//===- CodeGenPrepare.cpp - Prepare a function for code generation --------===// // // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. // See https://llvm.org/LICENSE.txt for license information. // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception // //===----------------------------------------------------------------------===// // // This pass munges the code in the input function to better prepare it for // SelectionDAG-based code generation. This works around limitations in it's // basic-block-at-a-time approach. It should eventually be removed. // //===----------------------------------------------------------------------===// #include "llvm/CodeGen/CodeGenPrepare.h" #include "llvm/ADT/APInt.h" #include "llvm/ADT/ArrayRef.h" #include "llvm/ADT/DenseMap.h" #include "llvm/ADT/MapVector.h" #include "llvm/ADT/PointerIntPair.h" #include "llvm/ADT/STLExtras.h" #include "llvm/ADT/SmallPtrSet.h" #include "llvm/ADT/SmallVector.h" #include "llvm/ADT/Statistic.h" #include "llvm/Analysis/BlockFrequencyInfo.h" #include "llvm/Analysis/BranchProbabilityInfo.h" #include "llvm/Analysis/InstructionSimplify.h" #include "llvm/Analysis/LoopInfo.h" #include "llvm/Analysis/ProfileSummaryInfo.h" #include "llvm/Analysis/TargetLibraryInfo.h" #include "llvm/Analysis/TargetTransformInfo.h" #include "llvm/Analysis/ValueTracking.h" #include "llvm/Analysis/VectorUtils.h" #include "llvm/CodeGen/Analysis.h" #include "llvm/CodeGen/BasicBlockSectionsProfileReader.h" #include "llvm/CodeGen/ISDOpcodes.h" #include "llvm/CodeGen/SelectionDAGNodes.h" #include "llvm/CodeGen/TargetLowering.h" #include "llvm/CodeGen/TargetPassConfig.h" #include "llvm/CodeGen/TargetSubtargetInfo.h" #include "llvm/CodeGen/ValueTypes.h" #include "llvm/CodeGenTypes/MachineValueType.h" #include "llvm/Config/llvm-config.h" #include "llvm/IR/Argument.h" #include "llvm/IR/Attributes.h" #include "llvm/IR/BasicBlock.h" #include "llvm/IR/Constant.h" #include "llvm/IR/Constants.h" #include "llvm/IR/DataLayout.h" #include "llvm/IR/DebugInfo.h" #include "llvm/IR/DerivedTypes.h" #include "llvm/IR/Dominators.h" #include "llvm/IR/Function.h" #include "llvm/IR/GetElementPtrTypeIterator.h" #include "llvm/IR/GlobalValue.h" #include "llvm/IR/GlobalVariable.h" #include "llvm/IR/IRBuilder.h" #include "llvm/IR/InlineAsm.h" #include "llvm/IR/InstrTypes.h" #include "llvm/IR/Instruction.h" #include "llvm/IR/Instructions.h" #include "llvm/IR/IntrinsicInst.h" #include "llvm/IR/Intrinsics.h" #include "llvm/IR/IntrinsicsAArch64.h" #include "llvm/IR/LLVMContext.h" #include "llvm/IR/MDBuilder.h" #include "llvm/IR/Module.h" #include "llvm/IR/Operator.h" #include "llvm/IR/PatternMatch.h" #include "llvm/IR/ProfDataUtils.h" #include "llvm/IR/Statepoint.h" #include "llvm/IR/Type.h" #include "llvm/IR/Use.h" #include "llvm/IR/User.h" #include "llvm/IR/Value.h" #include "llvm/IR/ValueHandle.h" #include "llvm/IR/ValueMap.h" #include "llvm/InitializePasses.h" #include "llvm/Pass.h" #include "llvm/Support/BlockFrequency.h" #include "llvm/Support/BranchProbability.h" #include "llvm/Support/Casting.h" #include "llvm/Support/CommandLine.h" #include "llvm/Support/Compiler.h" #include "llvm/Support/Debug.h" #include "llvm/Support/ErrorHandling.h" #include "llvm/Support/MathExtras.h" #include "llvm/Support/raw_ostream.h" #include "llvm/Target/TargetMachine.h" #include "llvm/Target/TargetOptions.h" #include "llvm/Transforms/Utils/BasicBlockUtils.h" #include "llvm/Transforms/Utils/BypassSlowDivision.h" #include "llvm/Transforms/Utils/Local.h" #include "llvm/Transforms/Utils/SimplifyLibCalls.h" #include "llvm/Transforms/Utils/SizeOpts.h" #include #include #include #include #include #include #include #include #include using namespace llvm; using namespace llvm::PatternMatch; #define DEBUG_TYPE "codegenprepare" STATISTIC(NumBlocksElim, "Number of blocks eliminated"); STATISTIC(NumPHIsElim, "Number of trivial PHIs eliminated"); STATISTIC(NumGEPsElim, "Number of GEPs converted to casts"); STATISTIC(NumCmpUses, "Number of uses of Cmp expressions replaced with uses of " "sunken Cmps"); STATISTIC(NumCastUses, "Number of uses of Cast expressions replaced with uses " "of sunken Casts"); STATISTIC(NumMemoryInsts, "Number of memory instructions whose address " "computations were sunk"); STATISTIC(NumMemoryInstsPhiCreated, "Number of phis created when address " "computations were sunk to memory instructions"); STATISTIC(NumMemoryInstsSelectCreated, "Number of select created when address " "computations were sunk to memory instructions"); STATISTIC(NumExtsMoved, "Number of [s|z]ext instructions combined with loads"); STATISTIC(NumExtUses, "Number of uses of [s|z]ext instructions optimized"); STATISTIC(NumAndsAdded, "Number of and mask instructions added to form ext loads"); STATISTIC(NumAndUses, "Number of uses of and mask instructions optimized"); STATISTIC(NumRetsDup, "Number of return instructions duplicated"); STATISTIC(NumDbgValueMoved, "Number of debug value instructions moved"); STATISTIC(NumSelectsExpanded, "Number of selects turned into branches"); STATISTIC(NumStoreExtractExposed, "Number of store(extractelement) exposed"); static cl::opt DisableBranchOpts( "disable-cgp-branch-opts", cl::Hidden, cl::init(false), cl::desc("Disable branch optimizations in CodeGenPrepare")); static cl::opt DisableGCOpts("disable-cgp-gc-opts", cl::Hidden, cl::init(false), cl::desc("Disable GC optimizations in CodeGenPrepare")); static cl::opt DisableSelectToBranch("disable-cgp-select2branch", cl::Hidden, cl::init(false), cl::desc("Disable select to branch conversion.")); static cl::opt AddrSinkUsingGEPs("addr-sink-using-gep", cl::Hidden, cl::init(true), cl::desc("Address sinking in CGP using GEPs.")); static cl::opt EnableAndCmpSinking("enable-andcmp-sinking", cl::Hidden, cl::init(true), cl::desc("Enable sinkinig and/cmp into branches.")); static cl::opt DisableStoreExtract( "disable-cgp-store-extract", cl::Hidden, cl::init(false), cl::desc("Disable store(extract) optimizations in CodeGenPrepare")); static cl::opt StressStoreExtract( "stress-cgp-store-extract", cl::Hidden, cl::init(false), cl::desc("Stress test store(extract) optimizations in CodeGenPrepare")); static cl::opt DisableExtLdPromotion( "disable-cgp-ext-ld-promotion", cl::Hidden, cl::init(false), cl::desc("Disable ext(promotable(ld)) -> promoted(ext(ld)) optimization in " "CodeGenPrepare")); static cl::opt StressExtLdPromotion( "stress-cgp-ext-ld-promotion", cl::Hidden, cl::init(false), cl::desc("Stress test ext(promotable(ld)) -> promoted(ext(ld)) " "optimization in CodeGenPrepare")); static cl::opt DisablePreheaderProtect( "disable-preheader-prot", cl::Hidden, cl::init(false), cl::desc("Disable protection against removing loop preheaders")); static cl::opt ProfileGuidedSectionPrefix( "profile-guided-section-prefix", cl::Hidden, cl::init(true), cl::desc("Use profile info to add section prefix for hot/cold functions")); static cl::opt ProfileUnknownInSpecialSection( "profile-unknown-in-special-section", cl::Hidden, cl::desc("In profiling mode like sampleFDO, if a function doesn't have " "profile, we cannot tell the function is cold for sure because " "it may be a function newly added without ever being sampled. " "With the flag enabled, compiler can put such profile unknown " "functions into a special section, so runtime system can choose " "to handle it in a different way than .text section, to save " "RAM for example. ")); static cl::opt BBSectionsGuidedSectionPrefix( "bbsections-guided-section-prefix", cl::Hidden, cl::init(true), cl::desc("Use the basic-block-sections profile to determine the text " "section prefix for hot functions. Functions with " "basic-block-sections profile will be placed in `.text.hot` " "regardless of their FDO profile info. Other functions won't be " "impacted, i.e., their prefixes will be decided by FDO/sampleFDO " "profiles.")); static cl::opt FreqRatioToSkipMerge( "cgp-freq-ratio-to-skip-merge", cl::Hidden, cl::init(2), cl::desc("Skip merging empty blocks if (frequency of empty block) / " "(frequency of destination block) is greater than this ratio")); static cl::opt ForceSplitStore( "force-split-store", cl::Hidden, cl::init(false), cl::desc("Force store splitting no matter what the target query says.")); static cl::opt EnableTypePromotionMerge( "cgp-type-promotion-merge", cl::Hidden, cl::desc("Enable merging of redundant sexts when one is dominating" " the other."), cl::init(true)); static cl::opt DisableComplexAddrModes( "disable-complex-addr-modes", cl::Hidden, cl::init(false), cl::desc("Disables combining addressing modes with different parts " "in optimizeMemoryInst.")); static cl::opt AddrSinkNewPhis("addr-sink-new-phis", cl::Hidden, cl::init(false), cl::desc("Allow creation of Phis in Address sinking.")); static cl::opt AddrSinkNewSelects( "addr-sink-new-select", cl::Hidden, cl::init(true), cl::desc("Allow creation of selects in Address sinking.")); static cl::opt AddrSinkCombineBaseReg( "addr-sink-combine-base-reg", cl::Hidden, cl::init(true), cl::desc("Allow combining of BaseReg field in Address sinking.")); static cl::opt AddrSinkCombineBaseGV( "addr-sink-combine-base-gv", cl::Hidden, cl::init(true), cl::desc("Allow combining of BaseGV field in Address sinking.")); static cl::opt AddrSinkCombineBaseOffs( "addr-sink-combine-base-offs", cl::Hidden, cl::init(true), cl::desc("Allow combining of BaseOffs field in Address sinking.")); static cl::opt AddrSinkCombineScaledReg( "addr-sink-combine-scaled-reg", cl::Hidden, cl::init(true), cl::desc("Allow combining of ScaledReg field in Address sinking.")); static cl::opt EnableGEPOffsetSplit("cgp-split-large-offset-gep", cl::Hidden, cl::init(true), cl::desc("Enable splitting large offset of GEP.")); static cl::opt EnableICMP_EQToICMP_ST( "cgp-icmp-eq2icmp-st", cl::Hidden, cl::init(false), cl::desc("Enable ICMP_EQ to ICMP_S(L|G)T conversion.")); static cl::opt VerifyBFIUpdates("cgp-verify-bfi-updates", cl::Hidden, cl::init(false), cl::desc("Enable BFI update verification for " "CodeGenPrepare.")); static cl::opt OptimizePhiTypes("cgp-optimize-phi-types", cl::Hidden, cl::init(true), cl::desc("Enable converting phi types in CodeGenPrepare")); static cl::opt HugeFuncThresholdInCGPP("cgpp-huge-func", cl::init(10000), cl::Hidden, cl::desc("Least BB number of huge function.")); static cl::opt MaxAddressUsersToScan("cgp-max-address-users-to-scan", cl::init(100), cl::Hidden, cl::desc("Max number of address users to look at")); static cl::opt DisableDeletePHIs("disable-cgp-delete-phis", cl::Hidden, cl::init(false), cl::desc("Disable elimination of dead PHI nodes.")); namespace { enum ExtType { ZeroExtension, // Zero extension has been seen. SignExtension, // Sign extension has been seen. BothExtension // This extension type is used if we saw sext after // ZeroExtension had been set, or if we saw zext after // SignExtension had been set. It makes the type // information of a promoted instruction invalid. }; enum ModifyDT { NotModifyDT, // Not Modify any DT. ModifyBBDT, // Modify the Basic Block Dominator Tree. ModifyInstDT // Modify the Instruction Dominator in a Basic Block, // This usually means we move/delete/insert instruction // in a Basic Block. So we should re-iterate instructions // in such Basic Block. }; using SetOfInstrs = SmallPtrSet; using TypeIsSExt = PointerIntPair; using InstrToOrigTy = DenseMap; using SExts = SmallVector; using ValueToSExts = MapVector; class TypePromotionTransaction; class CodeGenPrepare { friend class CodeGenPrepareLegacyPass; const TargetMachine *TM = nullptr; const TargetSubtargetInfo *SubtargetInfo = nullptr; const TargetLowering *TLI = nullptr; const TargetRegisterInfo *TRI = nullptr; const TargetTransformInfo *TTI = nullptr; const BasicBlockSectionsProfileReader *BBSectionsProfileReader = nullptr; const TargetLibraryInfo *TLInfo = nullptr; LoopInfo *LI = nullptr; std::unique_ptr BFI; std::unique_ptr BPI; ProfileSummaryInfo *PSI = nullptr; /// As we scan instructions optimizing them, this is the next instruction /// to optimize. Transforms that can invalidate this should update it. BasicBlock::iterator CurInstIterator; /// Keeps track of non-local addresses that have been sunk into a block. /// This allows us to avoid inserting duplicate code for blocks with /// multiple load/stores of the same address. The usage of WeakTrackingVH /// enables SunkAddrs to be treated as a cache whose entries can be /// invalidated if a sunken address computation has been erased. ValueMap SunkAddrs; /// Keeps track of all instructions inserted for the current function. SetOfInstrs InsertedInsts; /// Keeps track of the type of the related instruction before their /// promotion for the current function. InstrToOrigTy PromotedInsts; /// Keep track of instructions removed during promotion. SetOfInstrs RemovedInsts; /// Keep track of sext chains based on their initial value. DenseMap SeenChainsForSExt; /// Keep track of GEPs accessing the same data structures such as structs or /// arrays that are candidates to be split later because of their large /// size. MapVector, SmallVector, int64_t>, 32>> LargeOffsetGEPMap; /// Keep track of new GEP base after splitting the GEPs having large offset. SmallSet, 2> NewGEPBases; /// Map serial numbers to Large offset GEPs. DenseMap, int> LargeOffsetGEPID; /// Keep track of SExt promoted. ValueToSExts ValToSExtendedUses; /// True if the function has the OptSize attribute. bool OptSize; /// DataLayout for the Function being processed. const DataLayout *DL = nullptr; /// Building the dominator tree can be expensive, so we only build it /// lazily and update it when required. std::unique_ptr DT; public: CodeGenPrepare(){}; CodeGenPrepare(const TargetMachine *TM) : TM(TM){}; /// If encounter huge function, we need to limit the build time. bool IsHugeFunc = false; /// FreshBBs is like worklist, it collected the updated BBs which need /// to be optimized again. /// Note: Consider building time in this pass, when a BB updated, we need /// to insert such BB into FreshBBs for huge function. SmallSet FreshBBs; void releaseMemory() { // Clear per function information. InsertedInsts.clear(); PromotedInsts.clear(); FreshBBs.clear(); BPI.reset(); BFI.reset(); } bool run(Function &F, FunctionAnalysisManager &AM); private: template void resetIteratorIfInvalidatedWhileCalling(BasicBlock *BB, F f) { // Substituting can cause recursive simplifications, which can invalidate // our iterator. Use a WeakTrackingVH to hold onto it in case this // happens. Value *CurValue = &*CurInstIterator; WeakTrackingVH IterHandle(CurValue); f(); // If the iterator instruction was recursively deleted, start over at the // start of the block. if (IterHandle != CurValue) { CurInstIterator = BB->begin(); SunkAddrs.clear(); } } // Get the DominatorTree, building if necessary. DominatorTree &getDT(Function &F) { if (!DT) DT = std::make_unique(F); return *DT; } void removeAllAssertingVHReferences(Value *V); bool eliminateAssumptions(Function &F); bool eliminateFallThrough(Function &F, DominatorTree *DT = nullptr); bool eliminateMostlyEmptyBlocks(Function &F); BasicBlock *findDestBlockOfMergeableEmptyBlock(BasicBlock *BB); bool canMergeBlocks(const BasicBlock *BB, const BasicBlock *DestBB) const; void eliminateMostlyEmptyBlock(BasicBlock *BB); bool isMergingEmptyBlockProfitable(BasicBlock *BB, BasicBlock *DestBB, bool isPreheader); bool makeBitReverse(Instruction &I); bool optimizeBlock(BasicBlock &BB, ModifyDT &ModifiedDT); bool optimizeInst(Instruction *I, ModifyDT &ModifiedDT); bool optimizeMemoryInst(Instruction *MemoryInst, Value *Addr, Type *AccessTy, unsigned AddrSpace); bool optimizeGatherScatterInst(Instruction *MemoryInst, Value *Ptr); bool optimizeInlineAsmInst(CallInst *CS); bool optimizeCallInst(CallInst *CI, ModifyDT &ModifiedDT); bool optimizeExt(Instruction *&I); bool optimizeExtUses(Instruction *I); bool optimizeLoadExt(LoadInst *Load); bool optimizeShiftInst(BinaryOperator *BO); bool optimizeFunnelShift(IntrinsicInst *Fsh); bool optimizeSelectInst(SelectInst *SI); bool optimizeShuffleVectorInst(ShuffleVectorInst *SVI); bool optimizeSwitchType(SwitchInst *SI); bool optimizeSwitchPhiConstants(SwitchInst *SI); bool optimizeSwitchInst(SwitchInst *SI); bool optimizeExtractElementInst(Instruction *Inst); bool dupRetToEnableTailCallOpts(BasicBlock *BB, ModifyDT &ModifiedDT); bool fixupDbgValue(Instruction *I); bool fixupDbgVariableRecord(DbgVariableRecord &I); bool fixupDbgVariableRecordsOnInst(Instruction &I); bool placeDbgValues(Function &F); bool placePseudoProbes(Function &F); bool canFormExtLd(const SmallVectorImpl &MovedExts, LoadInst *&LI, Instruction *&Inst, bool HasPromoted); bool tryToPromoteExts(TypePromotionTransaction &TPT, const SmallVectorImpl &Exts, SmallVectorImpl &ProfitablyMovedExts, unsigned CreatedInstsCost = 0); bool mergeSExts(Function &F); bool splitLargeGEPOffsets(); bool optimizePhiType(PHINode *Inst, SmallPtrSetImpl &Visited, SmallPtrSetImpl &DeletedInstrs); bool optimizePhiTypes(Function &F); bool performAddressTypePromotion( Instruction *&Inst, bool AllowPromotionWithoutCommonHeader, bool HasPromoted, TypePromotionTransaction &TPT, SmallVectorImpl &SpeculativelyMovedExts); bool splitBranchCondition(Function &F, ModifyDT &ModifiedDT); bool simplifyOffsetableRelocate(GCStatepointInst &I); bool tryToSinkFreeOperands(Instruction *I); bool replaceMathCmpWithIntrinsic(BinaryOperator *BO, Value *Arg0, Value *Arg1, CmpInst *Cmp, Intrinsic::ID IID); bool optimizeCmp(CmpInst *Cmp, ModifyDT &ModifiedDT); bool combineToUSubWithOverflow(CmpInst *Cmp, ModifyDT &ModifiedDT); bool combineToUAddWithOverflow(CmpInst *Cmp, ModifyDT &ModifiedDT); void verifyBFIUpdates(Function &F); bool _run(Function &F); }; class CodeGenPrepareLegacyPass : public FunctionPass { public: static char ID; // Pass identification, replacement for typeid CodeGenPrepareLegacyPass() : FunctionPass(ID) { initializeCodeGenPrepareLegacyPassPass(*PassRegistry::getPassRegistry()); } bool runOnFunction(Function &F) override; StringRef getPassName() const override { return "CodeGen Prepare"; } void getAnalysisUsage(AnalysisUsage &AU) const override { // FIXME: When we can selectively preserve passes, preserve the domtree. AU.addRequired(); AU.addRequired(); AU.addRequired(); AU.addRequired(); AU.addRequired(); AU.addUsedIfAvailable(); } }; } // end anonymous namespace char CodeGenPrepareLegacyPass::ID = 0; bool CodeGenPrepareLegacyPass::runOnFunction(Function &F) { if (skipFunction(F)) return false; auto TM = &getAnalysis().getTM(); CodeGenPrepare CGP(TM); CGP.DL = &F.getDataLayout(); CGP.SubtargetInfo = TM->getSubtargetImpl(F); CGP.TLI = CGP.SubtargetInfo->getTargetLowering(); CGP.TRI = CGP.SubtargetInfo->getRegisterInfo(); CGP.TLInfo = &getAnalysis().getTLI(F); CGP.TTI = &getAnalysis().getTTI(F); CGP.LI = &getAnalysis().getLoopInfo(); CGP.BPI.reset(new BranchProbabilityInfo(F, *CGP.LI)); CGP.BFI.reset(new BlockFrequencyInfo(F, *CGP.BPI, *CGP.LI)); CGP.PSI = &getAnalysis().getPSI(); auto BBSPRWP = getAnalysisIfAvailable(); CGP.BBSectionsProfileReader = BBSPRWP ? &BBSPRWP->getBBSPR() : nullptr; return CGP._run(F); } INITIALIZE_PASS_BEGIN(CodeGenPrepareLegacyPass, DEBUG_TYPE, "Optimize for code generation", false, false) INITIALIZE_PASS_DEPENDENCY(BasicBlockSectionsProfileReaderWrapperPass) INITIALIZE_PASS_DEPENDENCY(LoopInfoWrapperPass) INITIALIZE_PASS_DEPENDENCY(ProfileSummaryInfoWrapperPass) INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass) INITIALIZE_PASS_DEPENDENCY(TargetPassConfig) INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass) INITIALIZE_PASS_END(CodeGenPrepareLegacyPass, DEBUG_TYPE, "Optimize for code generation", false, false) FunctionPass *llvm::createCodeGenPrepareLegacyPass() { return new CodeGenPrepareLegacyPass(); } PreservedAnalyses CodeGenPreparePass::run(Function &F, FunctionAnalysisManager &AM) { CodeGenPrepare CGP(TM); bool Changed = CGP.run(F, AM); if (!Changed) return PreservedAnalyses::all(); PreservedAnalyses PA; PA.preserve(); PA.preserve(); PA.preserve(); return PA; } bool CodeGenPrepare::run(Function &F, FunctionAnalysisManager &AM) { DL = &F.getDataLayout(); SubtargetInfo = TM->getSubtargetImpl(F); TLI = SubtargetInfo->getTargetLowering(); TRI = SubtargetInfo->getRegisterInfo(); TLInfo = &AM.getResult(F); TTI = &AM.getResult(F); LI = &AM.getResult(F); BPI.reset(new BranchProbabilityInfo(F, *LI)); BFI.reset(new BlockFrequencyInfo(F, *BPI, *LI)); auto &MAMProxy = AM.getResult(F); PSI = MAMProxy.getCachedResult(*F.getParent()); BBSectionsProfileReader = AM.getCachedResult(F); return _run(F); } bool CodeGenPrepare::_run(Function &F) { bool EverMadeChange = false; OptSize = F.hasOptSize(); // Use the basic-block-sections profile to promote hot functions to .text.hot // if requested. if (BBSectionsGuidedSectionPrefix && BBSectionsProfileReader && BBSectionsProfileReader->isFunctionHot(F.getName())) { F.setSectionPrefix("hot"); } else if (ProfileGuidedSectionPrefix) { // The hot attribute overwrites profile count based hotness while profile // counts based hotness overwrite the cold attribute. // This is a conservative behabvior. if (F.hasFnAttribute(Attribute::Hot) || PSI->isFunctionHotInCallGraph(&F, *BFI)) F.setSectionPrefix("hot"); // If PSI shows this function is not hot, we will placed the function // into unlikely section if (1) PSI shows this is a cold function, or // (2) the function has a attribute of cold. else if (PSI->isFunctionColdInCallGraph(&F, *BFI) || F.hasFnAttribute(Attribute::Cold)) F.setSectionPrefix("unlikely"); else if (ProfileUnknownInSpecialSection && PSI->hasPartialSampleProfile() && PSI->isFunctionHotnessUnknown(F)) F.setSectionPrefix("unknown"); } /// This optimization identifies DIV instructions that can be /// profitably bypassed and carried out with a shorter, faster divide. if (!OptSize && !PSI->hasHugeWorkingSetSize() && TLI->isSlowDivBypassed()) { const DenseMap &BypassWidths = TLI->getBypassSlowDivWidths(); BasicBlock *BB = &*F.begin(); while (BB != nullptr) { // bypassSlowDivision may create new BBs, but we don't want to reapply the // optimization to those blocks. BasicBlock *Next = BB->getNextNode(); // F.hasOptSize is already checked in the outer if statement. if (!llvm::shouldOptimizeForSize(BB, PSI, BFI.get())) EverMadeChange |= bypassSlowDivision(BB, BypassWidths); BB = Next; } } // Get rid of @llvm.assume builtins before attempting to eliminate empty // blocks, since there might be blocks that only contain @llvm.assume calls // (plus arguments that we can get rid of). EverMadeChange |= eliminateAssumptions(F); // Eliminate blocks that contain only PHI nodes and an // unconditional branch. EverMadeChange |= eliminateMostlyEmptyBlocks(F); ModifyDT ModifiedDT = ModifyDT::NotModifyDT; if (!DisableBranchOpts) EverMadeChange |= splitBranchCondition(F, ModifiedDT); // Split some critical edges where one of the sources is an indirect branch, // to help generate sane code for PHIs involving such edges. EverMadeChange |= SplitIndirectBrCriticalEdges(F, /*IgnoreBlocksWithoutPHI=*/true); // If we are optimzing huge function, we need to consider the build time. // Because the basic algorithm's complex is near O(N!). IsHugeFunc = F.size() > HugeFuncThresholdInCGPP; // Transformations above may invalidate dominator tree and/or loop info. DT.reset(); LI->releaseMemory(); LI->analyze(getDT(F)); bool MadeChange = true; bool FuncIterated = false; while (MadeChange) { MadeChange = false; for (BasicBlock &BB : llvm::make_early_inc_range(F)) { if (FuncIterated && !FreshBBs.contains(&BB)) continue; ModifyDT ModifiedDTOnIteration = ModifyDT::NotModifyDT; bool Changed = optimizeBlock(BB, ModifiedDTOnIteration); if (ModifiedDTOnIteration == ModifyDT::ModifyBBDT) DT.reset(); MadeChange |= Changed; if (IsHugeFunc) { // If the BB is updated, it may still has chance to be optimized. // This usually happen at sink optimization. // For example: // // bb0: // %and = and i32 %a, 4 // %cmp = icmp eq i32 %and, 0 // // If the %cmp sink to other BB, the %and will has chance to sink. if (Changed) FreshBBs.insert(&BB); else if (FuncIterated) FreshBBs.erase(&BB); } else { // For small/normal functions, we restart BB iteration if the dominator // tree of the Function was changed. if (ModifiedDTOnIteration != ModifyDT::NotModifyDT) break; } } // We have iterated all the BB in the (only work for huge) function. FuncIterated = IsHugeFunc; if (EnableTypePromotionMerge && !ValToSExtendedUses.empty()) MadeChange |= mergeSExts(F); if (!LargeOffsetGEPMap.empty()) MadeChange |= splitLargeGEPOffsets(); MadeChange |= optimizePhiTypes(F); if (MadeChange) eliminateFallThrough(F, DT.get()); #ifndef NDEBUG if (MadeChange && VerifyLoopInfo) LI->verify(getDT(F)); #endif // Really free removed instructions during promotion. for (Instruction *I : RemovedInsts) I->deleteValue(); EverMadeChange |= MadeChange; SeenChainsForSExt.clear(); ValToSExtendedUses.clear(); RemovedInsts.clear(); LargeOffsetGEPMap.clear(); LargeOffsetGEPID.clear(); } NewGEPBases.clear(); SunkAddrs.clear(); if (!DisableBranchOpts) { MadeChange = false; // Use a set vector to get deterministic iteration order. The order the // blocks are removed may affect whether or not PHI nodes in successors // are removed. SmallSetVector WorkList; for (BasicBlock &BB : F) { SmallVector Successors(successors(&BB)); MadeChange |= ConstantFoldTerminator(&BB, true); if (!MadeChange) continue; for (BasicBlock *Succ : Successors) if (pred_empty(Succ)) WorkList.insert(Succ); } // Delete the dead blocks and any of their dead successors. MadeChange |= !WorkList.empty(); while (!WorkList.empty()) { BasicBlock *BB = WorkList.pop_back_val(); SmallVector Successors(successors(BB)); DeleteDeadBlock(BB); for (BasicBlock *Succ : Successors) if (pred_empty(Succ)) WorkList.insert(Succ); } // Merge pairs of basic blocks with unconditional branches, connected by // a single edge. if (EverMadeChange || MadeChange) MadeChange |= eliminateFallThrough(F); EverMadeChange |= MadeChange; } if (!DisableGCOpts) { SmallVector Statepoints; for (BasicBlock &BB : F) for (Instruction &I : BB) if (auto *SP = dyn_cast(&I)) Statepoints.push_back(SP); for (auto &I : Statepoints) EverMadeChange |= simplifyOffsetableRelocate(*I); } // Do this last to clean up use-before-def scenarios introduced by other // preparatory transforms. EverMadeChange |= placeDbgValues(F); EverMadeChange |= placePseudoProbes(F); #ifndef NDEBUG if (VerifyBFIUpdates) verifyBFIUpdates(F); #endif return EverMadeChange; } bool CodeGenPrepare::eliminateAssumptions(Function &F) { bool MadeChange = false; for (BasicBlock &BB : F) { CurInstIterator = BB.begin(); while (CurInstIterator != BB.end()) { Instruction *I = &*(CurInstIterator++); if (auto *Assume = dyn_cast(I)) { MadeChange = true; Value *Operand = Assume->getOperand(0); Assume->eraseFromParent(); resetIteratorIfInvalidatedWhileCalling(&BB, [&]() { RecursivelyDeleteTriviallyDeadInstructions(Operand, TLInfo, nullptr); }); } } } return MadeChange; } /// An instruction is about to be deleted, so remove all references to it in our /// GEP-tracking data strcutures. void CodeGenPrepare::removeAllAssertingVHReferences(Value *V) { LargeOffsetGEPMap.erase(V); NewGEPBases.erase(V); auto GEP = dyn_cast(V); if (!GEP) return; LargeOffsetGEPID.erase(GEP); auto VecI = LargeOffsetGEPMap.find(GEP->getPointerOperand()); if (VecI == LargeOffsetGEPMap.end()) return; auto &GEPVector = VecI->second; llvm::erase_if(GEPVector, [=](auto &Elt) { return Elt.first == GEP; }); if (GEPVector.empty()) LargeOffsetGEPMap.erase(VecI); } // Verify BFI has been updated correctly by recomputing BFI and comparing them. void LLVM_ATTRIBUTE_UNUSED CodeGenPrepare::verifyBFIUpdates(Function &F) { DominatorTree NewDT(F); LoopInfo NewLI(NewDT); BranchProbabilityInfo NewBPI(F, NewLI, TLInfo); BlockFrequencyInfo NewBFI(F, NewBPI, NewLI); NewBFI.verifyMatch(*BFI); } /// Merge basic blocks which are connected by a single edge, where one of the /// basic blocks has a single successor pointing to the other basic block, /// which has a single predecessor. bool CodeGenPrepare::eliminateFallThrough(Function &F, DominatorTree *DT) { bool Changed = false; // Scan all of the blocks in the function, except for the entry block. // Use a temporary array to avoid iterator being invalidated when // deleting blocks. SmallVector Blocks; for (auto &Block : llvm::drop_begin(F)) Blocks.push_back(&Block); SmallSet Preds; for (auto &Block : Blocks) { auto *BB = cast_or_null(Block); if (!BB) continue; // If the destination block has a single pred, then this is a trivial // edge, just collapse it. BasicBlock *SinglePred = BB->getSinglePredecessor(); // Don't merge if BB's address is taken. if (!SinglePred || SinglePred == BB || BB->hasAddressTaken()) continue; // Make an effort to skip unreachable blocks. if (DT && !DT->isReachableFromEntry(BB)) continue; BranchInst *Term = dyn_cast(SinglePred->getTerminator()); if (Term && !Term->isConditional()) { Changed = true; LLVM_DEBUG(dbgs() << "To merge:\n" << *BB << "\n\n\n"); // Merge BB into SinglePred and delete it. MergeBlockIntoPredecessor(BB, /* DTU */ nullptr, LI, /* MSSAU */ nullptr, /* MemDep */ nullptr, /* PredecessorWithTwoSuccessors */ false, DT); Preds.insert(SinglePred); if (IsHugeFunc) { // Update FreshBBs to optimize the merged BB. FreshBBs.insert(SinglePred); FreshBBs.erase(BB); } } } // (Repeatedly) merging blocks into their predecessors can create redundant // debug intrinsics. for (const auto &Pred : Preds) if (auto *BB = cast_or_null(Pred)) RemoveRedundantDbgInstrs(BB); return Changed; } /// Find a destination block from BB if BB is mergeable empty block. BasicBlock *CodeGenPrepare::findDestBlockOfMergeableEmptyBlock(BasicBlock *BB) { // If this block doesn't end with an uncond branch, ignore it. BranchInst *BI = dyn_cast(BB->getTerminator()); if (!BI || !BI->isUnconditional()) return nullptr; // If the instruction before the branch (skipping debug info) isn't a phi // node, then other stuff is happening here. BasicBlock::iterator BBI = BI->getIterator(); if (BBI != BB->begin()) { --BBI; while (isa(BBI)) { if (BBI == BB->begin()) break; --BBI; } if (!isa(BBI) && !isa(BBI)) return nullptr; } // Do not break infinite loops. BasicBlock *DestBB = BI->getSuccessor(0); if (DestBB == BB) return nullptr; if (!canMergeBlocks(BB, DestBB)) DestBB = nullptr; return DestBB; } /// Eliminate blocks that contain only PHI nodes, debug info directives, and an /// unconditional branch. Passes before isel (e.g. LSR/loopsimplify) often split /// edges in ways that are non-optimal for isel. Start by eliminating these /// blocks so we can split them the way we want them. bool CodeGenPrepare::eliminateMostlyEmptyBlocks(Function &F) { SmallPtrSet Preheaders; SmallVector LoopList(LI->begin(), LI->end()); while (!LoopList.empty()) { Loop *L = LoopList.pop_back_val(); llvm::append_range(LoopList, *L); if (BasicBlock *Preheader = L->getLoopPreheader()) Preheaders.insert(Preheader); } bool MadeChange = false; // Copy blocks into a temporary array to avoid iterator invalidation issues // as we remove them. // Note that this intentionally skips the entry block. SmallVector Blocks; for (auto &Block : llvm::drop_begin(F)) { // Delete phi nodes that could block deleting other empty blocks. if (!DisableDeletePHIs) MadeChange |= DeleteDeadPHIs(&Block, TLInfo); Blocks.push_back(&Block); } for (auto &Block : Blocks) { BasicBlock *BB = cast_or_null(Block); if (!BB) continue; BasicBlock *DestBB = findDestBlockOfMergeableEmptyBlock(BB); if (!DestBB || !isMergingEmptyBlockProfitable(BB, DestBB, Preheaders.count(BB))) continue; eliminateMostlyEmptyBlock(BB); MadeChange = true; } return MadeChange; } bool CodeGenPrepare::isMergingEmptyBlockProfitable(BasicBlock *BB, BasicBlock *DestBB, bool isPreheader) { // Do not delete loop preheaders if doing so would create a critical edge. // Loop preheaders can be good locations to spill registers. If the // preheader is deleted and we create a critical edge, registers may be // spilled in the loop body instead. if (!DisablePreheaderProtect && isPreheader && !(BB->getSinglePredecessor() && BB->getSinglePredecessor()->getSingleSuccessor())) return false; // Skip merging if the block's successor is also a successor to any callbr // that leads to this block. // FIXME: Is this really needed? Is this a correctness issue? for (BasicBlock *Pred : predecessors(BB)) { if (isa(Pred->getTerminator()) && llvm::is_contained(successors(Pred), DestBB)) return false; } // Try to skip merging if the unique predecessor of BB is terminated by a // switch or indirect branch instruction, and BB is used as an incoming block // of PHIs in DestBB. In such case, merging BB and DestBB would cause ISel to // add COPY instructions in the predecessor of BB instead of BB (if it is not // merged). Note that the critical edge created by merging such blocks wont be // split in MachineSink because the jump table is not analyzable. By keeping // such empty block (BB), ISel will place COPY instructions in BB, not in the // predecessor of BB. BasicBlock *Pred = BB->getUniquePredecessor(); if (!Pred || !