtual bool runOnSCC(const std::vector &SCC); static char ID; // Pass identification, replacement for typeid ArgPromotion(unsigned maxElements = 3) : CallGraphSCCPass(&ID), maxElements(maxElements) {} /// A vector used to hold the indices of a single GEP instruction typedef std::vector IndicesVector; private: bool PromoteArguments(CallGraphNode *CGN); bool isSafeToPromoteArgument(Argument *Arg, bool isByVal) const; Function *DoPromotion(Function *F, SmallPtrSet &ArgsToPromote, SmallPtrSet &ByValArgsToTransform); /// The maximum number of elements to expand, or 0 for unlimited. unsigned maxElements; }; } char ArgPromotion::ID = 0; static RegisterPass X("argpromotion", "Promote 'by reference' arguments to scalars"); Pass *llvm::createArgumentPromotionPass(unsigned maxElements) { return new ArgPromotion(maxElements); } bool ArgPromotion::runOnSCC(const std::vector &SCC) { bool Changed = false, LocalChange; do { // Iterate until we stop promoting from this SCC. LocalChange = false; // Attempt to promote arguments from all functions in this SCC. for (unsigned i = 0, e = SCC.size(); i != e; ++i) LocalChange |= PromoteArguments(SCC[i]); Changed |= LocalChange; // Remember that we changed something. } while (LocalChange); return Changed; } /// PromoteArguments - This method checks the specified function to see if there /// are any promotable arguments and if it is safe to promote the function (for /// example, all callers are direct). If safe to promote some arguments, it /// calls the DoPromotion method. /// bool ArgPromotion::PromoteArguments(CallGraphNode *CGN) { Function *F = CGN->getFunction(); // Make sure that it is local to this module. if (!F || !F->hasInternalLinkage()) return false; // First check: see if there are any pointer arguments! If not, quick exit. SmallVector, 16> PointerArgs; unsigned ArgNo = 0; for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(); I != E; ++I, ++ArgNo) if (isa(I->getType())) PointerArgs.push_back(std::pair(I, ArgNo)); if (PointerArgs.empty()) return false; // Second check: make sure that all callers are direct callers. We can't // transform functions that have indirect callers. for (Value::use_iterator UI = F->use_begin(), E = F->use_end(); UI != E; ++UI) { CallSite CS = CallSite::get(*UI); if (!CS.getInstruction()) // "Taking the address" of the function return false; // Ensure that this call site is CALLING the function, not passing it as // an argument. if (UI.getOperandNo() != 0) return false; } // Check to see which arguments are promotable. If an argument is promotable, // add it to ArgsToPromote. SmallPtrSet ArgsToPromote; SmallPtrSet ByValArgsToTransform; for (unsigned i = 0; i != PointerArgs.size(); ++i) { bool isByVal = F->paramHasAttr(PointerArgs[i].second+1, Attribute::ByVal); // If this is a byval argument, and if the aggregate type is small, just // pass the elements, which is always safe. Argument *PtrArg = PointerArgs[i].first; if (isByVal) { const Type *AgTy = cast(PtrArg->getType())->getElementType(); if (const StructType *STy = dyn_cast(AgTy)) { if (maxElements > 0 && STy->getNumElements() > maxElements) { DOUT << "argpromotion disable promoting argument '" << PtrArg->getName() << "' because it would require adding more " << "than " << maxElements << " arguments to the function.\n"; } else { // If all the elements are single-value types, we can promote it. bool AllSimple = true; for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) if (!STy->getElementType(i)->isSingleValueType()) { AllSimple = false; break; } // Safe to transform, don't even bother trying to "promote" it. // Passing the elements as a scalar will allow scalarrepl to hack on // the new alloca we introduce. if (AllSimple) { ByValArgsToTransform.insert(PtrArg); continue; } } } } // Otherwise, see if we can promote the pointer to its value. if (isSafeToPromoteArgument(PtrArg, isByVal)) ArgsToPromote.insert(PtrArg); } // No promotable pointer arguments. if (ArgsToPromote.empty() && ByValArgsToTransform.empty()) return false; Function *NewF = DoPromotion(F, ArgsToPromote, ByValArgsToTransform); // Update the call graph to know that the function has been transformed. getAnalysis().changeFunction(F, NewF); return true; } /// IsAlwaysValidPointer - Return true if the specified pointer is always legal /// to load. static bool IsAlwaysValidPointer(Value *V) { if (isa(V) || isa(V)) return true; if (GetElementPtrInst *GEP = dyn_cast(V)) return IsAlwaysValidPointer(GEP->getOperand(0)); if (ConstantExpr *CE = dyn_cast(V)) if (CE->getOpcode() == Instruction::GetElementPtr) return IsAlwaysValidPointer(CE->getOperand(0)); return false; } /// AllCalleesPassInValidPointerForArgument - Return true if we can prove that /// all callees pass in a valid pointer for the specified function argument. static bool AllCalleesPassInValidPointerForArgument(Argument *Arg) { Function *Callee = Arg->getParent(); unsigned ArgNo = std::distance(Callee->arg_begin(), Function::arg_iterator(Arg)); // Look at all call sites of the function. At this pointer we know we only // have direct callees. for (Value::use_iterator UI = Callee->use_begin(), E = Callee->use_end(); UI != E; ++UI) { CallSite CS = CallSite::get(*UI); assert(CS.getInstruction() && "Should only have direct calls!"); if (!IsAlwaysValidPointer(CS.getArgument(ArgNo))) return false; } return true; } /// Returns true if Prefix is a prefix of longer. That means, Longer has a size /// that is greater than or equal to the size of prefix, and each of the /// elements in Prefix is the same as the corresponding elements in Longer. /// /// This means it also returns true when Prefix and Longer are equal! static bool IsPrefix(const ArgPromotion::IndicesVector &Prefix, const ArgPromotion::IndicesVector &Longer) { if (Prefix.size() > Longer.size()) return false; for (unsigned i = 0, e = Prefix.size(); i != e; ++i) if (Prefix[i] != Longer[i]) return false; return true; } /// Checks if Indices, or a prefix of Indices, is in Set. static bool PrefixIn(const ArgPromotion::IndicesVector &Indices, std::set &Set) { std::set::iterator Low; Low = Set.upper_bound(Indices); if (Low != Set.begin()) Low--; // Low is now the last element smaller than or equal to Indices. This means // it points to a prefix of Indices (possibly Indices itself), if such // prefix exists. // // This load is safe if any prefix of its operands is safe to load. return Low != Set.end() && IsPrefix(*Low, Indices); } /// Mark the given indices (ToMark) as safe in the the given set of indices /// (Safe). Marking safe usually means adding ToMark to Safe. However, if there /// is already a prefix of Indices in Safe, Indices are implicitely marked safe /// already. Furthermore, any indices that Indices is itself a prefix of, are /// removed from Safe (since they are implicitely safe because of Indices now). static void MarkIndicesSafe(const ArgPromotion::IndicesVector &ToMark, std::set &Safe) { std::set::iterator Low; Low = Safe.upper_bound(ToMark); // Guard against the case where Safe is empty if (Low != Safe.begin()) Low--; // Low is now the last element smaller than or equal to Indices. This // means it points to a prefix of Indices (possibly Indices itself), if // such prefix exists. if (Low != Safe.end()) { if (IsPrefix(*Low, ToMark)) // If there is already a prefix of these indices (or exactly these // indices) marked a safe, don't bother adding these indices return; // Increment Low, so we can use it as a "insert before" hint ++Low; } // Insert Low = Safe.insert(Low, ToMark); ++Low; // If there we're a prefix of longer index list(s), remove those std::set::iterator End = Safe.end(); while (Low != End && IsPrefix(ToMark, *Low)) { std::set::iterator Remove = Low; ++Low; Safe.erase(Remove); } } /// isSafeToPromoteArgument - As you might guess from the name of this method, /// it checks to see if it is both safe and useful to promote the argument. /// This method limits promotion of