diff options
Diffstat (limited to 'lib/Transforms/IPO')
22 files changed, 567 insertions, 402 deletions
diff --git a/lib/Transforms/IPO/ArgumentPromotion.cpp b/lib/Transforms/IPO/ArgumentPromotion.cpp index b94dd69..385544a 100644 --- a/lib/Transforms/IPO/ArgumentPromotion.cpp +++ b/lib/Transforms/IPO/ArgumentPromotion.cpp @@ -31,21 +31,21 @@ #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/LLVMContext.h" +#include "llvm/ADT/DepthFirstIterator.h" +#include "llvm/ADT/Statistic.h" +#include "llvm/ADT/StringExtras.h" #include "llvm/Analysis/AliasAnalysis.h" #include "llvm/Analysis/CallGraph.h" -#include "llvm/Support/CallSite.h" +#include "llvm/Analysis/CallGraphSCCPass.h" +#include "llvm/IR/Constants.h" +#include "llvm/IR/DerivedTypes.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/LLVMContext.h" +#include "llvm/IR/Module.h" #include "llvm/Support/CFG.h" +#include "llvm/Support/CallSite.h" #include "llvm/Support/Debug.h" #include "llvm/Support/raw_ostream.h" -#include "llvm/ADT/DepthFirstIterator.h" -#include "llvm/ADT/Statistic.h" -#include "llvm/ADT/StringExtras.h" #include <set> using namespace llvm; @@ -153,7 +153,8 @@ CallGraphNode *ArgPromotion::PromoteArguments(CallGraphNode *CGN) { SmallPtrSet<Argument*, 8> ArgsToPromote; SmallPtrSet<Argument*, 8> ByValArgsToTransform; for (unsigned i = 0; i != PointerArgs.size(); ++i) { - bool isByVal = F->paramHasAttr(PointerArgs[i].second+1, Attribute::ByVal); + bool isByVal=F->getAttributes(). + hasAttribute(PointerArgs[i].second+1, Attribute::ByVal); Argument *PtrArg = PointerArgs[i].first; Type *AgTy = cast<PointerType>(PtrArg->getType())->getElementType(); @@ -510,15 +511,17 @@ CallGraphNode *ArgPromotion::DoPromotion(Function *F, // what the new GEP/Load instructions we are inserting look like. std::map<IndicesVector, LoadInst*> OriginalLoads; - // Attributes - Keep track of the parameter attributes for the arguments + // Attribute - 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<AttributeWithIndex, 8> AttributesVec; - const AttrListPtr &PAL = F->getAttributes(); + const AttributeSet &PAL = F->getAttributes(); // Add any return attributes. - if (Attributes attrs = PAL.getRetAttributes()) - AttributesVec.push_back(AttributeWithIndex::get(0, attrs)); + Attribute attrs = PAL.getRetAttributes(); + if (attrs.hasAttributes()) + AttributesVec.push_back(AttributeWithIndex::get(AttributeSet::ReturnIndex, + attrs)); // First, determine the new argument list unsigned ArgIndex = 1; @@ -534,7 +537,8 @@ CallGraphNode *ArgPromotion::DoPromotion(Function *F, } else if (!ArgsToPromote.count(I)) { // Unchanged argument Params.push_back(I->getType()); - if (Attributes attrs = PAL.getParamAttributes(ArgIndex)) + Attribute attrs = PAL.getParamAttributes(ArgIndex); + if (attrs.hasAttributes()) AttributesVec.push_back(AttributeWithIndex::get(Params.size(), attrs)); } else if (I->use_empty()) { // Dead argument (which are always marked as promotable) @@ -587,19 +591,13 @@ CallGraphNode *ArgPromotion::DoPromotion(Function *F, } // Add any function attributes. - if (Attributes attrs = PAL.getFnAttributes()) - AttributesVec.push_back(AttributeWithIndex::get(~0, attrs)); + attrs = PAL.getFnAttributes(); + if (attrs.hasAttributes()) + AttributesVec.push_back(AttributeWithIndex::get(AttributeSet::FunctionIndex, + attrs)); 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::getInt32Ty(F->getContext())); - } - // Construct the new function type using the new arguments. FunctionType *NFTy = FunctionType::get(RetTy, Params, FTy->isVarArg()); @@ -613,7 +611,7 @@ CallGraphNode *ArgPromotion::DoPromotion(Function *F, // Recompute the parameter attributes list based on the new arguments for // the function. - NF->setAttributes(AttrListPtr::get(AttributesVec)); + NF->setAttributes(AttributeSet::get(F->getContext(), AttributesVec)); AttributesVec.clear(); F->getParent()->getFunctionList().insert(F, NF); @@ -638,11 +636,13 @@ CallGraphNode *ArgPromotion::DoPromotion(Function *F, CallSite CS(F->use_back()); assert(CS.getCalledFunction() == F); Instruction *Call = CS.getInstruction(); - const AttrListPtr &CallPAL = CS.getAttributes(); + const AttributeSet &CallPAL = CS.getAttributes(); // Add any return attributes. - if (Attributes attrs = CallPAL.getRetAttributes()) - AttributesVec.push_back(AttributeWithIndex::get(0, attrs)); + Attribute attrs = CallPAL.getRetAttributes(); + if (attrs.hasAttributes()) + AttributesVec.push_back(AttributeWithIndex::get(AttributeSet::ReturnIndex, + attrs)); // Loop over the operands, inserting GEP and loads in the caller as // appropriate. @@ -653,7 +653,8 @@ CallGraphNode *ArgPromotion::DoPromotion(Function *F, if (!ArgsToPromote.count(I) && !ByValArgsToTransform.count(I)) { Args.push_back(*AI); // Unmodified argument - if (Attributes Attrs = CallPAL.getParamAttributes(ArgIndex)) + Attribute Attrs = CallPAL.getParamAttributes(ArgIndex); + if (Attrs.hasAttributes()) AttributesVec.push_back(AttributeWithIndex::get(Args.size(), Attrs)); } else if (ByValArgsToTransform.count(I)) { @@ -711,30 +712,32 @@ CallGraphNode *ArgPromotion::DoPromotion(Function *F, } } - if (ExtraArgHack) - Args.push_back(Constant::getNullValue(Type::getInt32Ty(F->getContext()))); - // Push any varargs arguments on the list. for (; AI != CS.arg_end(); ++AI, ++ArgIndex) { Args.push_back(*AI); - if (Attributes Attrs = CallPAL.getParamAttributes(ArgIndex)) + Attribute Attrs = CallPAL.getParamAttributes(ArgIndex); + if (Attrs.hasAttributes()) AttributesVec.push_back(AttributeWithIndex::get(Args.size(), Attrs)); } // Add any function attributes. - if (Attributes attrs = CallPAL.getFnAttributes()) - AttributesVec.push_back(AttributeWithIndex::get(~0, attrs)); + attrs = CallPAL.getFnAttributes(); + if (attrs.hasAttributes()) + AttributesVec.push_back(AttributeWithIndex::get(AttributeSet::FunctionIndex, + attrs)); Instruction *New; if (InvokeInst *II = dyn_cast<InvokeInst>(Call)) { New = InvokeInst::Create(NF, II->getNormalDest(), II->getUnwindDest(), Args, "", Call); cast<InvokeInst>(New)->setCallingConv(CS.getCallingConv()); - cast<InvokeInst>(New)->setAttributes(AttrListPtr::get(AttributesVec)); + cast<InvokeInst>(New)->setAttributes(AttributeSet::get(II->getContext(), + AttributesVec)); } else { New = CallInst::Create(NF, Args, "", Call); cast<CallInst>(New)->setCallingConv(CS.getCallingConv()); - cast<CallInst>(New)->setAttributes(AttrListPtr::get(AttributesVec)); + cast<CallInst>(New)->setAttributes(AttributeSet::get(New->getContext(), + AttributesVec)); if (cast<CallInst>(Call)->isTailCall()) cast<CallInst>(New)->setTailCall(); } @@ -870,16 +873,9 @@ CallGraphNode *ArgPromotion::DoPromotion(Function *F, } // Increment I2 past all of the arguments added for this promoted pointer. - for (unsigned i = 0, e = ArgIndices.size(); i != e; ++i) - ++I2; + std::advance(I2, ArgIndices.size()); } - // Notify the alias analysis implementation that we inserted a new argument. - if (ExtraArgHack) - AA.copyValue(Constant::getNullValue(Type::getInt32Ty(F->getContext())), - NF->arg_begin()); - - // Tell the alias analysis that the old function is about to disappear. AA.replaceWithNewValue(F, NF); diff --git a/lib/Transforms/IPO/BarrierNoopPass.cpp b/lib/Transforms/IPO/BarrierNoopPass.cpp new file mode 100644 index 0000000..2e32240 --- /dev/null +++ b/lib/Transforms/IPO/BarrierNoopPass.cpp @@ -0,0 +1,47 @@ +//===- BarrierNoopPass.cpp - A barrier pass for the pass manager ----------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// NOTE: DO NOT USE THIS IF AVOIDABLE +// +// This pass is a nonce pass intended to allow manipulation of the implicitly +// nesting pass manager. For example, it can be used to cause a CGSCC pass +// manager to be closed prior to running a new collection of function passes. +// +// FIXME: This is a huge HACK. This should be removed when the pass manager's +// nesting is made explicit instead of implicit. +// +//===----------------------------------------------------------------------===// + +#include "llvm/Pass.h" +#include "llvm/Transforms/IPO.h" +using namespace llvm; + +namespace { +/// \brief A nonce module pass used to place a barrier in a pass manager. +/// +/// There is no mechanism for ending a CGSCC pass manager once one is started. +/// This prevents extension points from having clear deterministic ordering +/// when they are phrased as non-module passes. +class BarrierNoop : public ModulePass { +public: + static char ID; // Pass identification. + + BarrierNoop() : ModulePass(ID) { + initializeBarrierNoopPass(*PassRegistry::getPassRegistry()); + } + + bool runOnModule(Module &M) { return false; } +}; +} + +ModulePass *llvm::createBarrierNoopPass() { return new BarrierNoop(); } + +char BarrierNoop::ID = 0; +INITIALIZE_PASS(BarrierNoop, "barrier", "A No-Op Barrier Pass", + false, false) diff --git a/lib/Transforms/IPO/CMakeLists.txt b/lib/Transforms/IPO/CMakeLists.txt index 3f6b1de..90c1c33 100644 --- a/lib/Transforms/IPO/CMakeLists.txt +++ b/lib/Transforms/IPO/CMakeLists.txt @@ -1,5 +1,6 @@ add_llvm_library(LLVMipo ArgumentPromotion.cpp + BarrierNoopPass.cpp ConstantMerge.cpp DeadArgumentElimination.cpp ExtractGV.cpp diff --git a/lib/Transforms/IPO/ConstantMerge.cpp b/lib/Transforms/IPO/ConstantMerge.cpp index d8fae8a..8336d3a 100644 --- a/lib/Transforms/IPO/ConstantMerge.cpp +++ b/lib/Transforms/IPO/ConstantMerge.cpp @@ -19,15 +19,15 @@ #define DEBUG_TYPE "constmerge" #include "llvm/Transforms/IPO.h" -#include "llvm/Constants.h" -#include "llvm/DerivedTypes.h" -#include "llvm/Module.h" -#include "llvm/Pass.h" -#include "llvm/Target/TargetData.h" #include "llvm/ADT/DenseMap.h" #include "llvm/ADT/PointerIntPair.h" #include "llvm/ADT/SmallPtrSet.h" #include "llvm/ADT/Statistic.h" +#include "llvm/IR/Constants.h" +#include "llvm/IR/DataLayout.h" +#include "llvm/IR/DerivedTypes.h" +#include "llvm/IR/Module.h" +#include "llvm/Pass.h" using namespace llvm; STATISTIC(NumMerged, "Number of global constants merged"); @@ -50,7 +50,7 @@ namespace { // alignment to a concrete value. unsigned getAlignment(GlobalVariable *GV) const; - const TargetData *TD; + const DataLayout *TD; }; } @@ -98,7 +98,7 @@ unsigned ConstantMerge::getAlignment(GlobalVariable *GV) const { } bool ConstantMerge::runOnModule(Module &M) { - TD = getAnalysisIfAvailable<TargetData>(); + TD = getAnalysisIfAvailable<DataLayout>(); // Find all the globals that are marked "used". These cannot be merged. SmallPtrSet<const GlobalValue*, 8> UsedGlobals; @@ -107,7 +107,7 @@ bool ConstantMerge::runOnModule(Module &M) { // Map unique <constants, has-unknown-alignment> pairs to globals. We don't // want to merge globals of unknown alignment with those of explicit - // alignment. If we have TargetData, we always know the alignment. + // alignment. If we have DataLayout, we always know the alignment. DenseMap<PointerIntPair<Constant*, 1, bool>, GlobalVariable*> CMap; // Replacements - This vector contains a list of replacements to perform. diff --git a/lib/Transforms/IPO/DeadArgumentElimination.cpp b/lib/Transforms/IPO/DeadArgumentElimination.cpp index fd23a93..ff040e7 100644 --- a/lib/Transforms/IPO/DeadArgumentElimination.cpp +++ b/lib/Transforms/IPO/DeadArgumentElimination.cpp @@ -19,20 +19,23 @@ #define DEBUG_TYPE "deadargelim" #include "llvm/Transforms/IPO.h" -#include "llvm/CallingConv.h" -#include "llvm/Constant.h" -#include "llvm/DerivedTypes.h" -#include "llvm/Instructions.h" -#include "llvm/IntrinsicInst.h" -#include "llvm/LLVMContext.h" -#include "llvm/Module.h" +#include "llvm/ADT/DenseMap.h" +#include "llvm/ADT/SmallVector.h" +#include "llvm/ADT/Statistic.h" +#include "llvm/ADT/StringExtras.h" +#include "llvm/DIBuilder.h" +#include "llvm/DebugInfo.h" +#include "llvm/IR/CallingConv.h" +#include "llvm/IR/Constant.h" +#include "llvm/IR/DerivedTypes.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/IntrinsicInst.h" +#include "llvm/IR/LLVMContext.h" +#include "llvm/IR/Module.h" #include "llvm/Pass.h" #include "llvm/Support/CallSite.h" #include "llvm/Support/Debug.h" #include "llvm/Support/raw_ostream.h" -#include "llvm/ADT/SmallVector.h" -#include "llvm/ADT/Statistic.h" -#include "llvm/ADT/StringExtras.h" #include <map> #include <set> using namespace llvm; @@ -121,6 +124,15 @@ namespace { typedef SmallVector<RetOrArg, 5> UseVector; + // Map each LLVM function to corresponding metadata with debug info. If + // the function is replaced with another one, we should patch the pointer + // to LLVM function in metadata. + // As the code generation for module is finished (and DIBuilder is + // finalized) we assume that subprogram descriptors won't be changed, and + // they are stored in map for short duration anyway. + typedef DenseMap<Function*, DISubprogram> FunctionDIMap; + FunctionDIMap FunctionDIs; + protected: // DAH uses this to specify a different ID. explicit DAE(char &ID) : ModulePass(ID) {} @@ -141,6 +153,7 @@ namespace { unsigned RetValNum = 0); Liveness SurveyUses(const Value *V, UseVector &MaybeLiveUses); + void CollectFunctionDIs(Module &M); void SurveyFunction(const Function &F); void MarkValue(const RetOrArg &RA, Liveness L, const UseVector &MaybeLiveUses); @@ -180,6 +193,33 @@ INITIALIZE_PASS(DAH, "deadarghaX0r", ModulePass *llvm::createDeadArgEliminationPass() { return new DAE(); } ModulePass *llvm::createDeadArgHackingPass() { return new DAH(); } +/// CollectFunctionDIs - Map each function in the module to its debug info +/// descriptor. +void DAE::CollectFunctionDIs(Module &M) { + FunctionDIs.clear(); + + for (Module::named_metadata_iterator I = M.named_metadata_begin(), + E = M.named_metadata_end(); I != E; ++I) { + NamedMDNode &NMD = *I; + for (unsigned MDIndex = 0, MDNum = NMD.getNumOperands(); + MDIndex < MDNum; ++MDIndex) { + MDNode *Node = NMD.getOperand(MDIndex); + if (!DIDescriptor(Node).isCompileUnit()) + continue; + DICompileUnit CU(Node); + const DIArray &SPs = CU.getSubprograms(); + for (unsigned SPIndex = 0, SPNum = SPs.getNumElements(); + SPIndex < SPNum; ++SPIndex) { + DISubprogram SP(SPs.getElement(SPIndex)); + if (!SP.Verify()) + continue; + if (Function *F = SP.getFunction()) + FunctionDIs[F] = SP; + } + } + } +} + /// DeleteDeadVarargs - If this is an function that takes a ... list, and if /// llvm.vastart is never called, the varargs list is dead for the function. bool DAE::DeleteDeadVarargs(Function &Fn) { @@ -231,14 +271,16 @@ bool DAE::DeleteDeadVarargs(Function &Fn) { Args.assign(CS.arg_begin(), CS.arg_begin() + NumArgs); // Drop any attributes that were on the vararg arguments. - AttrListPtr PAL = CS.getAttributes(); + AttributeSet PAL = CS.getAttributes(); if (!PAL.isEmpty() && PAL.getSlot(PAL.getNumSlots() - 1).Index > NumArgs) { SmallVector<AttributeWithIndex, 8> AttributesVec; for (unsigned i = 0; PAL.getSlot(i).Index <= NumArgs; ++i) AttributesVec.push_back(PAL.getSlot(i)); - if (Attributes FnAttrs = PAL.getFnAttributes()) - AttributesVec.push_back(AttributeWithIndex::get(~0, FnAttrs)); - PAL = AttrListPtr::get(AttributesVec); + Attribute FnAttrs = PAL.getFnAttributes(); + if (FnAttrs.hasAttributes()) + AttributesVec.push_back(AttributeWithIndex::get(AttributeSet::FunctionIndex, + FnAttrs)); + PAL = AttributeSet::get(Fn.getContext(), AttributesVec); } Instruction *New; @@ -284,6 +326,11 @@ bool DAE::DeleteDeadVarargs(Function &Fn) { I2->takeName(I); } + // Patch the pointer to LLVM function in debug info descriptor. + FunctionDIMap::iterator DI = FunctionDIs.find(&Fn); + if (DI != FunctionDIs.end()) + DI->second.replaceFunction(NF); + // Finally, nuke the old function. Fn.eraseFromParent(); return true; @@ -651,11 +698,11 @@ bool DAE::RemoveDeadStuffFromFunction(Function *F) { // Set up to build a new list of parameter attributes. SmallVector<AttributeWithIndex, 8> AttributesVec; - const AttrListPtr &PAL = F->getAttributes(); + const AttributeSet &PAL = F->getAttributes(); // The existing function return attributes. - Attributes RAttrs = PAL.getRetAttributes(); - Attributes FnAttrs = PAL.getFnAttributes(); + Attribute RAttrs = PAL.getRetAttributes(); + Attribute FnAttrs = PAL.getFnAttributes(); // Find out the new return value. @@ -717,13 +764,17 @@ bool DAE::RemoveDeadStuffFromFunction(Function *F) { // here. Currently, this should not be possible, but special handling might be // required when new return value attributes are added. if (NRetTy->isVoidTy()) - RAttrs &= ~Attribute::typeIncompatible(NRetTy); + RAttrs = + Attribute::get(NRetTy->getContext(), AttrBuilder(RAttrs). + removeAttributes(Attribute::typeIncompatible(NRetTy))); else - assert((RAttrs & Attribute::typeIncompatible(NRetTy)) == 0 - && "Return attributes no longer compatible?"); + assert(!AttrBuilder(RAttrs). + hasAttributes(Attribute::typeIncompatible(NRetTy)) && + "Return attributes no longer compatible?"); - if (RAttrs) - AttributesVec.push_back(AttributeWithIndex::get(0, RAttrs)); + if (RAttrs.hasAttributes()) + AttributesVec.push_back(AttributeWithIndex::get(AttributeSet::ReturnIndex, + RAttrs)); // Remember which arguments are still alive. SmallVector<bool, 10> ArgAlive(FTy->getNumParams(), false); @@ -740,7 +791,8 @@ bool DAE::RemoveDeadStuffFromFunction(Function *F) { // Get the original parameter attributes (skipping the first one, that is // for the return value. - if (Attributes Attrs = PAL.getParamAttributes(i + 1)) + Attribute Attrs = PAL.getParamAttributes(i + 1); + if (Attrs.hasAttributes()) AttributesVec.push_back(AttributeWithIndex::get(Params.size(), Attrs)); } else { ++NumArgumentsEliminated; @@ -749,11 +801,12 @@ bool DAE::RemoveDeadStuffFromFunction(Function *F) { } } - if (FnAttrs != Attribute::None) - AttributesVec.push_back(AttributeWithIndex::get(~0, FnAttrs)); + if (FnAttrs.hasAttributes()) + AttributesVec.push_back(AttributeWithIndex::get(AttributeSet::FunctionIndex, + FnAttrs)); // Reconstruct the AttributesList based on the vector we constructed. - AttrListPtr NewPAL = AttrListPtr::get(AttributesVec); + AttributeSet NewPAL = AttributeSet::get(F->getContext(), AttributesVec); // Create the new function type based on the recomputed parameters. FunctionType *NFTy = FunctionType::get(NRetTy, Params, FTy->isVarArg()); @@ -780,15 +833,18 @@ bool DAE::RemoveDeadStuffFromFunction(Function *F) { Instruction *Call = CS.getInstruction(); AttributesVec.clear(); - const AttrListPtr &CallPAL = CS.getAttributes(); + const AttributeSet &CallPAL = CS.getAttributes(); // The call return attributes. - Attributes RAttrs = CallPAL.getRetAttributes(); - Attributes FnAttrs = CallPAL.getFnAttributes(); + Attribute RAttrs = CallPAL.getRetAttributes(); + Attribute FnAttrs = CallPAL.getFnAttributes(); // Adjust in case the function was changed to return void. - RAttrs &= ~Attribute::typeIncompatible(NF->getReturnType()); - if (RAttrs) - AttributesVec.push_back(AttributeWithIndex::get(0, RAttrs)); + RAttrs = + Attribute::get(NF->getContext(), AttrBuilder(RAttrs). + removeAttributes(Attribute::typeIncompatible(NF->getReturnType()))); + if (RAttrs.hasAttributes()) + AttributesVec.push_back(AttributeWithIndex::get(AttributeSet::ReturnIndex, + RAttrs)); // Declare these outside of the loops, so we can reuse them for the second // loop, which loops the varargs. @@ -800,22 +856,25 @@ bool DAE::RemoveDeadStuffFromFunction(Function *F) { if (ArgAlive[i]) { Args.push_back(*I); // Get original parameter attributes, but skip return attributes. - if (Attributes Attrs = CallPAL.getParamAttributes(i + 1)) + Attribute Attrs = CallPAL.getParamAttributes(i + 1); + if (Attrs.hasAttributes()) AttributesVec.push_back(AttributeWithIndex::get(Args.size(), Attrs)); } // Push any varargs arguments on the list. Don't forget their attributes. for (CallSite::arg_iterator E = CS.arg_end(); I != E; ++I, ++i) { Args.push_back(*I); - if (Attributes