(isa(Pred->getTerminator()) || isa(Pred->getTerminator()))) return true; if (BB->getTerminator() != BB->getFirstNonPHIOrDbg()) return true; // We use a simple cost heuristic which determine skipping merging is // profitable if the cost of skipping merging is less than the cost of // merging : Cost(skipping merging) < Cost(merging BB), where the // Cost(skipping merging) is Freq(BB) * (Cost(Copy) + Cost(Branch)), and // the Cost(merging BB) is Freq(Pred) * Cost(Copy). // Assuming Cost(Copy) == Cost(Branch), we could simplify it to : // Freq(Pred) / Freq(BB) > 2. // Note that if there are multiple empty blocks sharing the same incoming // value for the PHIs in the DestBB, we consider them together. In such // case, Cost(merging BB) will be the sum of their frequencies. if (!isa(DestBB->begin())) return true; SmallPtrSet SameIncomingValueBBs; // Find all other incoming blocks from which incoming values of all PHIs in // DestBB are the same as the ones from BB. for (BasicBlock *DestBBPred : predecessors(DestBB)) { if (DestBBPred == BB) continue; if (llvm::all_of(DestBB->phis(), [&](const PHINode &DestPN) { return DestPN.getIncomingValueForBlock(BB) == DestPN.getIncomingValueForBlock(DestBBPred); })) SameIncomingValueBBs.insert(DestBBPred); } // See if all BB's incoming values are same as the value from Pred. In this // case, no reason to skip merging because COPYs are expected to be place in // Pred already. if (SameIncomingValueBBs.count(Pred)) return true; BlockFrequency PredFreq = BFI->getBlockFreq(Pred); BlockFrequency BBFreq = BFI->getBlockFreq(BB); for (auto *SameValueBB : SameIncomingValueBBs) if (SameValueBB->getUniquePredecessor() == Pred && DestBB == findDestBlockOfMergeableEmptyBlock(SameValueBB)) BBFreq += BFI->getBlockFreq(SameValueBB); std::optional Limit = BBFreq.mul(FreqRatioToSkipMerge); return !Limit || PredFreq <= *Limit; } /// Return true if we can merge BB into DestBB if there is a single /// unconditional branch between them, and BB contains no other non-phi /// instructions. bool CodeGenPrepare::canMergeBlocks(const BasicBlock *BB, const BasicBlock *DestBB) const { // We only want to eliminate blocks whose phi nodes are used by phi nodes in // the successor. If there are more complex condition (e.g. preheaders), // don't mess around with them. for (const PHINode &PN : BB->phis()) { for (const User *U : PN.users()) { const Instruction *UI = cast(U); if (UI->getParent() != DestBB || !isa(UI)) return false; // If User is inside DestBB block and it is a PHINode then check // incoming value. If incoming value is not from BB then this is // a complex condition (e.g. preheaders) we want to avoid here. if (UI->getParent() == DestBB) { if (const PHINode *UPN = dyn_cast(UI)) for (unsigned I = 0, E = UPN->getNumIncomingValues(); I != E; ++I) { Instruction *Insn = dyn_cast(UPN->getIncomingValue(I)); if (Insn && Insn->getParent() == BB && Insn->getParent() != UPN->getIncomingBlock(I)) return false; } } } } // If BB and DestBB contain any common predecessors, then the phi nodes in BB // and DestBB may have conflicting incoming values for the block. If so, we // can't merge the block. const PHINode *DestBBPN = dyn_cast(DestBB->begin()); if (!DestBBPN) return true; // no conflict. // Collect the preds of BB. SmallPtrSet BBPreds; if (const PHINode *BBPN = dyn_cast(BB->begin())) { // It is faster to get preds from a PHI than with pred_iterator. for (unsigned i = 0, e = BBPN->getNumIncomingValues(); i != e; ++i) BBPreds.insert(BBPN->getIncomingBlock(i)); } else { BBPreds.insert(pred_begin(BB), pred_end(BB)); } // Walk the preds of DestBB. for (unsigned i = 0, e = DestBBPN->getNumIncomingValues(); i != e; ++i) { BasicBlock *Pred = DestBBPN->getIncomingBlock(i); if (BBPreds.count(Pred)) { // Common predecessor? for (const PHINode &PN : DestBB->phis()) { const Value *V1 = PN.getIncomingValueForBlock(Pred); const Value *V2 = PN.getIncomingValueForBlock(BB); // If V2 is a phi node in BB, look up what the mapped value will be. if (const PHINode *V2PN = dyn_cast(V2)) if (V2PN->getParent() == BB) V2 = V2PN->getIncomingValueForBlock(Pred); // If there is a conflict, bail out. if (V1 != V2) return false; } } } return true; } /// Replace all old uses with new ones, and push the updated BBs into FreshBBs. static void replaceAllUsesWith(Value *Old, Value *New, SmallSet &FreshBBs, bool IsHuge) { auto *OldI = dyn_cast(Old); if (OldI) { for (Value::user_iterator UI = OldI->user_begin(), E = OldI->user_end(); UI != E; ++UI) { Instruction *User = cast(*UI); if (IsHuge) FreshBBs.insert(User->getParent()); } } Old->replaceAllUsesWith(New); } /// Eliminate a basic block that has only phi's and an unconditional branch in /// it. void CodeGenPrepare::eliminateMostlyEmptyBlock(BasicBlock *BB) { BranchInst *BI = cast(BB->getTerminator()); BasicBlock *DestBB = BI->getSuccessor(0); LLVM_DEBUG(dbgs() << "MERGING MOSTLY EMPTY BLOCKS - BEFORE:\n" << *BB << *DestBB); // If the destination block has a single pred, then this is a trivial edge, // just collapse it. if (BasicBlock *SinglePred = DestBB->getSinglePredecessor()) { if (SinglePred != DestBB) { assert(SinglePred == BB && "Single predecessor not the same as predecessor"); // Merge DestBB into SinglePred/BB and delete it. MergeBlockIntoPredecessor(DestBB); // Note: BB(=SinglePred) will not be deleted on this path. // DestBB(=its single successor) is the one that was deleted. LLVM_DEBUG(dbgs() << "AFTER:\n" << *SinglePred << "\n\n\n"); if (IsHugeFunc) { // Update FreshBBs to optimize the merged BB. FreshBBs.insert(SinglePred); FreshBBs.erase(DestBB); } return; } } // Otherwise, we have multiple predecessors of BB. Update the PHIs in DestBB // to handle the new incoming edges it is about to have. for (PHINode &PN : DestBB->phis()) { // Remove the incoming value for BB, and remember it. Value *InVal = PN.removeIncomingValue(BB, false); // Two options: either the InVal is a phi node defined in BB or it is some // value that dominates BB. PHINode *InValPhi = dyn_cast(InVal); if (InValPhi && InValPhi->getParent() == BB) { // Add all of the input values of the input PHI as inputs of this phi. for (unsigned i = 0, e = InValPhi->getNumIncomingValues(); i != e; ++i) PN.addIncoming(InValPhi->getIncomingValue(i), InValPhi->getIncomingBlock(i)); } else { // Otherwise, add one instance of the dominating value for each edge that // we will be adding. if (PHINode *BBPN = dyn_cast(BB->begin())) { for (unsigned i = 0, e = BBPN->getNumIncomingValues(); i != e; ++i) PN.addIncoming(InVal, BBPN->getIncomingBlock(i)); } else { for (BasicBlock *Pred : predecessors(BB)) PN.addIncoming(InVal, Pred); } } } // The PHIs are now updated, change everything that refers to BB to use // DestBB and remove BB. BB->replaceAllUsesWith(DestBB); BB->eraseFromParent(); ++NumBlocksElim; LLVM_DEBUG(dbgs() << "AFTER:\n" << *DestBB << "\n\n\n"); } // Computes a map of base pointer relocation instructions to corresponding // derived pointer relocation instructions given a vector of all relocate calls static void computeBaseDerivedRelocateMap( const SmallVectorImpl &AllRelocateCalls, MapVector> &RelocateInstMap) { // Collect information in two maps: one primarily for locating the base object // while filling the second map; the second map is the final structure holding // a mapping between Base and corresponding Derived relocate calls MapVector, GCRelocateInst *> RelocateIdxMap; for (auto *ThisRelocate : AllRelocateCalls) { auto K = std::make_pair(ThisRelocate->getBasePtrIndex(), ThisRelocate->getDerivedPtrIndex()); RelocateIdxMap.insert(std::make_pair(K, ThisRelocate)); } for (auto &Item : RelocateIdxMap) { std::pair Key = Item.first; if (Key.first == Key.second) // Base relocation: nothing to insert continue; GCRelocateInst *I = Item.second; auto BaseKey = std::make_pair(Key.first, Key.first); // We're iterating over RelocateIdxMap so we cannot modify it. auto MaybeBase = RelocateIdxMap.find(BaseKey); if (MaybeBase == RelocateIdxMap.end()) // TODO: We might want to insert a new base object relocate and gep off // that, if there are enough derived object relocates. continue; RelocateInstMap[MaybeBase->second].push_back(I); } } // Accepts a GEP and extracts the operands into a vector provided they're all // small integer constants static bool getGEPSmallConstantIntOffsetV(GetElementPtrInst *GEP, SmallVectorImpl &OffsetV) { for (unsigned i = 1; i < GEP->getNumOperands(); i++) { // Only accept small constant integer operands auto *Op = dyn_cast(GEP->getOperand(i)); if (!Op || Op->getZExtValue() > 20) return false; } for (unsigned i = 1; i < GEP->getNumOperands(); i++) OffsetV.push_back(GEP->getOperand(i)); return true; } // Takes a RelocatedBase (base pointer relocation instruction) and Targets to // replace, computes a replacement, and affects it. static bool simplifyRelocatesOffABase(GCRelocateInst *RelocatedBase, const SmallVectorImpl &Targets) { bool MadeChange = false; // We must ensure the relocation of derived pointer is defined after // relocation of base pointer. If we find a relocation corresponding to base // defined earlier than relocation of base then we move relocation of base // right before found relocation. We consider only relocation in the same // basic block as relocation of base. Relocations from other basic block will // be skipped by optimization and we do not care about them. for (auto R = RelocatedBase->getParent()->getFirstInsertionPt(); &*R != RelocatedBase; ++R) if (auto *RI = dyn_cast(R)) if (RI->getStatepoint() == RelocatedBase->getStatepoint()) if (RI->getBasePtrIndex() == RelocatedBase->getBasePtrIndex()) { RelocatedBase->moveBefore(RI); MadeChange = true; break; } for (GCRelocateInst *ToReplace : Targets) { assert(ToReplace->getBasePtrIndex() == RelocatedBase->getBasePtrIndex() && "Not relocating a derived object of the original base object"); if (ToReplace->getBasePtrIndex() == ToReplace->getDerivedPtrIndex()) { // A duplicate relocate call. TODO: coalesce duplicates. continue; } if (RelocatedBase->getParent() != ToReplace->getParent()) { // Base and derived relocates are in different basic blocks. // In this case transform is only valid when base dominates derived // relocate. However it would be too expensive to check dominance // for each such relocate, so we skip the whole transformation. continue; } Value *Base = ToReplace->getBasePtr(); auto *Derived = dyn_cast(ToReplace->getDerivedPtr()); if (!Derived || Derived->getPointerOperand() != Base) continue; SmallVector OffsetV; if (!getGEPSmallConstantIntOffsetV(Derived, OffsetV)) continue; // Create a Builder and replace the target callsite with a gep assert(RelocatedBase->getNextNode() && "Should always have one since it's not a terminator"); // Insert after RelocatedBase IRBuilder<> Builder(RelocatedBase->getNextNode()); Builder.SetCurrentDebugLocation(ToReplace->getDebugLoc()); // If gc_relocate does not match the actual type, cast it to the right type. // In theory, there must be a bitcast after gc_relocate if the type does not // match, and we should reuse it to get the derived pointer. But it could be // cases like this: // bb1: // ... // %g1 = call coldcc i8 addrspace(1)* // @llvm.experimental.gc.relocate.p1i8(...) br label %merge // // bb2: // ... // %g2 = call coldcc i8 addrspace(1)* // @llvm.experimental.gc.relocate.p1i8(...) br label %merge // // merge: // %p1 = phi i8 addrspace(1)* [ %g1, %bb1 ], [ %g2, %bb2 ] // %cast = bitcast i8 addrspace(1)* %p1 in to i32 addrspace(1)* // // In this case, we can not find the bitcast any more. So we insert a new // bitcast no matter there is already one or not. In this way, we can handle // all cases, and the extra bitcast should be optimized away in later // passes. Value *ActualRelocatedBase = RelocatedBase; if (RelocatedBase->getType() != Base->getType()) { ActualRelocatedBase = Builder.CreateBitCast(RelocatedBase, Base->getType()); } Value *Replacement = Builder.CreateGEP(Derived->getSourceElementType(), ActualRelocatedBase, ArrayRef(OffsetV)); Replacement->takeName(ToReplace); // If the newly generated derived pointer's type does not match the original // derived pointer's type, cast the new derived pointer to match it. Same // reasoning as above. Value *ActualReplacement = Replacement; if (Replacement->getType() != ToReplace->getType()) { ActualReplacement = Builder.CreateBitCast(Replacement, ToReplace->getType()); } ToReplace->replaceAllUsesWith(ActualReplacement); ToReplace->eraseFromParent(); MadeChange = true; } return MadeChange; } // Turns this: // // %base = ... // %ptr = gep %base + 15 // %tok = statepoint (%fun, i32 0, i32 0, i32 0, %base, %ptr) // %base' = relocate(%tok, i32 4, i32 4) // %ptr' = relocate(%tok, i32 4, i32 5) // %val = load %ptr' // // into this: // // %base = ... // %ptr = gep %base + 15 // %tok = statepoint (%fun, i32 0, i32 0, i32 0, %base, %ptr) // %base' = gc.relocate(%tok, i32 4, i32 4) // %ptr' = gep %base' + 15 // %val = load %ptr' bool CodeGenPrepare::simplifyOffsetableRelocate(GCStatepointInst &I) { bool MadeChange = false; SmallVector AllRelocateCalls; for (auto *U : I.users()) if (GCRelocateInst *Relocate = dyn_cast(U)) // Collect all the relocate calls associated with a statepoint AllRelocateCalls.push_back(Relocate); // We need at least one base pointer relocation + one derived pointer // relocation to mangle if (AllRelocateCalls.size() < 2) return false; // RelocateInstMap is a mapping from the base relocate instruction to the // corresponding derived relocate instructions MapVector> RelocateInstMap; computeBaseDerivedRelocateMap(AllRelocateCalls, RelocateInstMap); if (RelocateInstMap.empty()) return false; for (auto &Item : RelocateInstMap) // Item.first is the RelocatedBase to offset against // Item.second is the vector of Targets to replace MadeChange = simplifyRelocatesOffABase(Item.first, Item.second); return MadeChange; } /// Sink the specified cast instruction into its user blocks. static bool SinkCast(CastInst *CI) { BasicBlock *DefBB = CI->getParent(); /// InsertedCasts - Only insert a cast in each block once. DenseMap InsertedCasts; bool MadeChange = false; for (Value::user_iterator UI = CI->user_begin(), E = CI->user_end(); UI != E;) { Use &TheUse = UI.getUse(); Instruction *User = cast(*UI); // Figure out which BB this cast is used in. For PHI's this is the // appropriate predecessor block. BasicBlock *UserBB = User->getParent(); if (PHINode *PN = dyn_cast(User)) { UserBB = PN->getIncomingBlock(TheUse); } // Preincrement use iterator so we don't invalidate it. ++UI; // The first insertion point of a block containing an EH pad is after the // pad. If the pad is the user, we cannot sink the cast past the pad. if (User->isEHPad()) continue; // If the block selected to receive the cast is an EH pad that does not // allow non-PHI instructions before the terminator, we can't sink the // cast. if (UserBB->getTerminator()->isEHPad()) continue; // If this user is in the same block as the cast, don't change the cast. if (UserBB == DefBB) continue; // If we have already inserted a cast into this block, use it. CastInst *&InsertedCast = InsertedCasts[UserBB]; if (!InsertedCast) { BasicBlock::iterator InsertPt = UserBB->getFirstInsertionPt(); assert(InsertPt != UserBB->end()); InsertedCast = cast(CI->clone()); InsertedCast->insertBefore(*UserBB, InsertPt); } // Replace a use of the cast with a use of the new cast. TheUse = InsertedCast; MadeChange = true; ++NumCastUses; } // If we removed all uses, nuke the cast. if (CI->use_empty()) { salvageDebugInfo(*CI); CI->eraseFromParent(); MadeChange = true; } return MadeChange; } /// If the specified cast instruction is a noop copy (e.g. it's casting from /// one pointer type to another, i32->i8 on PPC), sink it into user blocks to /// reduce the number of virtual registers that must be created and coalesced. /// /// Return true if any changes are made. static bool OptimizeNoopCopyExpression(CastInst *CI, const TargetLowering &TLI, const DataLayout &DL) { // Sink only "cheap" (or nop) address-space casts. This is a weaker condition // than sinking only nop casts, but is helpful on some platforms. if (auto *ASC = dyn_cast(CI)) { if (!TLI.isFreeAddrSpaceCast(ASC->getSrcAddressSpace(), ASC->getDestAddressSpace())) return false; } // If this is a noop copy, EVT SrcVT = TLI.getValueType(DL, CI->getOperand(0)->getType()); EVT DstVT = TLI.getValueType(DL, CI->getType()); // This is an fp<->int conversion? if (SrcVT.isInteger() != DstVT.isInteger()) return false; // If this is an extension, it will be a zero or sign extension, which // isn't a noop. if (SrcVT.bitsLT(DstVT)) return false; // If these values will be promoted, find out what they will be promoted // to. This helps us consider truncates on PPC as noop copies when they // are. if (TLI.getTypeAction(CI->getContext(), SrcVT) == TargetLowering::TypePromoteInteger) SrcVT = TLI.getTypeToTransformTo(CI->getContext(), SrcVT); if (TLI.getTypeAction(CI->getContext(), DstVT) == TargetLowering::TypePromoteInteger) DstVT = TLI.getTypeToTransformTo(CI->getContext(), DstVT); // If, after promotion, these are the same types, this is a noop copy. if (SrcVT != DstVT) return false; return SinkCast(CI); } // Match a simple increment by constant operation. Note that if a sub is // matched, the step is negated (as if the step had been canonicalized to // an add, even though we leave the instruction alone.) static bool matchIncrement(const Instruction *IVInc, Instruction *&LHS, Constant *&Step) { if (match(IVInc, m_Add(m_Instruction(LHS), m_Constant(Step))) || match(IVInc, m_ExtractValue<0>(m_Intrinsic( m_Instruction(LHS), m_Constant(Step))))) return true; if (match(IVInc, m_Sub(m_Instruction(LHS), m_Constant(Step))) || match(IVInc, m_ExtractValue<0>(m_Intrinsic( m_Instruction(LHS), m_Constant(Step))))) { Step = ConstantExpr::getNeg(Step); return true; } return false; } /// If given \p PN is an inductive variable with value IVInc coming from the /// backedge, and on each iteration it gets increased by Step, return pair /// . Otherwise, return std::nullopt. static std::optional> getIVIncrement(const PHINode *PN, const LoopInfo *LI) { const Loop *L = LI->getLoopFor(PN->getParent()); if (!L || L->getHeader() != PN->getParent() || !L->getLoopLatch()) return std::nullopt; auto *IVInc = dyn_cast(PN->getIncomingValueForBlock(L->getLoopLatch())); if (!IVInc || LI->getLoopFor(IVInc->getParent()) != L) return std::nullopt; Instruction *LHS = nullptr; Constant *Step = nullptr; if (matchIncrement(IVInc, LHS, Step) && LHS == PN) return std::make_pair(IVInc, Step); return std::nullopt; } static bool isIVIncrement(const Value *V, const LoopInfo *LI) { auto *I = dyn_cast(V); if (!I) return false; Instruction *LHS = nullptr; Constant *Step = nullptr; if (!matchIncrement(I, LHS, Step)) return false; if (auto *PN = dyn_cast(LHS)) if (auto IVInc = getIVIncrement(PN, LI)) return IVInc->first == I; return false; } bool CodeGenPrepare::replaceMathCmpWithIntrinsic(BinaryOperator *BO, Value *Arg0, Value *Arg1, CmpInst *Cmp, Intrinsic::ID IID) { auto IsReplacableIVIncrement = [this, &Cmp](BinaryOperator *BO) { if (!isIVIncrement(BO, LI)) return false; const Loop *L = LI->getLoopFor(BO->getParent()); assert(L && "L should not be null after isIVIncrement()"); // Do not risk on moving increment into a child loop. if (LI->getLoopFor(Cmp->getParent()) != L) return false; // Finally, we need to ensure that the insert point will dominate all // existing uses of the increment. auto &DT = getDT(*BO->getParent()->getParent()); if (DT.dominates(Cmp->getParent(), BO->getParent())) // If we're moving up the dom tree, all uses are trivially dominated. // (This is the common case for code produced by LSR.) return true; // Otherwise, special case the single use in the phi recurrence. return BO->hasOneUse() && DT.dominates(Cmp->getParent(), L->getLoopLatch()); }; if (BO->getParent() != Cmp->getParent() && !IsReplacableIVIncrement(BO)) { // We used to use a dominator tree here to allow multi-block optimization. // But that was problematic because: // 1. It could cause a perf regression by hoisting the math op into the // critical path. // 2. It could cause a perf regression by creating a value that was live // across multiple blocks and increasing register pressure. // 3. Use of a dominator tree could cause large compile-time regression. // This is because we recompute the DT on every change in the main CGP // run-loop. The recomputing is probably unnecessary in many cases, so if // that was fixed, using a DT here would be ok. // // There is one important particular case we still want to handle: if BO is // the IV increment. Important properties that make it profitable: // - We can speculate IV increment anywhere in the loop (as long as the // indvar Phi is its only user); // - Upon computing Cmp, we effectively compute something equivalent to the // IV increment (despite it loops differently in the IR). So moving it up // to the cmp point does not really increase register pressure. return false; } // We allow matching the canonical IR (add X, C) back to (usubo X, -C). if (BO->getOpcode() == Instruction::Add && IID == Intrinsic::usub_with_overflow) { assert(isa(Arg1) && "Unexpected input for usubo"); Arg1 = ConstantExpr::getNeg(cast(Arg1)); } // Insert at the first instruction of the pair. Instruction *InsertPt = nullptr; for (Instruction &Iter : *Cmp->getParent()) { // If BO is an XOR, it is not guaranteed that it comes after both inputs to // the overflow intrinsic are defined. if ((BO->getOpcode() != Instruction::Xor && &Iter == BO) || &Iter == Cmp) { InsertPt = &Iter; break; } } assert(InsertPt != nullptr && "Parent block did not contain cmp or binop"); IRBuilder<> Builder(InsertPt); Value *MathOV = Builder.CreateBinaryIntrinsic(IID, Arg0, Arg1); if (BO->getOpcode() != Instruction::Xor) { Value *Math = Builder.CreateExtractValue(MathOV, 0, "math"); replaceAllUsesWith(BO, Math, FreshBBs, IsHugeFunc); } else assert(BO->hasOneUse() && "Patterns with XOr should use the BO only in the compare"); Value *OV = Builder.CreateExtractValue(MathOV, 1, "ov"); replaceAllUsesWith(Cmp, OV, FreshBBs, IsHugeFunc); Cmp->eraseFromParent(); BO->eraseFromParent(); return true; } /// Match special-case patterns that check for unsigned add overflow. static bool matchUAddWithOverflowConstantEdgeCases(CmpInst *Cmp, BinaryOperator *&Add) { // Add = add A, 1; Cmp = icmp eq A,-1 (overflow if A is max val) // Add = add A,-1; Cmp = icmp ne A, 0 (overflow if A is non-zero) Value *A = Cmp->getOperand(0), *B = Cmp->getOperand(1); // We are not expecting non-canonical/degenerate code. Just bail out. if (isa(A)) return false; ICmpInst::Predicate Pred = Cmp->getPredicate(); if (Pred == ICmpInst::ICMP_EQ && match(B, m_AllOnes())) B = ConstantInt::get(B->getType(), 1); else if (Pred == ICmpInst::ICMP_NE && match(B, m_ZeroInt())) B = ConstantInt::get(B->getType(), -1); else return false; // Check the users of the variable operand of the compare looking for an add // with the adjusted constant. for (User *U : A->users()) { if (match(U, m_Add(m_Specific(A), m_Specific(B)))) { Add = cast(U); return true; } } return false; } /// Try to combine the compare into a call to the llvm.uadd.with.overflow /// intrinsic. Return true if any changes were made. bool CodeGenPrepare::combineToUAddWithOverflow(CmpInst *Cmp, ModifyDT &ModifiedDT) { bool EdgeCase = false; Value *A, *B; BinaryOperator *Add; if (!match(Cmp, m_UAddWithOverflow(m_Value(A), m_Value(B), m_BinOp(Add)))) { if (!matchUAddWithOverflowConstantEdgeCases(Cmp, Add)) return false; // Set A and B in case we match matchUAddWithOverflowConstantEdgeCases. A = Add->getOperand(0); B = Add->getOperand(1); EdgeCase = true; } if (!TLI->shouldFormOverflowOp(ISD::UADDO, TLI->getValueType(*DL, Add->getType()), Add->hasNUsesOrMore(EdgeCase ? 1 : 2))) return false; // We don't want to move around uses of condition values this late, so we // check if it is legal to create the call to the intrinsic in the basic // block containing the icmp. if (Add->getParent() != Cmp->getParent() && !Add->hasOneUse()) return false; if (!replaceMathCmpWithIntrinsic(Add, A, B, Cmp, Intrinsic::uadd_with_overflow)) return false; // Reset callers - do not crash by iterating over a dead instruction. ModifiedDT = ModifyDT::ModifyInstDT; return true; } bool CodeGenPrepare::combineToUSubWithOverflow(CmpInst *Cmp, ModifyDT &ModifiedDT) { // We are not expecting non-canonical/degenerate code. Just bail out. Value *A = Cmp->getOperand(0), *B = Cmp->getOperand(1); if (isa(A) && isa(B)) return false; // Convert (A u> B) to (A u< B) to simplify pattern matching. ICmpInst::Predicate Pred = Cmp->getPredicate(); if (Pred == ICmpInst::ICMP_UGT) { std::swap(A, B); Pred = ICmpInst::ICMP_ULT; } // Convert special-case: (A == 0) is the same as (A u< 1). if (Pred == ICmpInst::ICMP_EQ && match(B, m_ZeroInt())) { B = ConstantInt::get(B->getType(), 1); Pred = ICmpInst::ICMP_ULT; } // Convert special-case: (A != 0) is the same as (0 u< A). if (Pred == ICmpInst::ICMP_NE && match(B, m_ZeroInt())) { std::swap(A, B); Pred = ICmpInst::ICMP_ULT; } if (Pred != ICmpInst::ICMP_ULT) return false; // Walk the users of a variable operand of a compare looking for a subtract or // add with that same operand. Also match the 2nd operand of the compare to // the add/sub, but that may be a negated constant operand of an add. Value *CmpVariableOperand = isa(A) ? B : A; BinaryOperator *Sub = nullptr; for (User *U : CmpVariableOperand->users()) { // A - B, A u< B --> usubo(A, B) if (match(U, m_Sub(m_Specific(A), m_Specific(B)))) { Sub = cast(U); break; } // A + (-C), A u< C (canonicalized form of (sub A, C)) const APInt *CmpC, *AddC; if (match(U, m_Add(m_Specific(A), m_APInt(AddC))) && match(B, m_APInt(CmpC)) && *AddC == -(*CmpC)) { Sub = cast(U); break; } } if (!Sub) return false; if (!TLI->shouldFormOverflowOp(ISD::USUBO, TLI->getValueType(*DL, Sub->getType()), Sub->hasNUsesOrMore(1))) return false; if (!replaceMathCmpWithIntrinsic(Sub, Sub->getOperand(0), Sub->getOperand(1), Cmp, Intrinsic::usub_with_overflow)) return false; // Reset callers - do not crash by iterating over a dead instruction. ModifiedDT = ModifyDT::ModifyInstDT; return true; } /// Sink the given CmpInst into user blocks to reduce the number of virtual /// registers that must be created and coalesced. This is a clear win except on /// targets with multiple condition code registers (PowerPC), where it might /// lose; some adjustment may be wanted there. /// /// Return true if any changes are made. static bool sinkCmpExpression(CmpInst *Cmp, const TargetLowering &TLI) { if (TLI.hasMultipleConditionRegisters()) return false; // Avoid sinking soft-FP comparisons, since this can move them into a loop. if (TLI.useSoftFloat() && isa(Cmp)) return false; // Only insert a cmp in each block once. DenseMap InsertedCmps; bool MadeChange = false; for (Value::user_iterator UI = Cmp->user_begin(), E = Cmp->user_end(); UI != E;) { Use &TheUse = UI.getUse(); Instruction *User = cast(*UI); // Preincrement use iterator so we don't invalidate it. ++UI; // Don't bother for PHI nodes. if (isa(User)) continue; // Figure out which BB this cmp is used in. BasicBlock *UserBB = User->getParent(); BasicBlock *DefBB = Cmp->getParent(); // If this user is in the same block as the cmp, don't change the cmp. if (UserBB == DefBB) continue; // If we have already inserted a cmp into this block, use it. CmpInst *&InsertedCmp = InsertedCmps[UserBB]; if (!InsertedCmp) { BasicBlock::iterator InsertPt = UserBB->getFirstInsertionPt(); assert(InsertPt != UserBB->end()); InsertedCmp = CmpInst::Create(Cmp->getOpcode(), Cmp->getPredicate(), Cmp->getOperand(0), Cmp->getOperand(1), ""); InsertedCmp->insertBefore(*UserBB, InsertPt); // Propagate the debug info. InsertedCmp->setDebugLoc(Cmp->getDebugLoc()); } // Replace a use of the cmp with a use of the new cmp. TheUse = InsertedCmp; MadeChange = true; ++NumCmpUses; } // If we removed all uses, nuke the cmp. if (Cmp->use_empty()) { Cmp->eraseFromParent(); MadeChange = true; } return MadeChange; } /// For pattern like: /// /// DomCond = icmp sgt/slt CmpOp0, CmpOp1 (might not be in DomBB) /// ... /// DomBB: /// ... /// br DomCond, TrueBB, CmpBB /// CmpBB: (with DomBB being the single predecessor) /// ... /// Cmp = icmp eq CmpOp0, CmpOp1 /// ... /// /// It would use two comparison on targets that lowering of icmp sgt/slt is /// different from lowering of icmp eq (PowerPC). This function try to convert /// 'Cmp = icmp eq CmpOp0, CmpOp1' to ' Cmp = icmp slt/sgt CmpOp0, CmpOp1'. /// After that, DomCond and Cmp can use the same comparison so reduce one /// comparison. /// /// Return true if any changes are made. static bool foldICmpWithDominatingICmp(CmpInst *Cmp, const TargetLowering &TLI) { if (!EnableICMP_EQToICMP_ST && TLI.isEqualityCmpFoldedWithSignedCmp()) return false; ICmpInst::Predicate Pred = Cmp->getPredicate(); if (Pred != ICmpInst::ICMP_EQ) return false; // If icmp eq has users other than BranchInst and SelectInst, converting it to // icmp slt/sgt would introduce more redundant LLVM IR. for (User *U : Cmp->users()) { if (isa(U)) continue; if (isa(U) && cast(U)->getCondition() == Cmp) continue; return false; } // This is a cheap/incomplete check for dominance - just match a single // predecessor with a conditional branch. BasicBlock *CmpBB = Cmp->getParent(); BasicBlock *DomBB = CmpBB->getSinglePredecessor(); if (!DomBB) return false; // We want to ensure that the only way control gets to the comparison of // interest is that a less/greater than comparison on the same operands is // false. Value *DomCond; BasicBlock *TrueBB, *FalseBB; if (!match(DomBB->getTerminator(), m_Br(m_Value(DomCond), TrueBB, FalseBB))) return false; if (CmpBB != FalseBB) return false; Value *CmpOp0 = Cmp->getOperand(0), *CmpOp1 = Cmp->getOperand(1); ICmpInst::Predicate DomPred; if (!match(DomCond, m_ICmp(DomPred, m_Specific(CmpOp0), m_Specific(CmpOp1)))) return false; if (DomPred != ICmpInst::ICMP_SGT && DomPred != ICmpInst::ICMP_SLT) return false; // Convert the equality comparison to the opposite of the dominating // comparison and swap the direction for all branch/select users. // We have conceptually converted: // Res = (a < b) ? : (a == b) ? : ; // to // Res = (a < b) ? : (a > b) ? : ; // And similarly for branches. for (User *U : Cmp->users()) { if (auto *BI = dyn_cast(U)) { assert(BI->isConditional() && "Must be conditional"); BI->swapSuccessors(); continue; } if (auto *SI = dyn_cast(U)) { // Swap operands SI->swapValues(); SI->swapProfMetadata(); continue; } llvm_unreachable("Must be a branch or a select"); } Cmp->setPredicate(CmpInst::getSwappedPredicate(DomPred)); return true; } /// Many architectures use the same instruction for both subtract and cmp. Try /// to swap cmp operands to match subtract operations to allow for CSE. static bool swapICmpOperandsToExposeCSEOpportunities(CmpInst *Cmp) { Value *Op0 = Cmp->getOperand(0); Value *Op1 = Cmp->getOperand(1); if (!Op0->getType()->isIntegerTy() || isa(Op0) || isa(Op1) || Op0 == Op1) return false; // If a subtract already has the same operands as a compare, swapping would be // bad. If a subtract has the same operands as a compare but in reverse order, // then swapping is good. int GoodToSwap = 0; unsigned NumInspected = 0; for (const User *U : Op0->users()) { // Avoid walking many users. if (++NumInspected > 128) return false; if (match(U, m_Sub(m_Specific(Op1), m_Specific(Op0)))) GoodToSwap++; else if (match(U, m_Sub(m_Specific(Op0), m_Specific(Op1)))) GoodToSwap--; } if (GoodToSwap > 0) { Cmp->swapOperands(); return true; } return false; } static bool foldFCmpToFPClassTest(CmpInst *Cmp, const TargetLowering &TLI, const DataLayout &DL) { FCmpInst *FCmp = dyn_cast(Cmp); if (!FCmp) return false; // Don't fold if the target offers free fabs and the predicate is legal. EVT VT = TLI.getValueType(DL, Cmp->getOperand(0)->getType()); if (TLI.isFAbsFree(VT) && TLI.isCondCodeLegal(getFCmpCondCode(FCmp->getPredicate()), VT.getSimpleVT())) return false; // Reverse the canonicalization if it is a FP class test auto ShouldReverseTransform = [](FPClassTest ClassTest) { return ClassTest == fcInf || ClassTest == (fcInf | fcNan); }; auto [ClassVal, ClassTest] = fcmpToClassTest(FCmp->getPredicate(), *FCmp->getParent()->getParent(), FCmp->getOperand(0), FCmp->getOperand(1)); if (!ClassVal) return false; if (!ShouldReverseTransform(ClassTest) && !ShouldReverseTransform(~ClassTest)) return false; IRBuilder<> Builder(Cmp); Value *IsFPClass = Builder.createIsFPClass(ClassVal, ClassTest); Cmp->replaceAllUsesWith(IsFPClass); RecursivelyDeleteTriviallyDeadInstructions(Cmp); return true; } bool CodeGenPrepare::optimizeCmp(CmpInst *Cmp, ModifyDT &ModifiedDT) { if (sinkCmpExpression(Cmp, *TLI)) return true; if (combineToUAddWithOverflow(Cmp, ModifiedDT)) return true; if (combineToUSubWithOverflow(Cmp, ModifiedDT)) return true; if (foldICmpWithDominatingICmp(Cmp, *TLI)) return true; if (swapICmpOperandsToExposeCSEOpportunities(Cmp)) return true; if (foldFCmpToFPClassTest(Cmp, *TLI, *DL)) return true; return false; } /// Duplicate and sink the given 'and' instruction into user blocks where it is /// used in a compare to allow isel to generate better code for targets where /// this operation can be combined. /// /// Return true if any changes are made. static bool sinkAndCmp0Expression(Instruction *AndI, const TargetLowering &TLI, SetOfInstrs &InsertedInsts) { // Double-check that we're not trying to optimize an instruction that was // already optimized by some other part of this pass. assert(!InsertedInsts.count(AndI) && "Attempting to optimize already optimized and instruction"); (void)InsertedInsts; // Nothing to do for single use in same basic block. if (AndI->hasOneUse() && AndI->getParent() == cast(*AndI->user_begin())->getParent()) return false; // Try to avoid cases where sinking/duplicating is likely to increase register // pressure. if (!isa(AndI->getOperand(0)) && !isa(AndI->getOperand(1)) && AndI->getOperand(0)->hasOneUse() && AndI->getOperand(1)->hasOneUse()) return false; for (auto *U : AndI->users()) { Instruction *User = cast(U); // Only sink 'and' feeding icmp with 0. if (!isa(User)) return false; auto *CmpC = dyn_cast(User->getOperand(1)); if (!CmpC || !CmpC->isZero()) return false; } if (!TLI.isMaskAndCmp0FoldingBeneficial(*AndI)) return false; LLVM_DEBUG(dbgs() << "found 'and' feeding only icmp 0;\n"); LLVM_DEBUG(AndI->getParent()->dump()); // Push the 'and' into the same block as the icmp 0. There should only be // one (icmp (and, 0)) in each block, since CSE/GVN should have removed any // others, so we don't need to keep track of which BBs we insert into. for (Value::user_iterator UI = AndI->user_begin(), E = AndI->user_end(); UI != E;) { Use &TheUse = UI.getUse(); Instruction *User = cast(*UI); // Preincrement use iterator so we don't invalidate it. ++UI; LLVM_DEBUG(dbgs() << "sinking 'and' use: " << *User << "\n"); // Keep the 'and' in the same place if the use is already in the same block. Instruction *InsertPt = User->getParent() == AndI->getParent() ? AndI : User; Instruction *InsertedAnd = BinaryOperator::Create( Instruction::And, AndI->getOperand(0), AndI->getOperand(1), "", InsertPt->getIterator()); // Propagate the debug info. InsertedAnd->setDebugLoc(AndI->getDebugLoc()); // Replace a use of the 'and' with a use of the new 'and'. TheUse = InsertedAnd; ++NumAndUses; LLVM_DEBUG(User->getParent()->dump()); } // We removed all uses, nuke the and. AndI->eraseFromParent(); return true; } /// Check if the candidates could be combined with a shift instruction, which /// includes: /// 1. Truncate instruction /// 2. And instruction and the imm is a mask of the low bits: /// imm & (imm+1) == 0 static bool