aggregates to only promote up to three /// elements of the aggregate in order to avoid exploding the number of /// arguments passed in. bool ArgPromotion::isSafeToPromoteArgument(Argument *Arg, bool isByVal) const { typedef std::set GEPIndicesSet; // Quick exit for unused arguments if (Arg->use_empty()) return true; // We can only promote this argument if all of the uses are loads, or are GEP // instructions (with constant indices) that are subsequently loaded. // // Promoting the argument causes it to be loaded in the caller // unconditionally. This is only safe if we can prove that either the load // would have happened in the callee anyway (ie, there is a load in the entry // block) or the pointer passed in at every call site is guaranteed to be // valid. // In the former case, invalid loads can happen, but would have happened // anyway, in the latter case, invalid loads won't happen. This prevents us // from introducing an invalid load that wouldn't have happened in the // original code. // // This set will contain all sets of indices that are loaded in the entry // block, and thus are safe to unconditionally load in the caller. GEPIndicesSet SafeToUnconditionallyLoad; // This set contains all the sets of indices that we are planning to promote. // This makes it possible to limit the number of arguments added. GEPIndicesSet ToPromote; // If the pointer is always valid, any load with first index 0 is valid. if(isByVal || AllCalleesPassInValidPointerForArgument(Arg)) SafeToUnconditionallyLoad.insert(IndicesVector(1, 0)); // First, iterate the entry block and mark loads of (geps of) arguments as // safe. BasicBlock *EntryBlock = Arg->getParent()->begin(); // Declare this here so we can reuse it IndicesVector Indices; for (BasicBlock::iterator I = EntryBlock->begin(), E = EntryBlock->end(); I != E; ++I) if (LoadInst *LI = dyn_cast(I)) { Value *V = LI->getPointerOperand(); if (GetElementPtrInst *GEP = dyn_cast(V)) { V = GEP->getPointerOperand(); if (V == Arg) { // This load actually loads (part of) Arg? Check the indices then. Indices.reserve(GEP->getNumIndices()); for (User::op_iterator II = GEP->idx_begin(), IE = GEP->idx_end(); II != IE; ++II) if (ConstantInt *CI = dyn_cast(*II)) Indices.push_back(CI->getSExtValue()); else // We found a non-constant GEP index for this argument? Bail out // right away, can't promote this argument at all. return false; // Indices checked out, mark them as safe MarkIndicesSafe(Indices, SafeToUnconditionallyLoad); Indices.clear(); } } else if (V == Arg) { // Direct loads are equivalent to a GEP with a single 0 index. MarkIndicesSafe(IndicesVector(1, 0), SafeToUnconditionallyLoad); } } // Now, iterate all uses of the argument to see if there are any uses that are // not (GEP+)loads, or any (GEP+)loads that are not safe to promote. SmallVector Loads; IndicesVector Operands; for (Value::use_iterator UI = Arg->use_begin(), E = Arg->use_end(); UI != E; ++UI) { Operands.clear(); if (LoadInst *LI = dyn_cast(*UI)) { if (LI->isVolatile()) return false; // Don't hack volatile loads Loads.push_back(LI); // Direct loads are equivalent to a GEP with a zero index and then a load. Operands.push_back(0); } else if (GetElementPtrInst *GEP = dyn_cast(*UI)) { if (GEP->use_empty()) { // Dead GEP's cause trouble later. Just remove them if we run into // them. getAnalysis().deleteValue(GEP); GEP->eraseFromParent(); // TODO: This runs the above loop over and over again for dead GEPS // Couldn't we just do increment the UI iterator earlier and erase the // use? return isSafeToPromoteArgument(Arg, isByVal); } // Ensure that all of the indices are constants. for (User::op_iterator i = GEP->idx_begin(), e = GEP->idx_end(); i != e; ++i) if (ConstantInt *C = dyn_cast(*i)) Operands.push_back(C->getSExtValue()); else return false; // Not a constant operand GEP! // Ensure that the only users of the GEP are load instructions. for (Value::use_iterator UI = GEP->use_begin(), E = GEP->use_end(); UI != E; ++UI) if (LoadInst *LI = dyn_cast(*UI)) { if (LI->isVolatile()) return false; // Don't hack volatile loads Loads.push_back(LI); } else { // Other uses than load? return false; } } else { return false; // Not a load or a GEP. } // Now, see if it is safe to promote this load / loads of this GEP. Loading // is safe if Operands, or a prefix of Operands, is marked as safe. if (!PrefixIn(Operands, SafeToUnconditionallyLoad)) return false; // See if we are already promoting a load with these indices. If not, check // to make sure that we aren't promoting too many elements. If so, nothing // to do. if (ToPromote.find(Operands) == ToPromote.end()) { if (maxElements > 0 && ToPromote.size() == maxElements) { DOUT << "argpromotion not promoting argument '" << Arg->getName() << "' because it would require adding more " << "than " << maxElements << " arguments to the function.\n"; // We limit aggregate promotion to only promoting up to a fixed number // of elements of the aggregate. return false; } ToPromote.insert(Operands); } } if (Loads.empty()) return true; // No users, this is a dead argument. // Okay, now we know that the argument is only used by load instructions and // it is safe to unconditionally perform all of them. Use alias analysis to // check to see if the pointer is guaranteed to not be modified from entry of // the function to each of the load instructions. // Because there could be several/many load instructions, remember which // blocks we know to be transparent to the load. SmallPtrSet TranspBlocks; AliasAnalysis &AA = getAnalysis(); TargetData &TD = getAnalysis(); for (unsigned i = 0, e = Loads.size(); i != e; ++i) { // Check to see if the load is invalidated from the start of the block to // the load itself. LoadInst *Load = Loads[i]; BasicBlock *BB = Load->getParent(); const PointerType *LoadTy = cast(Load->getPointerOperand()->getType()); unsigned LoadSize = (unsigned)TD.getTypeStoreSize(LoadTy->getElementType()); if (AA.canInstructionRangeModify(BB->front(), *Load, Arg, LoadSize)) return false; // Pointer is invalidated! // Now check every path from the entry block to the load for transparency. // To do this, we perform a depth first search on the inverse CFG from the // loading block. for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) for (idf_ext_iterator > I = idf_ext_begin(*PI, TranspBlocks), E = idf_ext_end(*PI, TranspBlocks); I != E; ++I) if (AA.canBasicBlockModify(**I, Arg, LoadSize)) return false; } // If the path from the entry of the function to each load is free of // instructions that potentially invalidate the load, we can make the // transformation! return true; } /// DoPromotion - This method actually performs the promotion of the specified /// arguments, and returns the new function. At this point, we know that it's /// safe to do so. Function *ArgPromotion::DoPromotion(Function *F, SmallPtrSet &ArgsToPromote, SmallPtrSet &ByValArgsToTransform) { // Start by computing a new prototype for the function, which is the same as // the old function, but has modified arguments. const FunctionType *FTy = F->getFunctionType(); std::vector Params; typedef std::set ScalarizeTable; // ScalarizedElements - If we are promoting a pointer that has elements // accessed out of it, keep track of which elements are accessed so that we // can add one argument for each. // // Arguments that are directly loaded will have a zero element value here, to // handle cases where there are both a direct load and GEP accesses. // std::map ScalarizedElements; // OriginalLoads - Keep track of a representative load instruction from the // original function so that we can tell the alias analysis implementation // what the new GEP/Load instructions we are inserting look like. std::map OriginalLoads; // Attributes - Keep track of the parameter attributes for the arguments // that we are *not* promoting. For the ones that we do promote, the parameter // attributes are lost SmallVector