Attrs = CallPAL.getParamAttributes(i + 1)) + Attribute Attrs = CallPAL.getParamAttributes(i + 1); + if (Attrs.hasAttributes()) AttributesVec.push_back(AttributeWithIndex::get(Args.size(), Attrs)); } - if (FnAttrs != Attribute::None) - AttributesVec.push_back(AttributeWithIndex::get(~0, FnAttrs)); + if (FnAttrs.hasAttributes()) + AttributesVec.push_back(AttributeWithIndex::get(AttributeSet::FunctionIndex, + FnAttrs)); // Reconstruct the AttributesList based on the vector we constructed. - AttrListPtr NewCallPAL = AttrListPtr::get(AttributesVec); + AttributeSet NewCallPAL = AttributeSet::get(F->getContext(), AttributesVec); Instruction *New; if (InvokeInst *II = dyn_cast<InvokeInst>(Call)) { @@ -952,6 +1011,11 @@ bool DAE::RemoveDeadStuffFromFunction(Function *F) { BB->getInstList().erase(RI); } + // Patch the pointer to LLVM function in debug info descriptor. + FunctionDIMap::iterator DI = FunctionDIs.find(F); + if (DI != FunctionDIs.end()) + DI->second.replaceFunction(NF); + // Now that the old function is dead, delete it. F->eraseFromParent(); @@ -961,6 +1025,9 @@ bool DAE::RemoveDeadStuffFromFunction(Function *F) { bool DAE::runOnModule(Module &M) { bool Changed = false; + // Collect debug info descriptors for functions. + CollectFunctionDIs(M); + // First pass: Do a simple check to see if any functions can have their "..." // removed. We can do this if they never call va_start. This loop cannot be // fused with the next loop, because deleting a function invalidates diff --git a/lib/Transforms/IPO/ExtractGV.cpp b/lib/Transforms/IPO/ExtractGV.cpp index 4c7f0ed..8a6bfc6 100644 --- a/lib/Transforms/IPO/ExtractGV.cpp +++ b/lib/Transforms/IPO/ExtractGV.cpp @@ -11,13 +11,13 @@ // //===----------------------------------------------------------------------===// -#include "llvm/Instructions.h" -#include "llvm/LLVMContext.h" -#include "llvm/Module.h" -#include "llvm/Pass.h" -#include "llvm/Constants.h" #include "llvm/Transforms/IPO.h" #include "llvm/ADT/SetVector.h" +#include "llvm/IR/Constants.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/LLVMContext.h" +#include "llvm/IR/Module.h" +#include "llvm/Pass.h" #include <algorithm> using namespace llvm; @@ -51,32 +51,75 @@ namespace { // Visit the GlobalVariables. for (Module::global_iterator I = M.global_begin(), E = M.global_end(); I != E; ++I) { - if (deleteStuff == (bool)Named.count(I) && !I->isDeclaration()) { - I->setInitializer(0); - } else { + bool Delete = + deleteStuff == (bool)Named.count(I) && !I->isDeclaration(); + if (!Delete) { if (I->hasAvailableExternallyLinkage()) continue; if (I->getName() == "llvm.global_ctors") continue; } - if (I->hasLocalLinkage()) + bool Local = I->hasLocalLinkage(); + if (Local) I->setVisibility(GlobalValue::HiddenVisibility); - I->setLinkage(GlobalValue::ExternalLinkage); + + if (Local || Delete) + I->setLinkage(GlobalValue::ExternalLinkage); + + if (Delete) + I->setInitializer(0); } // Visit the Functions. for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) { - if (deleteStuff == (bool)Named.count(I) && !I->isDeclaration()) { - I->deleteBody(); - } else { + bool Delete = + deleteStuff == (bool)Named.count(I) && !I->isDeclaration(); + if (!Delete) { if (I->hasAvailableExternallyLinkage()) continue; } - if (I->hasLocalLinkage()) + bool Local = I->hasLocalLinkage(); + if (Local) I->setVisibility(GlobalValue::HiddenVisibility); - I->setLinkage(GlobalValue::ExternalLinkage); + + if (Local || Delete) + I->setLinkage(GlobalValue::ExternalLinkage); + + if (Delete) + I->deleteBody(); + } + + // Visit the Aliases. + for (Module::alias_iterator I = M.alias_begin(), E = M.alias_end(); + I != E;) { + Module::alias_iterator CurI = I; + ++I; + + if (CurI->hasLocalLinkage()) { + CurI->setVisibility(GlobalValue::HiddenVisibility); + CurI->setLinkage(GlobalValue::ExternalLinkage); + } + + if (deleteStuff == (bool)Named.count(CurI)) { + Type *Ty = CurI->getType()->getElementType(); + + CurI->removeFromParent(); + llvm::Value *Declaration; + if (FunctionType *FTy = dyn_cast<FunctionType>(Ty)) { + Declaration = Function::Create(FTy, GlobalValue::ExternalLinkage, + CurI->getName(), &M); + + } else { + Declaration = + new GlobalVariable(M, Ty, false, GlobalValue::ExternalLinkage, + 0, CurI->getName()); + + } + CurI->replaceAllUsesWith(Declaration); + delete CurI; + } } return true; diff --git a/lib/Transforms/IPO/FunctionAttrs.cpp b/lib/Transforms/IPO/FunctionAttrs.cpp index f3f6228..e9bc4ad 100644 --- a/lib/Transforms/IPO/FunctionAttrs.cpp +++ b/lib/Transforms/IPO/FunctionAttrs.cpp @@ -20,17 +20,17 @@ #define DEBUG_TYPE "functionattrs" #include "llvm/Transforms/IPO.h" -#include "llvm/CallGraphSCCPass.h" -#include "llvm/GlobalVariable.h" -#include "llvm/IntrinsicInst.h" -#include "llvm/LLVMContext.h" -#include "llvm/Analysis/AliasAnalysis.h" -#include "llvm/Analysis/CallGraph.h" -#include "llvm/Analysis/CaptureTracking.h" #include "llvm/ADT/SCCIterator.h" +#include "llvm/ADT/SetVector.h" #include "llvm/ADT/SmallSet.h" #include "llvm/ADT/Statistic.h" -#include "llvm/ADT/UniqueVector.h" +#include "llvm/Analysis/AliasAnalysis.h" +#include "llvm/Analysis/CallGraph.h" +#include "llvm/Analysis/CallGraphSCCPass.h" +#include "llvm/Analysis/CaptureTracking.h" +#include "llvm/IR/GlobalVariable.h" +#include "llvm/IR/IntrinsicInst.h" +#include "llvm/IR/LLVMContext.h" #include "llvm/Support/InstIterator.h" using namespace llvm; @@ -212,10 +212,17 @@ bool FunctionAttrs::AddReadAttrs(const CallGraphSCC &SCC) { MadeChange = true; // Clear out any existing attributes. - F->removeAttribute(~0, Attribute::ReadOnly | Attribute::ReadNone); + AttrBuilder B; + B.addAttribute(Attribute::ReadOnly) + .addAttribute(Attribute::ReadNone); + F->removeAttribute(AttributeSet::FunctionIndex, + Attribute::get(F->getContext(), B)); // Add in the new attribute. - F->addAttribute(~0, ReadsMemory? Attribute::ReadOnly : Attribute::ReadNone); + B.clear(); + B.addAttribute(ReadsMemory ? Attribute::ReadOnly : Attribute::ReadNone); + F->addAttribute(AttributeSet::FunctionIndex, + Attribute::get(F->getContext(), B)); if (ReadsMemory) ++NumReadOnly; @@ -276,8 +283,6 @@ namespace { void tooManyUses() { Captured = true; } - bool shouldExplore(Use *U) { return true; } - bool captured(Use *U) { CallSite CS(U->getUser()); if (!CS.getInstruction()) { Captured = true; return true; } @@ -352,6 +357,9 @@ bool FunctionAttrs::AddNoCaptureAttrs(const CallGraphSCC &SCC) { ArgumentGraph AG; + AttrBuilder B; + B.addAttribute(Attribute::NoCapture); + // Check each function in turn, determining which pointer arguments are not // captured. for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I) { @@ -373,7 +381,7 @@ bool FunctionAttrs::AddNoCaptureAttrs(const CallGraphSCC &SCC) { for (Function::arg_iterator A = F->arg_begin(), E = F->arg_end(); A != E; ++A) { if (A->getType()->isPointerTy() && !A->hasNoCaptureAttr()) { - A->addAttr(Attribute::NoCapture); + A->addAttr(Attribute::get(F->getContext(), B)); ++NumNoCapture; Changed = true; } @@ -388,7 +396,7 @@ bool FunctionAttrs::AddNoCaptureAttrs(const CallGraphSCC &SCC) { if (!Tracker.Captured) { if (Tracker.Uses.empty()) { // If it's trivially not captured, mark it nocapture now. - A->addAttr(Attribute::NoCapture); + A->addAttr(Attribute::get(F->getContext(), B)); ++NumNoCapture; Changed = true; } else { @@ -421,7 +429,9 @@ bool FunctionAttrs::AddNoCaptureAttrs(const CallGraphSCC &SCC) { // eg. "void f(int* x) { if (...) f(x); }" if (ArgumentSCC[0]->Uses.size() == 1 && ArgumentSCC[0]->Uses[0] == ArgumentSCC[0]) { - ArgumentSCC[0]->Definition->addAttr(Attribute::NoCapture); + ArgumentSCC[0]-> + Definition-> + addAttr(Attribute::get(ArgumentSCC[0]->Definition->getContext(), B)); ++NumNoCapture; Changed = true; } @@ -463,7 +473,7 @@ bool FunctionAttrs::AddNoCaptureAttrs(const CallGraphSCC &SCC) { for (unsigned i = 0, e = ArgumentSCC.size(); i != e; ++i) { Argument *A = ArgumentSCC[i]->Definition; - A->addAttr(Attribute::NoCapture); + A->addAttr(Attribute::get(A->getContext(), B)); ++NumNoCapture; Changed = true; } @@ -476,13 +486,13 @@ bool FunctionAttrs::AddNoCaptureAttrs(const CallGraphSCC &SCC) { /// or a pointer that doesn't alias any other pointer visible to the caller. bool FunctionAttrs::IsFunctionMallocLike(Function *F, SmallPtrSet<Function*, 8> &SCCNodes) const { - UniqueVector<Value *> FlowsToReturn; + SmallSetVector<Value *, 8> FlowsToReturn; for (Function::iterator I = F->begin(), E = F->end(); I != E; ++I) if (ReturnInst *Ret = dyn_cast<ReturnInst>(I->getTerminator())) FlowsToReturn.insert(Ret->getReturnValue()); for (unsigned i = 0; i != FlowsToReturn.size(); ++i) { - Value *RetVal = FlowsToReturn[i+1]; // UniqueVector[0] is reserved. + Value *RetVal = FlowsToReturn[i]; if (Constant *C = dyn_cast<Constant>(RetVal)) { if (!C->isNullValue() && !isa<UndefValue>(C)) diff --git a/lib/Transforms/IPO/GlobalDCE.cpp b/lib/Transforms/IPO/GlobalDCE.cpp index 18c1c7b..dc99492 100644 --- a/lib/Transforms/IPO/GlobalDCE.cpp +++ b/lib/Transforms/IPO/GlobalDCE.cpp @@ -17,11 +17,11 @@ #define DEBUG_TYPE "globaldce" #include "llvm/Transforms/IPO.h" -#include "llvm/Constants.h" -#include "llvm/Module.h" -#include "llvm/Pass.h" #include "llvm/ADT/SmallPtrSet.h" #include "llvm/ADT/Statistic.h" +#include "llvm/IR/Constants.h" +#include "llvm/IR/Module.h" +#include "llvm/Pass.h" using namespace llvm; STATISTIC(NumAliases , "Number of global aliases removed"); diff --git a/lib/Transforms/IPO/GlobalOpt.cpp b/lib/Transforms/IPO/GlobalOpt.cpp index b888e95..abd37c2 100644 --- a/lib/Transforms/IPO/GlobalOpt.cpp +++ b/lib/Transforms/IPO/GlobalOpt.cpp @@ -15,29 +15,29 @@ #define DEBUG_TYPE "globalopt" #include "llvm/Transforms/IPO.h" -#include "llvm/CallingConv.h" -#include "llvm/Constants.h" -#include "llvm/DerivedTypes.h" -#include "llvm/Instructions.h" -#include "llvm/IntrinsicInst.h" -#include "llvm/Module.h" -#include "llvm/Operator.h" -#include "llvm/Pass.h" +#include "llvm/ADT/DenseMap.h" +#include "llvm/ADT/STLExtras.h" +#include "llvm/ADT/SmallPtrSet.h" +#include "llvm/ADT/SmallVector.h" +#include "llvm/ADT/Statistic.h" #include "llvm/Analysis/ConstantFolding.h" #include "llvm/Analysis/MemoryBuiltins.h" -#include "llvm/Target/TargetData.h" -#include "llvm/Target/TargetLibraryInfo.h" +#include "llvm/IR/CallingConv.h" +#include "llvm/IR/Constants.h" +#include "llvm/IR/DataLayout.h" +#include "llvm/IR/DerivedTypes.