isExtractBitsCandidateUse(Instruction *User) { if (!isa(User)) { if (User->getOpcode() != Instruction::And || !isa(User->getOperand(1))) return false; const APInt &Cimm = cast(User->getOperand(1))->getValue(); if ((Cimm & (Cimm + 1)).getBoolValue()) return false; } return true; } /// Sink both shift and truncate instruction to the use of truncate's BB. static bool SinkShiftAndTruncate(BinaryOperator *ShiftI, Instruction *User, ConstantInt *CI, DenseMap &InsertedShifts, const TargetLowering &TLI, const DataLayout &DL) { BasicBlock *UserBB = User->getParent(); DenseMap InsertedTruncs; auto *TruncI = cast(User); bool MadeChange = false; for (Value::user_iterator TruncUI = TruncI->user_begin(), TruncE = TruncI->user_end(); TruncUI != TruncE;) { Use &TruncTheUse = TruncUI.getUse(); Instruction *TruncUser = cast(*TruncUI); // Preincrement use iterator so we don't invalidate it. ++TruncUI; int ISDOpcode = TLI.InstructionOpcodeToISD(TruncUser->getOpcode()); if (!ISDOpcode) continue; // If the use is actually a legal node, there will not be an // implicit truncate. // FIXME: always querying the result type is just an // approximation; some nodes' legality is determined by the // operand or other means. There's no good way to find out though. if (TLI.isOperationLegalOrCustom( ISDOpcode, TLI.getValueType(DL, TruncUser->getType(), true))) continue; // Don't bother for PHI nodes. if (isa(TruncUser)) continue; BasicBlock *TruncUserBB = TruncUser->getParent(); if (UserBB == TruncUserBB) continue; BinaryOperator *&InsertedShift = InsertedShifts[TruncUserBB]; CastInst *&InsertedTrunc = InsertedTruncs[TruncUserBB]; if (!InsertedShift && !InsertedTrunc) { BasicBlock::iterator InsertPt = TruncUserBB->getFirstInsertionPt(); assert(InsertPt != TruncUserBB->end()); // Sink the shift if (ShiftI->getOpcode() == Instruction::AShr) InsertedShift = BinaryOperator::CreateAShr(ShiftI->getOperand(0), CI, ""); else InsertedShift = BinaryOperator::CreateLShr(ShiftI->getOperand(0), CI, ""); InsertedShift->setDebugLoc(ShiftI->getDebugLoc()); InsertedShift->insertBefore(*TruncUserBB, InsertPt); // Sink the trunc BasicBlock::iterator TruncInsertPt = TruncUserBB->getFirstInsertionPt(); TruncInsertPt++; // It will go ahead of any debug-info. TruncInsertPt.setHeadBit(true); assert(TruncInsertPt != TruncUserBB->end()); InsertedTrunc = CastInst::Create(TruncI->getOpcode(), InsertedShift, TruncI->getType(), ""); InsertedTrunc->insertBefore(*TruncUserBB, TruncInsertPt); InsertedTrunc->setDebugLoc(TruncI->getDebugLoc()); MadeChange = true; TruncTheUse = InsertedTrunc; } } return MadeChange; } /// Sink the shift *right* instruction into user blocks if the uses could /// potentially be combined with this shift instruction and generate BitExtract /// instruction. It will only be applied if the architecture supports BitExtract /// instruction. Here is an example: /// BB1: /// %x.extract.shift = lshr i64 %arg1, 32 /// BB2: /// %x.extract.trunc = trunc i64 %x.extract.shift to i16 /// ==> /// /// BB2: /// %x.extract.shift.1 = lshr i64 %arg1, 32 /// %x.extract.trunc = trunc i64 %x.extract.shift.1 to i16 /// /// CodeGen will recognize the pattern in BB2 and generate BitExtract /// instruction. /// Return true if any changes are made. static bool OptimizeExtractBits(BinaryOperator *ShiftI, ConstantInt *CI, const TargetLowering &TLI, const DataLayout &DL) { BasicBlock *DefBB = ShiftI->getParent(); /// Only insert instructions in each block once. DenseMap InsertedShifts; bool shiftIsLegal = TLI.isTypeLegal(TLI.getValueType(DL, ShiftI->getType())); bool MadeChange = false; for (Value::user_iterator UI = ShiftI->user_begin(), E = ShiftI->user_end(); UI != E;) { Use &TheUse = UI.getUse(); Instruction *User = cast(*UI); // Preincrement use iterator so we don't invalidate it. ++UI; // Don't bother for PHI nodes. if (isa(User)) continue; if (!isExtractBitsCandidateUse(User)) continue; BasicBlock *UserBB = User->getParent(); if (UserBB == DefBB) { // If the shift and truncate instruction are in the same BB. The use of // the truncate(TruncUse) may still introduce another truncate if not // legal. In this case, we would like to sink both shift and truncate // instruction to the BB of TruncUse. // for example: // BB1: // i64 shift.result = lshr i64 opnd, imm // trunc.result = trunc shift.result to i16 // // BB2: // ----> We will have an implicit truncate here if the architecture does // not have i16 compare. // cmp i16 trunc.result, opnd2 // if (isa(User) && shiftIsLegal // If the type of the truncate is legal, no truncate will be // introduced in other basic blocks. && (!TLI.isTypeLegal(TLI.getValueType(DL, User->getType())))) MadeChange = SinkShiftAndTruncate(ShiftI, User, CI, InsertedShifts, TLI, DL); continue; } // If we have already inserted a shift into this block, use it. BinaryOperator *&InsertedShift = InsertedShifts[UserBB]; if (!InsertedShift) { BasicBlock::iterator InsertPt = UserBB->getFirstInsertionPt(); assert(InsertPt != UserBB->end()); if (ShiftI->getOpcode() == Instruction::AShr) InsertedShift = BinaryOperator::CreateAShr(ShiftI->getOperand(0), CI, ""); else InsertedShift = BinaryOperator::CreateLShr(ShiftI->getOperand(0), CI, ""); InsertedShift->insertBefore(*UserBB, InsertPt); InsertedShift->setDebugLoc(ShiftI->getDebugLoc()); MadeChange = true; } // Replace a use of the shift with a use of the new shift. TheUse = InsertedShift; } // If we removed all uses, or there are none, nuke the shift. if (ShiftI->use_empty()) { salvageDebugInfo(*ShiftI); ShiftI->eraseFromParent(); MadeChange = true; } return MadeChange; } /// If counting leading or trailing zeros is an expensive operation and a zero /// input is defined, add a check for zero to avoid calling the intrinsic. /// /// We want to transform: /// %z = call i64 @llvm.cttz.i64(i64 %A, i1 false) /// /// into: /// entry: /// %cmpz = icmp eq i64 %A, 0 /// br i1 %cmpz, label %cond.end, label %cond.false /// cond.false: /// %z = call i64 @llvm.cttz.i64(i64 %A, i1 true) /// br label %cond.end /// cond.end: /// %ctz = phi i64 [ 64, %entry ], [ %z, %cond.false ] /// /// If the transform is performed, return true and set ModifiedDT to true. static bool despeculateCountZeros(IntrinsicInst *CountZeros, LoopInfo &LI, const TargetLowering *TLI, const DataLayout *DL, ModifyDT &ModifiedDT, SmallSet &FreshBBs, bool IsHugeFunc) { // If a zero input is undefined, it doesn't make sense to despeculate that. if (match(CountZeros->getOperand(1), m_One())) return false; // If it's cheap to speculate, there's nothing to do. Type *Ty = CountZeros->getType(); auto IntrinsicID = CountZeros->getIntrinsicID(); if ((IntrinsicID == Intrinsic::cttz && TLI->isCheapToSpeculateCttz(Ty)) || (IntrinsicID == Intrinsic::ctlz && TLI->isCheapToSpeculateCtlz(Ty))) return false; // Only handle legal scalar cases. Anything else requires too much work. unsigned SizeInBits = Ty->getScalarSizeInBits(); if (Ty->isVectorTy() || SizeInBits > DL->getLargestLegalIntTypeSizeInBits()) return false; // Bail if the value is never zero. Use &Op = CountZeros->getOperandUse(0); if (isKnownNonZero(Op, *DL)) return false; // The intrinsic will be sunk behind a compare against zero and branch. BasicBlock *StartBlock = CountZeros->getParent(); BasicBlock *CallBlock = StartBlock->splitBasicBlock(CountZeros, "cond.false"); if (IsHugeFunc) FreshBBs.insert(CallBlock); // Create another block after the count zero intrinsic. A PHI will be added // in this block to select the result of the intrinsic or the bit-width // constant if the input to the intrinsic is zero. BasicBlock::iterator SplitPt = std::next(BasicBlock::iterator(CountZeros)); // Any debug-info after CountZeros should not be included. SplitPt.setHeadBit(true); BasicBlock *EndBlock = CallBlock->splitBasicBlock(SplitPt, "cond.end"); if (IsHugeFunc) FreshBBs.insert(EndBlock); // Update the LoopInfo. The new blocks are in the same loop as the start // block. if (Loop *L = LI.getLoopFor(StartBlock)) { L->addBasicBlockToLoop(CallBlock, LI); L->addBasicBlockToLoop(EndBlock, LI); } // Set up a builder to create a compare, conditional branch, and PHI. IRBuilder<> Builder(CountZeros->getContext()); Builder.SetInsertPoint(StartBlock->getTerminator()); Builder.SetCurrentDebugLocation(CountZeros->getDebugLoc()); // Replace the unconditional branch that was created by the first split with // a compare against zero and a conditional branch. Value *Zero = Constant::getNullValue(Ty); // Avoid introducing branch on poison. This also replaces the ctz operand. if (!isGuaranteedNotToBeUndefOrPoison(Op)) Op = Builder.CreateFreeze(Op, Op->getName() + ".fr"); Value *Cmp = Builder.CreateICmpEQ(Op, Zero, "cmpz"); Builder.CreateCondBr(Cmp, EndBlock, CallBlock); StartBlock->getTerminator()->eraseFromParent(); // Create a PHI in the end block to select either the output of the intrinsic // or the bit width of the operand. Builder.SetInsertPoint(EndBlock, EndBlock->begin()); PHINode *PN = Builder.CreatePHI(Ty, 2, "ctz"); replaceAllUsesWith(CountZeros, PN, FreshBBs, IsHugeFunc); Value *BitWidth = Builder.getInt(APInt(SizeInBits, SizeInBits)); PN->addIncoming(BitWidth, StartBlock); PN->addIncoming(CountZeros, CallBlock); // We are explicitly handling the zero case, so we can set the intrinsic's // undefined zero argument to 'true'. This will also prevent reprocessing the // intrinsic; we only despeculate when a zero input is defined. CountZeros->setArgOperand(1, Builder.getTrue()); ModifiedDT = ModifyDT::ModifyBBDT; return true; } bool CodeGenPrepare::optimizeCallInst(CallInst *CI, ModifyDT &ModifiedDT) { BasicBlock *BB = CI->getParent(); // Lower inline assembly if we can. // If we found an inline asm expession, and if the target knows how to // lower it to normal LLVM code, do so now. if (CI->isInlineAsm()) { if (TLI->ExpandInlineAsm(CI)) { // Avoid invalidating the iterator. CurInstIterator = BB->begin(); // Avoid processing instructions out of order, which could cause // reuse before a value is defined. SunkAddrs.clear(); return true; } // Sink address computing for memory operands into the block. if (optimizeInlineAsmInst(CI)) return true; } // Align the pointer arguments to this call if the target thinks it's a good // idea unsigned MinSize; Align PrefAlign; if (TLI->shouldAlignPointerArgs(CI, MinSize, PrefAlign)) { for (auto &Arg : CI->args()) { // We want to align both objects whose address is used directly and // objects whose address is used in casts and GEPs, though it only makes // sense for GEPs if the offset is a multiple of the desired alignment and // if size - offset meets the size threshold. if (!Arg->getType()->isPointerTy()) continue; APInt Offset(DL->getIndexSizeInBits( cast(Arg->getType())->getAddressSpace()), 0); Value *Val = Arg->stripAndAccumulateInBoundsConstantOffsets(*DL, Offset); uint64_t Offset2 = Offset.getLimitedValue(); if (!isAligned(PrefAlign, Offset2)) continue; AllocaInst *AI; if ((AI = dyn_cast(Val)) && AI->getAlign() < PrefAlign && DL->getTypeAllocSize(AI->getAllocatedType()) >= MinSize + Offset2) AI->setAlignment(PrefAlign); // Global variables can only be aligned if they are defined in this // object (i.e. they are uniquely initialized in this object), and // over-aligning global variables that have an explicit section is // forbidden. GlobalVariable *GV; if ((GV = dyn_cast(Val)) && GV->canIncreaseAlignment() && GV->getPointerAlignment(*DL) < PrefAlign && DL->getTypeAllocSize(GV->getValueType()) >= MinSize + Offset2) GV->setAlignment(PrefAlign); } } // If this is a memcpy (or similar) then we may be able to improve the // alignment. if (MemIntrinsic *MI = dyn_cast(CI)) { Align DestAlign = getKnownAlignment(MI->getDest(), *DL); MaybeAlign MIDestAlign = MI->getDestAlign(); if (!MIDestAlign || DestAlign > *MIDestAlign) MI->setDestAlignment(DestAlign); if (MemTransferInst *MTI = dyn_cast(MI)) { MaybeAlign MTISrcAlign = MTI->getSourceAlign(); Align SrcAlign = getKnownAlignment(MTI->getSource(), *DL); if (!MTISrcAlign || SrcAlign > *MTISrcAlign) MTI->setSourceAlignment(SrcAlign); } } // If we have a cold call site, try to sink addressing computation into the // cold block. This interacts with our handling for loads and stores to // ensure that we can fold all uses of a potential addressing computation // into their uses. TODO: generalize this to work over profiling data if (CI->hasFnAttr(Attribute::Cold) && !OptSize && !llvm::shouldOptimizeForSize(BB, PSI, BFI.get())) for (auto &Arg : CI->args()) { if (!Arg->getType()->isPointerTy()) continue; unsigned AS = Arg->getType()->getPointerAddressSpace(); if (optimizeMemoryInst(CI, Arg, Arg->getType(), AS)) return true; } IntrinsicInst *II = dyn_cast(CI); if (II) { switch (II->getIntrinsicID()) { default: break; case Intrinsic::assume: llvm_unreachable("llvm.assume should have been removed already"); case Intrinsic::allow_runtime_check: case Intrinsic::allow_ubsan_check: case Intrinsic::experimental_widenable_condition: { // Give up on future widening opportunities so that we can fold away dead // paths and merge blocks before going into block-local instruction // selection. if (II->use_empty()) { II->eraseFromParent(); return true; } Constant *RetVal = ConstantInt::getTrue(II->getContext()); resetIteratorIfInvalidatedWhileCalling(BB, [&]() { replaceAndRecursivelySimplify(CI, RetVal, TLInfo, nullptr); }); return true; } case Intrinsic::objectsize: llvm_unreachable("llvm.objectsize.* should have been lowered already"); case Intrinsic::is_constant: llvm_unreachable("llvm.is.constant.* should have been lowered already"); case Intrinsic::aarch64_stlxr: case Intrinsic::aarch64_stxr: { ZExtInst *ExtVal = dyn_cast(CI->getArgOperand(0)); if (!ExtVal || !ExtVal->hasOneUse() || ExtVal->getParent() == CI->getParent()) return false; // Sink a zext feeding stlxr/stxr before it, so it can be folded into it. ExtVal->moveBefore(CI); // Mark this instruction as "inserted by CGP", so that other // optimizations don't touch it. InsertedInsts.insert(ExtVal); return true; } case Intrinsic::launder_invariant_group: case Intrinsic::strip_invariant_group: { Value *ArgVal = II->getArgOperand(0); auto it = LargeOffsetGEPMap.find(II); if (it != LargeOffsetGEPMap.end()) { // Merge entries in LargeOffsetGEPMap to reflect the RAUW. // Make sure not to have to deal with iterator invalidation // after possibly adding ArgVal to LargeOffsetGEPMap. auto GEPs = std::move(it->second); LargeOffsetGEPMap[ArgVal].append(GEPs.begin(), GEPs.end()); LargeOffsetGEPMap.erase(II); } replaceAllUsesWith(II, ArgVal, FreshBBs, IsHugeFunc); II->eraseFromParent(); return true; } case Intrinsic::cttz: case Intrinsic::ctlz: // If counting zeros is expensive, try to avoid it. return despeculateCountZeros(II, *LI, TLI, DL, ModifiedDT, FreshBBs, IsHugeFunc); case Intrinsic::fshl: case Intrinsic::fshr: return optimizeFunnelShift(II); case Intrinsic::dbg_assign: case Intrinsic::dbg_value: return fixupDbgValue(II); case Intrinsic::masked_gather: return optimizeGatherScatterInst(II, II->getArgOperand(0)); case Intrinsic::masked_scatter: return optimizeGatherScatterInst(II, II->getArgOperand(1)); } SmallVector PtrOps; Type *AccessTy; if (TLI->getAddrModeArguments(II, PtrOps, AccessTy)) while (!PtrOps.empty()) { Value *PtrVal = PtrOps.pop_back_val(); unsigned AS = PtrVal->getType()->getPointerAddressSpace(); if (optimizeMemoryInst(II, PtrVal, AccessTy, AS)) return true; } } // From here on out we're working with named functions. if (!CI->getCalledFunction()) return false; // Lower all default uses of _chk calls. This is very similar // to what InstCombineCalls does, but here we are only lowering calls // to fortified library functions (e.g. __memcpy_chk) that have the default // "don't know" as the objectsize. Anything else should be left alone. FortifiedLibCallSimplifier Simplifier(TLInfo, true); IRBuilder<> Builder(CI); if (Value *V = Simplifier.optimizeCall(CI, Builder)) { replaceAllUsesWith(CI, V, FreshBBs, IsHugeFunc); CI->eraseFromParent(); return true; } return false; } static bool isIntrinsicOrLFToBeTailCalled(const TargetLibraryInfo *TLInfo, const CallInst *CI) { assert(CI && CI->use_empty()); if (const auto *II = dyn_cast(CI)) switch (II->getIntrinsicID()) { case Intrinsic::memset: case Intrinsic::memcpy: case Intrinsic::memmove: return true; default: return false; } LibFunc LF; Function *Callee = CI->getCalledFunction(); if (Callee && TLInfo && TLInfo->getLibFunc(*Callee, LF)) switch (LF) { case LibFunc_strcpy: case LibFunc_strncpy: case LibFunc_strcat: case LibFunc_strncat: return true; default: return false; } return false; } /// Look for opportunities to duplicate return instructions to the predecessor /// to enable tail call optimizations. The case it is currently looking for is /// the following one. Known intrinsics or library function that may be tail /// called are taken into account as well. /// @code /// bb0: /// %tmp0 = tail call i32 @f0() /// br label %return /// bb1: /// %tmp1 = tail call i32 @f1() /// br label %return /// bb2: /// %tmp2 = tail call i32 @f2() /// br label %return /// return: /// %retval = phi i32 [ %tmp0, %bb0 ], [ %tmp1, %bb1 ], [ %tmp2, %bb2 ] /// ret i32 %retval /// @endcode /// /// => /// /// @code /// bb0: /// %tmp0 = tail call i32 @f0() /// ret i32 %tmp0 /// bb1: /// %tmp1 = tail call i32 @f1() /// ret i32 %tmp1 /// bb2: /// %tmp2 = tail call i32 @f2() /// ret i32 %tmp2 /// @endcode bool CodeGenPrepare::dupRetToEnableTailCallOpts(BasicBlock *BB, ModifyDT &ModifiedDT) { if (!BB->getTerminator()) return false; ReturnInst *RetI = dyn_cast(BB->getTerminator()); if (!RetI) return false; assert(LI->getLoopFor(BB) == nullptr && "A return block cannot be in a loop"); PHINode *PN = nullptr; ExtractValueInst *EVI = nullptr; BitCastInst *BCI = nullptr; Value *V = RetI->getReturnValue(); if (V) { BCI = dyn_cast(V); if (BCI) V = BCI->getOperand(0); EVI = dyn_cast(V); if (EVI) { V = EVI->getOperand(0); if (!llvm::all_of(EVI->indices(), [](unsigned idx) { return idx == 0; })) return false; } PN = dyn_cast(V); } if (PN && PN->getParent() != BB) return false; auto isLifetimeEndOrBitCastFor = [](const Instruction *Inst) { const BitCastInst *BC = dyn_cast(Inst); if (BC && BC->hasOneUse()) Inst = BC->user_back(); if (const IntrinsicInst *II = dyn_cast(Inst)) return II->getIntrinsicID() == Intrinsic::lifetime_end; return false; }; // Make sure there are no instructions between the first instruction // and return. const Instruction *BI = BB->getFirstNonPHI(); // Skip over debug and the bitcast. while (isa(BI) || BI == BCI || BI == EVI || isa(BI) || isLifetimeEndOrBitCastFor(BI)) BI = BI->getNextNode(); if (BI != RetI) return false; /// Only dup the ReturnInst if the CallInst is likely to be emitted as a tail /// call. const Function *F = BB->getParent(); SmallVector TailCallBBs; if (PN) { for (unsigned I = 0, E = PN->getNumIncomingValues(); I != E; ++I) { // Look through bitcasts. Value *IncomingVal = PN->getIncomingValue(I)->stripPointerCasts(); CallInst *CI = dyn_cast(IncomingVal); BasicBlock *PredBB = PN->getIncomingBlock(I); // Make sure the phi value is indeed produced by the tail call. if (CI && CI->hasOneUse() && CI->getParent() == PredBB && TLI->mayBeEmittedAsTailCall(CI) && attributesPermitTailCall(F, CI, RetI, *TLI)) { TailCallBBs.push_back(PredBB); } else { // Consider the cases in which the phi value is indirectly produced by // the tail call, for example when encountering memset(), memmove(), // strcpy(), whose return value may have been optimized out. In such // cases, the value needs to be the first function argument. // // bb0: // tail call void @llvm.memset.p0.i64(ptr %0, i8 0, i64 %1) // br label %return // return: // %phi = phi ptr [ %0, %bb0 ], [ %2, %entry ] if (PredBB && PredBB->getSingleSuccessor() == BB) CI = dyn_cast_or_null( PredBB->getTerminator()->getPrevNonDebugInstruction(true)); if (CI && CI->use_empty() && isIntrinsicOrLFToBeTailCalled(TLInfo, CI) && IncomingVal == CI->getArgOperand(0) && TLI->mayBeEmittedAsTailCall(CI) && attributesPermitTailCall(F, CI, RetI, *TLI)) TailCallBBs.push_back(PredBB); } } } else { SmallPtrSet VisitedBBs; for (BasicBlock *Pred : predecessors(BB)) { if (!VisitedBBs.insert(Pred).second) continue; if (Instruction *I = Pred->rbegin()->getPrevNonDebugInstruction(true)) { CallInst *CI = dyn_cast(I); if (CI && CI->use_empty() && TLI->mayBeEmittedAsTailCall(CI) && attributesPermitTailCall(F, CI, RetI, *TLI)) { // Either we return void or the return value must be the first // argument of a known intrinsic or library function. if (!V || isa(V) || (isIntrinsicOrLFToBeTailCalled(TLInfo, CI) && V == CI->getArgOperand(0))) { TailCallBBs.push_back(Pred); } } } } } bool Changed = false; for (auto const &TailCallBB : TailCallBBs) { // Make sure the call instruction is followed by an unconditional branch to // the return block. BranchInst *BI = dyn_cast(TailCallBB->getTerminator()); if (!BI || !BI->isUnconditional() || BI->getSuccessor(0) != BB) continue; // Duplicate the return into TailCallBB. (void)FoldReturnIntoUncondBranch(RetI, BB, TailCallBB); assert(!VerifyBFIUpdates || BFI->getBlockFreq(BB) >= BFI->getBlockFreq(TailCallBB)); BFI->setBlockFreq(BB, (BFI->getBlockFreq(BB) - BFI->getBlockFreq(TailCallBB))); ModifiedDT = ModifyDT::ModifyBBDT; Changed = true; ++NumRetsDup; } // If we eliminated all predecessors of the block, delete the block now. if (Changed && !BB->hasAddressTaken() && pred_empty(BB)) BB->eraseFromParent(); return Changed; } //===----------------------------------------------------------------------===// // Memory Optimization //===----------------------------------------------------------------------===// namespace { /// This is an extended version of TargetLowering::AddrMode /// which holds actual Value*'s for register values. struct ExtAddrMode : public TargetLowering::AddrMode { Value *BaseReg = nullptr; Value *ScaledReg = nullptr; Value *OriginalValue = nullptr; bool InBounds = true; enum FieldName { NoField = 0x00, BaseRegField = 0x01, BaseGVField = 0x02, BaseOffsField = 0x04, ScaledRegField = 0x08, ScaleField = 0x10, MultipleFields = 0xff }; ExtAddrMode() = default; void print(raw_ostream &OS) const; void dump() const; FieldName compare(const ExtAddrMode &other) { // First check that the types are the same on each field, as differing types // is something we can't cope with later on. if (BaseReg && other.BaseReg && BaseReg->getType() != other.BaseReg->getType()) return MultipleFields; if (BaseGV && other.BaseGV && BaseGV->getType() != other.BaseGV->getType()) return MultipleFields; if (ScaledReg && other.ScaledReg && ScaledReg->getType() != other.ScaledReg->getType()) return MultipleFields; // Conservatively reject 'inbounds' mismatches. if (InBounds != other.InBounds) return MultipleFields; // Check each field to see if it differs. unsigned Result = NoField; if (BaseReg != other.BaseReg) Result |= BaseRegField; if (BaseGV != other.BaseGV) Result |= BaseGVField; if (BaseOffs != other.BaseOffs) Result |= BaseOffsField; if (ScaledReg != other.ScaledReg) Result |= ScaledRegField; // Don't count 0 as being a different scale, because that actually means // unscaled (which will already be counted by having no ScaledReg). if (Scale && other.Scale && Scale != other.Scale) Result |= ScaleField; if (llvm::popcount(Result) > 1) return MultipleFields; else return static_cast(Result); } // An AddrMode is trivial if it involves no calculation i.e. it is just a base // with no offset. bool isTrivial() { // An AddrMode is (BaseGV + BaseReg + BaseOffs + ScaleReg * Scale) so it is // trivial if at most one of these terms is nonzero, except that BaseGV and // BaseReg both being zero actually means a null pointer value, which we // consider to be 'non-zero' here. return !BaseOffs && !Scale && !(BaseGV && BaseReg); } Value *GetFieldAsValue(FieldName Field, Type *IntPtrTy) { switch (Field) { default: return nullptr; case BaseRegField: return BaseReg; case BaseGVField: return BaseGV; case ScaledRegField: return ScaledReg; case BaseOffsField: return ConstantInt::get(IntPtrTy, BaseOffs); } } void SetCombinedField(FieldName Field, Value *V, const SmallVectorImpl &AddrModes) { switch (Field) { default: llvm_unreachable("Unhandled fields are expected to be rejected earlier"); break; case ExtAddrMode::BaseRegField: BaseReg = V; break; case ExtAddrMode::BaseGVField: // A combined BaseGV is an Instruction, not a GlobalValue, so it goes // in the BaseReg field. assert(BaseReg == nullptr); BaseReg = V; BaseGV = nullptr; break; case ExtAddrMode::ScaledRegField: ScaledReg = V; // If we have a mix of scaled and unscaled addrmodes then we want scale // to be the scale and not zero. if (!Scale) for (const ExtAddrMode &AM : AddrModes) if (AM.Scale) { Scale = AM.Scale; break; } break; case ExtAddrMode::BaseOffsField: // The offset is no longer a constant, so it goes in ScaledReg with a // scale of 1. assert(ScaledReg == nullptr); ScaledReg = V; Scale = 1; BaseOffs = 0; break; } } }; #ifndef NDEBUG static inline raw_ostream &operator<<(raw_ostream &OS, const ExtAddrMode &AM) { AM.print(OS); return OS; } #endif #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) void ExtAddrMode::print(raw_ostream &OS) const { bool NeedPlus = false; OS << "["; if (InBounds) OS << "inbounds "; if (BaseGV) { OS << "GV:"; BaseGV->printAsOperand(OS, /*PrintType=*/false); NeedPlus = true; } if (BaseOffs) { OS << (NeedPlus ? " + " : "") << BaseOffs; NeedPlus = true; } if (BaseReg) { OS << (NeedPlus ? " + " : "") << "Base:"; BaseReg->printAsOperand(OS, /*PrintType=*/false); NeedPlus = true; } if (Scale) { OS << (NeedPlus ? " + " : "") << Scale << "*"; ScaledReg->printAsOperand(OS, /*PrintType=*/false); } OS << ']'; } LLVM_DUMP_METHOD void ExtAddrMode::dump() const { print(dbgs()); dbgs() << '\n'; } #endif } // end anonymous namespace namespace { /// This class provides transaction based operation on the IR. /// Every change made through this class is recorded in the internal state and /// can be undone (rollback) until commit is called. /// CGP does not check if instructions could be speculatively executed when /// moved. Preserving the original location would pessimize the debugging /// experience, as well as negatively impact the quality of sample PGO. class TypePromotionTransaction { /// This represents the common interface of the individual transaction. /// Each class implements the logic for doing one specific modification on /// the IR via the TypePromotionTransaction. class TypePromotionAction { protected: /// The Instruction modified. Instruction *Inst; public: /// Constructor of the action. /// The constructor performs the related action on the IR. TypePromotionAction(Instruction *Inst) : Inst(Inst) {} virtual ~TypePromotionAction() = default; /// Undo the modification done by this action. /// When this method is called, the IR must be in the same state as it was /// before this action was applied. /// \pre Undoing the action works if and only if the IR is in the exact same /// state as it was directly after this action was applied. virtual void undo() = 0; /// Advocate every change made by this action. /// When the results on the IR of the action are to be kept, it is important /// to call this function, otherwise hidden information may be kept forever. virtual void commit() { // Nothing to be done, this action is not doing anything. } }; /// Utility to remember the position of an instruction. class InsertionHandler { /// Position of an instruction. /// Either an instruction: /// - Is the first in a basic block: BB is used. /// - Has a previous instruction: PrevInst is used. union { Instruction *PrevInst; BasicBlock *BB; } Point; std::optional BeforeDbgRecord = std::nullopt; /// Remember whether or not the instruction had a previous instruction. bool HasPrevInstruction; public: /// Record the position of \p Inst. InsertionHandler(Instruction *Inst) { HasPrevInstruction = (Inst != &*(Inst->getParent()->begin())); BasicBlock *BB = Inst->getParent(); // Record where we would have to re-insert the instruction in the sequence // of DbgRecords, if we ended up reinserting. if (BB->IsNewDbgInfoFormat) BeforeDbgRecord = Inst->getDbgReinsertionPosition(); if (HasPrevInstruction) { Point.PrevInst = &*std::prev(Inst->getIterator()); } else { Point.BB = BB; } } /// Insert \p Inst at the recorded position. void insert(Instruction *Inst) { if (HasPrevInstruction) { if (Inst->getParent()) Inst->removeFromParent(); Inst->insertAfter(&*Point.PrevInst); } else { BasicBlock::iterator Position = Point.BB->getFirstInsertionPt(); if (Inst->getParent()) Inst->moveBefore(*Point.BB, Position); else Inst->insertBefore(*Point.BB, Position); } Inst->getParent()->reinsertInstInDbgRecords(Inst, BeforeDbgRecord); } }; /// Move an instruction before another. class InstructionMoveBefore : public TypePromotionAction { /// Original position of the instruction. InsertionHandler Position; public: /// Move \p Inst before \p Before. InstructionMoveBefore(Instruction *Inst, Instruction *Before) : TypePromotionAction(Inst), Position(Inst) { LLVM_DEBUG(dbgs() << "Do: move: " << *Inst << "\nbefore: " << *Before << "\n"); Inst->moveBefore(Before); } /// Move the instruction back to its original position. void undo() override { LLVM_DEBUG(dbgs() << "Undo: moveBefore: " << *Inst << "\n"); Position.insert(Inst); } }; /// Set the operand of an instruction with a new value. class OperandSetter : public TypePromotionAction { /// Original operand of the instruction. Value *Origin; /// Index of the modified instruction. unsigned Idx; public: /// Set \p Idx operand of \p Inst with \p NewVal. OperandSetter(Instruction *Inst, unsigned Idx, Value *NewVal) : TypePromotionAction(Inst), Idx(Idx) { LLVM_DEBUG(dbgs() << "Do: setOperand: " << Idx << "\n" << "for:" << *Inst << "\n" << "with:" << *NewVal << "\n"); Origin = Inst->getOperand(Idx); Inst->setOperand(Idx, NewVal); } /// Restore the original value of the instruction. void undo() override { LLVM_DEBUG(dbgs() << "Undo: setOperand:" << Idx << "\n" << "for: " << *Inst << "\n" << "with: " << *Origin << "\n"); Inst->setOperand(Idx, Origin); } }; /// Hide the operands of an instruction. /// Do as if this instruction was not using any of its operands. class OperandsHider : public TypePromotionAction { /// The list of original operands. SmallVector OriginalValues; public: /// Remove \p Inst from the uses of the operands of \p Inst. OperandsHider(Instruction *Inst) : TypePromotionAction(Inst) { LLVM_DEBUG(dbgs() << "Do: OperandsHider: " << *Inst << "\n"); unsigned NumOpnds = Inst->getNumOperands(); OriginalValues.reserve(NumOpnds); for (unsigned It = 0; It < NumOpnds; ++It) { // Save the current operand. Value *Val = Inst->getOperand(It); OriginalValues.push_back(Val); // Set a dummy one. // We could use OperandSetter here, but that would imply an overhead // that we are not willing to pay. Inst->setOperand(It, UndefValue::get(Val->getType())); } } /// Restore the original list of uses. void undo() override { LLVM_DEBUG(dbgs() << "Undo: OperandsHider: " << *Inst << "\n"); for (unsigned It = 0, EndIt = OriginalValues.size(); It != EndIt; ++It) Inst->setOperand(It, OriginalValues[It]); } }; /// Build a truncate instruction. class TruncBuilder : public TypePromotionAction { Value *Val; public: /// Build a truncate instruction of \p Opnd producing a \p Ty /// result. /// trunc Opnd to Ty. TruncBuilder(Instruction *Opnd, Type *Ty) : TypePromotionAction(Opnd) { IRBuilder<> Builder(Opnd); Builder.SetCurrentDebugLocation(DebugLoc()); Val = Builder.CreateTrunc(Opnd, Ty, "promoted"); LLVM_DEBUG(dbgs() << "Do: TruncBuilder: " << *Val << "\n"); } /// Get the built value. Value *getBuiltValue() { return Val; } /// Remove the built instruction. void undo() override { LLVM_DEBUG(dbgs() << "Undo: TruncBuilder: " << *Val << "\n"); if (Instruction *IVal = dyn_cast(Val)) IVal->eraseFromParent(); } }; /// Build a sign extension instruction. class SExtBuilder : public TypePromotionAction { Value *Val; public: /// Build a sign extension instruction of \p Opnd producing a \p Ty /// result. /// sext Opnd to Ty. SExtBuilder(Instruction *InsertPt, Value *Opnd, Type *Ty) : TypePromotionAction(InsertPt) { IRBuilder<> Builder(InsertPt); Val = Builder.CreateSExt(Opnd, Ty, "promoted"); LLVM_DEBUG(dbgs() << "Do: SExtBuilder: " << *Val << "\n"); } /// Get the built value. Value *getBuiltValue() { return Val; } /// Remove the built instruction. void undo() override { LLVM_DEBUG(dbgs() << "Undo: SExtBuilder: " << *Val << "\n"); if (Instruction *IVal = dyn_cast(Val)) IVal->eraseFromParent(); } }; /// Build a zero extension instruction. class ZExtBuilder : public TypePromotionAction { Value *Val; public: /// Build a zero extension instruction of \p Opnd producing a \p Ty /// result. /// zext Opnd to Ty. ZExtBuilder(Instruction *InsertPt, Value *Opnd, Type *Ty) : TypePromotionAction(InsertPt) { IRBuilder<> Builder(InsertPt); Builder.SetCurrentDebugLocation(DebugLoc()); Val = Builder.CreateZExt(Opnd, Ty, "promoted"); LLVM_DEBUG(dbgs() << "Do: ZExtBuilder: " << *Val << "\n"); } /// Get the built value. Value *getBuiltValue() { return Val; } /// Remove the built instruction. void undo() override { LLVM_DEBUG(dbgs() << "Undo: ZExtBuilder: " << *Val << "\n"); if (Instruction *IVal = dyn_cast(Val)) IVal->eraseFromParent(); } }; /// Mutate an instruction to another type. class TypeMutator : public TypePromotionAction { /// Record the original type. Type *OrigTy; public: /// Mutate the type of \p Inst into \p NewTy. TypeMutator(Instruction *Inst, Type *NewTy) : TypePromotionAction(Inst), OrigTy(Inst->getType()) { LLVM_DEBUG(dbgs() << "Do: MutateType: " << *Inst << " with " << *NewTy << "\n"); Inst->mutateType(NewTy); } /// Mutate the instruction back to its original type. void undo() override { LLVM_DEBUG(dbgs() << "Undo: MutateType: " << *Inst << " with " << *OrigTy << "\n"); Inst->mutateType(OrigTy); } }; /// Replace the uses of an instruction by another instruction. class UsesReplacer : public TypePromotionAction { /// Helper structure to keep track of the replaced uses. struct InstructionAndIdx { /// The instruction using the instruction. Instruction *Inst; /// The index where this instruction is used for Inst. unsigned Idx; InstructionAndIdx(Instruction *Inst, unsigned Idx) : Inst(Inst), Idx(Idx) {} }; /// Keep track of the original uses (pair Instruction, Index). SmallVector OriginalUses; /// Keep track of the debug users. SmallVector DbgValues; /// And non-instruction