AttributesVec; const AttrListPtr &PAL = F->getAttributes(); // Add any return attributes. if (Attributes attrs = PAL.getRetAttributes()) AttributesVec.push_back(AttributeWithIndex::get(0, attrs)); // First, determine the new argument list unsigned ArgIndex = 1; for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(); I != E; ++I, ++ArgIndex) { if (ByValArgsToTransform.count(I)) { // Simple byval argument? Just add all the struct element types. const Type *AgTy = cast(I->getType())->getElementType(); const StructType *STy = cast(AgTy); for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) Params.push_back(STy->getElementType(i)); ++NumByValArgsPromoted; } else if (!ArgsToPromote.count(I)) { // Unchanged argument Params.push_back(I->getType()); if (Attributes attrs = PAL.getParamAttributes(ArgIndex)) AttributesVec.push_back(AttributeWithIndex::get(Params.size(), attrs)); } else if (I->use_empty()) { // Dead argument (which are always marked as promotable) ++NumArgumentsDead; } else { // Okay, this is being promoted. This means that the only uses are loads // or GEPs which are only used by loads // In this table, we will track which indices are loaded from the argument // (where direct loads are tracked as no indices). ScalarizeTable &ArgIndices = ScalarizedElements[I]; for (Value::use_iterator UI = I->use_begin(), E = I->use_end(); UI != E; ++UI) { Instruction *User = cast(*UI); assert(isa(User) || isa(User)); IndicesVector Indices; Indices.reserve(User->getNumOperands() - 1); // Since loads will only have a single operand, and GEPs only a single // non-index operand, this will record direct loads without any indices, // and gep+loads with the GEP indices. for (User::op_iterator II = User->op_begin() + 1, IE = User->op_end(); II != IE; ++II) Indices.push_back(cast(*II)->getSExtValue()); // GEPs with a single 0 index can be merged with direct loads if (Indices.size() == 1 && Indices.front() == 0) Indices.clear(); ArgIndices.insert(Indices); LoadInst *OrigLoad; if (LoadInst *L = dyn_cast(User)) OrigLoad = L; else // Take any load, we will use it only to update Alias Analysis OrigLoad = cast(User->use_back()); OriginalLoads[Indices] = OrigLoad; } // Add a parameter to the function for each element passed in. for (ScalarizeTable::iterator SI = ArgIndices.begin(), E = ArgIndices.end(); SI != E; ++SI) { Params.push_back(GetElementPtrInst::getIndexedType(I->getType(), &*SI->begin(), SI->size())); assert(Params.back()); } if (ArgIndices.size() == 1 && ArgIndices.begin()->empty()) ++NumArgumentsPromoted; else ++NumAggregatesPromoted; } } // Add any function attributes. if (Attributes attrs = PAL.getFnAttributes()) AttributesVec.push_back(AttributeWithIndex::get(~0, attrs)); const Type *RetTy = FTy->getReturnType(); // Work around LLVM bug PR56: the CWriter cannot emit varargs functions which // have zero fixed arguments. bool ExtraArgHack = false; if (Params.empty() && FTy->isVarArg()) { ExtraArgHack = true; Params.push_back(Type::Int32Ty); } // Construct the new function type using the new arguments. FunctionType *NFTy = FunctionType::get(RetTy, Params, FTy->isVarArg()); // Create the new function body and insert it into the module... Function *NF = Function::Create(NFTy, F->getLinkage(), F->getName()); NF->copyAttributesFrom(F); // Recompute the parameter attributes list based on the new arguments for // the function. NF->setAttributes(AttrListPtr::get(AttributesVec.begin(), AttributesVec.end())); AttributesVec.clear(); F->getParent()->getFunctionList().insert(F, NF); NF->takeName(F); // Get the alias analysis information that we need to update to reflect our // changes. AliasAnalysis &AA = getAnalysis(); // Get the callgraph information that we need to update to reflect our // changes. CallGraph &CG = getAnalysis(); // Loop over all of the callers of the function, transforming the call sites // to pass in the loaded pointers. // SmallVector Args; while (!F->use_empty()) { CallSite CS = CallSite::get(F->use_back()); Instruction *Call = CS.getInstruction(); const AttrListPtr &CallPAL = CS.getAttributes(); // Add any return attributes. if (Attributes attrs = CallPAL.getRetAttributes()) AttributesVec.push_back(AttributeWithIndex::get(0, attrs)); // Loop over the operands, inserting GEP and loads in the caller as // appropriate. CallSite::arg_iterator AI = CS.arg_begin(); ArgIndex = 1; for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(); I != E; ++I, ++AI, ++ArgIndex) if (!ArgsToPromote.count(I) && !ByValArgsToTransform.count(I)) { Args.push_back(*AI); // Unmodified argument if (Attributes Attrs = CallPAL.getParamAttributes(ArgIndex)) AttributesVec.push_back(AttributeWithIndex::get(Args.size(), Attrs)); } else if (ByValArgsToTransform.count(I)) { // Emit a GEP and load for each element of the struct. const Type *AgTy = cast(I->getType())->getElementType(); const StructType *STy = cast(AgTy); Value *Idxs[2] = { ConstantInt::get(Type::Int32Ty, 0), 0 }; for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { Idxs[1] = ConstantInt::get(Type::Int32Ty, i); Value *Idx = GetElementPtrInst::Create(*AI, Idxs, Idxs+2, (*AI)->getName()+"."+utostr(i), Call); // TODO: Tell AA about the new values? Args.push_back(new LoadInst(Idx, Idx->getName()+".val", Call)); } } else if (!I->use_empty()) { // Non-dead argument: insert GEPs and loads as appropriate. ScalarizeTable &ArgIndices = ScalarizedElements[I]; // Store the Value* version of the indices in here, but declare it now // for reuse std::vector Ops; for (ScalarizeTable::iterator SI = ArgIndices.begin(), E = ArgIndices.end(); SI != E; ++SI) { Value *V = *AI; LoadInst *OrigLoad = OriginalLoads[*SI]; if (!SI->empty()) { Ops.reserve(SI->size()); const Type *ElTy = V->getType(); for (IndicesVector::const_iterator II = SI->begin(), IE = SI->end(); II != IE; ++II) { // Use i32 to index structs, and i64 for others (pointers/arrays). // This satisfies GEP constraints. const Type *IdxTy = (isa(ElTy) ? Type::Int32Ty : Type::Int64Ty); Ops.push_back(ConstantInt::get(IdxTy, *II)); // Keep track of the type we're currently indexing ElTy = cast(ElTy)->getTypeAtIndex(*II); } // And create a GEP to extract those indices V = GetElementPtrInst::Create(V, Ops.begin(), Ops.end(), V->getName()+".idx", Call); Ops.clear(); AA.copyValue(OrigLoad->getOperand(0), V); } Args.push_back(new LoadInst(V, V->getName()+".val", Call)); AA.copyValue(OrigLoad, Args.back()); } } if (ExtraArgHack) Args.push_back(Constant::getNullValue(Type::Int32Ty)); // Push any varargs arguments on the list for (; AI != CS.arg_end(); ++AI, ++ArgIndex) { Args.push_back(*AI); if (Attributes Attrs = CallPAL.getParamAttributes(ArgIndex)) AttributesVec.push_back(AttributeWithIndex::get(Args.size(), Attrs)); } // Add any function attributes. if (Attributes attrs = CallPAL.getFnAttributes()) AttributesVec.push_back(AttributeWithIndex::get(~0, attrs)); Instruction *New; if (InvokeInst *II = dyn_cast(Call)) { New = InvokeInst::Create(NF, II->getNormalDest(), II->getUnwindDest(), Args.begin(), Args.end(), "", Call); cast(New)->setCallingConv(CS.getCallingConv()); cast(New)->setAttributes(AttrListPtr::get(AttributesVec.begin(), AttributesVec.end())); } else { New = CallInst::Create(NF, Args.begin(), Args.end(), "", Call); cast(New)->setCallingConv(CS.getCallingConv()); cast(New)->setAttributes(AttrListPtr::get(AttributesVec.begin(), AttributesVec.end())); if (cast(Call)->isTailCall()) cast(New)->setTailCall(); } Args.clear(); AttributesVec.clear(); // Update the alias analysis implementation to know that we are replacing // the old call with a new one. AA.replaceWithNewValue(Call, New); // Update the callgraph to know that the callsite has been transformed. CG[Call->getParent()->getParent()]->replaceCallSite(Call, New); if (!Call->use_empty()) { Call->replaceAllUsesWith(New); New->takeName(Call); } // Finally, remove the old call from the program, reducing the use-count of // F. Call->eraseFromParent(); } // Since