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/IntrinsicInst.h" +#include "llvm/IR/Module.h" +#include "llvm/IR/Operator.h" +#include "llvm/Pass.h" #include "llvm/Support/CallSite.h" #include "llvm/Support/Debug.h" #include "llvm/Support/ErrorHandling.h" #include "llvm/Support/GetElementPtrTypeIterator.h" #include "llvm/Support/MathExtras.h" #include "llvm/Support/raw_ostream.h" -#include "llvm/ADT/DenseMap.h" -#include "llvm/ADT/SmallPtrSet.h" -#include "llvm/ADT/SmallVector.h" -#include "llvm/ADT/Statistic.h" -#include "llvm/ADT/STLExtras.h" +#include "llvm/Target/TargetLibraryInfo.h" #include <algorithm> using namespace llvm; @@ -83,7 +83,7 @@ namespace { const GlobalStatus &GS); bool OptimizeEmptyGlobalCXXDtors(Function *CXAAtExitFn); - TargetData *TD; + DataLayout *TD; TargetLibraryInfo *TLI; }; } @@ -148,17 +148,13 @@ struct GlobalStatus { /// an instruction (e.g. a constant expr or GV initializer). bool HasNonInstructionUser; - /// HasPHIUser - Set to true if this global has a user that is a PHI node. - bool HasPHIUser; - /// AtomicOrdering - Set to the strongest atomic ordering requirement. AtomicOrdering Ordering; GlobalStatus() : isCompared(false), isLoaded(false), StoredType(NotStored), StoredOnceValue(0), AccessingFunction(0), HasMultipleAccessingFunctions(false), - HasNonInstructionUser(false), HasPHIUser(false), - Ordering(NotAtomic) {} + HasNonInstructionUser(false), Ordering(NotAtomic) {} }; } @@ -200,11 +196,11 @@ static bool AnalyzeGlobal(const Value *V, GlobalStatus &GS, const User *U = *UI; if (const ConstantExpr *CE = dyn_cast<ConstantExpr>(U)) { GS.HasNonInstructionUser = true; - + // If the result of the constantexpr isn't pointer type, then we won't // know to expect it in various places. Just reject early. if (!isa<PointerType>(CE->getType())) return true; - + if (AnalyzeGlobal(CE, GS, PHIUsers)) return true; } else if (const Instruction *I = dyn_cast<Instruction>(U)) { if (!GS.HasMultipleAccessingFunctions) { @@ -225,6 +221,7 @@ static bool AnalyzeGlobal(const Value *V, GlobalStatus &GS, // Don't hack on volatile stores. if (SI->isVolatile()) return true; + GS.Ordering = StrongerOrdering(GS.Ordering, SI->getOrdering()); // If this is a direct store to the global (i.e., the global is a scalar @@ -234,6 +231,14 @@ static bool AnalyzeGlobal(const Value *V, GlobalStatus &GS, if (const GlobalVariable *GV = dyn_cast<GlobalVariable>( SI->getOperand(1))) { Value *StoredVal = SI->getOperand(0); + + if (Constant *C = dyn_cast<Constant>(StoredVal)) { + if (C->isThreadDependent()) { + // The stored value changes between threads; don't track it. + return true; + } + } + if (StoredVal == GV->getInitializer()) { if (GS.StoredType < GlobalStatus::isInitializerStored) GS.StoredType = GlobalStatus::isInitializerStored; @@ -265,7 +270,6 @@ static bool AnalyzeGlobal(const Value *V, GlobalStatus &GS, // have to be careful about infinite recursion. if (PHIUsers.insert(PN)) // Not already visited. if (AnalyzeGlobal(I, GS, PHIUsers)) return true; - GS.HasPHIUser = true; } else if (isa<CmpInst>(I)) { GS.isCompared = true; } else if (const MemTransferInst *MTI = dyn_cast<MemTransferInst>(I)) { @@ -464,7 +468,7 @@ static bool CleanupPointerRootUsers(GlobalVariable *GV, /// quick scan over the use list to clean up the easy and obvious cruft. This /// returns true if it made a change. static bool CleanupConstantGlobalUsers(Value *V, Constant *Init, - TargetData *TD, TargetLibraryInfo *TLI) { + DataLayout *TD, TargetLibraryInfo *TLI) { bool Changed = false; for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI != E;) { User *U = *UI++; @@ -656,7 +660,7 @@ static bool GlobalUsersSafeToSRA(GlobalValue *GV) { /// behavior of the program in a more fine-grained way. We have determined that /// this transformation is safe already. We return the first global variable we /// insert so that the caller can reprocess it. -static GlobalVariable *SRAGlobal(GlobalVariable *GV, const TargetData &TD) { +static GlobalVariable *SRAGlobal(GlobalVariable *GV, const DataLayout &TD) { // Make sure this global only has simple uses that we can SRA. if (!GlobalUsersSafeToSRA(GV)) return 0; @@ -932,7 +936,7 @@ static bool OptimizeAwayTrappingUsesOfValue(Value *V, Constant *NewV) { /// if the loaded value is dynamically null, then we know that they cannot be /// reachable with a null optimize away the load. static bool OptimizeAwayTrappingUsesOfLoads(GlobalVariable *GV, Constant *LV, - TargetData *TD, + DataLayout *TD, TargetLibraryInfo *TLI) { bool Changed = false; @@ -962,7 +966,9 @@ static bool OptimizeAwayTrappingUsesOfLoads(GlobalVariable *GV, Constant *LV, // If we get here we could have other crazy uses that are transitively // loaded. assert((isa<PHINode>(GlobalUser) || isa<SelectInst>(GlobalUser) || - isa<ConstantExpr>(GlobalUser) || isa<CmpInst>(GlobalUser)) && + isa<ConstantExpr>(GlobalUser) || isa<CmpInst>(GlobalUser) || + isa<BitCastInst>(GlobalUser) || + isa<GetElementPtrInst>(GlobalUser)) && "Only expect load and stores!"); } } @@ -994,7 +1000,7 @@ static bool OptimizeAwayTrappingUsesOfLoads(GlobalVariable *GV, Constant *LV, /// ConstantPropUsersOf - Walk the use list of V, constant folding all of the /// instructions that are foldable. static void ConstantPropUsersOf(Value *V, - TargetData *TD, TargetLibraryInfo *TLI) { + DataLayout *TD, TargetLibraryInfo *TLI) { for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI != E; ) if (Instruction *I = dyn_cast<Instruction>(*UI++)) if (Constant *NewC = ConstantFoldInstruction(I, TD, TLI)) { @@ -1017,7 +1023,7 @@ static GlobalVariable *OptimizeGlobalAddressOfMalloc(GlobalVariable *GV, CallInst *CI, Type *AllocTy, ConstantInt *NElements, - TargetData *TD, + DataLayout *TD, TargetLibraryInfo *TLI) { DEBUG(errs() << "PROMOTING GLOBAL: " << *GV << " CALL = " << *CI << '\n'); @@ -1466,7 +1472,7 @@ static void RewriteUsesOfLoadForHeapSRoA(LoadInst *Load, /// PerformHeapAllocSRoA - CI is an allocation of an array of structures. Break /// it up into multiple allocations of arrays of the fields. static GlobalVariable *PerformHeapAllocSRoA(GlobalVariable *GV, CallInst *CI, - Value *NElems, TargetData *TD, + Value *NElems, DataLayout *TD, const TargetLibraryInfo *TLI) { DEBUG(dbgs() << "SROA HEAP ALLOC: " << *GV << " MALLOC = " << *CI << '\n'); Type *MAT = getMallocAllocatedType(CI, TLI); @@ -1658,7 +1664,7 @@ static bool TryToOptimizeStoreOfMallocToGlobal(GlobalVariable *GV, Type *AllocTy, AtomicOrdering Ordering, Module::global_iterator &GVI, - TargetData *TD, + DataLayout *TD, TargetLibraryInfo *TLI) { if (!TD) return false; @@ -1757,7 +1763,7 @@ static bool TryToOptimizeStoreOfMallocToGlobal(GlobalVariable *GV, static bool OptimizeOnceStoredGlobal(GlobalVariable *GV, Value *StoredOnceVal, AtomicOrdering Ordering, Module::global_iterator &GVI, - TargetData *TD, TargetLibraryInfo *TLI) { + DataLayout *TD, TargetLibraryInfo *TLI) { // Ignore no-op GEPs and bitcasts. StoredOnceVal = StoredOnceVal->stripPointerCasts(); @@ -2000,7 +2006,7 @@ bool GlobalOpt::ProcessInternalGlobal(GlobalVariable *GV, ++NumMarked; return true; } else if (!GV->getInitializer()->getType()->isSingleValueType()) { - if (TargetData *TD = getAnalysisIfAvailable<TargetData>()) + if (DataLayout *TD = getAnalysisIfAvailable<DataLayout>()) if (GlobalVariable *FirstNewGV = SRAGlobal(GV, *TD)) { GVI = FirstNewGV; // Don't skip the newly produced globals! return true; @@ -2059,25 +2065,26 @@ static void ChangeCalleesToFastCall(Function *F) { } } -static AttrListPtr StripNest(const AttrListPtr &Attrs) { +static AttributeSet StripNest(LLVMContext &C, const AttributeSet &Attrs) { for (unsigned i = 0, e = Attrs.getNumSlots(); i != e; ++i) { - if ((Attrs.getSlot(i).Attrs & Attribute::Nest) == 0) + if (!Attrs.getSlot(i).Attrs.hasAttribute(Attribute::Nest)) continue; // There can be only one. - return Attrs.removeAttr(Attrs.getSlot(i).Index, Attribute::Nest); + return Attrs.removeAttr(C, Attrs.getSlot(i).Index, + Attribute::get(C, Attribute::Nest)); } return Attrs; } static void RemoveNestAttribute(Function *F) { - F->setAttributes(StripNest(F->getAttributes())); + F->setAttributes(StripNest(F->getContext(), F->getAttributes())); for (Value::use_iterator UI = F->use_begin(), E = F->use_end(); UI != E;++UI){ if (isa<BlockAddress>(*UI)) continue; CallSite User(cast<Instruction>(*UI)); - User.setAttributes(StripNest(User.getAttributes())); + User.setAttributes(StripNest(F->getContext(), User.getAttributes())); } } @@ -2145,7 +2152,7 @@ bool GlobalOpt::OptimizeGlobalVars(Module &M) { GlobalVariable *GlobalOpt::FindGlobalCtors(Module &M) { GlobalVariable *GV = M.getGlobalVariable("llvm.global_ctors"); if (GV == 0) return 0; - + // Verify that the initializer is simple enough for us to handle. We are // only allowed to optimize the initializer if it is unique. if (!GV->hasUniqueInitializer()) return 0; @@ -2251,10 +2258,10 @@ static GlobalVariable *InstallGlobalCtors(GlobalVariable *GCL, } -static inline bool +static inline bool isSimpleEnoughValueToCommit(Constant *C, SmallPtrSet<Constant*, 8> &SimpleConstants, - const TargetData *TD); + const DataLayout *TD); /// isSimpleEnoughValueToCommit - Return true if the specified constant can be @@ -2267,13 +2274,13 @@ isSimpleEnoughValueToCommit(Constant *C, /// time. static bool isSimpleEnoughValueToCommitHelper(Constant *C, SmallPtrSet<Constant*, 