debug-users too. SmallVector DbgVariableRecords; /// Keep track of the new value so that we can undo it by replacing /// instances of the new value with the original value. Value *New; using use_iterator = SmallVectorImpl::iterator; public: /// Replace all the use of \p Inst by \p New. UsesReplacer(Instruction *Inst, Value *New) : TypePromotionAction(Inst), New(New) { LLVM_DEBUG(dbgs() << "Do: UsersReplacer: " << *Inst << " with " << *New << "\n"); // Record the original uses. for (Use &U : Inst->uses()) { Instruction *UserI = cast(U.getUser()); OriginalUses.push_back(InstructionAndIdx(UserI, U.getOperandNo())); } // Record the debug uses separately. They are not in the instruction's // use list, but they are replaced by RAUW. findDbgValues(DbgValues, Inst, &DbgVariableRecords); // Now, we can replace the uses. Inst->replaceAllUsesWith(New); } /// Reassign the original uses of Inst to Inst. void undo() override { LLVM_DEBUG(dbgs() << "Undo: UsersReplacer: " << *Inst << "\n"); for (InstructionAndIdx &Use : OriginalUses) Use.Inst->setOperand(Use.Idx, Inst); // RAUW has replaced all original uses with references to the new value, // including the debug uses. Since we are undoing the replacements, // the original debug uses must also be reinstated to maintain the // correctness and utility of debug value instructions. for (auto *DVI : DbgValues) DVI->replaceVariableLocationOp(New, Inst); // Similar story with DbgVariableRecords, the non-instruction // representation of dbg.values. for (DbgVariableRecord *DVR : DbgVariableRecords) DVR->replaceVariableLocationOp(New, Inst); } }; /// Remove an instruction from the IR. class InstructionRemover : public TypePromotionAction { /// Original position of the instruction. InsertionHandler Inserter; /// Helper structure to hide all the link to the instruction. In other /// words, this helps to do as if the instruction was removed. OperandsHider Hider; /// Keep track of the uses replaced, if any. UsesReplacer *Replacer = nullptr; /// Keep track of instructions removed. SetOfInstrs &RemovedInsts; public: /// Remove all reference of \p Inst and optionally replace all its /// uses with New. /// \p RemovedInsts Keep track of the instructions removed by this Action. /// \pre If !Inst->use_empty(), then New != nullptr InstructionRemover(Instruction *Inst, SetOfInstrs &RemovedInsts, Value *New = nullptr) : TypePromotionAction(Inst), Inserter(Inst), Hider(Inst), RemovedInsts(RemovedInsts) { if (New) Replacer = new UsesReplacer(Inst, New); LLVM_DEBUG(dbgs() << "Do: InstructionRemover: " << *Inst << "\n"); RemovedInsts.insert(Inst); /// The instructions removed here will be freed after completing /// optimizeBlock() for all blocks as we need to keep track of the /// removed instructions during promotion. Inst->removeFromParent(); } ~InstructionRemover() override { delete Replacer; } InstructionRemover &operator=(const InstructionRemover &other) = delete; InstructionRemover(const InstructionRemover &other) = delete; /// Resurrect the instruction and reassign it to the proper uses if /// new value was provided when build this action. void undo() override { LLVM_DEBUG(dbgs() << "Undo: InstructionRemover: " << *Inst << "\n"); Inserter.insert(Inst); if (Replacer) Replacer->undo(); Hider.undo(); RemovedInsts.erase(Inst); } }; public: /// Restoration point. /// The restoration point is a pointer to an action instead of an iterator /// because the iterator may be invalidated but not the pointer. using ConstRestorationPt = const TypePromotionAction *; TypePromotionTransaction(SetOfInstrs &RemovedInsts) : RemovedInsts(RemovedInsts) {} /// Advocate every changes made in that transaction. Return true if any change /// happen. bool commit(); /// Undo all the changes made after the given point. void rollback(ConstRestorationPt Point); /// Get the current restoration point. ConstRestorationPt getRestorationPoint() const; /// \name API for IR modification with state keeping to support rollback. /// @{ /// Same as Instruction::setOperand. void setOperand(Instruction *Inst, unsigned Idx, Value *NewVal); /// Same as Instruction::eraseFromParent. void eraseInstruction(Instruction *Inst, Value *NewVal = nullptr); /// Same as Value::replaceAllUsesWith. void replaceAllUsesWith(Instruction *Inst, Value *New); /// Same as Value::mutateType. void mutateType(Instruction *Inst, Type *NewTy); /// Same as IRBuilder::createTrunc. Value *createTrunc(Instruction *Opnd, Type *Ty); /// Same as IRBuilder::createSExt. Value *createSExt(Instruction *Inst, Value *Opnd, Type *Ty); /// Same as IRBuilder::createZExt. Value *createZExt(Instruction *Inst, Value *Opnd, Type *Ty); private: /// The ordered list of actions made so far. SmallVector, 16> Actions; using CommitPt = SmallVectorImpl>::iterator; SetOfInstrs &RemovedInsts; }; } // end anonymous namespace void TypePromotionTransaction::setOperand(Instruction *Inst, unsigned Idx, Value *NewVal) { Actions.push_back(std::make_unique( Inst, Idx, NewVal)); } void TypePromotionTransaction::eraseInstruction(Instruction *Inst, Value *NewVal) { Actions.push_back( std::make_unique( Inst, RemovedInsts, NewVal)); } void TypePromotionTransaction::replaceAllUsesWith(Instruction *Inst, Value *New) { Actions.push_back( std::make_unique(Inst, New)); } void TypePromotionTransaction::mutateType(Instruction *Inst, Type *NewTy) { Actions.push_back( std::make_unique(Inst, NewTy)); } Value *TypePromotionTransaction::createTrunc(Instruction *Opnd, Type *Ty) { std::unique_ptr Ptr(new TruncBuilder(Opnd, Ty)); Value *Val = Ptr->getBuiltValue(); Actions.push_back(std::move(Ptr)); return Val; } Value *TypePromotionTransaction::createSExt(Instruction *Inst, Value *Opnd, Type *Ty) { std::unique_ptr Ptr(new SExtBuilder(Inst, Opnd, Ty)); Value *Val = Ptr->getBuiltValue(); Actions.push_back(std::move(Ptr)); return Val; } Value *TypePromotionTransaction::createZExt(Instruction *Inst, Value *Opnd, Type *Ty) { std::unique_ptr Ptr(new ZExtBuilder(Inst, Opnd, Ty)); Value *Val = Ptr->getBuiltValue(); Actions.push_back(std::move(Ptr)); return Val; } TypePromotionTransaction::ConstRestorationPt TypePromotionTransaction::getRestorationPoint() const { return !Actions.empty() ? Actions.back().get() : nullptr; } bool TypePromotionTransaction::commit() { for (std::unique_ptr &Action : Actions) Action->commit(); bool Modified = !Actions.empty(); Actions.clear(); return Modified; } void TypePromotionTransaction::rollback( TypePromotionTransaction::ConstRestorationPt Point) { while (!Actions.empty() && Point != Actions.back().get()) { std::unique_ptr Curr = Actions.pop_back_val(); Curr->undo(); } } namespace { /// A helper class for matching addressing modes. /// /// This encapsulates the logic for matching the target-legal addressing modes. class AddressingModeMatcher { SmallVectorImpl &AddrModeInsts; const TargetLowering &TLI; const TargetRegisterInfo &TRI; const DataLayout &DL; const LoopInfo &LI; const std::function getDTFn; /// AccessTy/MemoryInst - This is the type for the access (e.g. double) and /// the memory instruction that we're computing this address for. Type *AccessTy; unsigned AddrSpace; Instruction *MemoryInst; /// This is the addressing mode that we're building up. This is /// part of the return value of this addressing mode matching stuff. ExtAddrMode &AddrMode; /// The instructions inserted by other CodeGenPrepare optimizations. const SetOfInstrs &InsertedInsts; /// A map from the instructions to their type before promotion. InstrToOrigTy &PromotedInsts; /// The ongoing transaction where every action should be registered. TypePromotionTransaction &TPT; // A GEP which has too large offset to be folded into the addressing mode. std::pair, int64_t> &LargeOffsetGEP; /// This is set to true when we should not do profitability checks. /// When true, IsProfitableToFoldIntoAddressingMode always returns true. bool IgnoreProfitability; /// True if we are optimizing for size. bool OptSize = false; ProfileSummaryInfo *PSI; BlockFrequencyInfo *BFI; AddressingModeMatcher( SmallVectorImpl &AMI, const TargetLowering &TLI, const TargetRegisterInfo &TRI, const LoopInfo &LI, const std::function getDTFn, Type *AT, unsigned AS, Instruction *MI, ExtAddrMode &AM, const SetOfInstrs &InsertedInsts, InstrToOrigTy &PromotedInsts, TypePromotionTransaction &TPT, std::pair, int64_t> &LargeOffsetGEP, bool OptSize, ProfileSummaryInfo *PSI, BlockFrequencyInfo *BFI) : AddrModeInsts(AMI), TLI(TLI), TRI(TRI), DL(MI->getDataLayout()), LI(LI), getDTFn(getDTFn), AccessTy(AT), AddrSpace(AS), MemoryInst(MI), AddrMode(AM), InsertedInsts(InsertedInsts), PromotedInsts(PromotedInsts), TPT(TPT), LargeOffsetGEP(LargeOffsetGEP), OptSize(OptSize), PSI(PSI), BFI(BFI) { IgnoreProfitability = false; } public: /// Find the maximal addressing mode that a load/store of V can fold, /// give an access type of AccessTy. This returns a list of involved /// instructions in AddrModeInsts. /// \p InsertedInsts The instructions inserted by other CodeGenPrepare /// optimizations. /// \p PromotedInsts maps the instructions to their type before promotion. /// \p The ongoing transaction where every action should be registered. static ExtAddrMode Match(Value *V, Type *AccessTy, unsigned AS, Instruction *MemoryInst, SmallVectorImpl &AddrModeInsts, const TargetLowering &TLI, const LoopInfo &LI, const std::function getDTFn, const TargetRegisterInfo &TRI, const SetOfInstrs &InsertedInsts, InstrToOrigTy &PromotedInsts, TypePromotionTransaction &TPT, std::pair, int64_t> &LargeOffsetGEP, bool OptSize, ProfileSummaryInfo *PSI, BlockFrequencyInfo *BFI) { ExtAddrMode Result; bool Success = AddressingModeMatcher(AddrModeInsts, TLI, TRI, LI, getDTFn, AccessTy, AS, MemoryInst, Result, InsertedInsts, PromotedInsts, TPT, LargeOffsetGEP, OptSize, PSI, BFI) .matchAddr(V, 0); (void)Success; assert(Success && "Couldn't select *anything*?"); return Result; } private: bool matchScaledValue(Value *ScaleReg, int64_t Scale, unsigned Depth); bool matchAddr(Value *Addr, unsigned Depth); bool matchOperationAddr(User *AddrInst, unsigned Opcode, unsigned Depth, bool *MovedAway = nullptr); bool isProfitableToFoldIntoAddressingMode(Instruction *I, ExtAddrMode &AMBefore, ExtAddrMode &AMAfter); bool valueAlreadyLiveAtInst(Value *Val, Value *KnownLive1, Value *KnownLive2); bool isPromotionProfitable(unsigned NewCost, unsigned OldCost, Value *PromotedOperand) const; }; class PhiNodeSet; /// An iterator for PhiNodeSet. class PhiNodeSetIterator { PhiNodeSet *const Set; size_t CurrentIndex = 0; public: /// The constructor. Start should point to either a valid element, or be equal /// to the size of the underlying SmallVector of the PhiNodeSet. PhiNodeSetIterator(PhiNodeSet *const Set, size_t Start); PHINode *operator*() const; PhiNodeSetIterator &operator++(); bool operator==(const PhiNodeSetIterator &RHS) const; bool operator!=(const PhiNodeSetIterator &RHS) const; }; /// Keeps a set of PHINodes. /// /// This is a minimal set implementation for a specific use case: /// It is very fast when there are very few elements, but also provides good /// performance when there are many. It is similar to SmallPtrSet, but also /// provides iteration by insertion order, which is deterministic and stable /// across runs. It is also similar to SmallSetVector, but provides removing /// elements in O(1) time. This is achieved by not actually removing the element /// from the underlying vector, so comes at the cost of using more memory, but /// that is fine, since PhiNodeSets are used as short lived objects. class PhiNodeSet { friend class PhiNodeSetIterator; using MapType = SmallDenseMap; using iterator = PhiNodeSetIterator; /// Keeps the elements in the order of their insertion in the underlying /// vector. To achieve constant time removal, it never deletes any element. SmallVector NodeList; /// Keeps the elements in the underlying set implementation. This (and not the /// NodeList defined above) is the source of truth on whether an element /// is actually in the collection. MapType NodeMap; /// Points to the first valid (not deleted) element when the set is not empty /// and the value is not zero. Equals to the size of the underlying vector /// when the set is empty. When the value is 0, as in the beginning, the /// first element may or may not be valid. size_t FirstValidElement = 0; public: /// Inserts a new element to the collection. /// \returns true if the element is actually added, i.e. was not in the /// collection before the operation. bool insert(PHINode *Ptr) { if (NodeMap.insert(std::make_pair(Ptr, NodeList.size())).second) { NodeList.push_back(Ptr); return true; } return false; } /// Removes the element from the collection. /// \returns whether the element is actually removed, i.e. was in the /// collection before the operation. bool erase(PHINode *Ptr) { if (NodeMap.erase(Ptr)) { SkipRemovedElements(FirstValidElement); return true; } return false; } /// Removes all elements and clears the collection. void clear() { NodeMap.clear(); NodeList.clear(); FirstValidElement = 0; } /// \returns an iterator that will iterate the elements in the order of /// insertion. iterator begin() { if (FirstValidElement == 0) SkipRemovedElements(FirstValidElement); return PhiNodeSetIterator(this, FirstValidElement); } /// \returns an iterator that points to the end of the collection. iterator end() { return PhiNodeSetIterator(this, NodeList.size()); } /// Returns the number of elements in the collection. size_t size() const { return NodeMap.size(); } /// \returns 1 if the given element is in the collection, and 0 if otherwise. size_t count(PHINode *Ptr) const { return NodeMap.count(Ptr); } private: /// Updates the CurrentIndex so that it will point to a valid element. /// /// If the element of NodeList at CurrentIndex is valid, it does not /// change it. If there are no more valid elements, it updates CurrentIndex /// to point to the end of the NodeList. void SkipRemovedElements(size_t &CurrentIndex) { while (CurrentIndex < NodeList.size()) { auto it = NodeMap.find(NodeList[CurrentIndex]); // If the element has been deleted and added again later, NodeMap will // point to a different index, so CurrentIndex will still be invalid. if (it != NodeMap.end() && it->second == CurrentIndex) break; ++CurrentIndex; } } }; PhiNodeSetIterator::PhiNodeSetIterator(PhiNodeSet *const Set, size_t Start) : Set(Set), CurrentIndex(Start) {} PHINode *PhiNodeSetIterator::operator*() const { assert(CurrentIndex < Set->NodeList.size() && "PhiNodeSet access out of range"); return Set->NodeList[CurrentIndex]; } PhiNodeSetIterator &PhiNodeSetIterator::operator++() { assert(CurrentIndex < Set->NodeList.size() && "PhiNodeSet access out of range"); ++CurrentIndex; Set->SkipRemovedElements(CurrentIndex); return *this; } bool PhiNodeSetIterator::operator==(const PhiNodeSetIterator &RHS) const { return CurrentIndex == RHS.CurrentIndex; } bool PhiNodeSetIterator::operator!=(const PhiNodeSetIterator &RHS) const { return !((*this) == RHS); } /// Keep track of simplification of Phi nodes. /// Accept the set of all phi nodes and erase phi node from this set /// if it is simplified. class SimplificationTracker { DenseMap Storage; const SimplifyQuery &SQ; // Tracks newly created Phi nodes. The elements are iterated by insertion // order. PhiNodeSet AllPhiNodes; // Tracks newly created Select nodes. SmallPtrSet AllSelectNodes; public: SimplificationTracker(const SimplifyQuery &sq) : SQ(sq) {} Value *Get(Value *V) { do { auto SV = Storage.find(V); if (SV == Storage.end()) return V; V = SV->second; } while (true); } Value *Simplify(Value *Val) { SmallVector WorkList; SmallPtrSet Visited; WorkList.push_back(Val); while (!WorkList.empty()) { auto *P = WorkList.pop_back_val(); if (!Visited.insert(P).second) continue; if (auto *PI = dyn_cast(P)) if (Value *V = simplifyInstruction(cast(PI), SQ)) { for (auto *U : PI->users()) WorkList.push_back(cast(U)); Put(PI, V); PI->replaceAllUsesWith(V); if (auto *PHI = dyn_cast(PI)) AllPhiNodes.erase(PHI); if (auto *Select = dyn_cast(PI)) AllSelectNodes.erase(Select); PI->eraseFromParent(); } } return Get(Val); } void Put(Value *From, Value *To) { Storage.insert({From, To}); } void ReplacePhi(PHINode *From, PHINode *To) { Value *OldReplacement = Get(From); while (OldReplacement != From) { From = To; To = dyn_cast(OldReplacement); OldReplacement = Get(From); } assert(To && Get(To) == To && "Replacement PHI node is already replaced."); Put(From, To); From->replaceAllUsesWith(To); AllPhiNodes.erase(From); From->eraseFromParent(); } PhiNodeSet &newPhiNodes() { return AllPhiNodes; } void insertNewPhi(PHINode *PN) { AllPhiNodes.insert(PN); } void insertNewSelect(SelectInst *SI) { AllSelectNodes.insert(SI); } unsigned countNewPhiNodes() const { return AllPhiNodes.size(); } unsigned countNewSelectNodes() const { return AllSelectNodes.size(); } void destroyNewNodes(Type *CommonType) { // For safe erasing, replace the uses with dummy value first. auto *Dummy = PoisonValue::get(CommonType); for (auto *I : AllPhiNodes) { I->replaceAllUsesWith(Dummy); I->eraseFromParent(); } AllPhiNodes.clear(); for (auto *I : AllSelectNodes) { I->replaceAllUsesWith(Dummy); I->eraseFromParent(); } AllSelectNodes.clear(); } }; /// A helper class for combining addressing modes. class AddressingModeCombiner { typedef DenseMap FoldAddrToValueMapping; typedef std::pair PHIPair; private: /// The addressing modes we've collected. SmallVector AddrModes; /// The field in which the AddrModes differ, when we have more than one. ExtAddrMode::FieldName DifferentField = ExtAddrMode::NoField; /// Are the AddrModes that we have all just equal to their original values? bool AllAddrModesTrivial = true; /// Common Type for all different fields in addressing modes. Type *CommonType = nullptr; /// SimplifyQuery for simplifyInstruction utility. const SimplifyQuery &SQ; /// Original Address. Value *Original; /// Common value among addresses Value *CommonValue = nullptr; public: AddressingModeCombiner(const SimplifyQuery &_SQ, Value *OriginalValue) : SQ(_SQ), Original(OriginalValue) {} ~AddressingModeCombiner() { eraseCommonValueIfDead(); } /// Get the combined AddrMode const ExtAddrMode &getAddrMode() const { return AddrModes[0]; } /// Add a new AddrMode if it's compatible with the AddrModes we already /// have. /// \return True iff we succeeded in doing so. bool addNewAddrMode(ExtAddrMode &NewAddrMode) { // Take note of if we have any non-trivial AddrModes, as we need to detect // when all AddrModes are trivial as then we would introduce a phi or select // which just duplicates what's already there. AllAddrModesTrivial = AllAddrModesTrivial && NewAddrMode.isTrivial(); // If this is the first addrmode then everything is fine. if (AddrModes.empty()) { AddrModes.emplace_back(NewAddrMode); return true; } // Figure out how different this is from the other address modes, which we // can do just by comparing against the first one given that we only care // about the cumulative difference. ExtAddrMode::FieldName ThisDifferentField = AddrModes[0].compare(NewAddrMode); if (DifferentField == ExtAddrMode::NoField) DifferentField = ThisDifferentField; else if (DifferentField != ThisDifferentField) DifferentField = ExtAddrMode::MultipleFields; // If NewAddrMode differs in more than one dimension we cannot handle it. bool CanHandle = DifferentField != ExtAddrMode::MultipleFields; // If Scale Field is different then we reject. CanHandle = CanHandle && DifferentField != ExtAddrMode::ScaleField; // We also must reject the case when base offset is different and // scale reg is not null, we cannot handle this case due to merge of // different offsets will be used as ScaleReg. CanHandle = CanHandle && (DifferentField != ExtAddrMode::BaseOffsField || !NewAddrMode.ScaledReg); // We also must reject the case when GV is different and BaseReg installed // due to we want to use base reg as a merge of GV values. CanHandle = CanHandle && (DifferentField != ExtAddrMode::BaseGVField || !NewAddrMode.HasBaseReg); // Even if NewAddMode is the same we still need to collect it due to // original value is different. And later we will need all original values // as anchors during finding the common Phi node. if (CanHandle) AddrModes.emplace_back(NewAddrMode); else AddrModes.clear(); return CanHandle; } /// Combine the addressing modes we've collected into a single /// addressing mode. /// \return True iff we successfully combined them or we only had one so /// didn't need to combine them anyway. bool combineAddrModes() { // If we have no AddrModes then they can't be combined. if (AddrModes.size() == 0) return false; // A single AddrMode can trivially be combined. if (AddrModes.size() == 1 || DifferentField == ExtAddrMode::NoField) return true; // If the AddrModes we collected are all just equal to the value they are // derived from then combining them wouldn't do anything useful. if (AllAddrModesTrivial) return false; if (!addrModeCombiningAllowed()) return false; // Build a map between to // value of base register. // Bail out if there is no common type. FoldAddrToValueMapping Map; if (!initializeMap(Map)) return false; CommonValue = findCommon(Map); if (CommonValue) AddrModes[0].SetCombinedField(DifferentField, CommonValue, AddrModes); return CommonValue != nullptr; } private: /// `CommonValue` may be a placeholder inserted by us. /// If the placeholder is not used, we should remove this dead instruction. void eraseCommonValueIfDead() { if (CommonValue && CommonValue->getNumUses() == 0) if (Instruction *CommonInst = dyn_cast(CommonValue)) CommonInst->eraseFromParent(); } /// Initialize Map with anchor values. For address seen /// we set the value of different field saw in this address. /// At the same time we find a common type for different field we will /// use to create new Phi/Select nodes. Keep it in CommonType field. /// Return false if there is no common type found. bool initializeMap(FoldAddrToValueMapping &Map) { // Keep track of keys where the value is null. We will need to replace it // with constant null when we know the common type. SmallVector NullValue; Type *IntPtrTy = SQ.DL.getIntPtrType(AddrModes[0].OriginalValue->getType()); for (auto &AM : AddrModes) { Value *DV = AM.GetFieldAsValue(DifferentField, IntPtrTy); if (DV) { auto *Type = DV->getType(); if (CommonType && CommonType != Type) return false; CommonType = Type; Map[AM.OriginalValue] = DV; } else { NullValue.push_back(AM.OriginalValue); } } assert(CommonType && "At least one non-null value must be!"); for (auto *V : NullValue) Map[V] = Constant::getNullValue(CommonType); return true; } /// We have mapping between value A and other value B where B was a field in /// addressing mode represented by A. Also we have an original value C /// representing an address we start with. Traversing from C through phi and /// selects we ended up with A's in a map. This utility function tries to find /// a value V which is a field in addressing mode C and traversing through phi /// nodes and selects we will end up in corresponded values B in a map. /// The utility will create a new Phi/Selects if needed. // The simple example looks as follows: // BB1: // p1 = b1 + 40 // br cond BB2, BB3 // BB2: // p2 = b2 + 40 // br BB3 // BB3: // p = phi [p1, BB1], [p2, BB2] // v = load p // Map is // p1 -> b1 // p2 -> b2 // Request is // p -> ? // The function tries to find or build phi [b1, BB1], [b2, BB2] in BB3. Value *findCommon(FoldAddrToValueMapping &Map) { // Tracks the simplification of newly created phi nodes. The reason we use // this mapping is because we will add new created Phi nodes in AddrToBase. // Simplification of Phi nodes is recursive, so some Phi node may // be simplified after we added it to AddrToBase. In reality this // simplification is possible only if original phi/selects were not // simplified yet. // Using this mapping we can find the current value in AddrToBase. SimplificationTracker ST(SQ); // First step, DFS to create PHI nodes for all intermediate blocks. // Also fill traverse order for the second step. SmallVector TraverseOrder; InsertPlaceholders(Map, TraverseOrder, ST); // Second Step, fill new nodes by merged values and simplify if possible. FillPlaceholders(Map, TraverseOrder, ST); if (!AddrSinkNewSelects && ST.countNewSelectNodes() > 0) { ST.destroyNewNodes(CommonType); return nullptr; } // Now we'd like to match New Phi nodes to existed ones. unsigned PhiNotMatchedCount = 0; if (!MatchPhiSet(ST, AddrSinkNewPhis, PhiNotMatchedCount)) { ST.destroyNewNodes(CommonType); return nullptr; } auto *Result = ST.Get(Map.find(Original)->second); if (Result) { NumMemoryInstsPhiCreated += ST.countNewPhiNodes() + PhiNotMatchedCount; NumMemoryInstsSelectCreated += ST.countNewSelectNodes(); } return Result; } /// Try to match PHI node to Candidate. /// Matcher tracks the matched Phi nodes. bool MatchPhiNode(PHINode *PHI, PHINode *Candidate, SmallSetVector &Matcher, PhiNodeSet &PhiNodesToMatch) { SmallVector WorkList; Matcher.insert({PHI, Candidate}); SmallSet MatchedPHIs; MatchedPHIs.insert(PHI); WorkList.push_back({PHI, Candidate}); SmallSet Visited; while (!WorkList.empty()) { auto Item = WorkList.pop_back_val(); if (!Visited.insert(Item).second) continue; // We iterate over all incoming values to Phi to compare them. // If values are different and both of them Phi and the first one is a // Phi we added (subject to match) and both of them is in the same basic // block then we can match our pair if values match. So we state that // these values match and add it to work list to verify that. for (auto *B : Item.first->blocks()) { Value *FirstValue = Item.first->getIncomingValueForBlock(B); Value *SecondValue = Item.second->getIncomingValueForBlock(B); if (FirstValue == SecondValue) continue; PHINode *FirstPhi = dyn_cast(FirstValue); PHINode *SecondPhi = dyn_cast(SecondValue); // One of them is not Phi or // The first one is not Phi node from the set we'd like to match or // Phi nodes from different basic blocks then // we will not be able to match. if (!FirstPhi || !SecondPhi || !PhiNodesToMatch.count(FirstPhi) || FirstPhi->getParent() != SecondPhi->getParent()) return false; // If we already matched them then continue. if (Matcher.count({FirstPhi, SecondPhi})) continue; // So the values are different and does not match. So we need them to // match. (But we register no more than one match per PHI node, so that // we won't later try to replace them twice.) if (MatchedPHIs.insert(FirstPhi).second) Matcher.insert({FirstPhi, SecondPhi}); // But me must check it. WorkList.push_back({FirstPhi, SecondPhi}); } } return true; } /// For the given set of PHI nodes (in the SimplificationTracker) try /// to find their equivalents. /// Returns false if this matching fails and creation of new Phi is disabled. bool MatchPhiSet(SimplificationTracker &ST, bool AllowNewPhiNodes, unsigned &PhiNotMatchedCount) { // Matched and PhiNodesToMatch iterate their elements in a deterministic // order, so the replacements (ReplacePhi) are also done in a deterministic // order. SmallSetVector Matched; SmallPtrSet WillNotMatch; PhiNodeSet &PhiNodesToMatch = ST.newPhiNodes(); while (PhiNodesToMatch.size()) { PHINode *PHI = *PhiNodesToMatch.begin(); // Add us, if no Phi nodes in the basic block we do not match. WillNotMatch.clear(); WillNotMatch.insert(PHI); // Traverse all Phis until we found equivalent or fail to do that. bool IsMatched = false; for (auto &P : PHI->getParent()->phis()) { // Skip new Phi nodes. if (PhiNodesToMatch.count(&P)) continue; if ((IsMatched = MatchPhiNode(PHI, &P, Matched, PhiNodesToMatch))) break; // If it does not match, collect all Phi nodes from matcher. // if we end up with no match, them all these Phi nodes will not match // later. for (auto M : Matched) WillNotMatch.insert(M.first); Matched.clear(); } if (IsMatched) { // Replace all matched values and erase them. for (auto MV : Matched) ST.ReplacePhi(MV.first, MV.second); Matched.clear(); continue; } // If we are not allowed to create new nodes then bail out. if (!AllowNewPhiNodes) return false; // Just remove all seen values in matcher. They will not match anything. PhiNotMatchedCount += WillNotMatch.size(); for (auto *P : WillNotMatch) PhiNodesToMatch.erase(P); } return true; } /// Fill the placeholders with values from predecessors and simplify them. void FillPlaceholders(FoldAddrToValueMapping &Map, SmallVectorImpl &TraverseOrder, SimplificationTracker &ST) { while (!TraverseOrder.empty()) { Value *Current = TraverseOrder.pop_back_val(); assert(Map.contains(Current) && "No node to fill!!!"); Value *V = Map[Current]; if (SelectInst *Select = dyn_cast(V)) { // CurrentValue also must be Select. auto *CurrentSelect = cast(Current); auto *TrueValue = CurrentSelect->getTrueValue(); assert(Map.contains(TrueValue) && "No True Value!"); Select->setTrueValue(ST.Get(Map[TrueValue])); auto *FalseValue = CurrentSelect->getFalseValue(); assert(Map.contains(FalseValue) && "No False Value!"); Select->setFalseValue(ST.Get(Map[FalseValue])); } else { // Must be a Phi node then. auto *PHI = cast(V); // Fill the Phi node with values from predecessors. for (auto *B : predecessors(PHI->getParent())) { Value *PV = cast(Current)->getIncomingValueForBlock(B); assert(Map.contains(PV) && "No predecessor Value!"); PHI->addIncoming(ST.Get(Map[PV]), B); } } Map[Current] = ST.Simplify(V); } } /// Starting from original value recursively iterates over def-use chain up to /// known ending values represented in a map. For each traversed phi/select /// inserts a placeholder Phi or Select. /// Reports all new created Phi/Select nodes by adding them to set. /// Also reports and order in what values have been traversed. void InsertPlaceholders(FoldAddrToValueMapping &Map, SmallVectorImpl &TraverseOrder, SimplificationTracker &ST) { SmallVector Worklist; assert((isa(Original) || isa(Original)) && "Address must be a Phi or Select node"); auto *Dummy = PoisonValue::get(CommonType); Worklist.push_back(Original); while (!Worklist.empty()) { Value *Current = Worklist.pop_back_val(); // if it is already visited or it is an ending value then skip it. if (Map.contains(Current)) continue; TraverseOrder.push_back(Current); // CurrentValue must be a Phi node or select. All others must be covered // by anchors. if (SelectInst *CurrentSelect = dyn_cast(Current)) { // Is it OK to get metadata from OrigSelect?! // Create a Select placeholder with dummy value. SelectInst *Select = SelectInst::Create(CurrentSelect->getCondition(), Dummy, Dummy, CurrentSelect->getName(), CurrentSelect->getIterator(), CurrentSelect); Map[Current] = Select; ST.insertNewSelect(Select); // We are interested in True and False values. Worklist.push_back(CurrentSelect->getTrueValue()); Worklist.push_back(CurrentSelect->getFalseValue()); } else { // It must be a Phi node then. PHINode *CurrentPhi = cast(Current); unsigned PredCount = CurrentPhi->getNumIncomingValues(); PHINode *PHI = PHINode::Create(CommonType, PredCount, "sunk_phi", CurrentPhi->getIterator()); Map[Current] = PHI; ST.insertNewPhi(PHI); append_range(Worklist, CurrentPhi->incoming_values()); } } } bool addrModeCombiningAllowed() { if (DisableComplexAddrModes) return false; switch (DifferentField) { default: return false; case ExtAddrMode::BaseRegField: return AddrSinkCombineBaseReg; case ExtAddrMode::BaseGVField: return AddrSinkCombineBaseGV; case ExtAddrMode::BaseOffsField: return AddrSinkCombineBaseOffs; case ExtAddrMode::ScaledRegField: return AddrSinkCombineScaledReg; } } }; } // end anonymous namespace /// Try adding ScaleReg*Scale to the current addressing mode. /// Return true and update AddrMode if this addr mode is legal for the target, /// false if not. bool AddressingModeMatcher::matchScaledValue(Value *ScaleReg, int64_t Scale, unsigned Depth) { // If Scale is 1, then this is the same as adding ScaleReg to the addressing // mode. Just process that directly. if (Scale == 1) return matchAddr(ScaleReg, Depth); // If the scale is 0, it takes nothing to add this. if (Scale == 0) return true; // If we already have a scale of this value, we can add to it, otherwise, we // need an available scale field. if (AddrMode.Scale != 0 && AddrMode.ScaledReg != ScaleReg) return false; ExtAddrMode TestAddrMode = AddrMode; // Add scale to turn X*4+X*3 -> X*7. This could also do things like // [A+B + A*7] -> [B+A*8]. TestAddrMode.Scale += Scale; TestAddrMode.ScaledReg = ScaleReg; // If the new address isn't legal, bail out. if (!TLI.isLegalAddressingMode(DL, TestAddrMode, AccessTy, AddrSpace)) return false; // It was legal, so commit it. AddrMode = TestAddrMode; // Okay, we decided that we can add ScaleReg+Scale to AddrMode. Check now // to see if ScaleReg is actually X+C. If so, we can turn this into adding // X*Scale + C*Scale to addr mode. If we found available IV increment, do not // go any further: we can reuse it and cannot eliminate it. ConstantInt *CI = nullptr; Value *AddLHS = nullptr; if (isa(ScaleReg) && // not a constant expr. match(ScaleReg, m_Add(m_Value(AddLHS), m_ConstantInt(CI))) && !isIVIncrement(ScaleReg, &LI) && CI->getValue().isSignedIntN(64)) { TestAddrMode.InBounds = false; TestAddrMode.ScaledReg = AddLHS; TestAddrMode.BaseOffs += CI->getSExtValue() * TestAddrMode.Scale; // If this addressing mode is legal, commit it and remember that we folded // this instruction. if (TLI.isLegalAddressingMode(DL, TestAddrMode, AccessTy, AddrSpace)) { AddrModeInsts.push_back(cast(ScaleReg)); AddrMode = TestAddrMode; return true; } // Restore status quo. TestAddrMode = AddrMode; } // If this is an add recurrence with a constant step, return the increment // instruction and the canonicalized step. auto GetConstantStep = [this](const Value *V) -> std::optional> { auto *PN = dyn_cast(V); if (!PN) return std::nullopt; auto IVInc = getIVIncrement(PN, &LI); if (!IVInc) return std::nullopt; // TODO: The result of the intrinsics above is two-complement. However when // IV inc is expressed as add or sub, iv.next is potentially a poison value. // If it has nuw or nsw flags, we need to make sure that these flags are // inferrable at the point of memory instruction. Otherwise we are replacing // well-defined two-complement computation with poison. Currently, to avoid // potentially complex analysis needed to prove this, we reject such cases. if (auto *OIVInc = dyn_cast(IVInc->first)) if (OIVInc->hasNoSignedWrap() || OIVInc->hasNoUnsignedWrap()) return std::nullopt; if (auto *ConstantStep = dyn_cast(IVInc->second)) return std::make_pair(IVInc->first, ConstantStep->getValue()); return std::nullopt; }; // Try to account for the following special case: // 1. ScaleReg is an inductive variable; // 2. We use it with non-zero offset; // 3. IV's increment is available at the point of memory instruction. // // In this case, we may reuse the IV increment instead of the IV Phi to // achieve the following advantages: // 1. If IV step matches the offset, we will have no need in the offset; // 2. Even if they don't match, we will reduce the overlap of living IV // and IV increment, that will potentially lead to better register // assignment. if (AddrMode.BaseOffs) { if (auto IVStep = GetConstantStep(ScaleReg)) { Instruction *IVInc = IVStep->first; // The following assert is important to ensure a lack of infinite loops. // This transforms is (intentionally) the inverse of the one just above. // If they don't agree on the definition of an increment, we'd alternate // back and forth indefinitely. assert(isIVIncrement(IVInc, &LI) && "implied by GetConstantStep"); APInt Step = IVStep->second; APInt Offset = Step * AddrMode.Scale; if (Offset.isSignedIntN(64)) { TestAddrMode.InBounds = false; TestAddrMode.ScaledReg = IVInc; TestAddrMode.BaseOffs -= Offset.getLimitedValue(); // If this addressing mode is legal, commit it.. // (Note that we defer the (expensive) domtree base legality check // to the very last possible point.) if (TLI.isLegalAddressingMode(DL, TestAddrMode, AccessTy, AddrSpace) && getDTFn().dominates(IVInc, MemoryInst)) { AddrModeInsts.push_back(cast(IVInc)); AddrMode = TestAddrMode; return true; } // Restore status quo. TestAddrMode = AddrMode; } } } // Otherwise, just return what we have. return true; } /// This is a little filter, which returns true if an addressing computation /// involving I might be folded into a load/store accessing it. /// This doesn't need to be perfect, but needs to accept at least /// the set of instructions that MatchOperationAddr can. static bool MightBeFoldableInst(Instruction *I) { switch (I->getOpcode()) { case Instruction::BitCast: case Instruction::AddrSpaceCast: // Don't touch identity bitcasts. if (I->getType() == I->getOperand(0)->getType()) return false; return I->getType()->isIntOrPtrTy(); case Instruction::PtrToInt: // PtrToInt is always a noop, as we know that the int type is pointer sized. return true; case Instruction::IntToPtr: // We know the input is intptr_t, so this is foldable. return true; case Instruction::Add: return true; case Instruction::Mul: case Instruction::Shl: // Can only handle X*C and X << C. return isa(I->getOperand(1)); case Instruction::GetElementPtr: return true; default: return false; } } /// Check whether or not \p Val is a legal instruction for \p TLI. /// \note \p Val is assumed to be the product of some type promotion. /// Therefore if \p Val has an undefined state in \p TLI, this is assumed /// to be legal, as the non-promoted value would have had the same state. static bool isPromotedInstructionLegal(const TargetLowering &TLI, const DataLayout &DL, Value *Val) { Instruction *PromotedInst = dyn_cast(Val); if (!PromotedInst) return false; int ISDOpcode = TLI.InstructionOpcodeToISD(PromotedInst->getOpcode()); // If the ISDOpcode is undefined, it was undefined before the promotion. if (!ISDOpcode) return true; // Otherwise, check if the promoted instruction is legal or not. return TLI.isOperationLegalOrCustom( ISDOpcode, TLI.getValueType(DL, PromotedInst->getType())); } namespace { /// Hepler class to perform type promotion. class TypePromotionHelper { /// Utility function to add a promoted instruction \p ExtOpnd to /// \p PromotedInsts and record the type of extension we have seen. static void addPromotedInst(InstrToOrigTy &PromotedInsts, Instruction *ExtOpnd, bool IsSExt) { ExtType ExtTy = IsSExt ? SignExtension : ZeroExtension; InstrToOrigTy::iterator It = PromotedInsts.find(ExtOpnd); if (It != PromotedInsts.end()) { // If the new extension is same as original, the information in // PromotedInsts[ExtOpnd] is still correct. if (It->second.getInt() == ExtTy) return; // Now the new extension is different from old extension, we make // the type information invalid by setting extension type to // BothExtension. ExtTy = BothExtension; } PromotedInsts[ExtOpnd] = TypeIsSExt(ExtOpnd->getType(), ExtTy); } /// Utility function to query the original type of instruction \p Opnd /// with a matched extension type. If the extension doesn't match, we /// cannot use the information we had on the original type. /// BothExtension doesn't match any extension type. static const Type *getOrigType(const InstrToOrigTy &PromotedInsts, Instruction *Opnd, bool IsSExt) { ExtType ExtTy = IsSExt ? SignExtension : ZeroExtension; InstrToOrigTy::const_iterator It = PromotedInsts.find(Opnd); if (It != PromotedInsts.end() && It->second.getInt() == ExtTy) return It->second.getPointer(); return nullptr; } /// Utility function to check whether or not a sign or zero extension /// of \p Inst with \p ConsideredExtType can be moved through \p Inst by /// either using the operands of \p Inst or promoting \p Inst. /// The type of the extension is defined by \p IsSExt. /// In other words, check if: /// ext (Ty Inst opnd1 opnd2 ... opndN) to ConsideredExtType. /// #1 Promotion applies: /// ConsideredExtType Inst (ext opnd1 to ConsideredExtType, ...). /// #2 Operand reuses: /// ext opnd1 to ConsideredExtType. /// \p PromotedInsts maps the instructions to their type before promotion. static bool canGetThrough(const Instruction *Inst, Type *ConsideredExtType, const InstrToOrigTy &PromotedInsts, bool IsSExt); /// Utility function to determine if \p OpIdx should be promoted when /// promoting \p Inst. static bool shouldExtOperand(const Instruction *Inst, int OpIdx) { return !(isa(Inst) && OpIdx == 0); } /// Utility function to promote the operand of \p Ext when this /// operand is a promotable trunc or sext or zext. /// \p PromotedInsts maps the instructions to their type before promotion. /// \p CreatedInstsCost[out] contains the cost of all instructions /// created to promote the operand of Ext. /// Newly added extensions are inserted in \p Exts. /// Newly added truncates are inserted in \p Truncs. /// Should never be called directly. /// \return The promoted value which is used instead of Ext. static Value *promoteOperandForTruncAndAnyExt( Instruction *Ext, TypePromotionTransaction &TPT, InstrToOrigTy &PromotedInsts, unsigned &CreatedInstsCost, SmallVectorImpl *Exts, SmallVectorImpl *Truncs, const TargetLowering &TLI); /// Utility function to promote the operand of \p Ext when this /// operand is promotable and is not a supported trunc or sext. /// \p PromotedInsts maps the instructions to their type before promotion. /// \p CreatedInstsCost[out] contains the cost of all the instructions /// created to promote the operand of Ext. /// Newly added extensions are inserted in \p Exts. /// Newly added truncates are inserted in \p Truncs. /// Should never be called directly. /// \return The promoted value which is used instead of Ext. static Value *promoteOperandForOther(Instruction *Ext, TypePromotionTransaction &TPT, InstrToOrigTy &PromotedInsts, unsigned &CreatedInstsCost, SmallVectorImpl *Exts, SmallVectorImpl *Truncs, const TargetLowering &TLI, bool IsSExt); /// \see promoteOperandForOther. static Value *signExtendOperandForOther( Instruction *Ext, TypePromotionTransaction &TPT, InstrToOrigTy &PromotedInsts, unsigned &CreatedInstsCost, SmallVectorImpl *Exts, SmallVectorImpl *Truncs, const TargetLowering &TLI) { return promoteOperandForOther(Ext, TPT, PromotedInsts, CreatedInstsCost, Exts, Truncs, TLI, true); } /// \see promoteOperandForOther. static Value *zeroExtendOperandForOther( Instruction *Ext, TypePromotionTransaction &TPT, InstrToOrigTy &PromotedInsts, unsigned &CreatedInstsCost, SmallVectorImpl *Exts, SmallVectorImpl *Truncs, const TargetLowering &TLI) { return promoteOperandForOther(Ext, TPT, PromotedInsts, CreatedInstsCost, Exts, Truncs, TLI, false); } public: /// Type for the utility function that promotes the operand of Ext. using Action = Value *(*)(Instruction *Ext, TypePromotionTransaction &TPT, InstrToOrigTy &PromotedInsts, unsigned &CreatedInstsCost, SmallVectorImpl *Exts, SmallVectorImpl *Truncs, const TargetLowering &TLI); /// Given a sign/zero extend instruction \p Ext, return the appropriate /// action to promote the operand of \p Ext instead of using Ext. /// \return NULL if no promotable action is possible with the current /// sign extension. /// \p InsertedInsts keeps track of all the instructions inserted by the /// other CodeGenPrepare optimizations. This information is important /// because we do not want to promote these instructions as CodeGenPrepare /// will reinsert them later. Thus creating an infinite loop: create/remove. /// \p PromotedInsts maps the instructions to their type before promotion. static Action getAction(Instruction *Ext, const SetOfInstrs &InsertedInsts, const TargetLowering &TLI, const InstrToOrigTy &PromotedInsts); }; } // end anonymous namespace bool TypePromotionHelper::canGetThrough(const Instruction *Inst, Type *ConsideredExtType, const InstrToOrigTy &PromotedInsts, bool IsSExt) { // The promotion helper does not know how to deal with vector types yet. // To be able to fix that, we would need to fix the places where we // statically extend, e.g., constants and such. if (Inst->getType()->isVectorTy()) return false; // We can always get through zext. if (isa(Inst)) return true; // sext(sext) is ok too. if (IsSExt && isa(Inst)) return true; // We can get through binary operator, if it is legal. In other words, the // binary operator must have a nuw or nsw flag. if (const auto *BinOp = dyn_cast(Inst)) if (isa(BinOp) && ((!IsSExt && BinOp->hasNoUnsignedWrap()) || (IsSExt && BinOp->hasNoSignedWrap()))) return true; // ext(and(opnd, cst)) --> and(ext(opnd), ext(cst)) if ((Inst->getOpcode() == Instruction::And || Inst->getOpcode() == Instruction::Or)) return true; // ext(xor(opnd, cst)) --> xor(ext(opnd), ext(cst)) if (Inst->getOpcode() == Instruction::Xor) { // Make sure it is not a NOT. if (const auto *Cst = dyn_cast(Inst->getOperand(1))) if (!Cst->getValue().isAllOnes()) return true; } // zext(shrl(opnd, cst)) --> shrl(zext(opnd), zext(cst)) // It may change a poisoned value into a regular value, like // zext i32 (shrl i8 %val, 12) --> shrl i32 (zext i8 %val), 12 // poisoned value regular value // It should be OK since undef covers valid value. if (Inst->getOpcode() == Instruction::LShr && !IsSExt) return true; // and(ext(shl(opnd, cst)), cst) --> and(shl(ext(opnd), ext(cst)), cst) // It may change a poisoned value into a regular value, like // zext i32 (shl i8 %val, 12) --> shl i32 (zext i8 %val), 12 // poisoned value regular value // It should be OK since undef covers valid value. if (Inst->getOpcode() == Instruction::Shl && Inst->hasOneUse()) { const auto *ExtInst = cast(*Inst->user_begin()); if (ExtInst->hasOneUse()) { const auto *AndInst = dyn_cast(*ExtInst->user_begin()); if (AndInst && AndInst->getOpcode() == Instruction::And) { const auto *Cst = dyn_cast(AndInst->getOperand(1)); if (Cst && Cst->getValue().isIntN(Inst->getType()->getIntegerBitWidth())) return true; } } } // Check if we can do the following simplification. // ext(trunc(opnd)) --> ext(opnd) if (!isa(Inst)) return false; Value *OpndVal = Inst->getOperand(0); // Check if we can use this operand in the extension. // If the type is larger than the result type of the extension, we cannot. if (!OpndVal->getType()->isIntegerTy() || OpndVal->getType()->getIntegerBitWidth() > ConsideredExtType->getIntegerBitWidth()) return false; // If the operand of the truncate is not an instruction, we will not have // any information on the dropped bits. // (Actually we could for constant but it is not worth the extra logic). Instruction *Opnd = dyn_cast(OpndVal); if (!Opnd) return false; // Check if the source of the type is narrow enough. // I.e., check that trunc just drops extended bits of the same kind of // the extension. // #1 get the type of the operand and check the kind of the extended bits. const Type *OpndType = getOrigType(PromotedInsts, Opnd, IsSExt); if (OpndType) ; else if ((IsSExt && isa(Opnd)) || (!IsSExt && isa(Opnd))) OpndType = Opnd->getOperand(0)->getType(); else return false; // #2 check that the truncate just drops extended bits. return Inst->getType()->getIntegerBitWidth() >= OpndType->getIntegerBitWidth(); } TypePromotionHelper::Action TypePromotionHelper::getAction( Instruction *Ext, const SetOfInstrs &InsertedInsts, const TargetLowering &TLI, const InstrToOrigTy &PromotedInsts) { assert((isa(Ext) || isa(Ext)) && "Unexpected instruction type"); Instruction *ExtOpnd = dyn_cast(Ext->getOperand(0)); Type *ExtTy = Ext->getType(); bool IsSExt = isa(Ext); // If the operand of the extension is not an instruction, we cannot // get through. // If it, check we can get through. if (!ExtOpnd || !canGetThrough(ExtOpnd, ExtTy, PromotedInsts, IsSExt)) return nullptr; // Do not promote if the operand has been added by codegenprepare. // Otherwise, it means we are undoing an optimization that is likely to be // redone, thus causing potential infinite loop. if (isa(ExtOpnd) && InsertedInsts.count(ExtOpnd)) return nullptr; // SExt or Trunc instructions. // Return the related handler. if (isa(ExtOpnd) || isa(ExtOpnd) || isa(ExtOpnd)) return promoteOperandForTruncAndAnyExt; // Regular instruction. // Abort early if we will have to insert non-free instructions. if (!ExtOpnd->hasOneUse() && !TLI.isTruncateFree(ExtTy, ExtOpnd->getType())) return nullptr; return IsSExt ? signExtendOperandForOther : zeroExtendOperandForOther; } Value *TypePromotionHelper::promoteOperandForTruncAndAnyExt( Instruction *SExt, TypePromotionTransaction &TPT, InstrToOrigTy &PromotedInsts, unsigned &CreatedInstsCost, SmallVectorImpl *Exts, SmallVectorImpl *Truncs, const TargetLowering &TLI) { // By construction, the operand of SExt is an instruction. Otherwise we cannot // get through it and this method should not be called. Instruction *SExtOpnd = cast(SExt->getOperand(0)); Value *ExtVal = SExt; bool HasMergedNonFreeExt = false; if (isa(SExtOpnd)) { // Replace s|zext(zext(opnd)) // => zext(opnd). HasMergedNonFreeExt = !TLI.isExtFree(SExtOpnd); Value *ZExt = TPT.createZExt(SExt, SExtOpnd->getOperand(0), SExt->getType()); TPT.replaceAllUsesWith(SExt, ZExt); TPT.eraseInstruction(SExt); ExtVal = ZExt; } else { // Replace z|sext(trunc(opnd)) or sext(sext(opnd)) // => z|sext(opnd). TPT.setOperand(SExt, 0, SExtOpnd->getOperand(0)); } CreatedInstsCost = 0; // Remove dead code. if (SExtOpnd->use_empty()) TPT.eraseInstruction(SExtOpnd); // Check if the extension is still needed. Instruction *ExtInst = dyn_cast(ExtVal); if (!ExtInst || ExtInst->getType() != ExtInst->getOperand(0)->getType()) { if (ExtInst) { if (Exts) Exts->push_back(ExtInst); CreatedInstsCost = !TLI.isExtFree(ExtInst) && !HasMergedNonFreeExt; } return ExtVal; } // At this point we have: ext ty opnd to ty. // Reassign the uses of ExtInst to the opnd and remove ExtInst. Value *NextVal = ExtInst->getOperand(0); TPT.eraseInstruction(ExtInst, NextVal); return NextVal; } Value *TypePromotionHelper::promoteOperandForOther( Instruction *Ext, TypePromotionTransaction &TPT, InstrToOrigTy &PromotedInsts, unsigned &CreatedInstsCost, SmallVectorImpl *Exts, SmallVectorImpl *Truncs, const TargetLowering &TLI, bool IsSExt) { // By construction, the operand of Ext is an instruction. Otherwise we cannot // get through it and this method should not be called. Instruction *ExtOpnd = cast(Ext->getOperand(0)); CreatedInstsCost = 0; if (!ExtOpnd->hasOneUse()) { // ExtOpnd will be promoted. // All its uses, but Ext, will need to use a truncated value of the // promoted version. // Create the truncate now. Value *Trunc = TPT.createTrunc(Ext, ExtOpnd->getType()); if (Instruction *ITrunc = dyn_cast(Trunc)) { // Insert it just after the definition. ITrunc->moveAfter(ExtOpnd); if (Truncs) Truncs->push_back(ITrunc); } TPT.replaceAllUsesWith(ExtOpnd, Trunc); // Restore the operand of Ext (which has been replaced by the previous call // to replaceAllUsesWith) to avoid creating a cycle trunc <-> sext. TPT.setOperand(Ext, 0, ExtOpnd); } // Get through the Instruction: // 1. Update its type. // 2. Replace the uses of Ext by Inst. // 3. Extend each operand that needs to be extended. // Remember the original type of the instruction before promotion. // This is useful to know that the high bits are sign extended bits. addPromotedInst(PromotedInsts, ExtOpnd, IsSExt); // Step #1. TPT.mutateType(ExtOpnd, Ext->getType()); // Step #2. TPT.replaceAllUsesWith(Ext, ExtOpnd); // Step #3. LLVM_DEBUG(dbgs() << "Propagate Ext to operands\n"); for (int OpIdx = 0, EndOpIdx = ExtOpnd->getNumOperands(); OpIdx != EndOpIdx; ++OpIdx) { LLVM_DEBUG(dbgs() << "Operand:\n" << *(ExtOpnd->getOperand(OpIdx)) << '\n'); if (ExtOpnd->getOperand(OpIdx)->getType() == Ext->getType() || !shouldExtOperand(ExtOpnd, OpIdx)) { LLVM_DEBUG(dbgs() << "No need to propagate\n"); continue; } // Check if we can statically extend the operand. Value *Opnd = ExtOpnd->getOperand(OpIdx); if (const ConstantInt *Cst = dyn_cast(Opnd)) { LLVM_DEBUG(dbgs() << "Statically extend\n"); unsigned BitWidth = Ext->getType()->getIntegerBitWidth(); APInt CstVal = IsSExt ? Cst->getValue().sext(BitWidth) : Cst->getValue().zext(BitWidth); TPT.setOperand(ExtOpnd, OpIdx, ConstantInt::get(Ext->getType(), CstVal)); continue; } // UndefValue are typed, so we have to statically sign extend them. if (isa(Opnd)) { LLVM_DEBUG(dbgs() << "Statically extend\n"); TPT.setOperand(ExtOpnd, OpIdx, UndefValue::get(Ext->getType())); continue; } // Otherwise we have to explicitly sign extend the operand. Value *ValForExtOpnd = IsSExt ? TPT.createSExt(ExtOpnd, Opnd, Ext->getType()) : TPT.createZExt(ExtOpnd, Opnd, Ext->getType()); TPT.setOperand(ExtOpnd, OpIdx, ValForExtOpnd); Instruction *InstForExtOpnd = dyn_cast(ValForExtOpnd); if (!InstForExtOpnd) continue; if (Exts) Exts->push_back(InstForExtOpnd); CreatedInstsCost += !TLI.isExtFree(InstForExtOpnd); } LLVM_DEBUG(dbgs() << "Extension is useless now\n"); TPT.eraseInstruction(Ext); return ExtOpnd; } /// Check whether or not promoting an instruction to a wider type is profitable. /// \p NewCost gives the cost of extension instructions created by the /// promotion. /// \p OldCost gives the cost of extension instructions before the promotion /// plus the number of instructions that have been /// matched in the addressing mode the promotion. /// \p PromotedOperand is the value that has been promoted. /// \return True if the promotion is profitable, false otherwise. bool AddressingModeMatcher::isPromotionProfitable( unsigned NewCost, unsigned OldCost, Value *PromotedOperand) const { LLVM_DEBUG(dbgs() << "OldCost: " << OldCost << "\tNewCost: " << NewCost << '\n'); // The cost of the new extensions is greater than the cost of the // old extension plus what we folded. // This is not profitable. if (NewCost > OldCost) return false; if (NewCost < OldCost) return true; // The promotion is neutral but it may help folding the sign extension in // loads for instance. // Check that we did not create an illegal instruction. return isPromotedInstructionLegal(TLI, DL, PromotedOperand); } /// Given an instruction or constant expr, see if we can fold the operation /// into the addressing mode. If so, update the addressing mode and return /// true, otherwise return false without modifying AddrMode. /// If \p MovedAway is not NULL, it contains the information of whether or /// not AddrInst has to be folded into the addressing mode on success. /// If \p MovedAway == true, \p AddrInst will not be part of the addressing /// because it has been moved away. /// Thus AddrInst must not be added in the matched instructions. /// This state can happen when AddrInst is a sext, since it may be moved away. /// Therefore, AddrInst may not be valid when MovedAway is true and it must /// not be referenced anymore. bool AddressingModeMatcher::matchOperationAddr(User *AddrInst, unsigned Opcode, unsigned Depth, bool *MovedAway) { // Avoid exponential behavior on extremely deep expression trees. if (Depth >= 5) return false; // By default, all matched instructions stay in place. if (MovedAway) *MovedAway = false; switch (Opcode) { case Instruction::PtrToInt: // PtrToInt is always a noop, as we know that the int type is pointer sized. return matchAddr(AddrInst->getOperand(0), Depth); case Instruction::IntToPtr: { auto AS = AddrInst->getType()->getPointerAddressSpace(); auto PtrTy = MVT::getIntegerVT(DL.getPointerSizeInBits(AS)); // This inttoptr is a no-op if the integer type is pointer sized. if (TLI.getValueType(DL, AddrInst->getOperand(0)->getType()) == PtrTy) return matchAddr(AddrInst->getOperand(0), Depth); return false; } case Instruction::BitCast: // BitCast is always a noop, and we can handle it as long as it is // int->int or pointer->pointer (we don't want int<->fp or something). if (AddrInst->getOperand(0)->getType()->isIntOrPtrTy() && // Don't touch identity bitcasts. These were probably put here by LSR, // and we don't want to mess around with them. Assume it knows what it // is doing. AddrInst->getOperand(0)->getType() != AddrInst->getType()) return matchAddr(AddrInst->getOperand(0), Depth); return false; case Instruction::AddrSpaceCast: { unsigned SrcAS = AddrInst->getOperand(0)->getType()->getPointerAddressSpace(); unsigned DestAS = AddrInst->getType()->getPointerAddressSpace(); if (TLI.getTargetMachine().isNoopAddrSpaceCast(SrcAS, DestAS)) return matchAddr(AddrInst->getOperand(0), Depth); return false; } case Instruction::Add: { // Check to see if we can merge in one operand, then the other. If so, we // win. ExtAddrMode BackupAddrMode = AddrMode; unsigned OldSize = AddrModeInsts.size(); // Start a transaction at this point. // The LHS may match but not the RHS. // Therefore, we need a higher level restoration point to undo partially // matched operation. TypePromotionTransaction::ConstRestorationPt LastKnownGood = TPT.getRestorationPoint(); // Try to match an integer constant second to increase its chance of ending // up in `BaseOffs`, resp. decrease its chance of ending up in `BaseReg`. int First = 0, Second = 1; if (isa(AddrInst->getOperand(First)) && !isa(AddrInst->getOperand(Second))) std::swap(First, Second); AddrMode.InBounds = false; if (matchAddr(AddrInst->getOperand(First), Depth + 1) && matchAddr(AddrInst->getOperand(Second), Depth + 1)) return true; // Restore the old addr mode info. AddrMode = BackupAddrMode; AddrModeInsts.resize(OldSize); TPT.rollback(LastKnownGood); // Otherwise this was over-aggressive. Try merging operands in the opposite // order. if (matchAddr(AddrInst->getOperand(Second), Depth + 1) && matchAddr(AddrInst->getOperand(First), Depth + 1)) return true; // Otherwise we definitely can't merge the ADD in. AddrMode = BackupAddrMode; AddrModeInsts.resize(OldSize); TPT.rollback(LastKnownGood); break; } // case Instruction::Or: // TODO: We can handle "Or Val, Imm" iff this OR is equivalent to an ADD. // break; case Instruction::Mul: case Instruction::Shl: { // Can only handle X*C and X << C. AddrMode.InBounds = false; ConstantInt *RHS = dyn_cast(AddrInst->getOperand(1)); if (!RHS || RHS->getBitWidth() > 64) return false; int64_t Scale = Opcode == Instruction::Shl ? 1LL << RHS->getLimitedValue(RHS->getBitWidth() - 1) : RHS->getSExtValue(); return matchScaledValue(AddrInst->getOperand(0), Scale, Depth); } case Instruction::GetElementPtr: { // Scan the GEP. We check it if it contains constant offsets and at most // one variable offset. int VariableOperand = -1; unsigned VariableScale = 0; int64_t ConstantOffset = 0; gep_type_iterator GTI = gep_type_begin(AddrInst); for (unsigned i = 1, e = AddrInst->getNumOperands(); i != e; ++i, ++GTI) { if (StructType *STy = GTI.getStructTypeOrNull()) { const StructLayout *SL = DL.getStructLayout(STy); unsigned Idx = cast(AddrInst->getOperand(i))->getZExtValue(); ConstantOffset += SL->getElementOffset(Idx); } else { TypeSize TS = GTI.getSequentialElementStride(DL); if (TS.isNonZero()) { // The optimisations below currently only work for fixed offsets. if (TS.isScalable()) return false; int64_t TypeSize = TS.getFixedValue(); if (ConstantInt *CI = dyn_cast(AddrInst->getOperand(i))) { const APInt &CVal = CI->getValue(); if (CVal.getSignificantBits() <= 64) { ConstantOffset += CVal.getSExtValue() * TypeSize; continue; } } // We only allow one variable index at the moment. if (VariableOperand != -1) return false; // Remember the variable index. VariableOperand = i; VariableScale = TypeSize; } } } // A common case is for the GEP to only do a constant offset. In this case, // just add it to the disp field and check validity. if (VariableOperand == -1) { AddrMode.BaseOffs += ConstantOffset; if (matchAddr(AddrInst->getOperand(0), Depth + 1)) { if (!cast(AddrInst)->isInBounds()) AddrMode.InBounds = false; return true; } AddrMode.BaseOffs -= ConstantOffset; if (EnableGEPOffsetSplit && isa(AddrInst) && TLI.shouldConsiderGEPOffsetSplit() && Depth == 0 && ConstantOffset > 0) { // Record GEPs with non-zero offsets as candidates for splitting in // the event that the offset cannot fit into the r+i addressing mode. // Simple and common case that only one GEP is used in calculating the // address for the memory access. Value *Base = AddrInst->getOperand(0); auto *BaseI = dyn_cast(Base); auto *GEP = cast(AddrInst); if (isa(Base) || isa(Base) || (BaseI && !isa(BaseI) && !isa(BaseI))) { // Make sure the parent block allows inserting non-PHI instructions // before the terminator. BasicBlock *Parent = BaseI ? BaseI->getParent() : &GEP->getFunction()->getEntryBlock(); if (!Parent->getTerminator()->isEHPad()) LargeOffsetGEP = std::make_pair(GEP, ConstantOffset); } } return false; } // Save the valid addressing mode in case we can't match. ExtAddrMode BackupAddrMode = AddrMode; unsigned OldSize = AddrModeInsts.size(); // See if the scale and offset amount is valid for this target. AddrMode.BaseOffs += ConstantOffset; if (!cast(AddrInst)->isInBounds()) AddrMode.InBounds = false; // Match the base operand of the GEP. if (!matchAddr(AddrInst->getOperand(0), Depth + 1)) { // If it couldn't be matched, just stuff the value in a register. if (AddrMode.HasBaseReg) { AddrMode = BackupAddrMode; AddrModeInsts.resize(OldSize); return false; } AddrMode.HasBaseReg = true; AddrMode.BaseReg = AddrInst->getOperand(0); } // Match the remaining variable portion of the GEP. if (!matchScaledValue(AddrInst->getOperand(VariableOperand), VariableScale, Depth)) { // If it couldn't be matched, try stuffing the base into a register // instead of matching it, and retrying the match of the scale. AddrMode = BackupAddrMode; AddrModeInsts.resize(OldSize); if (AddrMode.HasBaseReg) return false; AddrMode.HasBaseReg = true; AddrMode.BaseReg = AddrInst->getOperand(0); AddrMode.BaseOffs += ConstantOffset; if (!matchScaledValue(AddrInst->getOperand(VariableOperand), VariableScale, Depth)) { // If even that didn't work, bail. AddrMode = BackupAddrMode; AddrModeInsts.resize(OldSize); return false; } } return true; } case Instruction::SExt: case Instruction::ZExt: { Instruction *Ext = dyn_cast(AddrInst); if (!Ext) return false; // Try to move this ext out of the way of the addressing mode. // Ask for a method for doing so. TypePromotionHelper::Action TPH = TypePromotionHelper::getAction(Ext, InsertedInsts, TLI, PromotedInsts); if (!TPH) return false; TypePromotionTransaction::ConstRestorationPt LastKnownGood = TPT.getRestorationPoint(); unsigned CreatedInstsCost = 0; unsigned ExtCost = !TLI.isExtFree(Ext); Value *PromotedOperand = TPH(Ext, TPT, PromotedInsts, CreatedInstsCost, nullptr, nullptr, TLI); // SExt has been moved away. // Thus either it will be rematched later in the recursive calls or it is // gone. Anyway, we must not fold it into the addressing mode at this point. // E.g., // op = add opnd, 1 // idx = ext op // addr = gep base, idx // is now: // promotedOpnd = ext opnd <- no match here // op = promoted_add promotedOpnd, 1 <- match (later in recursive calls) // addr = gep base, op <- match if (MovedAway) *MovedAway = true; assert(PromotedOperand && "TypePromotionHelper should have filtered out those cases"); ExtAddrMode BackupAddrMode = AddrMode; unsigned OldSize = AddrModeInsts.size(); if (!matchAddr(PromotedOperand, Depth) || // The total of the new cost is equal to the cost of the created // instructions. // The total of the old cost is equal to the cost of the extension plus // what we have saved in the addressing mode. !isPromotionProfitable(CreatedInstsCost, ExtCost + (AddrModeInsts.size() - OldSize), PromotedOperand)) { AddrMode = BackupAddrMode; AddrModeInsts.resize(OldSize); LLVM_DEBUG(dbgs() << "Sign extension does not pay off: rollback\n"); TPT.rollback(LastKnownGood); return false; } return true; } case Instruction::Call: if (IntrinsicInst *II = dyn_cast(AddrInst)) { if (II->getIntrinsicID() == Intrinsic::threadlocal_address) { GlobalValue &GV = cast(*II->getArgOperand(0)); if (TLI.addressingModeSupportsTLS(GV)) return matchAddr(AddrInst->getOperand(0), Depth); } } break; } return false; } /// If we can, try to add the value of 'Addr' into the current addressing mode. /// If Addr can't be added to AddrMode this returns false and leaves AddrMode /// unmodified. This assumes that Addr is either a pointer type or intptr_t /// for the target. /// bool AddressingModeMatcher::matchAddr(Value *Addr, unsigned Depth) { // Start a transaction at this point that we will rollback if the matching // fails. TypePromotionTransaction::ConstRestorationPt LastKnownGood = TPT.getRestorationPoint(); if (ConstantInt *CI = dyn_cast(Addr)) { if (CI->getValue().isSignedIntN(64)) { // Fold in immediates if legal for the target. AddrMode.BaseOffs += CI->getSExtValue(); if (TLI.isLegalAddressingMode(DL, AddrMode, AccessTy, AddrSpace)) return true; AddrMode.BaseOffs -= CI->getSExtValue(); } } else if (GlobalValue *GV = dyn_cast(Addr)) { // If this is a global variable, try to fold it into the addressing mode. if (!AddrMode.BaseGV) { AddrMode.BaseGV = GV; if (TLI.isLegalAddressingMode(DL, AddrMode, AccessTy, AddrSpace)) return true; AddrMode.BaseGV = nullptr; } } else if (Instruction *I = dyn_cast(Addr)) { ExtAddrMode BackupAddrMode = AddrMode; unsigned OldSize = AddrModeInsts.size(); // Check to see if it is possible to fold this operation. bool MovedAway = false; if (matchOperationAddr(I, I->getOpcode(), Depth, &MovedAway)) { // This instruction may have been moved away. If so, there is nothing // to check here. if (MovedAway) return true; // Okay, it's possible to fold this. Check to see if it is actually // *profitable* to do so. We use a simple cost model to avoid increasing // register pressure too much. if (I->hasOneUse() || isProfitableToFoldIntoAddressingMode(I, BackupAddrMode, AddrMode)) { AddrModeInsts.push_back(I); return true; } // It isn't profitable to do this, roll back. AddrMode = BackupAddrMode; AddrModeInsts.resize(OldSize); TPT.rollback(LastKnownGood); } } else if (ConstantExpr *CE = dyn_cast(Addr)) { if (matchOperationAddr(CE, CE->getOpcode(), Depth)) return true; TPT.rollback(LastKnownGood); } else if (isa(Addr)) { // Null pointer gets folded without affecting the addressing mode. return true; } // Worse case, the target should support [reg] addressing modes. :) if (!AddrMode.HasBaseReg) { AddrMode.HasBaseReg = true; AddrMode.BaseReg = Addr; // Still check for legality in case the target supports [imm] but not [i+r]. if (TLI.isLegalAddressingMode(DL, AddrMode, AccessTy, AddrSpace)) return true; AddrMode.HasBaseReg = false; AddrMode.BaseReg = nullptr; } // If the base register is already taken, see if we can do [r+r]. if (AddrMode.Scale == 0) { AddrMode.Scale = 1; AddrMode.ScaledReg = Addr; if (TLI.isLegalAddressingMode(DL, AddrMode, AccessTy, AddrSpace)) return true; AddrMode.Scale = 0; AddrMode.ScaledReg = nullptr; } // Couldn't match. TPT.rollback(LastKnownGood); return false; } /// Check to see if all uses of OpVal by the specified inline asm call are due /// to memory operands. If so, return true, otherwise return false. static bool IsOperandAMemoryOperand(CallInst *CI, InlineAsm *IA, Value *OpVal, const TargetLowering &TLI, const TargetRegisterInfo &TRI) { const Function *F = CI->getFunction(); TargetLowering::AsmOperandInfoVector TargetConstraints = TLI.ParseConstraints(F->getDataLayout(), &TRI, *CI); for (TargetLowering::AsmOperandInfo &OpInfo : TargetConstraints) { // Compute the constraint code and ConstraintType to use. TLI.ComputeConstraintToUse(OpInfo, SDValue()); // If this asm operand is our Value*, and if it isn't an indirect memory // operand, we can't fold it! TODO: Also handle C_Address? if (OpInfo.CallOperandVal == OpVal && (OpInfo.ConstraintType != TargetLowering::C_Memory || !OpInfo.isIndirect)) return false; } return true; } /// Recursively walk all the uses of I until we find a memory use. /// If we find an obviously non-foldable instruction, return true. /// Add accessed addresses and types to MemoryUses. static bool FindAllMemoryUses( Instruction *I, SmallVectorImpl> &MemoryUses, SmallPtrSetImpl &ConsideredInsts, const TargetLowering &TLI, const TargetRegisterInfo &TRI, bool OptSize, ProfileSummaryInfo *PSI, BlockFrequencyInfo *BFI, unsigned &SeenInsts) { // If we already considered this instruction, we're done. if (!ConsideredInsts.insert(I).second) return false; // If this is an obviously unfoldable instruction, bail out. if (!MightBeFoldableInst(I)) return true; // Loop over all the uses, recursively processing them. for (Use &U : I->uses()) { // Conservatively return true if we're seeing a large number or a deep chain // of users. This avoids excessive compilation times in pathological cases. if (SeenInsts++ >= MaxAddressUsersToScan) return true; Instruction *UserI = cast(U.getUser()); if (LoadInst *LI = dyn_cast(UserI)) { MemoryUses.push_back({&U, LI->getType()}); continue; } if (StoreInst *SI = dyn_cast(UserI)) { if (U.getOperandNo() != StoreInst::getPointerOperandIndex()) return true; // Storing addr, not into addr. MemoryUses.push_back({&U, SI->getValueOperand()->getType()}); continue; } if (AtomicRMWInst *RMW = dyn_cast(UserI)) { if (U.getOperandNo() != AtomicRMWInst::getPointerOperandIndex()) return