we have now created the new function, splice the body of the old // function right into the new function, leaving the old rotting hulk of the // function empty. NF->getBasicBlockList().splice(NF->begin(), F->getBasicBlockList()); // Loop over the argument list, transfering uses of the old arguments over to // the new arguments, also transfering over the names as well. // for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(), I2 = NF->arg_begin(); I != E; ++I) { if (!ArgsToPromote.count(I) && !ByValArgsToTransform.count(I)) { // If this is an unmodified argument, move the name and users over to the // new version. I->replaceAllUsesWith(I2); I2->takeName(I); AA.replaceWithNewValue(I, I2); ++I2; continue; } if (ByValArgsToTransform.count(I)) { // In the callee, we create an alloca, and store each of the new incoming // arguments into the alloca. Instruction *InsertPt = NF->begin()->begin(); // Just add all the struct element types. const Type *AgTy = cast(I->getType())->getElementType(); Value *TheAlloca = new AllocaInst(AgTy, 0, "", InsertPt); const StructType *STy = cast(AgTy); Value *Idxs[2] = { ConstantInt::get(Type::Int32Ty, 0), 0 }; for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) { Idxs[1] = ConstantInt::get(Type::Int32Ty, i); std::string Name = TheAlloca->getName()+"."+utostr(i); Value *Idx = GetElementPtrInst::Create(TheAlloca, Idxs, Idxs+2, Name, InsertPt); I2->setName(I->getName()+"."+utostr(i)); new StoreInst(I2++, Idx, InsertPt); } // Anything that used the arg should now use the alloca. I->replaceAllUsesWith(TheAlloca); TheAlloca->takeName(I); AA.replaceWithNewValue(I, TheAlloca); continue; } if (I->use_empty()) { AA.deleteValue(I); continue; } // Otherwise, if we promoted this argument, then all users are load // instructions (or GEPs with only load users), and all loads should be // using the new argument that we added. ScalarizeTable &ArgIndices = ScalarizedElements[I]; while (!I->use_empty()) { if (LoadInst *LI = dyn_cast(I->use_back())) { assert(ArgIndices.begin()->empty() && "Load element should sort to front!"); I2->setName(I->getName()+".val"); LI->replaceAllUsesWith(I2); AA.replaceWithNewValue(LI, I2); LI->eraseFromParent(); DOUT << "*** Promoted load of argument '" << I->getName() << "' in function '" << F->getName() << "'\n"; } else { GetElementPtrInst *GEP = cast(I->use_back()); IndicesVector Operands; Operands.reserve(GEP->getNumIndices()); for (User::op_iterator II = GEP->idx_begin(), IE = GEP->idx_end(); II != IE; ++II) Operands.push_back(cast(*II)->getSExtValue()); // GEPs with a single 0 index can be merged with direct loads if (Operands.size() == 1 && Operands.front() == 0) Operands.clear(); Function::arg_iterator TheArg = I2; for (ScalarizeTable::iterator It = ArgIndices.begin(); *It != Operands; ++It, ++TheArg) { assert(It != ArgIndices.end() && "GEP not handled??"); } std::string NewName = I->getName(); for (unsigned i = 0, e = Operands.size(); i != e; ++i) { NewName += "." + utostr(Operands[i]); } NewName += ".val"; TheArg->setName(NewName); DOUT << "*** Promoted agg argument '" << TheArg->getName() << "' of function '" << NF->getName() << "'\n"; // All of the uses must be load instructions. Replace them all with // the argument specified by ArgNo. while (!GEP->use_empty()) { LoadInst *L = cast(GEP->use_back()); L->replaceAllUsesWith(TheArg); AA.replaceWithNewValue(L, TheArg); L->eraseFromParent(); } AA.deleteValue(GEP); GEP->eraseFromParent(); } } // Increment I2 past all of the arguments added for this promoted pointer. for (unsigned i = 0, e = ArgIndices.size(); i != e; ++i) ++I2; } // Notify the alias analysis implementation that we inserted a new argument. if (ExtraArgHack) AA.copyValue(Constant::getNullValue(Type::Int32Ty), NF->arg_begin()); // Tell the alias analysis that the old function is about to disappear. AA.replaceWithNewValue(F, NF); // Now that the old function is dead, delete it. F->eraseFromParent(); return NF; }