8> &SimpleConstants, - const TargetData *TD) { + const DataLayout *TD) { // Simple integer, undef, constant aggregate zero, global addresses, etc are // all supported. if (C->getNumOperands() == 0 || isa<BlockAddress>(C) || isa<GlobalValue>(C)) return true; - + // Aggregate values are safe if all their elements are. if (isa<ConstantArray>(C) || isa<ConstantStruct>(C) || isa<ConstantVector>(C)) { @@ -2284,7 +2291,7 @@ static bool isSimpleEnoughValueToCommitHelper(Constant *C, } return true; } - + // We don't know exactly what relocations are allowed in constant expressions, // so we allow &global+constantoffset, which is safe and uniformly supported // across targets. @@ -2302,14 +2309,14 @@ static bool isSimpleEnoughValueToCommitHelper(Constant *C, TD->getTypeSizeInBits(CE->getOperand(0)->getType())) return false; return isSimpleEnoughValueToCommit(CE->getOperand(0), SimpleConstants, TD); - + // GEP is fine if it is simple + constant offset. case Instruction::GetElementPtr: for (unsigned i = 1, e = CE->getNumOperands(); i != e; ++i) if (!isa<ConstantInt>(CE->getOperand(i))) return false; return isSimpleEnoughValueToCommit(CE->getOperand(0), SimpleConstants, TD); - + case Instruction::Add: // We allow simple+cst. if (!isa<ConstantInt>(CE->getOperand(1))) @@ -2319,10 +2326,10 @@ static bool isSimpleEnoughValueToCommitHelper(Constant *C, return false; } -static inline bool +static inline bool isSimpleEnoughValueToCommit(Constant *C, SmallPtrSet<Constant*, 8> &SimpleConstants, - const TargetData *TD) { + const DataLayout *TD) { // If we already checked this constant, we win. if (!SimpleConstants.insert(C)) return true; // Check the constant. @@ -2367,7 +2374,7 @@ static bool isSimpleEnoughPointerToCommit(Constant *C) { return false; return ConstantFoldLoadThroughGEPConstantExpr(GV->getInitializer(), CE); - + // A constantexpr bitcast from a pointer to another pointer is a no-op, // and we know how to evaluate it by moving the bitcast from the pointer // operand to the value operand. @@ -2378,7 +2385,7 @@ static bool isSimpleEnoughPointerToCommit(Constant *C) { return cast<GlobalVariable>(CE->getOperand(0))->hasUniqueInitializer(); } } - + return false; } @@ -2408,7 +2415,7 @@ static Constant *EvaluateStoreInto(Constant *Init, Constant *Val, // Return the modified struct. return ConstantStruct::get(STy, Elts); } - + ConstantInt *CI = cast<ConstantInt>(Addr->getOperand(OpNo)); SequentialType *InitTy = cast<SequentialType>(Init->getType()); @@ -2453,7 +2460,7 @@ namespace { /// Once an evaluation call fails, the evaluation object should not be reused. class Evaluator { public: - Evaluator(const TargetData *TD, const TargetLibraryInfo *TLI) + Evaluator(const DataLayout *TD, const TargetLibraryInfo *TLI) : TD(TD), TLI(TLI) { ValueStack.push_back(new DenseMap<Value*, Constant*>); } @@ -2534,7 +2541,7 @@ private: /// simple enough to live in a static initializer of a global. SmallPtrSet<Constant*, 8> SimpleConstants; - const TargetData *TD; + const DataLayout *TD; const TargetLibraryInfo *TLI; }; @@ -2585,23 +2592,23 @@ bool Evaluator::EvaluateBlock(BasicBlock::iterator CurInst, if (!isSimpleEnoughPointerToCommit(Ptr)) // If this is too complex for us to commit, reject it. return false; - + Constant *Val = getVal(SI->getOperand(0)); // If this might be too difficult for the backend to handle (e.g. the addr // of one global variable divided by another) then we can't commit it. if (!isSimpleEnoughValueToCommit(Val, SimpleConstants, TD)) return false; - + if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Ptr)) if (CE->getOpcode() == Instruction::BitCast) { // If we're evaluating a store through a bitcast, then we need // to pull the bitcast off the pointer type and push it onto the // stored value. Ptr = CE->getOperand(0); - + Type *NewTy = cast<PointerType>(Ptr->getType())->getElementType(); - + // In order to push the bitcast onto the stored value, a bitcast // from NewTy to Val's type must be legal. If it's not, we can try // introspecting NewTy to find a legal conversion. @@ -2626,12 +2633,12 @@ bool Evaluator::EvaluateBlock(BasicBlock::iterator CurInst, return false; } } - + // If we found compatible types, go ahead and push the bitcast // onto the stored value. Val = ConstantExpr::getBitCast(Val, NewTy); } - + MutatedMemory[Ptr] = Val; } else if (BinaryOperator *BO = dyn_cast<BinaryOperator>(CurInst)) { InstResult = ConstantExpr::get(BO->getOpcode(), @@ -2793,7 +2800,7 @@ bool Evaluator::EvaluateBlock(BasicBlock::iterator CurInst, if (!CurInst->use_empty()) { if (ConstantExpr *CE = dyn_cast<ConstantExpr>(InstResult)) InstResult = ConstantFoldConstantExpression(CE, TD, TLI); - + setVal(CurInst, InstResult); } @@ -2872,14 +2879,14 @@ bool Evaluator::EvaluateFunction(Function *F, Constant *&RetVal, /// EvaluateStaticConstructor - Evaluate static constructors in the function, if /// we can. Return true if we can, false otherwise. -static bool EvaluateStaticConstructor(Function *F, const TargetData *TD, +static bool EvaluateStaticConstructor(Function *F, const DataLayout *TD, const TargetLibraryInfo *TLI) { // Call the function. Evaluator Eval(TD, TLI); Constant *RetValDummy; bool EvalSuccess = Eval.EvaluateFunction(F, RetValDummy, SmallVector<Constant*, 0>()); - + if (EvalSuccess) { // We succeeded at evaluation: commit the result. DEBUG(dbgs() << "FULLY EVALUATED GLOBAL CTOR FUNCTION '" @@ -2999,13 +3006,13 @@ static Function *FindCXAAtExit(Module &M, TargetLibraryInfo *TLI) { return 0; Function *Fn = M.getFunction(TLI->getName(LibFunc::cxa_atexit)); - + if (!Fn) return 0; FunctionType *FTy = Fn->getFunctionType(); - - // Checking that the function has the right return type, the right number of + + // Checking that the function has the right return type, the right number of // parameters and that they all have pointer types should be enough. if (!FTy->getReturnType()->isIntegerTy() || FTy->getNumParams() != 3 || @@ -3080,7 +3087,7 @@ bool GlobalOpt::OptimizeEmptyGlobalCXXDtors(Function *CXAAtExitFn) { // and remove them. bool Changed = false; - for (Function::use_iterator I = CXAAtExitFn->use_begin(), + for (Function::use_iterator I = CXAAtExitFn->use_begin(), E = CXAAtExitFn->use_end(); I != E;) { // We're only interested in calls. Theoretically, we could handle invoke // instructions as well, but neither llvm-gcc nor clang generate invokes @@ -3089,7 +3096,7 @@ bool GlobalOpt::OptimizeEmptyGlobalCXXDtors(Function *CXAAtExitFn) { if (!CI) continue; - Function *DtorFn = + Function *DtorFn = dyn_cast<Function>(CI->getArgOperand(0)->stripPointerCasts()); if (!DtorFn) continue; @@ -3113,7 +3120,7 @@ bool GlobalOpt::OptimizeEmptyGlobalCXXDtors(Function *CXAAtExitFn) { bool GlobalOpt::runOnModule(Module &M) { bool Changed = false; - TD = getAnalysisIfAvailable<TargetData>(); + TD = getAnalysisIfAvailable<DataLayout>(); TLI = &getAnalysis<TargetLibraryInfo>(); // Try to find the llvm.globalctors list. diff --git a/lib/Transforms/IPO/IPConstantPropagation.cpp b/lib/Transforms/IPO/IPConstantPropagation.cpp index d757e1f..4ac1dfc 100644 --- a/lib/Transforms/IPO/IPConstantPropagation.cpp +++ b/lib/Transforms/IPO/IPConstantPropagation.cpp @@ -17,14 +17,14 @@ #define DEBUG_TYPE "ipconstprop" #include "llvm/Transforms/IPO.h" -#include "llvm/Constants.h" -#include "llvm/Instructions.h" -#include "llvm/Module.h" -#include "llvm/Pass.h" +#include "llvm/ADT/SmallVector.h" +#include "llvm/ADT/Statistic.h" #include "llvm/Analysis/ValueTracking.h" +#include "llvm/IR/Constants.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/Module.h" +#include "llvm/Pass.h" #include "llvm/Support/CallSite.h" -#include "llvm/ADT/Statistic.h" -#include "llvm/ADT/SmallVector.h" using namespace llvm; STATISTIC(NumArgumentsProped, "Number of args turned into constants"); diff --git a/lib/Transforms/IPO/IPO.cpp b/lib/Transforms/IPO/IPO.cpp index 6233922..5d563d8 100644 --- a/lib/Transforms/IPO/IPO.cpp +++ b/lib/Transforms/IPO/IPO.cpp @@ -1,4 +1,4 @@ -//===-- Scalar.cpp --------------------------------------------------------===// +//===-- IPO.cpp -----------------------------------------------------------===// // // The LLVM Compiler Infrastructure // @@ -95,7 +95,10 @@ void LLVMAddIPSCCPPass(LLVMPassManagerRef PM) { } void LLVMAddInternalizePass(LLVMPassManagerRef PM, unsigned AllButMain) { - unwrap(PM)->add(createInternalizePass(AllButMain != 0)); + std::vector<const char *> Export; + if (AllButMain) + Export.push_back("main"); + unwrap(PM)->add(createInternalizePass(Export)); } void LLVMAddStripDeadPrototypesPass(LLVMPassManagerRef PM) { diff --git a/lib/Transforms/IPO/InlineAlways.cpp b/lib/Transforms/IPO/InlineAlways.cpp index 664ddf6..2971803 100644 --- a/lib/Transforms/IPO/InlineAlways.cpp +++ b/lib/Transforms/IPO/InlineAlways.cpp @@ -13,18 +13,18 @@ //===----------------------------------------------------------------------===// #define DEBUG_TYPE "inline" -#include "llvm/CallingConv.h" -#include "llvm/Instructions.h" -#include "llvm/IntrinsicInst.h" -#include "llvm/Module.h" -#include "llvm/Type.h" +#include "llvm/Transforms/IPO.h" +#include "llvm/ADT/SmallPtrSet.h" #include "llvm/Analysis/CallGraph.h" #include "llvm/Analysis/InlineCost.h" +#include "llvm/IR/CallingConv.h" +#include "llvm/IR/DataLayout.