true; // Storing addr, not into addr. MemoryUses.push_back({&U, RMW->getValOperand()->getType()}); continue; } if (AtomicCmpXchgInst *CmpX = dyn_cast(UserI)) { if (U.getOperandNo() != AtomicCmpXchgInst::getPointerOperandIndex()) return true; // Storing addr, not into addr. MemoryUses.push_back({&U, CmpX->getCompareOperand()->getType()}); continue; } if (CallInst *CI = dyn_cast(UserI)) { if (CI->hasFnAttr(Attribute::Cold)) { // If this is a cold call, we can sink the addressing calculation into // the cold path. See optimizeCallInst bool OptForSize = OptSize || llvm::shouldOptimizeForSize(CI->getParent(), PSI, BFI); if (!OptForSize) continue; } InlineAsm *IA = dyn_cast(CI->getCalledOperand()); if (!IA) return true; // If this is a memory operand, we're cool, otherwise bail out. if (!IsOperandAMemoryOperand(CI, IA, I, TLI, TRI)) return true; continue; } if (FindAllMemoryUses(UserI, MemoryUses, ConsideredInsts, TLI, TRI, OptSize, PSI, BFI, SeenInsts)) return true; } return false; } static bool FindAllMemoryUses( Instruction *I, SmallVectorImpl> &MemoryUses, const TargetLowering &TLI, const TargetRegisterInfo &TRI, bool OptSize, ProfileSummaryInfo *PSI, BlockFrequencyInfo *BFI) { unsigned SeenInsts = 0; SmallPtrSet ConsideredInsts; return FindAllMemoryUses(I, MemoryUses, ConsideredInsts, TLI, TRI, OptSize, PSI, BFI, SeenInsts); } /// Return true if Val is already known to be live at the use site that we're /// folding it into. If so, there is no cost to include it in the addressing /// mode. KnownLive1 and KnownLive2 are two values that we know are live at the /// instruction already. bool AddressingModeMatcher::valueAlreadyLiveAtInst(Value *Val, Value *KnownLive1, Value *KnownLive2) { // If Val is either of the known-live values, we know it is live! if (Val == nullptr || Val == KnownLive1 || Val == KnownLive2) return true; // All values other than instructions and arguments (e.g. constants) are live. if (!isa(Val) && !isa(Val)) return true; // If Val is a constant sized alloca in the entry block, it is live, this is // true because it is just a reference to the stack/frame pointer, which is // live for the whole function. if (AllocaInst *AI = dyn_cast(Val)) if (AI->isStaticAlloca()) return true; // Check to see if this value is already used in the memory instruction's // block. If so, it's already live into the block at the very least, so we // can reasonably fold it. return Val->isUsedInBasicBlock(MemoryInst->getParent()); } /// It is possible for the addressing mode of the machine to fold the specified /// instruction into a load or store that ultimately uses it. /// However, the specified instruction has multiple uses. /// Given this, it may actually increase register pressure to fold it /// into the load. For example, consider this code: /// /// X = ... /// Y = X+1 /// use(Y) -> nonload/store /// Z = Y+1 /// load Z /// /// In this case, Y has multiple uses, and can be folded into the load of Z /// (yielding load [X+2]). However, doing this will cause both "X" and "X+1" to /// be live at the use(Y) line. If we don't fold Y into load Z, we use one /// fewer register. Since Y can't be folded into "use(Y)" we don't increase the /// number of computations either. /// /// Note that this (like most of CodeGenPrepare) is just a rough heuristic. If /// X was live across 'load Z' for other reasons, we actually *would* want to /// fold the addressing mode in the Z case. This would make Y die earlier. bool AddressingModeMatcher::isProfitableToFoldIntoAddressingMode( Instruction *I, ExtAddrMode &AMBefore, ExtAddrMode &AMAfter) { if (IgnoreProfitability) return true; // AMBefore is the addressing mode before this instruction was folded into it, // and AMAfter is the addressing mode after the instruction was folded. Get // the set of registers referenced by AMAfter and subtract out those // referenced by AMBefore: this is the set of values which folding in this // address extends the lifetime of. // // Note that there are only two potential values being referenced here, // BaseReg and ScaleReg (global addresses are always available, as are any // folded immediates). Value *BaseReg = AMAfter.BaseReg, *ScaledReg = AMAfter.ScaledReg; // If the BaseReg or ScaledReg was referenced by the previous addrmode, their // lifetime wasn't extended by adding this instruction. if (valueAlreadyLiveAtInst(BaseReg, AMBefore.BaseReg, AMBefore.ScaledReg)) BaseReg = nullptr; if (valueAlreadyLiveAtInst(ScaledReg, AMBefore.BaseReg, AMBefore.ScaledReg)) ScaledReg = nullptr; // If folding this instruction (and it's subexprs) didn't extend any live // ranges, we're ok with it. if (!BaseReg && !ScaledReg) return true; // If all uses of this instruction can have the address mode sunk into them, // we can remove the addressing mode and effectively trade one live register // for another (at worst.) In this context, folding an addressing mode into // the use is just a particularly nice way of sinking it. SmallVector, 16> MemoryUses; if (FindAllMemoryUses(I, MemoryUses, TLI, TRI, OptSize, PSI, BFI)) return false; // Has a non-memory, non-foldable use! // Now that we know that all uses of this instruction are part of a chain of // computation involving only operations that could theoretically be folded // into a memory use, loop over each of these memory operation uses and see // if they could *actually* fold the instruction. The assumption is that // addressing modes are cheap and that duplicating the computation involved // many times is worthwhile, even on a fastpath. For sinking candidates // (i.e. cold call sites), this serves as a way to prevent excessive code // growth since most architectures have some reasonable small and fast way to // compute an effective address. (i.e LEA on x86) SmallVector MatchedAddrModeInsts; for (const std::pair &Pair : MemoryUses) { Value *Address = Pair.first->get(); Instruction *UserI = cast(Pair.first->getUser()); Type *AddressAccessTy = Pair.second; unsigned AS = Address->getType()->getPointerAddressSpace(); // Do a match against the root of this address, ignoring profitability. This // will tell us if the addressing mode for the memory operation will // *actually* cover the shared instruction. ExtAddrMode Result; std::pair, int64_t> LargeOffsetGEP(nullptr, 0); TypePromotionTransaction::ConstRestorationPt LastKnownGood = TPT.getRestorationPoint(); AddressingModeMatcher Matcher(MatchedAddrModeInsts, TLI, TRI, LI, getDTFn, AddressAccessTy, AS, UserI, Result, InsertedInsts, PromotedInsts, TPT, LargeOffsetGEP, OptSize, PSI, BFI); Matcher.IgnoreProfitability = true; bool Success = Matcher.matchAddr(Address, 0); (void)Success; assert(Success && "Couldn't select *anything*?"); // The match was to check the profitability, the changes made are not // part of the original matcher. Therefore, they should be dropped // otherwise the original matcher will not present the right state. TPT.rollback(LastKnownGood); // If the match didn't cover I, then it won't be shared by it. if (!is_contained(MatchedAddrModeInsts, I)) return false; MatchedAddrModeInsts.clear(); } return true; } /// Return true if the specified values are defined in a /// different basic block than BB. static bool IsNonLocalValue(Value *V, BasicBlock *BB) { if (Instruction *I = dyn_cast(V)) return I->getParent() != BB; return false; } /// Sink addressing mode computation immediate before MemoryInst if doing so /// can be done without increasing register pressure. The need for the /// register pressure constraint means this can end up being an all or nothing /// decision for all uses of the same addressing computation. /// /// Load and Store Instructions often have addressing modes that can do /// significant amounts of computation. As such, instruction selection will try /// to get the load or store to do as much computation as possible for the /// program. The problem is that isel can only see within a single block. As /// such, we sink as much legal addressing mode work into the block as possible. /// /// This method is used to optimize both load/store and inline asms with memory /// operands. It's also used to sink addressing computations feeding into cold /// call sites into their (cold) basic block. /// /// The motivation for handling sinking into cold blocks is that doing so can /// both enable other address mode sinking (by satisfying the register pressure /// constraint above), and reduce register pressure globally (by removing the /// addressing mode computation from the fast path entirely.). bool CodeGenPrepare::optimizeMemoryInst(Instruction *MemoryInst, Value *Addr, Type *AccessTy, unsigned AddrSpace) { Value *Repl = Addr; // Try to collapse single-value PHI nodes. This is necessary to undo // unprofitable PRE transformations. SmallVector worklist; SmallPtrSet Visited; worklist.push_back(Addr); // Use a worklist to iteratively look through PHI and select nodes, and // ensure that the addressing mode obtained from the non-PHI/select roots of // the graph are compatible. bool PhiOrSelectSeen = false; SmallVector AddrModeInsts; const SimplifyQuery SQ(*DL, TLInfo); AddressingModeCombiner AddrModes(SQ, Addr); TypePromotionTransaction TPT(RemovedInsts); TypePromotionTransaction::ConstRestorationPt LastKnownGood = TPT.getRestorationPoint(); while (!worklist.empty()) { Value *V = worklist.pop_back_val(); // We allow traversing cyclic Phi nodes. // In case of success after this loop we ensure that traversing through // Phi nodes ends up with all cases to compute address of the form // BaseGV + Base + Scale * Index + Offset // where Scale and Offset are constans and BaseGV, Base and Index // are exactly the same Values in all cases. // It means that BaseGV, Scale and Offset dominate our memory instruction // and have the same value as they had in address computation represented // as Phi. So we can safely sink address computation to memory instruction. if (!Visited.insert(V).second) continue; // For a PHI node, push all of its incoming values. if (PHINode *P = dyn_cast(V)) { append_range(worklist, P->incoming_values()); PhiOrSelectSeen = true; continue; } // Similar for select. if (SelectInst *SI = dyn_cast(V)) { worklist.push_back(SI->getFalseValue()); worklist.push_back(SI->getTrueValue()); PhiOrSelectSeen = true; continue; } // For non-PHIs, determine the addressing mode being computed. Note that // the result may differ depending on what other uses our candidate // addressing instructions might have. AddrModeInsts.clear(); std::pair, int64_t> LargeOffsetGEP(nullptr, 0); // Defer the query (and possible computation of) the dom tree to point of // actual use. It's expected that most address matches don't actually need // the domtree. auto getDTFn = [MemoryInst, this]() -> const DominatorTree & { Function *F = MemoryInst->getParent()->getParent(); return this->getDT(*F); }; ExtAddrMode NewAddrMode = AddressingModeMatcher::Match( V, AccessTy, AddrSpace, MemoryInst, AddrModeInsts, *TLI, *LI, getDTFn, *TRI, InsertedInsts, PromotedInsts, TPT, LargeOffsetGEP, OptSize, PSI, BFI.get()); GetElementPtrInst *GEP = LargeOffsetGEP.first; if (GEP && !NewGEPBases.count(GEP)) { // If splitting the underlying data structure can reduce the offset of a // GEP, collect the GEP. Skip the GEPs that are the new bases of // previously split data structures. LargeOffsetGEPMap[GEP->getPointerOperand()].push_back(LargeOffsetGEP); LargeOffsetGEPID.insert(std::make_pair(GEP, LargeOffsetGEPID.size())); } NewAddrMode.OriginalValue = V; if (!AddrModes.addNewAddrMode(NewAddrMode)) break; } // Try to combine the AddrModes we've collected. If we couldn't collect any, // or we have multiple but either couldn't combine them or combining them // wouldn't do anything useful, bail out now. if (!AddrModes.combineAddrModes()) { TPT.rollback(LastKnownGood); return false; } bool Modified = TPT.commit(); // Get the combined AddrMode (or the only AddrMode, if we only had one). ExtAddrMode AddrMode = AddrModes.getAddrMode(); // If all the instructions matched are already in this BB, don't do anything. // If we saw a Phi node then it is not local definitely, and if we saw a // select then we want to push the address calculation past it even if it's // already in this BB. if (!PhiOrSelectSeen && none_of(AddrModeInsts, [&](Value *V) { return IsNonLocalValue(V, MemoryInst->getParent()); })) { LLVM_DEBUG(dbgs() << "CGP: Found local addrmode: " << AddrMode << "\n"); return Modified; } // Insert this computation right after this user. Since our caller is // scanning from the top of the BB to the bottom, reuse of the expr are // guaranteed to happen later. IRBuilder<> Builder(MemoryInst); // Now that we determined the addressing expression we want to use and know // that we have to sink it into this block. Check to see if we have already // done this for some other load/store instr in this block. If so, reuse // the computation. Before attempting reuse, check if the address is valid // as it may have been erased. WeakTrackingVH SunkAddrVH = SunkAddrs[Addr]; Value *SunkAddr = SunkAddrVH.pointsToAliveValue() ? SunkAddrVH : nullptr; Type *IntPtrTy = DL->getIntPtrType(Addr->getType()); if (SunkAddr) { LLVM_DEBUG(dbgs() << "CGP: Reusing nonlocal addrmode: " << AddrMode << " for " << *MemoryInst << "\n"); if (SunkAddr->getType() != Addr->getType()) { if (SunkAddr->getType()->getPointerAddressSpace() != Addr->getType()->getPointerAddressSpace() && !DL->isNonIntegralPointerType(Addr->getType())) { // There are two reasons the address spaces might not match: a no-op // addrspacecast, or a ptrtoint/inttoptr pair. Either way, we emit a // ptrtoint/inttoptr pair to ensure we match the original semantics. // TODO: allow bitcast between different address space pointers with the // same size. SunkAddr = Builder.CreatePtrToInt(SunkAddr, IntPtrTy, "sunkaddr"); SunkAddr = Builder.CreateIntToPtr(SunkAddr, Addr->getType(), "sunkaddr"); } else SunkAddr = Builder.CreatePointerCast(SunkAddr, Addr->getType()); } } else if (AddrSinkUsingGEPs || (!AddrSinkUsingGEPs.getNumOccurrences() && SubtargetInfo->addrSinkUsingGEPs())) { // By default, we use the GEP-based method when AA is used later. This // prevents new inttoptr/ptrtoint pairs from degrading AA capabilities. LLVM_DEBUG(dbgs() << "CGP: SINKING nonlocal addrmode: " << AddrMode << " for " << *MemoryInst << "\n"); Value *ResultPtr = nullptr, *ResultIndex = nullptr; // First, find the pointer. if (AddrMode.BaseReg && AddrMode.BaseReg->getType()->isPointerTy()) { ResultPtr = AddrMode.BaseReg; AddrMode.BaseReg = nullptr; } if (AddrMode.Scale && AddrMode.ScaledReg->getType()->isPointerTy()) { // We can't add more than one pointer together, nor can we scale a // pointer (both of which seem meaningless). if (ResultPtr || AddrMode.Scale != 1) return Modified; ResultPtr = AddrMode.ScaledReg; AddrMode.Scale = 0; } // It is only safe to sign extend the BaseReg if we know that the math // required to create it did not overflow before we extend it. Since // the original IR value was tossed in favor of a constant back when // the AddrMode was created we need to bail out gracefully if widths // do not match instead of extending it. // // (See below for code to add the scale.) if (AddrMode.Scale) { Type *ScaledRegTy = AddrMode.ScaledReg->getType(); if (cast(IntPtrTy)->getBitWidth() > cast(ScaledRegTy)->getBitWidth()) return Modified; } GlobalValue *BaseGV = AddrMode.BaseGV; if (BaseGV != nullptr) { if (ResultPtr) return Modified; if (BaseGV->isThreadLocal()) { ResultPtr = Builder.CreateThreadLocalAddress(BaseGV); } else { ResultPtr = BaseGV; } } // If the real base value actually came from an inttoptr, then the matcher // will look through it and provide only the integer value. In that case, // use it here. if (!DL->isNonIntegralPointerType(Addr->getType())) { if (!ResultPtr && AddrMode.BaseReg) { ResultPtr = Builder.CreateIntToPtr(AddrMode.BaseReg, Addr->getType(), "sunkaddr"); AddrMode.BaseReg = nullptr; } else if (!ResultPtr && AddrMode.Scale == 1) { ResultPtr = Builder.CreateIntToPtr(AddrMode.ScaledReg, Addr->getType(), "sunkaddr"); AddrMode.Scale = 0; } } if (!ResultPtr && !AddrMode.BaseReg && !AddrMode.Scale && !AddrMode.BaseOffs) { SunkAddr = Constant::getNullValue(Addr->getType()); } else if (!ResultPtr) { return Modified; } else { Type *I8PtrTy = Builder.getPtrTy(Addr->getType()->getPointerAddressSpace()); // Start with the base register. Do this first so that subsequent address // matching finds it last, which will prevent it from trying to match it // as the scaled value in case it happens to be a mul. That would be // problematic if we've sunk a different mul for the scale, because then // we'd end up sinking both muls. if (AddrMode.BaseReg) { Value *V = AddrMode.BaseReg; if (V->getType() != IntPtrTy) V = Builder.CreateIntCast(V, IntPtrTy, /*isSigned=*/true, "sunkaddr"); ResultIndex = V; } // Add the scale value. if (AddrMode.Scale) { Value *V = AddrMode.ScaledReg; if (V->getType() == IntPtrTy) { // done. } else { assert(cast(IntPtrTy)->getBitWidth() < cast(V->getType())->getBitWidth() && "We can't transform if ScaledReg is too narrow"); V = Builder.CreateTrunc(V, IntPtrTy, "sunkaddr"); } if (AddrMode.Scale != 1) V = Builder.CreateMul(V, ConstantInt::get(IntPtrTy, AddrMode.Scale), "sunkaddr"); if (ResultIndex) ResultIndex = Builder.CreateAdd(ResultIndex, V, "sunkaddr"); else ResultIndex = V; } // Add in the Base Offset if present. if (AddrMode.BaseOffs) { Value *V = ConstantInt::get(IntPtrTy, AddrMode.BaseOffs); if (ResultIndex) { // We need to add this separately from the scale above to help with // SDAG consecutive load/store merging. if (ResultPtr->getType() != I8PtrTy) ResultPtr = Builder.CreatePointerCast(ResultPtr, I8PtrTy); ResultPtr = Builder.CreatePtrAdd(ResultPtr, ResultIndex, "sunkaddr", AddrMode.InBounds); } ResultIndex = V; } if (!ResultIndex) { SunkAddr = ResultPtr; } else { if (ResultPtr->getType() != I8PtrTy) ResultPtr = Builder.CreatePointerCast(ResultPtr, I8PtrTy); SunkAddr = Builder.CreatePtrAdd(ResultPtr, ResultIndex, "sunkaddr", AddrMode.InBounds); } if (SunkAddr->getType() != Addr->getType()) { if (SunkAddr->getType()->getPointerAddressSpace() != Addr->getType()->getPointerAddressSpace() && !DL->isNonIntegralPointerType(Addr->getType())) { // There are two reasons the address spaces might not match: a no-op // addrspacecast, or a ptrtoint/inttoptr pair. Either way, we emit a // ptrtoint/inttoptr pair to ensure we match the original semantics. // TODO: allow bitcast between different address space pointers with // the same size. SunkAddr = Builder.CreatePtrToInt(SunkAddr, IntPtrTy, "sunkaddr"); SunkAddr = Builder.CreateIntToPtr(SunkAddr, Addr->getType(), "sunkaddr"); } else SunkAddr = Builder.CreatePointerCast(SunkAddr, Addr->getType()); } } } else { // We'd require a ptrtoint/inttoptr down the line, which we can't do for // non-integral pointers, so in that case bail out now. Type *BaseTy = AddrMode.BaseReg ? AddrMode.BaseReg->getType() : nullptr; Type *ScaleTy = AddrMode.Scale ? AddrMode.ScaledReg->getType() : nullptr; PointerType *BasePtrTy = dyn_cast_or_null(BaseTy); PointerType *ScalePtrTy = dyn_cast_or_null(ScaleTy); if (DL->isNonIntegralPointerType(Addr->getType()) || (BasePtrTy && DL->isNonIntegralPointerType(BasePtrTy)) || (ScalePtrTy && DL->isNonIntegralPointerType(ScalePtrTy)) || (AddrMode.BaseGV && DL->isNonIntegralPointerType(AddrMode.BaseGV->getType()))) return Modified; LLVM_DEBUG(dbgs() << "CGP: SINKING nonlocal addrmode: " << AddrMode << " for " << *MemoryInst << "\n"); Type *IntPtrTy = DL->getIntPtrType(Addr->getType()); Value *Result = nullptr; // Start with the base register. Do this first so that subsequent address // matching finds it last, which will prevent it from trying to match it // as the scaled value in case it happens to be a mul. That would be // problematic if we've sunk a different mul for the scale, because then // we'd end up sinking both muls. if (AddrMode.BaseReg) { Value *V = AddrMode.BaseReg; if (V->getType()->isPointerTy()) V = Builder.CreatePtrToInt(V, IntPtrTy, "sunkaddr"); if (V->getType() != IntPtrTy) V = Builder.CreateIntCast(V, IntPtrTy, /*isSigned=*/true, "sunkaddr"); Result = V; } // Add the scale value. if (AddrMode.Scale) { Value *V = AddrMode.ScaledReg; if (V->getType() == IntPtrTy) { // done. } else if (V->getType()->isPointerTy()) { V = Builder.CreatePtrToInt(V, IntPtrTy, "sunkaddr"); } else if (cast(IntPtrTy)->getBitWidth() < cast(V->getType())->getBitWidth()) { V = Builder.CreateTrunc(V, IntPtrTy, "sunkaddr"); } else { // It is only safe to sign extend the BaseReg if we know that the math // required to create it did not overflow before we extend it. Since // the original IR value was tossed in favor of a constant back when // the AddrMode was created we need to bail out gracefully if widths // do not match instead of extending it. Instruction *I = dyn_cast_or_null(Result); if (I && (Result != AddrMode.BaseReg)) I->eraseFromParent(); return Modified; } if (AddrMode.Scale != 1) V = Builder.CreateMul(V, ConstantInt::get(IntPtrTy, AddrMode.Scale), "sunkaddr"); if (Result) Result = Builder.CreateAdd(Result, V, "sunkaddr"); else Result = V; } // Add in the BaseGV if present. GlobalValue *BaseGV = AddrMode.BaseGV; if (BaseGV != nullptr) { Value *BaseGVPtr; if (BaseGV->isThreadLocal()) { BaseGVPtr = Builder.CreateThreadLocalAddress(BaseGV); } else { BaseGVPtr = BaseGV; } Value *V = Builder.CreatePtrToInt(BaseGVPtr, IntPtrTy, "sunkaddr"); if (Result) Result = Builder.CreateAdd(Result, V, "sunkaddr"); else Result = V; } // Add in the Base Offset if present. if (AddrMode.BaseOffs) { Value *V = ConstantInt::get(IntPtrTy, AddrMode.BaseOffs); if (Result) Result = Builder.CreateAdd(Result, V, "sunkaddr"); else Result = V; } if (!Result) SunkAddr = Constant::getNullValue(Addr->getType()); else SunkAddr = Builder.CreateIntToPtr(Result, Addr->getType(), "sunkaddr"); } MemoryInst->replaceUsesOfWith(Repl, SunkAddr); // Store the newly computed address into the cache. In the case we reused a // value, this should be idempotent. SunkAddrs[Addr] = WeakTrackingVH(SunkAddr); // If we have no uses, recursively delete the value and all dead instructions // using it. if (Repl->use_empty()) { resetIteratorIfInvalidatedWhileCalling(CurInstIterator->getParent(), [&]() { RecursivelyDeleteTriviallyDeadInstructions( Repl, TLInfo, nullptr, [&](Value *V) { removeAllAssertingVHReferences(V); }); }); } ++NumMemoryInsts; return true; } /// Rewrite GEP input to gather/scatter to enable SelectionDAGBuilder to find /// a uniform base to use for ISD::MGATHER/MSCATTER. SelectionDAGBuilder can /// only handle a 2 operand GEP in the same basic block or a splat constant /// vector. The 2 operands to the GEP must have a scalar pointer and a vector /// index. /// /// If the existing GEP has a vector base pointer that is splat, we can look /// through the splat to find the scalar pointer. If we can't find a scalar /// pointer there's nothing we can do. /// /// If we have a GEP with more than 2 indices where the middle indices are all /// zeroes, we can replace it with 2 GEPs where the second has 2 operands. /// /// If the final index isn't a vector or is a splat, we can emit a scalar GEP /// followed by a GEP with an all zeroes vector index. This will enable /// SelectionDAGBuilder to use the scalar GEP as the uniform base and have a /// zero index. bool CodeGenPrepare::optimizeGatherScatterInst(Instruction *MemoryInst, Value *Ptr) { Value *NewAddr; if (const auto *GEP = dyn_cast(Ptr)) { // Don't optimize GEPs that don't have indices. if (!GEP->hasIndices()) return false; // If the GEP and the gather/scatter aren't in the same BB, don't optimize. // FIXME: We should support this by sinking the GEP. if (MemoryInst->getParent() != GEP->getParent()) return false; SmallVector Ops(GEP->operands()); bool RewriteGEP = false; if (Ops[0]->getType()->isVectorTy()) { Ops[0] = getSplatValue(Ops[0]); if (!Ops[0]) return false; RewriteGEP = true; } unsigned FinalIndex = Ops.size() - 1; // Ensure all but the last index is 0. // FIXME: This isn't strictly required. All that's required is that they are // all scalars or splats. for (unsigned i = 1; i < FinalIndex; ++i) { auto *C = dyn_cast(Ops[i]); if (!C) return false; if (isa(C->getType())) C = C->getSplatValue(); auto *CI = dyn_cast_or_null(C); if (!CI || !CI->isZero()) return false; // Scalarize the index if needed. Ops[i] = CI; } // Try to scalarize the final index. if (Ops[FinalIndex]->getType()->isVectorTy()) { if (Value *V = getSplatValue(Ops[FinalIndex])) { auto *C = dyn_cast(V); // Don't scalarize all zeros vector. if (!C || !C->isZero()) { Ops[FinalIndex] = V; RewriteGEP = true; } } } // If we made any changes or the we have extra operands, we need to generate // new instructions. if (!RewriteGEP && Ops.size() == 2) return false; auto NumElts = cast(Ptr->getType())->getElementCount(); IRBuilder<> Builder(MemoryInst); Type *SourceTy = GEP->getSourceElementType(); Type *ScalarIndexTy = DL->getIndexType(Ops[0]->getType()->getScalarType()); // If the final index isn't a vector, emit a scalar GEP containing all ops // and a vector GEP with all zeroes final index. if (!Ops[FinalIndex]->getType()->isVectorTy()) { NewAddr = Builder.CreateGEP(SourceTy, Ops[0], ArrayRef(Ops).drop_front()); auto *IndexTy = VectorType::get(ScalarIndexTy, NumElts); auto *SecondTy = GetElementPtrInst::getIndexedType( SourceTy, ArrayRef(Ops).drop_front()); NewAddr = Builder.CreateGEP(SecondTy, NewAddr, Constant::getNullValue(IndexTy)); } else { Value *Base = Ops[0]; Value *Index = Ops[FinalIndex]; // Create a scalar GEP if there are more than 2 operands. if (Ops.size() != 2) { // Replace the last index with 0. Ops[FinalIndex] = Constant::getNullValue(Ops[FinalIndex]->getType()->getScalarType()); Base = Builder.CreateGEP(SourceTy, Base, ArrayRef(Ops).drop_front()); SourceTy = GetElementPtrInst::getIndexedType( SourceTy, ArrayRef(Ops).drop_front()); } // Now create the GEP with scalar pointer and vector index. NewAddr = Builder.CreateGEP(SourceTy, Base, Index); } } else if (!isa(Ptr)) { // Not a GEP, maybe its a splat and we can create a GEP to enable // SelectionDAGBuilder to use it as a uniform base. Value *V = getSplatValue(Ptr); if (!V) return false; auto NumElts = cast(Ptr->getType())->getElementCount(); IRBuilder<> Builder(MemoryInst); // Emit a vector GEP with a scalar pointer and all 0s vector index. Type *ScalarIndexTy = DL->getIndexType(V->getType()->getScalarType()); auto *IndexTy = VectorType::get(ScalarIndexTy, NumElts); Type *ScalarTy; if (cast(MemoryInst)->getIntrinsicID() == Intrinsic::masked_gather) { ScalarTy = MemoryInst->getType()->getScalarType(); } else { assert(cast(MemoryInst)->getIntrinsicID() == Intrinsic::masked_scatter); ScalarTy = MemoryInst->getOperand(0)->getType()->getScalarType(); } NewAddr = Builder.CreateGEP(ScalarTy, V, Constant::getNullValue(IndexTy)); } else { // Constant, SelectionDAGBuilder knows to check if its a splat. return false; } MemoryInst->replaceUsesOfWith(Ptr, NewAddr); // If we have no uses, recursively delete the value and all dead instructions // using it. if (Ptr->use_empty()) RecursivelyDeleteTriviallyDeadInstructions( Ptr, TLInfo, nullptr, [&](Value *V) { removeAllAssertingVHReferences(V); }); return true; } /// If there are any memory operands, use OptimizeMemoryInst to sink their /// address computing into the block when possible / profitable. bool CodeGenPrepare::optimizeInlineAsmInst(CallInst *CS) { bool MadeChange = false; const TargetRegisterInfo *TRI = TM->getSubtargetImpl(*CS->getFunction())->getRegisterInfo(); TargetLowering::AsmOperandInfoVector TargetConstraints = TLI->ParseConstraints(*DL, TRI, *CS); unsigned ArgNo = 0; for (TargetLowering::AsmOperandInfo &OpInfo : TargetConstraints) { // Compute the constraint code and ConstraintType to use. TLI->ComputeConstraintToUse(OpInfo, SDValue()); // TODO: Also handle C_Address? if (OpInfo.ConstraintType == TargetLowering::C_Memory && OpInfo.isIndirect) { Value *OpVal = CS->getArgOperand(ArgNo++); MadeChange |= optimizeMemoryInst(CS, OpVal, OpVal->getType(), ~0u); } else if (OpInfo.Type == InlineAsm::isInput) ArgNo++; } return MadeChange; } /// Check if all the uses of \p Val are equivalent (or free) zero or /// sign extensions. static bool hasSameExtUse(Value *Val, const TargetLowering &TLI) { assert(!Val->use_empty() && "Input must have at least one use"); const Instruction *FirstUser = cast(*Val->user_begin()); bool IsSExt = isa(FirstUser); Type *ExtTy = FirstUser->getType(); for (const User *U : Val->users()) { const Instruction *UI = cast(U); if ((IsSExt && !isa(UI)) || (!IsSExt && !isa(UI))) return false; Type *CurTy = UI->getType(); // Same input and output types: Same instruction after CSE. if (CurTy == ExtTy) continue; // If IsSExt is true, we are in this situation: // a = Val // b = sext ty1 a to ty2 // c = sext ty1 a to ty3 // Assuming ty2 is shorter than ty3, this could be turned into: // a = Val // b = sext ty1 a to ty2 // c = sext ty2 b to ty3 // However, the last sext is not free. if (IsSExt) return false; // This is a ZExt, maybe this is free to extend from one type to another. // In that case, we would not account for a different use. Type *NarrowTy; Type *LargeTy; if (ExtTy->getScalarType()->getIntegerBitWidth() > CurTy->getScalarType()->getIntegerBitWidth()) { NarrowTy = CurTy; LargeTy = ExtTy; } else { NarrowTy = ExtTy; LargeTy = CurTy; } if (!TLI.isZExtFree(NarrowTy, LargeTy)) return false; } // All uses are the same or can be derived from one another for free. return true; } /// Try to speculatively promote extensions in \p Exts and continue /// promoting through newly promoted operands recursively as far as doing so is /// profitable. Save extensions profitably moved up, in \p ProfitablyMovedExts. /// When some promotion happened, \p TPT contains the proper state to revert /// them. /// /// \return true if some promotion happened, false otherwise. bool CodeGenPrepare::tryToPromoteExts( TypePromotionTransaction &TPT, const SmallVectorImpl &Exts, SmallVectorImpl &ProfitablyMovedExts, unsigned CreatedInstsCost) { bool Promoted = false; // Iterate over all the extensions to try to promote them. for (auto *I : Exts) { // Early check if we directly have ext(load). if (isa(I->getOperand(0))) { ProfitablyMovedExts.push_back(I); continue; } // Check whether or not we want to do any promotion. The reason we have // this check inside the for loop is to catch the case where an extension // is directly fed by a load because in such case the extension can be moved // up without any promotion on its operands. if (!TLI->enableExtLdPromotion() || DisableExtLdPromotion) return false; // Get the action to perform the promotion. TypePromotionHelper::Action TPH = TypePromotionHelper::getAction(I, InsertedInsts, *TLI, PromotedInsts); // Check if we can promote. if (!TPH) { // Save the current extension as we cannot move up through its operand. ProfitablyMovedExts.push_back(I); continue; } // Save the current state. TypePromotionTransaction::ConstRestorationPt LastKnownGood = TPT.getRestorationPoint(); SmallVector NewExts; unsigned NewCreatedInstsCost = 0; unsigned ExtCost = !TLI->isExtFree(I); // Promote. Value *PromotedVal = TPH(I, TPT, PromotedInsts, NewCreatedInstsCost, &NewExts, nullptr, *TLI); assert(PromotedVal && "TypePromotionHelper should have filtered out those cases"); // We would be able to merge only one extension in a load. // Therefore, if we have more than 1 new extension we heuristically // cut this search path, because it means we degrade the code quality. // With exactly 2, the transformation is neutral, because we will merge // one extension but leave one. However, we optimistically keep going, // because the new extension may be removed too. Also avoid replacing a // single free extension with multiple extensions, as this increases the // number of IR instructions while not providing any savings. long long TotalCreatedInstsCost = CreatedInstsCost + NewCreatedInstsCost; // FIXME: It would be possible to propagate a negative value instead of // conservatively ceiling it to 0. TotalCreatedInstsCost = std::max((long long)0, (TotalCreatedInstsCost - ExtCost)); if (!StressExtLdPromotion && (TotalCreatedInstsCost > 1 || !isPromotedInstructionLegal(*TLI, *DL, PromotedVal) || (ExtCost == 0 && NewExts.size() > 1))) { // This promotion is not profitable, rollback to the previous state, and // save the current extension in ProfitablyMovedExts as the latest // speculative promotion turned out to be unprofitable. TPT.rollback(LastKnownGood); ProfitablyMovedExts.push_back(I); continue; } // Continue promoting NewExts as far as doing so is profitable. SmallVector NewlyMovedExts; (void)tryToPromoteExts(TPT, NewExts, NewlyMovedExts, TotalCreatedInstsCost); bool NewPromoted = false; for (auto *ExtInst : NewlyMovedExts) { Instruction *MovedExt = cast(ExtInst); Value *ExtOperand = MovedExt->getOperand(0); // If we have reached to a load, we need this extra profitability check // as it could potentially be merged into an ext(load). if (isa(ExtOperand) && !