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/IntrinsicInst.h" +#include "llvm/IR/Module.h" +#include "llvm/IR/Type.h" #include "llvm/Support/CallSite.h" -#include "llvm/Transforms/IPO.h" #include "llvm/Transforms/IPO/InlinerPass.h" -#include "llvm/Target/TargetData.h" -#include "llvm/ADT/SmallPtrSet.h" using namespace llvm; @@ -32,6 +32,7 @@ namespace { // AlwaysInliner only inlines functions that are mark as "always inline". class AlwaysInliner : public Inliner { + InlineCostAnalyzer CA; public: // Use extremely low threshold. AlwaysInliner() : Inliner(ID, -2000000000, /*InsertLifetime*/true) { @@ -43,6 +44,10 @@ namespace { } static char ID; // Pass identification, replacement for typeid virtual InlineCost getInlineCost(CallSite CS); + + using llvm::Pass::doInitialization; + using llvm::Pass::doFinalization; + virtual bool doFinalization(CallGraph &CG) { return removeDeadFunctions(CG, /*AlwaysInlineOnly=*/true); } @@ -63,35 +68,6 @@ Pass *llvm::createAlwaysInlinerPass(bool InsertLifetime) { return new AlwaysInliner(InsertLifetime); } -/// \brief Minimal filter to detect invalid constructs for inlining. -static bool isInlineViable(Function &F) { - bool ReturnsTwice = F.hasFnAttr(Attribute::ReturnsTwice); - for (Function::iterator BI = F.begin(), BE = F.end(); BI != BE; ++BI) { - // Disallow inlining of functions which contain an indirect branch. - if (isa<IndirectBrInst>(BI->getTerminator())) - return false; - - for (BasicBlock::iterator II = BI->begin(), IE = BI->end(); II != IE; - ++II) { - CallSite CS(II); - if (!CS) - continue; - - // Disallow recursive calls. - if (&F == CS.getCalledFunction()) - return false; - - // Disallow calls which expose returns-twice to a function not previously - // attributed as such. - if (!ReturnsTwice && CS.isCall() && - cast<CallInst>(CS.getInstruction())->canReturnTwice()) - return false; - } - } - - return true; -} - /// \brief Get the inline cost for the always-inliner. /// /// The always inliner *only* handles functions which are marked with the @@ -106,27 +82,22 @@ static bool isInlineViable(Function &F) { /// likely not worth it in practice. InlineCost AlwaysInliner::getInlineCost(CallSite CS) { Function *Callee = CS.getCalledFunction(); - // We assume indirect calls aren't calling an always-inline function. - if (!Callee) return InlineCost::getNever(); - - // We can't inline calls to external functions. - // FIXME: We shouldn't even get here. - if (Callee->isDeclaration()) return InlineCost::getNever(); - - // Return never for anything not marked as always inline. - if (!Callee->hasFnAttr(Attribute::AlwaysInline)) - return InlineCost::getNever(); - // Do some minimal analysis to preclude non-viable functions. - if (!isInlineViable(*Callee)) - return InlineCost::getNever(); + // Only inline direct calls to functions with always-inline attributes + // that are viable for inlining. FIXME: We shouldn't even get here for + // declarations. + if (Callee && !Callee->isDeclaration() && + Callee->getAttributes().hasAttribute(AttributeSet::FunctionIndex, + Attribute::AlwaysInline) && + CA.isInlineViable(*Callee)) + return InlineCost::getAlways(); - // Otherwise, force inlining. - return InlineCost::getAlways(); + return InlineCost::getNever(); } // doInitialization - Initializes the vector of functions that have not // been annotated with the "always inline" attribute. bool AlwaysInliner::doInitialization(CallGraph &CG) { + CA.setDataLayout(getAnalysisIfAvailable<DataLayout>()); return false; } diff --git a/lib/Transforms/IPO/InlineSimple.cpp b/lib/Transforms/IPO/InlineSimple.cpp index 50038d8..9682923 100644 --- a/lib/Transforms/IPO/InlineSimple.cpp +++ b/lib/Transforms/IPO/InlineSimple.cpp @@ -12,17 +12,17 @@ //===----------------------------------------------------------------------===// #define DEBUG_TYPE "inline" -#include "llvm/CallingConv.h" -#include "llvm/Instructions.h" -#include "llvm/IntrinsicInst.h" -#include "llvm/Module.h" -#include "llvm/Type.h" +#include "llvm/Transforms/IPO.h" #include "llvm/Analysis/CallGraph.h" #include "llvm/Analysis/InlineCost.h" +#include "llvm/IR/CallingConv.h" +#include "llvm/IR/DataLayout.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/IntrinsicInst.h" +#include "llvm/IR/Module.h" +#include "llvm/IR/Type.h" #include "llvm/Support/CallSite.h" -#include "llvm/Transforms/IPO.h" #include "llvm/Transforms/IPO/InlinerPass.h" -#include "llvm/Target/TargetData.h" using namespace llvm; @@ -42,6 +42,7 @@ namespace { InlineCost getInlineCost(CallSite CS) { return CA.getInlineCost(CS, getInlineThreshold(CS)); } + using llvm::Pass::doInitialization; virtual bool doInitialization(CallGraph &CG); }; } @@ -62,7 +63,7 @@ Pass *llvm::createFunctionInliningPass(int Threshold) { // doInitialization - Initializes the vector of functions that have been // annotated with the noinline attribute. bool SimpleInliner::doInitialization(CallGraph &CG) { - CA.setTargetData(getAnalysisIfAvailable<TargetData>()); + CA.setDataLayout(getAnalysisIfAvailable<DataLayout>()); return false; } diff --git a/lib/Transforms/IPO/Inliner.cpp b/lib/Transforms/IPO/Inliner.cpp index 69a22fb..2187a2a 100644 --- a/lib/Transforms/IPO/Inliner.cpp +++ b/lib/Transforms/IPO/Inliner.cpp @@ -14,22 +14,22 @@ //===----------------------------------------------------------------------===// #define DEBUG_TYPE "inline" -#include "llvm/Module.h" -#include "llvm/Instructions.h" -#include "llvm/IntrinsicInst.h" +#include "llvm/Transforms/IPO/InlinerPass.h" +#include "llvm/ADT/SmallPtrSet.h" +#include "llvm/ADT/Statistic.h" #include "llvm/Analysis/CallGraph.h" #include "llvm/Analysis/InlineCost.h" -#include "llvm/Target/TargetData.h" -#include "llvm/Target/TargetLibraryInfo.h" -#include "llvm/Transforms/IPO/InlinerPass.h" -#include "llvm/Transforms/Utils/Cloning.h" -#include "llvm/Transforms/Utils/Local.h" +#include "llvm/IR/DataLayout.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/IntrinsicInst.h" +#include "llvm/IR/Module.h" #include "llvm/Support/CallSite.h" #include "llvm/Support/CommandLine.h" #include "llvm/Support/Debug.h" #include "llvm/Support/raw_ostream.h" -#include "llvm/ADT/SmallPtrSet.h" -#include "llvm/ADT/Statistic.h" +#include "llvm/Target/TargetLibraryInfo.h" +#include "llvm/Transforms/Utils/Cloning.h" +#include "llvm/Transforms/Utils/Local.h" using namespace llvm; STATISTIC(NumInlined, "Number of functions inlined"); @@ -64,8 +64,8 @@ Inliner::Inliner(char &ID, int Threshold, bool InsertLifetime) /// getAnalysisUsage - For this class, we declare that we require and preserve /// the call graph. If the derived class implements this method, it should /// always explicitly call the implementation here. -void Inliner::getAnalysisUsage(AnalysisUsage &Info) const { - CallGraphSCCPass::getAnalysisUsage(Info); +void Inliner::getAnalysisUsage(AnalysisUsage &AU) const { + CallGraphSCCPass::getAnalysisUsage(AU); } @@ -93,10 +93,13 @@ static bool InlineCallIfPossible(CallSite CS, InlineFunctionInfo &IFI, // If the inlined function had a higher stack protection level than the // calling function, then bump up the caller's stack protection level. - if (Callee->hasFnAttr(Attribute::StackProtectReq)) + if (Callee->getAttributes().hasAttribute(AttributeSet::FunctionIndex, + Attribute::StackProtectReq)) Caller->addFnAttr(Attribute::StackProtectReq); - else if (Callee->hasFnAttr(Attribute::StackProtect) && - !Caller->hasFnAttr(Attribute::StackProtectReq)) + else if (Callee->getAttributes().hasAttribute(AttributeSet::FunctionIndex, + Attribute::StackProtect) && + !Caller->getAttributes().hasAttribute(AttributeSet::FunctionIndex, + Attribute::StackProtectReq)) Caller->addFnAttr(Attribute::StackProtect); // Look at all of the allocas that we inlined through this call site. If we @@ -209,15 +212,21 @@ unsigned Inliner::getInlineThreshold(CallSite CS) const { // would decrease the threshold. Function *Caller = CS.getCaller(); bool OptSize = Caller && !Caller->isDeclaration() && - Caller->hasFnAttr(Attribute::OptimizeForSize); - if (!(InlineLimit.getNumOccurrences() > 0) && OptSize && OptSizeThreshold < thres) + Caller->getAttributes().hasAttribute(AttributeSet::FunctionIndex, + Attribute::OptimizeForSize); + if (!(InlineLimit.getNumOccurrences() > 0) && OptSize && + OptSizeThreshold < thres) thres = OptSizeThreshold; - // Listen to the inlinehint attribute when it would increase the threshold. + // Listen to the inlinehint attribute when it would increase the threshold + // and the caller does not need to minimize its size. Function *Callee = CS.getCalledFunction(); bool InlineHint = Callee && !Callee->isDeclaration() && - Callee->hasFnAttr(Attribute::InlineHint); - if (InlineHint && HintThreshold > thres) + Callee->getAttributes().hasAttribute(AttributeSet::FunctionIndex, + Attribute::InlineHint); + if (InlineHint && HintThreshold > thres + && !Caller->getAttributes().hasAttribute(AttributeSet::FunctionIndex, + Attribute::MinSize)) thres = HintThreshold; return thres; @@ -339,7 +348,7 @@ static bool InlineHistoryIncludes(Function *F, int InlineHistoryID, bool Inliner::runOnSCC(CallGraphSCC &SCC) { CallGraph &CG = getAnalysis<CallGraph>(); - const TargetData *TD = getAnalysisIfAvailable<TargetData>(); + const DataLayout *TD = getAnalysisIfAvailable<DataLayout>(); const TargetLibraryInfo *TLI = getAnalysisIfAvailable<TargetLibraryInfo>(); SmallPtrSet<Function*, 8> SCCFunctions; @@ -532,7 +541,9 @@ bool Inliner::removeDeadFunctions(CallGraph &CG, bool AlwaysInlineOnly) { // Handle the case when this function is called and we only want to care // about always-inline functions. This is a bit of a hack to share code // between here and the InlineAlways pass. - if (AlwaysInlineOnly && !F->hasFnAttr(Attribute::AlwaysInline)) + if (AlwaysInlineOnly && + !F->getAttributes().hasAttribute(AttributeSet::FunctionIndex, + Attribute::AlwaysInline)) continue; // If the only remaining users of the function are dead constants, remove diff --git a/lib/Transforms/IPO/Internalize.cpp b/lib/Transforms/IPO/Internalize.cpp index fb5869e..70d55b0 100644 --- a/lib/Transforms/IPO/Internalize.cpp +++ b/lib/Transforms/IPO/Internalize.cpp @@ -7,21 +7,21 @@ // //===----------------------------------------------------------------------===// // -// This pass loops over all of the functions in the input module, looking for a -// main function. If a main function is found, all other functions and all -// global variables with initializers are marked as internal. +// This pass loops over all of the functions and variables in the input module. +// If the function or variable is not in the list of external names given to +// the pass it is marked as internal. // //===----------------------------------------------------------------------===// #define DEBUG_TYPE "internalize" -#include "llvm/Analysis/CallGraph.h" #include "llvm/Transforms/IPO.h" +#include "llvm/ADT/Statistic.h" +#include "llvm/Analysis/CallGraph.h" +#include "llvm/IR/Module.h" #include "llvm/Pass.h" -#include "llvm/Module.h" #include "llvm/Support/CommandLine.h" #include "llvm/Support/Debug.h" #include "llvm/Support/raw_ostream.h" -#include "llvm/ADT/Statistic.h" #include <fstream> #include <set> using namespace llvm; @@ -45,13 +45,10 @@ APIList("internalize-public-api-list", cl::value_desc("list"), namespace { class InternalizePass : public ModulePass { std::set<std::string> ExternalNames; - /// If no api symbols were specified and a main function is defined, - /// assume the main function is the only API - bool AllButMain; public: static char ID; // Pass identification, replacement for typeid - explicit InternalizePass(bool AllButMain = true); - explicit InternalizePass(const std::vector <const char *>& exportList); + explicit InternalizePass(); + explicit InternalizePass(ArrayRef<const char *> exportList); void LoadFile(const char *Filename); virtual bool runOnModule(Module &M); @@ -66,8 +63,8 @@ char InternalizePass::ID = 0; INITIALIZE_PASS(InternalizePass, "internalize", "Internalize Global Symbols", false, false) -InternalizePass::InternalizePass(bool AllButMain) - : ModulePass(ID), AllButMain(AllButMain){ +InternalizePass::InternalizePass() + : ModulePass(ID) { initializeInternalizePassPass(*PassRegistry::getPassRegistry()); if (!APIFile.empty()) // If a filename is specified, use it. LoadFile(APIFile.c_str()); @@ -75,10 +72,10 @@ InternalizePass::InternalizePass(bool AllButMain) ExternalNames.insert(APIList.begin(), APIList.end()); } -InternalizePass::InternalizePass(const std::vector<const char *>&exportList) - : ModulePass(ID), AllButMain(false){ +InternalizePass::InternalizePass(ArrayRef<const char *> exportList) + : ModulePass(ID){ initializeInternalizePassPass(*PassRegistry::getPassRegistry()); - for(std::vector<const char *>::const_iterator itr = exportList.begin(); + for(ArrayRef<const char *>::const_iterator itr = exportList.begin(); itr != exportList.end(); itr++) { ExternalNames.insert(*itr); } @@ -103,23 +100,6 @@ void InternalizePass::LoadFile(const char *Filename) { bool InternalizePass::runOnModule(Module &M) { CallGraph *CG = getAnalysisIfAvailable<CallGraph>(); CallGraphNode *ExternalNode = CG ? CG->getExternalCallingNode() : 0; - - if (ExternalNames.empty()) { - // Return if we're not in 'all but main' mode and have no external api - if (!AllButMain) - return false; - // If no list or file of symbols was specified, check to see if there is a - // "main" symbol defined in the module. If so, use it, otherwise do not - // internalize the module, it must be a library or something. - // - Function *MainFunc = M.getFunction("main"); - if (MainFunc == 0 || MainFunc->isDeclaration()) - return false; // No main found, must be a library... - - // Preserve main, internalize all else. - ExternalNames.insert(MainFunc->getName()); - } - bool Changed = false; // Never internalize functions which code-gen might insert. @@ -189,10 +169,10 @@ bool InternalizePass::runOnModule(Module &M) { return Changed; } -ModulePass *llvm::createInternalizePass(bool AllButMain) { - return new InternalizePass(AllButMain); +ModulePass *llvm::createInternalizePass() { + return new InternalizePass(); } -ModulePass *llvm::createInternalizePass(const std::vector <const char *> &el) { +ModulePass *llvm::createInternalizePass(ArrayRef<const char *> el) { return new InternalizePass(el); } diff --git a/lib/Transforms/IPO/LoopExtractor.cpp b/lib/Transforms/IPO/LoopExtractor.cpp index 97d7cdc..8282a8e 100644 --- a/lib/Transforms/IPO/LoopExtractor.cpp +++ b/lib/Transforms/IPO/LoopExtractor.cpp @@ -16,16 +16,16 @@ #define DEBUG_TYPE "loop-extract" #include "llvm/Transforms/IPO.h" -#include "llvm/Instructions.h" -#include "llvm/Module.h" -#include "llvm/Pass.h" +#include "llvm/ADT/Statistic.h" #include "llvm/Analysis/Dominators.h" #include "llvm/Analysis/LoopPass.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/Module.h" +#include "llvm/Pass.h" #include "llvm/Support/CommandLine.h" #include "llvm/Transforms/Scalar.h" #include "llvm/Transforms/Utils/BasicBlockUtils.h" #include "llvm/Transforms/Utils/CodeExtractor.h" -#include "llvm/ADT/Statistic.h" #include <fstream> #include <set> using namespace llvm; diff --git a/lib/Transforms/IPO/MergeFunctions.cpp b/lib/Transforms/IPO/MergeFunctions.cpp index 9f70f66..892100f 100644 --- a/lib/Transforms/IPO/MergeFunctions.cpp +++ b/lib/Transforms/IPO/MergeFunctions.cpp @@ -45,25 +45,25 @@ #define DEBUG_TYPE "mergefunc" #include "llvm/Transforms/IPO.h" -#include "llvm/Constants.h" -#include "llvm/IRBuilder.h" -#include "llvm/InlineAsm.h" -#include "llvm/Instructions.h" -#include "llvm/LLVMContext.h" -#include "llvm/Module.h" -#include "llvm/Operator.h" -#include "llvm/Pass.h" #include "llvm/ADT/DenseSet.h" #include "llvm/ADT/FoldingSet.h" #include "llvm/ADT/STLExtras.h" #include "llvm/ADT/SmallSet.h" #include "llvm/ADT/Statistic.h" +#include "llvm/IR/Constants.h" +#include "llvm/IR/DataLayout.h" +#include "llvm/IR/IRBuilder.h" +#include "llvm/IR/InlineAsm.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/LLVMContext.h" +#include "llvm/IR/Module.h" +#include "llvm/IR/Operator.h" +#include "llvm/Pass.h" #include "llvm/Support/CallSite.h" #include "llvm/Support/Debug.h" #include "llvm/Support/ErrorHandling.h" #include "llvm/Support/ValueHandle.h" #include "llvm/Support/raw_ostream.h" -#include "llvm/Target/TargetData.h" #include <vector> using namespace llvm; @@ -92,19 +92,19 @@ static unsigned profileFunction(const Function *F) { namespace { /// ComparableFunction - A struct that pairs together functions with a -/// TargetData so that we can keep them together as elements in the DenseSet. +/// DataLayout so that we can keep them together as elements in the DenseSet. class ComparableFunction { public: static const ComparableFunction EmptyKey; static const ComparableFunction TombstoneKey; - static TargetData * const LookupOnly; + static DataLayout * const LookupOnly; - ComparableFunction(Function *Func, TargetData *TD) + ComparableFunction(Function *Func, DataLayout *TD) : Func(Func), Hash(profileFunction(Func)), TD(TD) {} Function *getFunc() const { return Func; } unsigned getHash() const { return Hash; } - TargetData *getTD() const { return TD; } + DataLayout *getTD() const { return TD; } // Drops AssertingVH reference to the function. Outside of debug mode, this // does nothing. @@ -120,13 +120,13 @@ private: AssertingVH<Function> Func; unsigned Hash; - TargetData *TD; + DataLayout *TD; }; const ComparableFunction ComparableFunction::EmptyKey = ComparableFunction(0); const ComparableFunction ComparableFunction::TombstoneKey = ComparableFunction(1); -TargetData *const ComparableFunction::LookupOnly = (TargetData*)(-1); +DataLayout *const ComparableFunction::LookupOnly = (DataLayout*)(-1); } @@ -150,12 +150,12 @@ namespace llvm { namespace { /// FunctionComparator - Compares two functions to determine whether or not -/// they will generate machine code with the same behaviour. TargetData is +/// they will generate machine code with the same behaviour. DataLayout is /// used if available. The comparator always fails conservatively (erring on the /// side of claiming that two functions are different). class FunctionComparator { public: - FunctionComparator(const TargetData *TD, const Function *F1, + FunctionComparator(const DataLayout *TD, const Function *F1, const Function *F2) : F1(F1), F2(F2), TD(TD) {} @@ -190,7 +190,7 @@ private: // The two functions undergoing comparison. const Function *F1, *F2; - const TargetData *TD; + const DataLayout *TD; DenseMap<const Value *, const Value *> id_map; DenseSet<const Value *> seen_values; @@ -346,13 +346,11 @@ bool FunctionComparator::isEquivalentGEP(const GEPOperator *GEP1, const GEPOperator *GEP2) { // When we have target data, we can reduce the GEP down to the value in bytes // added to the address. - if (TD && GEP1->hasAllConstantIndices() && GEP2->hasAllConstantIndices()) { - SmallVector<Value *, 8> Indices1(GEP1->idx_begin(), GEP1->idx_end()); - SmallVector<Value *, 8> Indices2(GEP2->idx_begin(), GEP2->idx_end()); - uint64_t Offset1 = TD->getIndexedOffset(GEP1->getPointerOperandType(), - Indices1); - uint64_t Offset2 = TD->getIndexedOffset(GEP2->getPointerOperandType(), - Indices2); + unsigned BitWidth = TD ? TD->getPointerSizeInBits() : 1; + APInt Offset1(BitWidth, 0), Offset2(BitWidth, 0); + if (TD && + GEP1->accumulateConstantOffset(*TD, Offset1) && + GEP2->accumulateConstantOffset(*TD, Offset2)) { return Offset1 == Offset2; } @@ -591,8 +589,8 @@ private: /// to modify it. FnSetType FnSet; - /// TargetData for more accurate GEP comparisons. May be NULL. - TargetData *TD; + /// DataLayout for more accurate GEP comparisons. May be NULL. + DataLayout *TD; /// Whether or not the target supports global aliases. bool HasGlobalAliases; @@ -609,7 +607,7 @@ ModulePass *llvm::createMergeFunctionsPass() { bool MergeFunctions::runOnModule(Module &M) { bool Changed = false; - TD = getAnalysisIfAvailable<TargetData>(); + TD = getAnalysisIfAvailable<DataLayout>(); for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) { if (!I->isDeclaration() && !I->hasAvailableExternallyLinkage()) diff --git a/lib/Transforms/IPO/PartialInlining.cpp b/lib/Transforms/IPO/PartialInlining.cpp index 9c9910b..fa518cb 100644 --- a/lib/Transforms/IPO/PartialInlining.cpp +++ b/lib/Transforms/IPO/PartialInlining.cpp @@ -14,14 +14,14 @@ #define DEBUG_TYPE "partialinlining" #include "llvm/Transforms/IPO.h" -#include "llvm/Instructions.h" -#include "llvm/Module.h" -#include "llvm/Pass.h" +#include "llvm/ADT/Statistic.h" #include "llvm/Analysis/Dominators.