(StressExtLdPromotion || NewCreatedInstsCost <= ExtCost || (ExtOperand->hasOneUse() || hasSameExtUse(ExtOperand, *TLI)))) continue; ProfitablyMovedExts.push_back(MovedExt); NewPromoted = true; } // If none of speculative promotions for NewExts is profitable, rollback // and save the current extension (I) as the last profitable extension. if (!NewPromoted) { TPT.rollback(LastKnownGood); ProfitablyMovedExts.push_back(I); continue; } // The promotion is profitable. Promoted = true; } return Promoted; } /// Merging redundant sexts when one is dominating the other. bool CodeGenPrepare::mergeSExts(Function &F) { bool Changed = false; for (auto &Entry : ValToSExtendedUses) { SExts &Insts = Entry.second; SExts CurPts; for (Instruction *Inst : Insts) { if (RemovedInsts.count(Inst) || !isa(Inst) || Inst->getOperand(0) != Entry.first) continue; bool inserted = false; for (auto &Pt : CurPts) { if (getDT(F).dominates(Inst, Pt)) { replaceAllUsesWith(Pt, Inst, FreshBBs, IsHugeFunc); RemovedInsts.insert(Pt); Pt->removeFromParent(); Pt = Inst; inserted = true; Changed = true; break; } if (!getDT(F).dominates(Pt, Inst)) // Give up if we need to merge in a common dominator as the // experiments show it is not profitable. continue; replaceAllUsesWith(Inst, Pt, FreshBBs, IsHugeFunc); RemovedInsts.insert(Inst); Inst->removeFromParent(); inserted = true; Changed = true; break; } if (!inserted) CurPts.push_back(Inst); } } return Changed; } // Splitting large data structures so that the GEPs accessing them can have // smaller offsets so that they can be sunk to the same blocks as their users. // For example, a large struct starting from %base is split into two parts // where the second part starts from %new_base. // // Before: // BB0: // %base = // // BB1: // %gep0 = gep %base, off0 // %gep1 = gep %base, off1 // %gep2 = gep %base, off2 // // BB2: // %load1 = load %gep0 // %load2 = load %gep1 // %load3 = load %gep2 // // After: // BB0: // %base = // %new_base = gep %base, off0 // // BB1: // %new_gep0 = %new_base // %new_gep1 = gep %new_base, off1 - off0 // %new_gep2 = gep %new_base, off2 - off0 // // BB2: // %load1 = load i32, i32* %new_gep0 // %load2 = load i32, i32* %new_gep1 // %load3 = load i32, i32* %new_gep2 // // %new_gep1 and %new_gep2 can be sunk to BB2 now after the splitting because // their offsets are smaller enough to fit into the addressing mode. bool CodeGenPrepare::splitLargeGEPOffsets() { bool Changed = false; for (auto &Entry : LargeOffsetGEPMap) { Value *OldBase = Entry.first; SmallVectorImpl, int64_t>> &LargeOffsetGEPs = Entry.second; auto compareGEPOffset = [&](const std::pair &LHS, const std::pair &RHS) { if (LHS.first == RHS.first) return false; if (LHS.second != RHS.second) return LHS.second < RHS.second; return LargeOffsetGEPID[LHS.first] < LargeOffsetGEPID[RHS.first]; }; // Sorting all the GEPs of the same data structures based on the offsets. llvm::sort(LargeOffsetGEPs, compareGEPOffset); LargeOffsetGEPs.erase(llvm::unique(LargeOffsetGEPs), LargeOffsetGEPs.end()); // Skip if all the GEPs have the same offsets. if (LargeOffsetGEPs.front().second == LargeOffsetGEPs.back().second) continue; GetElementPtrInst *BaseGEP = LargeOffsetGEPs.begin()->first; int64_t BaseOffset = LargeOffsetGEPs.begin()->second; Value *NewBaseGEP = nullptr; auto createNewBase = [&](int64_t BaseOffset, Value *OldBase, GetElementPtrInst *GEP) { LLVMContext &Ctx = GEP->getContext(); Type *PtrIdxTy = DL->getIndexType(GEP->getType()); Type *I8PtrTy = PointerType::get(Ctx, GEP->getType()->getPointerAddressSpace()); BasicBlock::iterator NewBaseInsertPt; BasicBlock *NewBaseInsertBB; if (auto *BaseI = dyn_cast(OldBase)) { // If the base of the struct is an instruction, the new base will be // inserted close to it. NewBaseInsertBB = BaseI->getParent(); if (isa(BaseI)) NewBaseInsertPt = NewBaseInsertBB->getFirstInsertionPt(); else if (InvokeInst *Invoke = dyn_cast(BaseI)) { NewBaseInsertBB = SplitEdge(NewBaseInsertBB, Invoke->getNormalDest(), DT.get(), LI); NewBaseInsertPt = NewBaseInsertBB->getFirstInsertionPt(); } else NewBaseInsertPt = std::next(BaseI->getIterator()); } else { // If the current base is an argument or global value, the new base // will be inserted to the entry block. NewBaseInsertBB = &BaseGEP->getFunction()->getEntryBlock(); NewBaseInsertPt = NewBaseInsertBB->getFirstInsertionPt(); } IRBuilder<> NewBaseBuilder(NewBaseInsertBB, NewBaseInsertPt); // Create a new base. Value *BaseIndex = ConstantInt::get(PtrIdxTy, BaseOffset); NewBaseGEP = OldBase; if (NewBaseGEP->getType() != I8PtrTy) NewBaseGEP = NewBaseBuilder.CreatePointerCast(NewBaseGEP, I8PtrTy); NewBaseGEP = NewBaseBuilder.CreatePtrAdd(NewBaseGEP, BaseIndex, "splitgep"); NewGEPBases.insert(NewBaseGEP); return; }; // Check whether all the offsets can be encoded with prefered common base. if (int64_t PreferBase = TLI->getPreferredLargeGEPBaseOffset( LargeOffsetGEPs.front().second, LargeOffsetGEPs.back().second)) { BaseOffset = PreferBase; // Create a new base if the offset of the BaseGEP can be decoded with one // instruction. createNewBase(BaseOffset, OldBase, BaseGEP); } auto *LargeOffsetGEP = LargeOffsetGEPs.begin(); while (LargeOffsetGEP != LargeOffsetGEPs.end()) { GetElementPtrInst *GEP = LargeOffsetGEP->first; int64_t Offset = LargeOffsetGEP->second; if (Offset != BaseOffset) { TargetLowering::AddrMode AddrMode; AddrMode.HasBaseReg = true; AddrMode.BaseOffs = Offset - BaseOffset; // The result type of the GEP might not be the type of the memory // access. if (!TLI->isLegalAddressingMode(*DL, AddrMode, GEP->getResultElementType(), GEP->getAddressSpace())) { // We need to create a new base if the offset to the current base is // too large to fit into the addressing mode. So, a very large struct // may be split into several parts. BaseGEP = GEP; BaseOffset = Offset; NewBaseGEP = nullptr; } } // Generate a new GEP to replace the current one. Type *PtrIdxTy = DL->getIndexType(GEP->getType()); if (!NewBaseGEP) { // Create a new base if we don't have one yet. Find the insertion // pointer for the new base first. createNewBase(BaseOffset, OldBase, GEP); } IRBuilder<> Builder(GEP); Value *NewGEP = NewBaseGEP; if (Offset != BaseOffset) { // Calculate the new offset for the new GEP. Value *Index = ConstantInt::get(PtrIdxTy, Offset - BaseOffset); NewGEP = Builder.CreatePtrAdd(NewBaseGEP, Index); } replaceAllUsesWith(GEP, NewGEP, FreshBBs, IsHugeFunc); LargeOffsetGEPID.erase(GEP); LargeOffsetGEP = LargeOffsetGEPs.erase(LargeOffsetGEP); GEP->eraseFromParent(); Changed = true; } } return Changed; } bool CodeGenPrepare::optimizePhiType( PHINode *I, SmallPtrSetImpl &Visited, SmallPtrSetImpl &DeletedInstrs) { // We are looking for a collection on interconnected phi nodes that together // only use loads/bitcasts and are used by stores/bitcasts, and the bitcasts // are of the same type. Convert the whole set of nodes to the type of the // bitcast. Type *PhiTy = I->getType(); Type *ConvertTy = nullptr; if (Visited.count(I) || (!I->getType()->isIntegerTy() && !I->getType()->isFloatingPointTy())) return false; SmallVector Worklist; Worklist.push_back(cast(I)); SmallPtrSet PhiNodes; SmallPtrSet Constants; PhiNodes.insert(I); Visited.insert(I); SmallPtrSet Defs; SmallPtrSet Uses; // This works by adding extra bitcasts between load/stores and removing // existing bicasts. If we have a phi(bitcast(load)) or a store(bitcast(phi)) // we can get in the situation where we remove a bitcast in one iteration // just to add it again in the next. We need to ensure that at least one // bitcast we remove are anchored to something that will not change back. bool AnyAnchored = false; while (!Worklist.empty()) { Instruction *II = Worklist.pop_back_val(); if (auto *Phi = dyn_cast(II)) { // Handle Defs, which might also be PHI's for (Value *V : Phi->incoming_values()) { if (auto *OpPhi = dyn_cast(V)) { if (!PhiNodes.count(OpPhi)) { if (!Visited.insert(OpPhi).second) return false; PhiNodes.insert(OpPhi); Worklist.push_back(OpPhi); } } else if (auto *OpLoad = dyn_cast(V)) { if (!OpLoad->isSimple()) return false; if (Defs.insert(OpLoad).second) Worklist.push_back(OpLoad); } else if (auto *OpEx = dyn_cast(V)) { if (Defs.insert(OpEx).second) Worklist.push_back(OpEx); } else if (auto *OpBC = dyn_cast(V)) { if (!ConvertTy) ConvertTy = OpBC->getOperand(0)->getType(); if (OpBC->getOperand(0)->getType() != ConvertTy) return false; if (Defs.insert(OpBC).second) { Worklist.push_back(OpBC); AnyAnchored |= !isa(OpBC->getOperand(0)) && !isa(OpBC->getOperand(0)); } } else if (auto *OpC = dyn_cast(V)) Constants.insert(OpC); else return false; } } // Handle uses which might also be phi's for (User *V : II->users()) { if (auto *OpPhi = dyn_cast(V)) { if (!PhiNodes.count(OpPhi)) { if (Visited.count(OpPhi)) return false; PhiNodes.insert(OpPhi); Visited.insert(OpPhi); Worklist.push_back(OpPhi); } } else if (auto *OpStore = dyn_cast(V)) { if (!OpStore->isSimple() || OpStore->getOperand(0) != II) return false; Uses.insert(OpStore); } else if (auto *OpBC = dyn_cast(V)) { if (!ConvertTy) ConvertTy = OpBC->getType(); if (OpBC->getType() != ConvertTy) return false; Uses.insert(OpBC); AnyAnchored |= any_of(OpBC->users(), [](User *U) { return !isa(U); }); } else { return false; } } } if (!ConvertTy || !AnyAnchored || !TLI->shouldConvertPhiType(PhiTy, ConvertTy)) return false; LLVM_DEBUG(dbgs() << "Converting " << *I << "\n and connected nodes to " << *ConvertTy << "\n"); // Create all the new phi nodes of the new type, and bitcast any loads to the // correct type. ValueToValueMap ValMap; for (ConstantData *C : Constants) ValMap[C] = ConstantExpr::getBitCast(C, ConvertTy); for (Instruction *D : Defs) { if (isa(D)) { ValMap[D] = D->getOperand(0); DeletedInstrs.insert(D); } else { BasicBlock::iterator insertPt = std::next(D->getIterator()); ValMap[D] = new BitCastInst(D, ConvertTy, D->getName() + ".bc", insertPt); } } for (PHINode *Phi : PhiNodes) ValMap[Phi] = PHINode::Create(ConvertTy, Phi->getNumIncomingValues(), Phi->getName() + ".tc", Phi->getIterator()); // Pipe together all the PhiNodes. for (PHINode *Phi : PhiNodes) { PHINode *NewPhi = cast(ValMap[Phi]); for (int i = 0, e = Phi->getNumIncomingValues(); i < e; i++) NewPhi->addIncoming(ValMap[Phi->getIncomingValue(i)], Phi->getIncomingBlock(i)); Visited.insert(NewPhi); } // And finally pipe up the stores and bitcasts for (Instruction *U : Uses) { if (isa(U)) { DeletedInstrs.insert(U); replaceAllUsesWith(U, ValMap[U->getOperand(0)], FreshBBs, IsHugeFunc); } else { U->setOperand(0, new BitCastInst(ValMap[U->getOperand(0)], PhiTy, "bc", U->getIterator())); } } // Save the removed phis to be deleted later. for (PHINode *Phi : PhiNodes) DeletedInstrs.insert(Phi); return true; } bool CodeGenPrepare::optimizePhiTypes(Function &F) { if (!OptimizePhiTypes) return false; bool Changed = false; SmallPtrSet Visited; SmallPtrSet DeletedInstrs; // Attempt to optimize all the phis in the functions to the correct type. for (auto &BB : F) for (auto &Phi : BB.phis()) Changed |= optimizePhiType(&Phi, Visited, DeletedInstrs); // Remove any old phi's that have been converted. for (auto *I : DeletedInstrs) { replaceAllUsesWith(I, PoisonValue::get(I->getType()), FreshBBs, IsHugeFunc); I->eraseFromParent(); } return Changed; } /// Return true, if an ext(load) can be formed from an extension in /// \p MovedExts. bool CodeGenPrepare::canFormExtLd( const SmallVectorImpl &MovedExts, LoadInst *&LI, Instruction *&Inst, bool HasPromoted) { for (auto *MovedExtInst : MovedExts) { if (isa(MovedExtInst->getOperand(0))) { LI = cast(MovedExtInst->getOperand(0)); Inst = MovedExtInst; break; } } if (!LI) return false; // If they're already in the same block, there's nothing to do. // Make the cheap checks first if we did not promote. // If we promoted, we need to check if it is indeed profitable. if (!HasPromoted && LI->getParent() == Inst->getParent()) return false; return TLI->isExtLoad(LI, Inst, *DL); } /// Move a zext or sext fed by a load into the same basic block as the load, /// unless conditions are unfavorable. This allows SelectionDAG to fold the /// extend into the load. /// /// E.g., /// \code /// %ld = load i32* %addr /// %add = add nuw i32 %ld, 4 /// %zext = zext i32 %add to i64 // \endcode /// => /// \code /// %ld = load i32* %addr /// %zext = zext i32 %ld to i64 /// %add = add nuw i64 %zext, 4 /// \encode /// Note that the promotion in %add to i64 is done in tryToPromoteExts(), which /// allow us to match zext(load i32*) to i64. /// /// Also, try to promote the computations used to obtain a sign extended /// value used into memory accesses. /// E.g., /// \code /// a = add nsw i32 b, 3 /// d = sext i32 a to i64 /// e = getelementptr ..., i64 d /// \endcode /// => /// \code /// f = sext i32 b to i64 /// a = add nsw i64 f, 3 /// e = getelementptr ..., i64 a /// \endcode /// /// \p Inst[in/out] the extension may be modified during the process if some /// promotions apply. bool CodeGenPrepare::optimizeExt(Instruction *&Inst) { bool AllowPromotionWithoutCommonHeader = false; /// See if it is an interesting sext operations for the address type /// promotion before trying to promote it, e.g., the ones with the right /// type and used in memory accesses. bool ATPConsiderable = TTI->shouldConsiderAddressTypePromotion( *Inst, AllowPromotionWithoutCommonHeader); TypePromotionTransaction TPT(RemovedInsts); TypePromotionTransaction::ConstRestorationPt LastKnownGood = TPT.getRestorationPoint(); SmallVector Exts; SmallVector SpeculativelyMovedExts; Exts.push_back(Inst); bool HasPromoted = tryToPromoteExts(TPT, Exts, SpeculativelyMovedExts); // Look for a load being extended. LoadInst *LI = nullptr; Instruction *ExtFedByLoad; // Try to promote a chain of computation if it allows to form an extended // load. if (canFormExtLd(SpeculativelyMovedExts, LI, ExtFedByLoad, HasPromoted)) { assert(LI && ExtFedByLoad && "Expect a valid load and extension"); TPT.commit(); // Move the extend into the same block as the load. ExtFedByLoad->moveAfter(LI); ++NumExtsMoved; Inst = ExtFedByLoad; return true; } // Continue promoting SExts if known as considerable depending on targets. if (ATPConsiderable && performAddressTypePromotion(Inst, AllowPromotionWithoutCommonHeader, HasPromoted, TPT, SpeculativelyMovedExts)) return true; TPT.rollback(LastKnownGood); return false; } // Perform address type promotion if doing so is profitable. // If AllowPromotionWithoutCommonHeader == false, we should find other sext // instructions that sign extended the same initial value. However, if // AllowPromotionWithoutCommonHeader == true, we expect promoting the // extension is just profitable. bool CodeGenPrepare::performAddressTypePromotion( Instruction *&Inst, bool AllowPromotionWithoutCommonHeader, bool HasPromoted, TypePromotionTransaction &TPT, SmallVectorImpl &SpeculativelyMovedExts) { bool Promoted = false; SmallPtrSet UnhandledExts; bool AllSeenFirst = true; for (auto *I : SpeculativelyMovedExts) { Value *HeadOfChain = I->getOperand(0); DenseMap::iterator AlreadySeen = SeenChainsForSExt.find(HeadOfChain); // If there is an unhandled SExt which has the same header, try to promote // it as well. if (AlreadySeen != SeenChainsForSExt.end()) { if (AlreadySeen->second != nullptr) UnhandledExts.insert(AlreadySeen->second); AllSeenFirst = false; } } if (!AllSeenFirst || (AllowPromotionWithoutCommonHeader && SpeculativelyMovedExts.size() == 1)) { TPT.commit(); if (HasPromoted) Promoted = true; for (auto *I : SpeculativelyMovedExts) { Value *HeadOfChain = I->getOperand(0); SeenChainsForSExt[HeadOfChain] = nullptr; ValToSExtendedUses[HeadOfChain].push_back(I); } // Update Inst as promotion happen. Inst = SpeculativelyMovedExts.pop_back_val(); } else { // This is the first chain visited from the header, keep the current chain // as unhandled. Defer to promote this until we encounter another SExt // chain derived from the same header. for (auto *I : SpeculativelyMovedExts) { Value *HeadOfChain = I->getOperand(0); SeenChainsForSExt[HeadOfChain] = Inst; } return false; } if (!AllSeenFirst && !UnhandledExts.empty()) for (auto *VisitedSExt : UnhandledExts) { if (RemovedInsts.count(VisitedSExt)) continue; TypePromotionTransaction TPT(RemovedInsts); SmallVector Exts; SmallVector Chains; Exts.push_back(VisitedSExt); bool HasPromoted = tryToPromoteExts(TPT, Exts, Chains); TPT.commit(); if (HasPromoted) Promoted = true; for (auto *I : Chains) { Value *HeadOfChain = I->getOperand(0); // Mark this as handled. SeenChainsForSExt[HeadOfChain] = nullptr; ValToSExtendedUses[HeadOfChain].push_back(I); } } return Promoted; } bool CodeGenPrepare::optimizeExtUses(Instruction *I) { BasicBlock *DefBB = I->getParent(); // If the result of a {s|z}ext and its source are both live out, rewrite all // other uses of the source with result of extension. Value *Src = I->getOperand(0); if (Src->hasOneUse()) return false; // Only do this xform if truncating is free. if (!TLI->isTruncateFree(I->getType(), Src->getType())) return false; // Only safe to perform the optimization if the source is also defined in // this block. if (!isa(Src) || DefBB != cast(Src)->getParent()) return false; bool DefIsLiveOut = false; for (User *U : I->users()) { Instruction *UI = cast(U); // Figure out which BB this ext is used in. BasicBlock *UserBB = UI->getParent(); if (UserBB == DefBB) continue; DefIsLiveOut = true; break; } if (!DefIsLiveOut) return false; // Make sure none of the uses are PHI nodes. for (User *U : Src->users()) { Instruction *UI = cast(U); BasicBlock *UserBB = UI->getParent(); if (UserBB == DefBB) continue; // Be conservative. We don't want this xform to end up introducing // reloads just before load / store instructions. if (isa(UI) || isa(UI) || isa(UI)) return false; } // InsertedTruncs - Only insert one trunc in each block once. DenseMap InsertedTruncs; bool MadeChange = false; for (Use &U : Src->uses()) { Instruction *User = cast(U.getUser()); // Figure out which BB this ext is used in. BasicBlock *UserBB = User->getParent(); if (UserBB == DefBB) continue; // Both src and def are live in this block. Rewrite the use. Instruction *&InsertedTrunc = InsertedTruncs[UserBB]; if (!InsertedTrunc) { BasicBlock::iterator InsertPt = UserBB->getFirstInsertionPt(); assert(InsertPt != UserBB->end()); InsertedTrunc = new TruncInst(I, Src->getType(), ""); InsertedTrunc->insertBefore(*UserBB, InsertPt); InsertedInsts.insert(InsertedTrunc); } // Replace a use of the {s|z}ext source with a use of the result. U = InsertedTrunc; ++NumExtUses; MadeChange = true; } return MadeChange; } // Find loads whose uses only use some of the loaded value's bits. Add an "and" // just after the load if the target can fold this into one extload instruction, // with the hope of eliminating some of the other later "and" instructions using // the loaded value. "and"s that are made trivially redundant by the insertion // of the new "and" are removed by this function, while others (e.g. those whose // path from the load goes through a phi) are left for isel to potentially // remove. // // For example: // // b0: // x = load i32 // ... // b1: // y = and x, 0xff // z = use y // // becomes: // // b0: // x = load i32 // x' = and x, 0xff // ... // b1: // z = use x' // // whereas: // // b0: // x1 = load i32 // ... // b1: // x2 = load i32 // ... // b2: // x = phi x1, x2 // y = and x, 0xff // // becomes (after a call to optimizeLoadExt for each load): // // b0: // x1 = load i32 // x1' = and x1, 0xff // ... // b1: // x2 = load i32 // x2' = and x2, 0xff // ... // b2: // x = phi x1', x2' // y = and x, 0xff bool CodeGenPrepare::optimizeLoadExt(LoadInst *Load) { if (!Load->isSimple() || !Load->getType()->isIntOrPtrTy()) return false; // Skip loads we've already transformed. if (Load->hasOneUse() && InsertedInsts.count(cast(*Load->user_begin()))) return false; // Look at all uses of Load, looking through phis, to determine how many bits // of the loaded value are needed. SmallVector WorkList; SmallPtrSet Visited; SmallVector AndsToMaybeRemove; for (auto *U : Load->users()) WorkList.push_back(cast(U)); EVT LoadResultVT = TLI->getValueType(*DL, Load->getType()); unsigned BitWidth = LoadResultVT.getSizeInBits(); // If the BitWidth is 0, do not try to optimize the type if (BitWidth == 0) return false; APInt DemandBits(BitWidth, 0); APInt WidestAndBits(BitWidth, 0); while (!WorkList.empty()) { Instruction *I = WorkList.pop_back_val(); // Break use-def graph loops. if (!Visited.insert(I).second) continue; // For a PHI node, push all of its users. if (auto *Phi = dyn_cast(I)) { for (auto *U : Phi->users()) WorkList.push_back(cast(U)); continue; } switch (I->getOpcode()) { case Instruction::And: { auto *AndC = dyn_cast(I->getOperand(1)); if (!AndC) return false; APInt AndBits = AndC->getValue(); DemandBits |= AndBits; // Keep track of the widest and mask we see. if (AndBits.ugt(WidestAndBits)) WidestAndBits = AndBits; if (AndBits == WidestAndBits && I->getOperand(0) == Load) AndsToMaybeRemove.push_back(I); break; } case Instruction::Shl: { auto *ShlC = dyn_cast(I->getOperand(1)); if (!ShlC) return false; uint64_t ShiftAmt = ShlC->getLimitedValue(BitWidth - 1); DemandBits.setLowBits(BitWidth - ShiftAmt); break; } case Instruction::Trunc: { EVT TruncVT = TLI->getValueType(*DL, I->getType()); unsigned TruncBitWidth = TruncVT.getSizeInBits(); DemandBits.setLowBits(TruncBitWidth); break; } default: return false; } } uint32_t ActiveBits = DemandBits.getActiveBits(); // Avoid hoisting (and (load x) 1) since it is unlikely to be folded by the // target even if isLoadExtLegal says an i1 EXTLOAD is valid. For example, // for the AArch64 target isLoadExtLegal(ZEXTLOAD, i32, i1) returns true, but // (and (load x) 1) is not matched as a single instruction, rather as a LDR // followed by an AND. // TODO: Look into removing this restriction by fixing backends to either // return false for isLoadExtLegal for i1 or have them select this pattern to // a single instruction. // // Also avoid hoisting if we didn't see any ands with the exact DemandBits // mask, since these are the only ands that will be removed by isel. if (ActiveBits <= 1 || !DemandBits.isMask(ActiveBits) || WidestAndBits != DemandBits) return false; LLVMContext &Ctx = Load->getType()->getContext(); Type *TruncTy = Type::getIntNTy(Ctx, ActiveBits); EVT TruncVT = TLI->getValueType(*DL, TruncTy); // Reject cases that won't be matched as extloads. if (!LoadResultVT.bitsGT(TruncVT) || !TruncVT.isRound() || !TLI->isLoadExtLegal(ISD::ZEXTLOAD, LoadResultVT, TruncVT)) return false; IRBuilder<> Builder(Load->getNextNonDebugInstruction()); auto *NewAnd = cast( Builder.CreateAnd(Load, ConstantInt::get(Ctx, DemandBits))); // Mark this instruction as "inserted by CGP", so that other // optimizations don't touch it. InsertedInsts.insert(NewAnd); // Replace all uses of load with new and (except for the use of load in the // new and itself). replaceAllUsesWith(Load, NewAnd, FreshBBs, IsHugeFunc); NewAnd->setOperand(0, Load); // Remove any and instructions that are now redundant. for (auto *And : AndsToMaybeRemove) // Check that the and mask is the same as the one we decided to put on the // new and. if (cast(And->getOperand(1))->getValue() == DemandBits) { replaceAllUsesWith(And, NewAnd, FreshBBs, IsHugeFunc); if (&*CurInstIterator == And) CurInstIterator = std::next(And->getIterator()); And->eraseFromParent(); ++NumAndUses; } ++NumAndsAdded; return true; } /// Check if V (an operand of a select instruction) is an expensive instruction /// that is only used once. static bool sinkSelectOperand(const TargetTransformInfo *TTI, Value *V) { auto *I = dyn_cast(V); // If it's safe to speculatively execute, then it should not have side // effects; therefore, it's safe to sink and possibly *not* execute. return I && I->hasOneUse() && isSafeToSpeculativelyExecute(I) && TTI->isExpensiveToSpeculativelyExecute(I); } /// Returns true if a SelectInst should be turned into an explicit branch. static bool isFormingBranchFromSelectProfitable(const TargetTransformInfo *TTI, const TargetLowering *TLI, SelectInst *SI) { // If even a predictable select is cheap, then a branch can't be cheaper. if (!TLI->isPredictableSelectExpensive()) return false; // FIXME: This should use the same heuristics as IfConversion to determine // whether a select is better represented as a branch. // If metadata tells us that the select condition is obviously predictable, // then we want to replace the select with a branch. uint64_t TrueWeight, FalseWeight; if (extractBranchWeights(*SI, TrueWeight, FalseWeight)) { uint64_t Max = std::max(TrueWeight, FalseWeight); uint64_t Sum = TrueWeight + FalseWeight; if (Sum != 0) { auto Probability = BranchProbability::getBranchProbability(Max, Sum); if (Probability > TTI->getPredictableBranchThreshold()) return true; } } CmpInst *Cmp = dyn_cast(SI->getCondition()); // If a branch is predictable, an out-of-order CPU can avoid blocking on its // comparison condition. If the compare has more than one use, there's // probably another cmov or setcc around, so it's not worth emitting a branch. if (!Cmp || !Cmp->hasOneUse()) return false; // If either operand of the select is expensive and only needed on one side // of the select, we should form a branch. if (sinkSelectOperand(TTI, SI->getTrueValue()) || sinkSelectOperand(TTI, SI->getFalseValue())) return true; return false; } /// If \p isTrue is true, return the true value of \p SI, otherwise return /// false value of \p SI. If the true/false value of \p SI is defined by any /// select instructions in \p Selects, look through the defining select /// instruction until the true/false value is not defined in \p Selects. static Value * getTrueOrFalseValue(SelectInst *SI, bool isTrue, const SmallPtrSet &Selects) { Value *V = nullptr; for (SelectInst *DefSI = SI; DefSI != nullptr && Selects.count(DefSI); DefSI = dyn_cast(V)) { assert(DefSI->getCondition() == SI->getCondition() && "The condition of DefSI does not match with SI"); V = (isTrue ? DefSI->getTrueValue() : DefSI->getFalseValue()); } assert(V && "Failed to get select true/false value"); return V; } bool CodeGenPrepare::optimizeShiftInst(BinaryOperator *Shift) { assert(Shift->isShift() && "Expected a shift"); // If this is (1) a vector shift, (2) shifts by scalars are cheaper than // general vector shifts, and (3) the shift amount is a select-of-splatted // values, hoist the shifts before the select: // shift Op0, (select Cond, TVal, FVal) --> // select Cond, (shift Op0, TVal), (shift Op0, FVal) // // This is inverting a generic IR transform when we know that the cost of a // general vector shift is more than the cost of 2 shift-by-scalars. // We can't do this effectively in SDAG because we may not be able to // determine if the select operands are splats from within a basic block. Type *Ty = Shift->getType(); if (!Ty->isVectorTy() || !TLI->isVectorShiftByScalarCheap(Ty)) return false; Value *Cond, *TVal, *FVal; if (!match(Shift->getOperand(1), m_OneUse(m_Select(m_Value(Cond), m_Value(TVal), m_Value(FVal))))) return false; if (!isSplatValue(TVal) || !isSplatValue(FVal)) return false; IRBuilder<> Builder(Shift); BinaryOperator::BinaryOps Opcode = Shift->getOpcode(); Value *NewTVal = Builder.CreateBinOp(Opcode, Shift->getOperand(0), TVal); Value *NewFVal = Builder.CreateBinOp(Opcode, Shift->getOperand(0), FVal); Value *NewSel = Builder.CreateSelect(Cond, NewTVal, NewFVal); replaceAllUsesWith(Shift, NewSel, FreshBBs, IsHugeFunc); Shift->eraseFromParent(); return true; } bool CodeGenPrepare::optimizeFunnelShift(IntrinsicInst *Fsh) { Intrinsic::ID Opcode = Fsh->getIntrinsicID(); assert((Opcode == Intrinsic::fshl || Opcode == Intrinsic::fshr) && "Expected a funnel shift"); // If this is (1) a vector funnel shift, (2) shifts by scalars are cheaper // than general vector shifts, and (3) the shift amount is select-of-splatted // values, hoist the funnel shifts before the select: // fsh Op0, Op1, (select Cond, TVal, FVal) --> // select Cond, (fsh Op0, Op1, TVal), (fsh Op0, Op1, FVal) // // This is inverting a generic IR transform when we know that the cost of a // general vector shift is more than the cost of 2 shift-by-scalars. // We can't do this effectively in SDAG because we may not be able to // determine if the select operands are splats from within a basic block. Type *Ty = Fsh->getType(); if (!Ty->isVectorTy() || !TLI->isVectorShiftByScalarCheap(Ty)) return false; Value *Cond, *TVal, *FVal; if (!match(Fsh->getOperand(2), m_OneUse(m_Select(m_Value(Cond), m_Value(TVal), m_Value(FVal))))) return false; if (!isSplatValue(TVal) || !isSplatValue(FVal)) return false; IRBuilder<> Builder(Fsh); Value *X = Fsh->getOperand(0), *Y = Fsh->getOperand(1); Value *NewTVal = Builder.CreateIntrinsic(Opcode, Ty, {X, Y, TVal}); Value *NewFVal = Builder.CreateIntrinsic(Opcode, Ty, {X, Y, FVal}); Value *NewSel = Builder.CreateSelect(Cond, NewTVal, NewFVal); replaceAllUsesWith(Fsh, NewSel, FreshBBs, IsHugeFunc); Fsh->eraseFromParent(); return true; } /// If we have a SelectInst that will likely profit from branch prediction, /// turn it into a branch. bool CodeGenPrepare::optimizeSelectInst(SelectInst *SI) { if (DisableSelectToBranch) return false; // If the SelectOptimize pass is enabled, selects have already been optimized. if (!getCGPassBuilderOption().DisableSelectOptimize) return false; // Find all consecutive select instructions that share the same condition. SmallVector ASI; ASI.push_back(SI); for (BasicBlock::iterator It = ++BasicBlock::iterator(SI); It != SI->getParent()->end(); ++It) { SelectInst *I = dyn_cast(&*It); if (I && SI->getCondition() == I->getCondition()) { ASI.push_back(I); } else { break; } } SelectInst *LastSI = ASI.back(); // Increment the current iterator to skip all the rest of select instructions // because they will be either "not lowered" or "all lowered" to branch. CurInstIterator = std::next(LastSI->getIterator()); // Examine debug-info attached to the consecutive select instructions. They // won't be individually optimised by optimizeInst, so we need to perform // DbgVariableRecord maintenence here instead. for (SelectInst *SI : ArrayRef(ASI).drop_front()) fixupDbgVariableRecordsOnInst(*SI); bool VectorCond = !SI->getCondition()->getType()->isIntegerTy(1); // Can we convert the 'select' to CF ? if (VectorCond || SI->getMetadata(LLVMContext::MD_unpredictable)) return false; TargetLowering::SelectSupportKind SelectKind; if (SI->getType()->isVectorTy()) SelectKind = TargetLowering::ScalarCondVectorVal; else SelectKind = TargetLowering::ScalarValSelect; if (TLI->isSelectSupported(SelectKind) && (!isFormingBranchFromSelectProfitable(TTI, TLI, SI) || OptSize || llvm::shouldOptimizeForSize(SI->getParent(), PSI, BFI.get()))) return false; // The DominatorTree needs to be rebuilt by any consumers after this // transformation. We simply reset here rather than setting the ModifiedDT // flag to avoid restarting the function walk in runOnFunction for each // select optimized. DT.reset(); // Transform a sequence like this: // start: // %cmp = cmp uge i32 %a, %b // %sel = select i1 %cmp, i32 %c, i32 %d // // Into: // start: // %cmp = cmp uge i32 %a, %b // %cmp.frozen = freeze %cmp // br i1 %cmp.frozen, label %select.true, label %select.false // select.true: // br label %select.end // select.false: // br label %select.end // select.end: // %sel = phi i32 [ %c, %select.true ], [ %d, %select.false ] // // %cmp should be frozen, otherwise it may introduce undefined behavior. // In addition, we may sink instructions that produce %c or %d from // the entry block into the destination(s) of the new branch. // If the true or false blocks do not contain a sunken instruction, that // block and its branch may be optimized away. In that case, one side of the // first branch will point directly to select.end, and the corresponding PHI // predecessor block will be the start block. // Collect values that go on the true side and the values that go on the false // side. SmallVector TrueInstrs, FalseInstrs; for (SelectInst *SI : ASI) { if (Value *V = SI->getTrueValue(); sinkSelectOperand(TTI, V)) TrueInstrs.push_back(cast(V)); if (Value *V = SI->getFalseValue(); sinkSelectOperand(TTI, V)) FalseInstrs.push_back(cast(V)); } // Split the select block, according to how many (if any) values go on each // side. BasicBlock *StartBlock = SI->getParent(); BasicBlock::iterator SplitPt = std::next(BasicBlock::iterator(LastSI)); // We should split before any debug-info. SplitPt.setHeadBit(true); IRBuilder<> IB(SI); auto *CondFr = IB.CreateFreeze(SI->getCondition(), SI->getName() + ".frozen"); BasicBlock *TrueBlock = nullptr; BasicBlock *FalseBlock = nullptr; BasicBlock *EndBlock = nullptr; BranchInst *TrueBranch = nullptr; BranchInst *FalseBranch = nullptr; if (TrueInstrs.size() == 0) { FalseBranch = cast(SplitBlockAndInsertIfElse( CondFr, SplitPt, false, nullptr, nullptr, LI)); FalseBlock = FalseBranch->getParent(); EndBlock = cast(FalseBranch->getOperand(0)); } else if (FalseInstrs.size() == 0) { TrueBranch = cast(SplitBlockAndInsertIfThen( CondFr, SplitPt, false, nullptr, nullptr, LI)); TrueBlock = TrueBranch->getParent(); EndBlock = cast(TrueBranch->getOperand(0)); } else { Instruction *ThenTerm = nullptr; Instruction *ElseTerm = nullptr; SplitBlockAndInsertIfThenElse(CondFr, SplitPt, &ThenTerm, &ElseTerm, nullptr, nullptr, LI); TrueBranch = cast(ThenTerm); FalseBranch = cast(ElseTerm); TrueBlock = TrueBranch->getParent(); FalseBlock = FalseBranch->getParent(); EndBlock = cast(TrueBranch->getOperand(0)); } EndBlock->setName("select.end"); if (TrueBlock) TrueBlock->setName("select.true.sink"); if (FalseBlock) FalseBlock->setName(FalseInstrs.size() == 0 ? "select.false" : "select.false.sink"); if (IsHugeFunc) { if (TrueBlock) FreshBBs.insert(TrueBlock); if (FalseBlock) FreshBBs.insert(FalseBlock); FreshBBs.insert(EndBlock); } BFI->setBlockFreq(EndBlock, BFI->getBlockFreq(StartBlock)); static const unsigned MD[] = { LLVMContext::MD_prof, LLVMContext::MD_unpredictable, LLVMContext::MD_make_implicit, LLVMContext::MD_dbg}; StartBlock->getTerminator()->copyMetadata(*SI, MD); // Sink expensive instructions into the conditional blocks to avoid executing // them speculatively. for (Instruction *I : TrueInstrs) I->moveBefore(TrueBranch); for (Instruction *I : FalseInstrs) I->moveBefore(FalseBranch); // If we did not create a new block for one of the 'true' or 'false' paths // of the condition, it means that side of the branch goes to the end block // directly and the path originates from the start block from the point of // view of the new PHI. if (TrueBlock == nullptr) TrueBlock = StartBlock; else if (FalseBlock == nullptr) FalseBlock = StartBlock; SmallPtrSet INS; INS.insert(ASI.begin(), ASI.end()); // Use reverse iterator because later select may use the value of the // earlier select, and we need to propagate value through earlier select // to get the PHI operand. for (SelectInst *SI : llvm::reverse(ASI)) { // The select itself is replaced with a PHI Node. PHINode *PN = PHINode::Create(SI->getType(), 2, ""); PN->insertBefore(EndBlock->begin()); PN->takeName(SI); PN->addIncoming(getTrueOrFalseValue(SI, true, INS), TrueBlock); PN->addIncoming(getTrueOrFalseValue(SI, false, INS), FalseBlock); PN->setDebugLoc(SI->getDebugLoc()); replaceAllUsesWith(SI, PN, FreshBBs, IsHugeFunc); SI->eraseFromParent(); INS.erase(SI); ++NumSelectsExpanded; } // Instruct OptimizeBlock to skip to the next block. CurInstIterator = StartBlock->end(); return true; } /// Some targets only accept certain types for splat inputs. For example a VDUP /// in MVE takes a GPR (integer) register, and the instruction that incorporate /// a VDUP (such as a VADD qd, qm, rm) also require a gpr register. bool CodeGenPrepare::optimizeShuffleVectorInst(ShuffleVectorInst *SVI) { // Accept shuf(insertelem(undef/poison, val, 0), undef/poison, <0,0,..