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/Module.h" +#include "llvm/Pass.h" +#include "llvm/Support/CFG.h" #include "llvm/Transforms/Utils/Cloning.h" #include "llvm/Transforms/Utils/CodeExtractor.h" -#include "llvm/ADT/Statistic.h" -#include "llvm/Support/CFG.h" using namespace llvm; STATISTIC(NumPartialInlined, "Number of functions partially inlined"); diff --git a/lib/Transforms/IPO/PassManagerBuilder.cpp b/lib/Transforms/IPO/PassManagerBuilder.cpp index 43b4ab5..6dc1773 100644 --- a/lib/Transforms/IPO/PassManagerBuilder.cpp +++ b/lib/Transforms/IPO/PassManagerBuilder.cpp @@ -14,32 +14,36 @@ #include "llvm/Transforms/IPO/PassManagerBuilder.h" - #include "llvm-c/Transforms/PassManagerBuilder.h" - -#include "llvm/PassManager.h" -#include "llvm/DefaultPasses.h" -#include "llvm/PassManager.h" +#include "llvm/ADT/SmallVector.h" #include "llvm/Analysis/Passes.h" #include "llvm/Analysis/Verifier.h" +#include "llvm/PassManager.h" #include "llvm/Support/CommandLine.h" +#include "llvm/Support/ManagedStatic.h" #include "llvm/Target/TargetLibraryInfo.h" +#include "llvm/Transforms/IPO.h" #include "llvm/Transforms/Scalar.h" #include "llvm/Transforms/Vectorize.h" -#include "llvm/Transforms/IPO.h" -#include "llvm/ADT/SmallVector.h" -#include "llvm/Support/ManagedStatic.h" using namespace llvm; static cl::opt<bool> -RunVectorization("vectorize", cl::desc("Run vectorization passes")); +RunLoopVectorization("vectorize-loops", + cl::desc("Run the Loop vectorization passes")); + +static cl::opt<bool> +RunBBVectorization("vectorize", cl::desc("Run the BB vectorization passes")); static cl::opt<bool> UseGVNAfterVectorization("use-gvn-after-vectorization", cl::init(false), cl::Hidden, cl::desc("Run GVN instead of Early CSE after vectorization passes")); +static cl::opt<bool> UseNewSROA("use-new-sroa", + cl::init(true), cl::Hidden, + cl::desc("Enable the new, experimental SROA pass")); + PassManagerBuilder::PassManagerBuilder() { OptLevel = 2; SizeLevel = 0; @@ -48,7 +52,8 @@ PassManagerBuilder::PassManagerBuilder() { DisableSimplifyLibCalls = false; DisableUnitAtATime = false; DisableUnrollLoops = false; - Vectorize = RunVectorization; + Vectorize = RunBBVectorization; + LoopVectorize = RunLoopVectorization; } PassManagerBuilder::~PassManagerBuilder() { @@ -100,7 +105,10 @@ void PassManagerBuilder::populateFunctionPassManager(FunctionPassManager &FPM) { addInitialAliasAnalysisPasses(FPM); FPM.add(createCFGSimplificationPass()); - FPM.add(createScalarReplAggregatesPass()); + if (UseNewSROA) + FPM.add(createSROAPass()); + else + FPM.add(createScalarReplAggregatesPass()); FPM.add(createEarlyCSEPass()); FPM.add(createLowerExpectIntrinsicPass()); } @@ -112,6 +120,14 @@ void PassManagerBuilder::populateModulePassManager(PassManagerBase &MPM) { MPM.add(Inliner); Inliner = 0; } + + // FIXME: This is a HACK! The inliner pass above implicitly creates a CGSCC + // pass manager, but we don't want to add extensions into that pass manager. + // To prevent this we must insert a no-op module pass to reset the pass + // manager to get the same behavior as EP_OptimizerLast in non-O0 builds. + if (!GlobalExtensions->empty() || !Extensions.empty()) + MPM.add(createBarrierNoopPass()); + addExtensionsToPM(EP_EnabledOnOptLevel0, MPM); return; } @@ -147,7 +163,10 @@ void PassManagerBuilder::populateModulePassManager(PassManagerBase &MPM) { // Start of function pass. // Break up aggregate allocas, using SSAUpdater. - MPM.add(createScalarReplAggregatesPass(-1, false)); + if (UseNewSROA) + MPM.add(createSROAPass(/*RequiresDomTree*/ false)); + else + MPM.add(createScalarReplAggregatesPass(-1, false)); MPM.add(createEarlyCSEPass()); // Catch trivial redundancies if (!DisableSimplifyLibCalls) MPM.add(createSimplifyLibCallsPass()); // Library Call Optimizations @@ -166,6 +185,10 @@ void PassManagerBuilder::populateModulePassManager(PassManagerBase &MPM) { MPM.add(createIndVarSimplifyPass()); // Canonicalize indvars MPM.add(createLoopIdiomPass()); // Recognize idioms like memset. MPM.add(createLoopDeletionPass()); // Delete dead loops + + if (LoopVectorize && OptLevel > 2) + MPM.add(createLoopVectorizePass()); + if (!DisableUnrollLoops) MPM.add(createLoopUnrollPass()); // Unroll small loops addExtensionsToPM(EP_LoopOptimizerEnd, MPM); @@ -201,13 +224,12 @@ void PassManagerBuilder::populateModulePassManager(PassManagerBase &MPM) { // FIXME: We shouldn't bother with this anymore. MPM.add(createStripDeadPrototypesPass()); // Get rid of dead prototypes - // GlobalOpt already deletes dead functions and globals, at -O3 try a + // GlobalOpt already deletes dead functions and globals, at -O2 try a // late pass of GlobalDCE. It is capable of deleting dead cycles. - if (OptLevel > 2) + if (OptLevel > 1) { MPM.add(createGlobalDCEPass()); // Remove dead fns and globals. - - if (OptLevel > 1) MPM.add(createConstantMergePass()); // Merge dup global constants + } } addExtensionsToPM(EP_OptimizerLast, MPM); } @@ -222,8 +244,11 @@ void PassManagerBuilder::populateLTOPassManager(PassManagerBase &PM, // Now that composite has been compiled, scan through the module, looking // for a main function. If main is defined, mark all other functions // internal. - if (Internalize) - PM.add(createInternalizePass(true)); + if (Internalize) { + std::vector<const char*> E; + E.push_back("main"); + PM.add(createInternalizePass(E)); + } // Propagate constants at call sites into the functions they call. This // opens opportunities for globalopt (and inlining) by substituting function @@ -265,7 +290,10 @@ void PassManagerBuilder::populateLTOPassManager(PassManagerBase &PM, PM.add(createInstructionCombiningPass()); PM.add(createJumpThreadingPass()); // Break up allocas - PM.add(createScalarReplAggregatesPass()); + if (UseNewSROA) + PM.add(createSROAPass()); + else + PM.add(createScalarReplAggregatesPass()); // Run a few AA driven optimizations here and now, to cleanup the code. PM.add(createFunctionAttrsPass()); // Add nocapture. @@ -289,7 +317,7 @@ void PassManagerBuilder::populateLTOPassManager(PassManagerBase &PM, PM.add(createGlobalDCEPass()); } -LLVMPassManagerBuilderRef LLVMPassManagerBuilderCreate(void) { +LLVMPassManagerBuilderRef LLVMPassManagerBuilderCreate() { PassManagerBuilder *PMB = new PassManagerBuilder(); return wrap(PMB); } diff --git a/lib/Transforms/IPO/PruneEH.cpp b/lib/Transforms/IPO/PruneEH.cpp index c8cc8fd..d872f0c 100644 --- a/lib/Transforms/IPO/PruneEH.cpp +++ b/lib/Transforms/IPO/PruneEH.cpp @@ -16,16 +16,16 @@ #define DEBUG_TYPE "prune-eh" #include "llvm/Transforms/IPO.h" -#include "llvm/CallGraphSCCPass.h" -#include "llvm/Constants.h" -#include "llvm/Function.h" -#include "llvm/LLVMContext.h" -#include "llvm/Instructions.h" -#include "llvm/IntrinsicInst.h" -#include "llvm/Analysis/CallGraph.h" #include "llvm/ADT/SmallPtrSet.h" #include "llvm/ADT/SmallVector.h" #include "llvm/ADT/Statistic.h" +#include "llvm/Analysis/CallGraph.h" +#include "llvm/Analysis/CallGraphSCCPass.h" +#include "llvm/IR/Constants.h" +#include "llvm/IR/Function.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/IntrinsicInst.h" +#include "llvm/IR/LLVMContext.h" #include "llvm/Support/CFG.h" #include <algorithm> using namespace llvm; @@ -137,16 +137,18 @@ bool PruneEH::runOnSCC(CallGraphSCC &SCC) { // If the SCC doesn't unwind or doesn't throw, note this fact. if (!SCCMightUnwind || !SCCMightReturn) for (CallGraphSCC::iterator I = SCC.begin(), E = SCC.end(); I != E; ++I) { - Attributes NewAttributes = Attribute::None; + AttrBuilder NewAttributes; if (!SCCMightUnwind) - NewAttributes |= Attribute::NoUnwind; + NewAttributes.addAttribute(Attribute::NoUnwind); if (!SCCMightReturn) - NewAttributes |= Attribute::NoReturn; + NewAttributes.addAttribute(Attribute::NoReturn); Function *F = (*I)->getFunction(); - const AttrListPtr &PAL = F->getAttributes(); - const AttrListPtr &NPAL = PAL.addAttr(~0, NewAttributes); + const AttributeSet &PAL = F->getAttributes(); + const AttributeSet &NPAL = PAL.addAttr(F->getContext(), ~0, + Attribute::get(F->getContext(), + NewAttributes)); if (PAL != NPAL) { MadeChange = true; F->setAttributes(NPAL); diff --git a/lib/Transforms/IPO/StripDeadPrototypes.cpp b/lib/Transforms/IPO/StripDeadPrototypes.cpp index b5f09ec..f00830a 100644 --- a/lib/Transforms/IPO/StripDeadPrototypes.cpp +++ b/lib/Transforms/IPO/StripDeadPrototypes.cpp @@ -16,9 +16,9 @@ #define DEBUG_TYPE "strip-dead-prototypes" #include "llvm/Transforms/IPO.h" -#include "llvm/Pass.h" -#include "llvm/Module.h" #include "llvm/ADT/Statistic.h" +#include "llvm/IR/Module.h" +#include "llvm/Pass.h" using namespace llvm; STATISTIC(NumDeadPrototypes, "Number of dead prototypes removed"); diff --git a/lib/Transforms/IPO/StripSymbols.cpp b/lib/Transforms/IPO/StripSymbols.cpp index 80bfc1c..5f8681f 100644 --- a/lib/Transforms/IPO/StripSymbols.cpp +++ b/lib/Transforms/IPO/StripSymbols.cpp @@ -21,17 +21,17 @@ //===----------------------------------------------------------------------===// #include "llvm/Transforms/IPO.h" -#include "llvm/Constants.h" +#include "llvm/ADT/DenseMap.h" +#include "llvm/ADT/SmallPtrSet.h" #include "llvm/DebugInfo.h" -#include "llvm/DerivedTypes.h" -#include "llvm/Instructions.h" -#include "llvm/Module.h" +#include "llvm/IR/Constants.h" +#include "llvm/IR/DerivedTypes.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/Module.h" +#include "llvm/IR/TypeFinder.h" +#include "llvm/IR/ValueSymbolTable.h" #include "llvm/Pass.h" -#include "llvm/TypeFinder.h" -#include "llvm/ValueSymbolTable.h" #include "llvm/Transforms/Utils/Local.h" -#include "llvm/ADT/DenseMap.h" -#include "llvm/ADT/SmallPtrSet.h" using namespace llvm; namespace { |