>) only if (!match(SVI, m_Shuffle(m_InsertElt(m_Undef(), m_Value(), m_ZeroInt()), m_Undef(), m_ZeroMask()))) return false; Type *NewType = TLI->shouldConvertSplatType(SVI); if (!NewType) return false; auto *SVIVecType = cast(SVI->getType()); assert(!NewType->isVectorTy() && "Expected a scalar type!"); assert(NewType->getScalarSizeInBits() == SVIVecType->getScalarSizeInBits() && "Expected a type of the same size!"); auto *NewVecType = FixedVectorType::get(NewType, SVIVecType->getNumElements()); // Create a bitcast (shuffle (insert (bitcast(..)))) IRBuilder<> Builder(SVI->getContext()); Builder.SetInsertPoint(SVI); Value *BC1 = Builder.CreateBitCast( cast(SVI->getOperand(0))->getOperand(1), NewType); Value *Shuffle = Builder.CreateVectorSplat(NewVecType->getNumElements(), BC1); Value *BC2 = Builder.CreateBitCast(Shuffle, SVIVecType); replaceAllUsesWith(SVI, BC2, FreshBBs, IsHugeFunc); RecursivelyDeleteTriviallyDeadInstructions( SVI, TLInfo, nullptr, [&](Value *V) { removeAllAssertingVHReferences(V); }); // Also hoist the bitcast up to its operand if it they are not in the same // block. if (auto *BCI = dyn_cast(BC1)) if (auto *Op = dyn_cast(BCI->getOperand(0))) if (BCI->getParent() != Op->getParent() && !isa(Op) && !Op->isTerminator() && !Op->isEHPad()) BCI->moveAfter(Op); return true; } bool CodeGenPrepare::tryToSinkFreeOperands(Instruction *I) { // If the operands of I can be folded into a target instruction together with // I, duplicate and sink them. SmallVector OpsToSink; if (!TLI->shouldSinkOperands(I, OpsToSink)) return false; // OpsToSink can contain multiple uses in a use chain (e.g. // (%u1 with %u1 = shufflevector), (%u2 with %u2 = zext %u1)). The dominating // uses must come first, so we process the ops in reverse order so as to not // create invalid IR. BasicBlock *TargetBB = I->getParent(); bool Changed = false; SmallVector ToReplace; Instruction *InsertPoint = I; DenseMap InstOrdering; unsigned long InstNumber = 0; for (const auto &I : *TargetBB) InstOrdering[&I] = InstNumber++; for (Use *U : reverse(OpsToSink)) { auto *UI = cast(U->get()); if (isa(UI)) continue; if (UI->getParent() == TargetBB) { if (InstOrdering[UI] < InstOrdering[InsertPoint]) InsertPoint = UI; continue; } ToReplace.push_back(U); } SetVector MaybeDead; DenseMap NewInstructions; for (Use *U : ToReplace) { auto *UI = cast(U->get()); Instruction *NI = UI->clone(); if (IsHugeFunc) { // Now we clone an instruction, its operands' defs may sink to this BB // now. So we put the operands defs' BBs into FreshBBs to do optimization. for (unsigned I = 0; I < NI->getNumOperands(); ++I) { auto *OpDef = dyn_cast(NI->getOperand(I)); if (!OpDef) continue; FreshBBs.insert(OpDef->getParent()); } } NewInstructions[UI] = NI; MaybeDead.insert(UI); LLVM_DEBUG(dbgs() << "Sinking " << *UI << " to user " << *I << "\n"); NI->insertBefore(InsertPoint); InsertPoint = NI; InsertedInsts.insert(NI); // Update the use for the new instruction, making sure that we update the // sunk instruction uses, if it is part of a chain that has already been // sunk. Instruction *OldI = cast(U->getUser()); if (NewInstructions.count(OldI)) NewInstructions[OldI]->setOperand(U->getOperandNo(), NI); else U->set(NI); Changed = true; } // Remove instructions that are dead after sinking. for (auto *I : MaybeDead) { if (!I->hasNUsesOrMore(1)) { LLVM_DEBUG(dbgs() << "Removing dead instruction: " << *I << "\n"); I->eraseFromParent(); } } return Changed; } bool CodeGenPrepare::optimizeSwitchType(SwitchInst *SI) { Value *Cond = SI->getCondition(); Type *OldType = Cond->getType(); LLVMContext &Context = Cond->getContext(); EVT OldVT = TLI->getValueType(*DL, OldType); MVT RegType = TLI->getPreferredSwitchConditionType(Context, OldVT); unsigned RegWidth = RegType.getSizeInBits(); if (RegWidth <= cast(OldType)->getBitWidth()) return false; // If the register width is greater than the type width, expand the condition // of the switch instruction and each case constant to the width of the // register. By widening the type of the switch condition, subsequent // comparisons (for case comparisons) will not need to be extended to the // preferred register width, so we will potentially eliminate N-1 extends, // where N is the number of cases in the switch. auto *NewType = Type::getIntNTy(Context, RegWidth); // Extend the switch condition and case constants using the target preferred // extend unless the switch condition is a function argument with an extend // attribute. In that case, we can avoid an unnecessary mask/extension by // matching the argument extension instead. Instruction::CastOps ExtType = Instruction::ZExt; // Some targets prefer SExt over ZExt. if (TLI->isSExtCheaperThanZExt(OldVT, RegType)) ExtType = Instruction::SExt; if (auto *Arg = dyn_cast(Cond)) { if (Arg->hasSExtAttr()) ExtType = Instruction::SExt; if (Arg->hasZExtAttr()) ExtType = Instruction::ZExt; } auto *ExtInst = CastInst::Create(ExtType, Cond, NewType); ExtInst->insertBefore(SI); ExtInst->setDebugLoc(SI->getDebugLoc()); SI->setCondition(ExtInst); for (auto Case : SI->cases()) { const APInt &NarrowConst = Case.getCaseValue()->getValue(); APInt WideConst = (ExtType == Instruction::ZExt) ? NarrowConst.zext(RegWidth) : NarrowConst.sext(RegWidth); Case.setValue(ConstantInt::get(Context, WideConst)); } return true; } bool CodeGenPrepare::optimizeSwitchPhiConstants(SwitchInst *SI) { // The SCCP optimization tends to produce code like this: // switch(x) { case 42: phi(42, ...) } // Materializing the constant for the phi-argument needs instructions; So we // change the code to: // switch(x) { case 42: phi(x, ...) } Value *Condition = SI->getCondition(); // Avoid endless loop in degenerate case. if (isa(*Condition)) return false; bool Changed = false; BasicBlock *SwitchBB = SI->getParent(); Type *ConditionType = Condition->getType(); for (const SwitchInst::CaseHandle &Case : SI->cases()) { ConstantInt *CaseValue = Case.getCaseValue(); BasicBlock *CaseBB = Case.getCaseSuccessor(); // Set to true if we previously checked that `CaseBB` is only reached by // a single case from this switch. bool CheckedForSinglePred = false; for (PHINode &PHI : CaseBB->phis()) { Type *PHIType = PHI.getType(); // If ZExt is free then we can also catch patterns like this: // switch((i32)x) { case 42: phi((i64)42, ...); } // and replace `(i64)42` with `zext i32 %x to i64`. bool TryZExt = PHIType->isIntegerTy() && PHIType->getIntegerBitWidth() > ConditionType->getIntegerBitWidth() && TLI->isZExtFree(ConditionType, PHIType); if (PHIType == ConditionType || TryZExt) { // Set to true to skip this case because of multiple preds. bool SkipCase = false; Value *Replacement = nullptr; for (unsigned I = 0, E = PHI.getNumIncomingValues(); I != E; I++) { Value *PHIValue = PHI.getIncomingValue(I); if (PHIValue != CaseValue) { if (!TryZExt) continue; ConstantInt *PHIValueInt = dyn_cast(PHIValue); if (!PHIValueInt || PHIValueInt->getValue() != CaseValue->getValue().zext(PHIType->getIntegerBitWidth())) continue; } if (PHI.getIncomingBlock(I) != SwitchBB) continue; // We cannot optimize if there are multiple case labels jumping to // this block. This check may get expensive when there are many // case labels so we test for it last. if (!CheckedForSinglePred) { CheckedForSinglePred = true; if (SI->findCaseDest(CaseBB) == nullptr) { SkipCase = true; break; } } if (Replacement == nullptr) { if (PHIValue == CaseValue) { Replacement = Condition; } else { IRBuilder<> Builder(SI); Replacement = Builder.CreateZExt(Condition, PHIType); } } PHI.setIncomingValue(I, Replacement); Changed = true; } if (SkipCase) break; } } } return Changed; } bool CodeGenPrepare::optimizeSwitchInst(SwitchInst *SI) { bool Changed = optimizeSwitchType(SI); Changed |= optimizeSwitchPhiConstants(SI); return Changed; } namespace { /// Helper class to promote a scalar operation to a vector one. /// This class is used to move downward extractelement transition. /// E.g., /// a = vector_op <2 x i32> /// b = extractelement <2 x i32> a, i32 0 /// c = scalar_op b /// store c /// /// => /// a = vector_op <2 x i32> /// c = vector_op a (equivalent to scalar_op on the related lane) /// * d = extractelement <2 x i32> c, i32 0 /// * store d /// Assuming both extractelement and store can be combine, we get rid of the /// transition. class VectorPromoteHelper { /// DataLayout associated with the current module. const DataLayout &DL; /// Used to perform some checks on the legality of vector operations. const TargetLowering &TLI; /// Used to estimated the cost of the promoted chain. const TargetTransformInfo &TTI; /// The transition being moved downwards. Instruction *Transition; /// The sequence of instructions to be promoted. SmallVector InstsToBePromoted; /// Cost of combining a store and an extract. unsigned StoreExtractCombineCost; /// Instruction that will be combined with the transition. Instruction *CombineInst = nullptr; /// The instruction that represents the current end of the transition. /// Since we are faking the promotion until we reach the end of the chain /// of computation, we need a way to get the current end of the transition. Instruction *getEndOfTransition() const { if (InstsToBePromoted.empty()) return Transition; return InstsToBePromoted.back(); } /// Return the index of the original value in the transition. /// E.g., for "extractelement <2 x i32> c, i32 1" the original value, /// c, is at index 0. unsigned getTransitionOriginalValueIdx() const { assert(isa(Transition) && "Other kind of transitions are not supported yet"); return 0; } /// Return the index of the index in the transition. /// E.g., for "extractelement <2 x i32> c, i32 0" the index /// is at index 1. unsigned getTransitionIdx() const { assert(isa(Transition) && "Other kind of transitions are not supported yet"); return 1; } /// Get the type of the transition. /// This is the type of the original value. /// E.g., for "extractelement <2 x i32> c, i32 1" the type of the /// transition is <2 x i32>. Type *getTransitionType() const { return Transition->getOperand(getTransitionOriginalValueIdx())->getType(); } /// Promote \p ToBePromoted by moving \p Def downward through. /// I.e., we have the following sequence: /// Def = Transition a to /// b = ToBePromoted Def, ... /// => /// b = ToBePromoted a, ... /// Def = Transition ToBePromoted to void promoteImpl(Instruction *ToBePromoted); /// Check whether or not it is profitable to promote all the /// instructions enqueued to be promoted. bool isProfitableToPromote() { Value *ValIdx = Transition->getOperand(getTransitionOriginalValueIdx()); unsigned Index = isa(ValIdx) ? cast(ValIdx)->getZExtValue() : -1; Type *PromotedType = getTransitionType(); StoreInst *ST = cast(CombineInst); unsigned AS = ST->getPointerAddressSpace(); // Check if this store is supported. if (!TLI.allowsMisalignedMemoryAccesses( TLI.getValueType(DL, ST->getValueOperand()->getType()), AS, ST->getAlign())) { // If this is not supported, there is no way we can combine // the extract with the store. return false; } // The scalar chain of computation has to pay for the transition // scalar to vector. // The vector chain has to account for the combining cost. enum TargetTransformInfo::TargetCostKind CostKind = TargetTransformInfo::TCK_RecipThroughput; InstructionCost ScalarCost = TTI.getVectorInstrCost(*Transition, PromotedType, CostKind, Index); InstructionCost VectorCost = StoreExtractCombineCost; for (const auto &Inst : InstsToBePromoted) { // Compute the cost. // By construction, all instructions being promoted are arithmetic ones. // Moreover, one argument is a constant that can be viewed as a splat // constant. Value *Arg0 = Inst->getOperand(0); bool IsArg0Constant = isa(Arg0) || isa(Arg0) || isa(Arg0); TargetTransformInfo::OperandValueInfo Arg0Info, Arg1Info; if (IsArg0Constant) Arg0Info.Kind = TargetTransformInfo::OK_UniformConstantValue; else Arg1Info.Kind = TargetTransformInfo::OK_UniformConstantValue; ScalarCost += TTI.getArithmeticInstrCost( Inst->getOpcode(), Inst->getType(), CostKind, Arg0Info, Arg1Info); VectorCost += TTI.getArithmeticInstrCost(Inst->getOpcode(), PromotedType, CostKind, Arg0Info, Arg1Info); } LLVM_DEBUG( dbgs() << "Estimated cost of computation to be promoted:\nScalar: " << ScalarCost << "\nVector: " << VectorCost << '\n'); return ScalarCost > VectorCost; } /// Generate a constant vector with \p Val with the same /// number of elements as the transition. /// \p UseSplat defines whether or not \p Val should be replicated /// across the whole vector. /// In other words, if UseSplat == true, we generate , /// otherwise we generate a vector with as many undef as possible: /// where \p Val is only /// used at the index of the extract. Value *getConstantVector(Constant *Val, bool UseSplat) const { unsigned ExtractIdx = std::numeric_limits::max(); if (!UseSplat) { // If we cannot determine where the constant must be, we have to // use a splat constant. Value *ValExtractIdx = Transition->getOperand(getTransitionIdx()); if (ConstantInt *CstVal = dyn_cast(ValExtractIdx)) ExtractIdx = CstVal->getSExtValue(); else UseSplat = true; } ElementCount EC = cast(getTransitionType())->getElementCount(); if (UseSplat) return ConstantVector::getSplat(EC, Val); if (!EC.isScalable()) { SmallVector ConstVec; UndefValue *UndefVal = UndefValue::get(Val->getType()); for (unsigned Idx = 0; Idx != EC.getKnownMinValue(); ++Idx) { if (Idx == ExtractIdx) ConstVec.push_back(Val); else ConstVec.push_back(UndefVal); } return ConstantVector::get(ConstVec); } else llvm_unreachable( "Generate scalable vector for non-splat is unimplemented"); } /// Check if promoting to a vector type an operand at \p OperandIdx /// in \p Use can trigger undefined behavior. static bool canCauseUndefinedBehavior(const Instruction *Use, unsigned OperandIdx) { // This is not safe to introduce undef when the operand is on // the right hand side of a division-like instruction. if (OperandIdx != 1) return false; switch (Use->getOpcode()) { default: return false; case Instruction::SDiv: case Instruction::UDiv: case Instruction::SRem: case Instruction::URem: return true; case Instruction::FDiv: case Instruction::FRem: return !Use->hasNoNaNs(); } llvm_unreachable(nullptr); } public: VectorPromoteHelper(const DataLayout &DL, const TargetLowering &TLI, const TargetTransformInfo &TTI, Instruction *Transition, unsigned CombineCost) : DL(DL), TLI(TLI), TTI(TTI), Transition(Transition), StoreExtractCombineCost(CombineCost) { assert(Transition && "Do not know how to promote null"); } /// Check if we can promote \p ToBePromoted to \p Type. bool canPromote(const Instruction *ToBePromoted) const { // We could support CastInst too. return isa(ToBePromoted); } /// Check if it is profitable to promote \p ToBePromoted /// by moving downward the transition through. bool shouldPromote(const Instruction *ToBePromoted) const { // Promote only if all the operands can be statically expanded. // Indeed, we do not want to introduce any new kind of transitions. for (const Use &U : ToBePromoted->operands()) { const Value *Val = U.get(); if (Val == getEndOfTransition()) { // If the use is a division and the transition is on the rhs, // we cannot promote the operation, otherwise we may create a // division by zero. if (canCauseUndefinedBehavior(ToBePromoted, U.getOperandNo())) return false; continue; } if (!isa(Val) && !isa(Val) && !isa(Val)) return false; } // Check that the resulting operation is legal. int ISDOpcode = TLI.InstructionOpcodeToISD(ToBePromoted->getOpcode()); if (!ISDOpcode) return false; return StressStoreExtract || TLI.isOperationLegalOrCustom( ISDOpcode, TLI.getValueType(DL, getTransitionType(), true)); } /// Check whether or not \p Use can be combined /// with the transition. /// I.e., is it possible to do Use(Transition) => AnotherUse? bool canCombine(const Instruction *Use) { return isa(Use); } /// Record \p ToBePromoted as part of the chain to be promoted. void enqueueForPromotion(Instruction *ToBePromoted) { InstsToBePromoted.push_back(ToBePromoted); } /// Set the instruction that will be combined with the transition. void recordCombineInstruction(Instruction *ToBeCombined) { assert(canCombine(ToBeCombined) && "Unsupported instruction to combine"); CombineInst = ToBeCombined; } /// Promote all the instructions enqueued for promotion if it is /// is profitable. /// \return True if the promotion happened, false otherwise. bool promote() { // Check if there is something to promote. // Right now, if we do not have anything to combine with, // we assume the promotion is not profitable. if (InstsToBePromoted.empty() || !CombineInst) return false; // Check cost. if (!StressStoreExtract && !isProfitableToPromote()) return false; // Promote. for (auto &ToBePromoted : InstsToBePromoted) promoteImpl(ToBePromoted); InstsToBePromoted.clear(); return true; } }; } // end anonymous namespace void VectorPromoteHelper::promoteImpl(Instruction *ToBePromoted) { // At this point, we know that all the operands of ToBePromoted but Def // can be statically promoted. // For Def, we need to use its parameter in ToBePromoted: // b = ToBePromoted ty1 a // Def = Transition ty1 b to ty2 // Move the transition down. // 1. Replace all uses of the promoted operation by the transition. // = ... b => = ... Def. assert(ToBePromoted->getType() == Transition->getType() && "The type of the result of the transition does not match " "the final type"); ToBePromoted->replaceAllUsesWith(Transition); // 2. Update the type of the uses. // b = ToBePromoted ty2 Def => b = ToBePromoted ty1 Def. Type *TransitionTy = getTransitionType(); ToBePromoted->mutateType(TransitionTy); // 3. Update all the operands of the promoted operation with promoted // operands. // b = ToBePromoted ty1 Def => b = ToBePromoted ty1 a. for (Use &U : ToBePromoted->operands()) { Value *Val = U.get(); Value *NewVal = nullptr; if (Val == Transition) NewVal = Transition->getOperand(getTransitionOriginalValueIdx()); else if (isa(Val) || isa(Val) || isa(Val)) { // Use a splat constant if it is not safe to use undef. NewVal = getConstantVector( cast(Val), isa(Val) || canCauseUndefinedBehavior(ToBePromoted, U.getOperandNo())); } else llvm_unreachable("Did you modified shouldPromote and forgot to update " "this?"); ToBePromoted->setOperand(U.getOperandNo(), NewVal); } Transition->moveAfter(ToBePromoted); Transition->setOperand(getTransitionOriginalValueIdx(), ToBePromoted); } /// Some targets can do store(extractelement) with one instruction. /// Try to push the extractelement towards the stores when the target /// has this feature and this is profitable. bool CodeGenPrepare::optimizeExtractElementInst(Instruction *Inst) { unsigned CombineCost = std::numeric_limits::max(); if (DisableStoreExtract || (!StressStoreExtract && !TLI->canCombineStoreAndExtract(Inst->getOperand(0)->getType(), Inst->getOperand(1), CombineCost))) return false; // At this point we know that Inst is a vector to scalar transition. // Try to move it down the def-use chain, until: // - We can combine the transition with its single use // => we got rid of the transition. // - We escape the current basic block // => we would need to check that we are moving it at a cheaper place and // we do not do that for now. BasicBlock *Parent = Inst->getParent(); LLVM_DEBUG(dbgs() << "Found an interesting transition: " << *Inst << '\n'); VectorPromoteHelper VPH(*DL, *TLI, *TTI, Inst, CombineCost); // If the transition has more than one use, assume this is not going to be // beneficial. while (Inst->hasOneUse()) { Instruction *ToBePromoted = cast(*Inst->user_begin()); LLVM_DEBUG(dbgs() << "Use: " << *ToBePromoted << '\n'); if (ToBePromoted->getParent() != Parent) { LLVM_DEBUG(dbgs() << "Instruction to promote is in a different block (" << ToBePromoted->getParent()->getName() << ") than the transition (" << Parent->getName() << ").\n"); return false; } if (VPH.canCombine(ToBePromoted)) { LLVM_DEBUG(dbgs() << "Assume " << *Inst << '\n' << "will be combined with: " << *ToBePromoted << '\n'); VPH.recordCombineInstruction(ToBePromoted); bool Changed = VPH.promote(); NumStoreExtractExposed += Changed; return Changed; } LLVM_DEBUG(dbgs() << "Try promoting.\n"); if (!VPH.canPromote(ToBePromoted) || !VPH.shouldPromote(ToBePromoted)) return false; LLVM_DEBUG(dbgs() << "Promoting is possible... Enqueue for promotion!\n"); VPH.enqueueForPromotion(ToBePromoted); Inst = ToBePromoted; } return false; } /// For the instruction sequence of store below, F and I values /// are bundled together as an i64 value before being stored into memory. /// Sometimes it is more efficient to generate separate stores for F and I, /// which can remove the bitwise instructions or sink them to colder places. /// /// (store (or (zext (bitcast F to i32) to i64), /// (shl (zext I to i64), 32)), addr) --> /// (store F, addr) and (store I, addr+4) /// /// Similarly, splitting for other merged store can also be beneficial, like: /// For pair of {i32, i32}, i64 store --> two i32 stores. /// For pair of {i32, i16}, i64 store --> two i32 stores. /// For pair of {i16, i16}, i32 store --> two i16 stores. /// For pair of {i16, i8}, i32 store --> two i16 stores. /// For pair of {i8, i8}, i16 store --> two i8 stores. /// /// We allow each target to determine specifically which kind of splitting is /// supported. /// /// The store patterns are commonly seen from the simple code snippet below /// if only std::make_pair(...) is sroa transformed before inlined into hoo. /// void goo(const std::pair &); /// hoo() { /// ... /// goo(std::make_pair(tmp, ftmp)); /// ... /// } /// /// Although we already have similar splitting in DAG Combine, we duplicate /// it in CodeGenPrepare to catch the case in which pattern is across /// multiple BBs. The logic in DAG Combine is kept to catch case generated /// during code expansion. static bool splitMergedValStore(StoreInst &SI, const DataLayout &DL, const TargetLowering &TLI) { // Handle simple but common cases only. Type *StoreType = SI.getValueOperand()->getType(); // The code below assumes shifting a value by , // whereas scalable vectors would have to be shifted by // <2log(vscale) + number of bits> in order to store the // low/high parts. Bailing out for now. if (StoreType->isScalableTy()) return false; if (!DL.typeSizeEqualsStoreSize(StoreType) || DL.getTypeSizeInBits(StoreType) == 0) return false; unsigned HalfValBitSize = DL.getTypeSizeInBits(StoreType) / 2; Type *SplitStoreType = Type::getIntNTy(SI.getContext(), HalfValBitSize); if (!DL.typeSizeEqualsStoreSize(SplitStoreType)) return false; // Don't split the store if it is volatile. if (SI.isVolatile()) return false; // Match the following patterns: // (store (or (zext LValue to i64), // (shl (zext HValue to i64), 32)), HalfValBitSize) // or // (store (or (shl (zext HValue to i64), 32)), HalfValBitSize) // (zext LValue to i64), // Expect both operands of OR and the first operand of SHL have only // one use. Value *LValue, *HValue; if (!match(SI.getValueOperand(), m_c_Or(m_OneUse(m_ZExt(m_Value(LValue))), m_OneUse(m_Shl(m_OneUse(m_ZExt(m_Value(HValue))), m_SpecificInt(HalfValBitSize)))))) return false; // Check LValue and HValue are int with size less or equal than 32. if (!LValue->getType()->isIntegerTy() || DL.getTypeSizeInBits(LValue->getType()) > HalfValBitSize || !HValue->getType()->isIntegerTy() || DL.getTypeSizeInBits(HValue->getType()) > HalfValBitSize) return false; // If LValue/HValue is a bitcast instruction, use the EVT before bitcast // as the input of target query. auto *LBC = dyn_cast(LValue); auto *HBC = dyn_cast(HValue); EVT LowTy = LBC ? EVT::getEVT(LBC->getOperand(0)->getType()) : EVT::getEVT(LValue->getType()); EVT HighTy = HBC ? EVT::getEVT(HBC->getOperand(0)->getType()) : EVT::getEVT(HValue->getType()); if (!ForceSplitStore && !TLI.isMultiStoresCheaperThanBitsMerge(LowTy, HighTy)) return false; // Start to split store. IRBuilder<> Builder(SI.getContext()); Builder.SetInsertPoint(&SI); // If LValue/HValue is a bitcast in another BB, create a new one in current // BB so it may be merged with the splitted stores by dag combiner. if (LBC && LBC->getParent() != SI.getParent()) LValue = Builder.CreateBitCast(LBC->getOperand(0), LBC->getType()); if (HBC && HBC->getParent() != SI.getParent()) HValue = Builder.CreateBitCast(HBC->getOperand(0), HBC->getType()); bool IsLE = SI.getDataLayout().isLittleEndian(); auto CreateSplitStore = [&](Value *V, bool Upper) { V = Builder.CreateZExtOrBitCast(V, SplitStoreType); Value *Addr = SI.getPointerOperand(); Align Alignment = SI.getAlign(); const bool IsOffsetStore = (IsLE && Upper) || (!IsLE && !Upper); if (IsOffsetStore) { Addr = Builder.CreateGEP( SplitStoreType, Addr, ConstantInt::get(Type::getInt32Ty(SI.getContext()), 1)); // When splitting the store in half, naturally one half will retain the // alignment of the original wider store, regardless of whether it was // over-aligned or not, while the other will require adjustment. Alignment = commonAlignment(Alignment, HalfValBitSize / 8); } Builder.CreateAlignedStore(V, Addr, Alignment); }; CreateSplitStore(LValue, false); CreateSplitStore(HValue, true); // Delete the old store. SI.eraseFromParent(); return true; } // Return true if the GEP has two operands, the first operand is of a sequential // type, and the second operand is a constant. static bool GEPSequentialConstIndexed(GetElementPtrInst *GEP) { gep_type_iterator I = gep_type_begin(*GEP); return GEP->getNumOperands() == 2 && I.isSequential() && isa(GEP->getOperand(1)); } // Try unmerging GEPs to reduce liveness interference (register pressure) across // IndirectBr edges. Since IndirectBr edges tend to touch on many blocks, // reducing liveness interference across those edges benefits global register // allocation. Currently handles only certain cases. // // For example, unmerge %GEPI and %UGEPI as below. // // ---------- BEFORE ---------- // SrcBlock: // ... // %GEPIOp = ... // ... // %GEPI = gep %GEPIOp, Idx // ... // indirectbr ... [ label %DstB0, label %DstB1, ... label %DstBi ... ] // (* %GEPI is alive on the indirectbr edges due to other uses ahead) // (* %GEPIOp is alive on the indirectbr edges only because of it's used by // %UGEPI) // // DstB0: ... (there may be a gep similar to %UGEPI to be unmerged) // DstB1: ... (there may be a gep similar to %UGEPI to be unmerged) // ... // // DstBi: // ... // %UGEPI = gep %GEPIOp, UIdx // ... // --------------------------- // // ---------- AFTER ---------- // SrcBlock: // ... (same as above) // (* %GEPI is still alive on the indirectbr edges) // (* %GEPIOp is no longer alive on the indirectbr edges as a result of the // unmerging) // ... // // DstBi: // ... // %UGEPI = gep %GEPI, (UIdx-Idx) // ... // --------------------------- // // The register pressure on the IndirectBr edges is reduced because %GEPIOp is // no longer alive on them. // // We try to unmerge GEPs here in CodGenPrepare, as opposed to limiting merging // of GEPs in the first place in InstCombiner::visitGetElementPtrInst() so as // not to disable further simplications and optimizations as a result of GEP // merging. // // Note this unmerging may increase the length of the data flow critical path // (the path from %GEPIOp to %UGEPI would go through %GEPI), which is a tradeoff // between the register pressure and the length of data-flow critical // path. Restricting this to the uncommon IndirectBr case would minimize the // impact of potentially longer critical path, if any, and the impact on compile // time. static bool tryUnmergingGEPsAcrossIndirectBr(GetElementPtrInst *GEPI, const TargetTransformInfo *TTI) { BasicBlock *SrcBlock = GEPI->getParent(); // Check that SrcBlock ends with an IndirectBr. If not, give up. The common // (non-IndirectBr) cases exit early here. if (!isa(SrcBlock->getTerminator())) return false; // Check that GEPI is a simple gep with a single constant index. if (!GEPSequentialConstIndexed(GEPI)) return false; ConstantInt *GEPIIdx = cast(GEPI->getOperand(1)); // Check that GEPI is a cheap one. if (TTI->getIntImmCost(GEPIIdx->getValue(), GEPIIdx->getType(), TargetTransformInfo::TCK_SizeAndLatency) > TargetTransformInfo::TCC_Basic) return false; Value *GEPIOp = GEPI->getOperand(0); // Check that GEPIOp is an instruction that's also defined in SrcBlock. if (!isa(GEPIOp)) return false; auto *GEPIOpI = cast(GEPIOp); if (GEPIOpI->getParent() != SrcBlock) return false; // Check that GEP is used outside the block, meaning it's alive on the // IndirectBr edge(s). if (llvm::none_of(GEPI->users(), [&](User *Usr) { if (auto *I = dyn_cast(Usr)) { if (I->getParent() != SrcBlock) { return true; } } return false; })) return false; // The second elements of the GEP chains to be unmerged. std::vector UGEPIs; // Check each user of GEPIOp to check if unmerging would make GEPIOp not alive // on IndirectBr edges. for (User *Usr : GEPIOp->users()) { if (Usr == GEPI) continue; // Check if Usr is an Instruction. If not, give up. if (!isa(Usr)) return false; auto *UI = cast(Usr); // Check if Usr in the same block as GEPIOp, which is fine, skip. if (UI->getParent() == SrcBlock) continue; // Check if Usr is a GEP. If not, give up. if (!isa(Usr)) return false; auto *UGEPI = cast(Usr); // Check if UGEPI is a simple gep with a single constant index and GEPIOp is // the pointer operand to it. If so, record it in the vector. If not, give // up. if (!GEPSequentialConstIndexed(UGEPI)) return false; if (UGEPI->getOperand(0) != GEPIOp) return false; if (UGEPI->getSourceElementType() != GEPI->getSourceElementType()) return false; if (GEPIIdx->getType() != cast(UGEPI->getOperand(1))->getType()) return false; ConstantInt *UGEPIIdx = cast(UGEPI->getOperand(1)); if (TTI->getIntImmCost(UGEPIIdx->getValue(), UGEPIIdx->getType(), TargetTransformInfo::TCK_SizeAndLatency) > TargetTransformInfo::TCC_Basic) return false; UGEPIs.push_back(UGEPI); } if (UGEPIs.size() == 0) return false; // Check the materializing cost of (Uidx-Idx). for (GetElementPtrInst *UGEPI : UGEPIs) { ConstantInt *UGEPIIdx = cast(UGEPI->getOperand(1)); APInt NewIdx = UGEPIIdx->getValue() - GEPIIdx->getValue(); InstructionCost ImmCost = TTI->getIntImmCost( NewIdx, GEPIIdx->getType(), TargetTransformInfo::TCK_SizeAndLatency); if (ImmCost > TargetTransformInfo::TCC_Basic) return false; } // Now unmerge between GEPI and UGEPIs. for (GetElementPtrInst *UGEPI : UGEPIs) { UGEPI->setOperand(0, GEPI); ConstantInt *UGEPIIdx = cast(UGEPI->getOperand(1)); Constant *NewUGEPIIdx = ConstantInt::get( GEPIIdx->getType(), UGEPIIdx->getValue() - GEPIIdx->getValue()); UGEPI->setOperand(1, NewUGEPIIdx); // If GEPI is not inbounds but UGEPI is inbounds, change UGEPI to not // inbounds to avoid UB. if (!GEPI->isInBounds()) { UGEPI->setIsInBounds(false); } } // After unmerging, verify that GEPIOp is actually only used in SrcBlock (not // alive on IndirectBr edges). assert(llvm::none_of(GEPIOp->users(), [&](User *Usr) { return cast(Usr)->getParent() != SrcBlock; }) && "GEPIOp is used outside SrcBlock"); return true; } static bool optimizeBranch(BranchInst *Branch, const TargetLowering &TLI, SmallSet &FreshBBs, bool IsHugeFunc) { // Try and convert // %c = icmp ult %x, 8 // br %c, bla, blb // %tc = lshr %x, 3 // to // %tc = lshr %x, 3 // %c = icmp eq %tc, 0 // br %c, bla, blb // Creating the cmp to zero can be better for the backend, especially if the // lshr produces flags that can be used automatically. if (!TLI.preferZeroCompareBranch() || !Branch->isConditional()) return false; ICmpInst *Cmp = dyn_cast(Branch->getCondition()); if (!Cmp || !isa(Cmp->getOperand(1)) || !Cmp->hasOneUse()) return false; Value *X = Cmp->getOperand(0); APInt CmpC = cast(Cmp->getOperand(1))->getValue(); for (auto *U : X->users()) { Instruction *UI = dyn_cast(U); // A quick dominance check if (!UI || (UI->getParent() != Branch->getParent() && UI->getParent() != Branch->getSuccessor(0) && UI->getParent() != Branch->getSuccessor(1)) || (UI->getParent() != Branch->getParent() && !UI->getParent()->getSinglePredecessor())) continue; if (CmpC.isPowerOf2() && Cmp->getPredicate() == ICmpInst::ICMP_ULT && match(UI, m_Shr(m_Specific(X), m_SpecificInt(CmpC.logBase2())))) { IRBuilder<> Builder(Branch); if (UI->getParent() != Branch->getParent()) UI->moveBefore(Branch); UI->dropPoisonGeneratingFlags(); Value *NewCmp = Builder.CreateCmp(ICmpInst::ICMP_EQ, UI, ConstantInt::get(UI->getType(), 0)); LLVM_DEBUG(dbgs() << "Converting " << *Cmp << "\n"); LLVM_DEBUG(dbgs() << " to compare on zero: " << *NewCmp << "\n"); replaceAllUsesWith(Cmp, NewCmp, FreshBBs, IsHugeFunc); return true; } if (Cmp->isEquality() && (match(UI, m_Add(m_Specific(X), m_SpecificInt(-CmpC))) || match(UI, m_Sub(m_Specific(X), m_SpecificInt(CmpC))))) { IRBuilder<> Builder(Branch); if (UI->getParent() != Branch->getParent()) UI->moveBefore(Branch); UI->dropPoisonGeneratingFlags(); Value *NewCmp = Builder.CreateCmp(Cmp->getPredicate(), UI, ConstantInt::get(UI->getType(), 0)); LLVM_DEBUG(dbgs() << "Converting " << *Cmp << "\n"); LLVM_DEBUG(dbgs() << " to compare on zero: " << *NewCmp << "\n"); replaceAllUsesWith(Cmp, NewCmp, FreshBBs, IsHugeFunc); return true; } } return false; } bool CodeGenPrepare::optimizeInst(Instruction *I, ModifyDT &ModifiedDT) { bool AnyChange = false; AnyChange = fixupDbgVariableRecordsOnInst(*I); // Bail out if we inserted the instruction to prevent optimizations from // stepping on each other's toes. if (InsertedInsts.count(I)) return AnyChange; // TODO: Move into the switch on opcode below here. if (PHINode *P = dyn_cast(I)) { // It is possible for very late stage optimizations (such as SimplifyCFG) // to introduce PHI nodes too late to be cleaned up. If we detect such a // trivial PHI, go ahead and zap it here. if (Value *V = simplifyInstruction(P, {*DL, TLInfo})) { LargeOffsetGEPMap.erase(P); replaceAllUsesWith(P, V, FreshBBs, IsHugeFunc); P->eraseFromParent(); ++NumPHIsElim; return true; } return AnyChange; } if (CastInst *CI = dyn_cast(I)) { // If the source of the cast is a constant, then this should have // already been constant folded. The only reason NOT to constant fold // it is if something (e.g. LSR) was careful to place the constant // evaluation in a block other than then one that uses it (e.g. to hoist // the address of globals out of a loop). If this is the case, we don't // want to forward-subst the cast. if (isa(CI->getOperand(0))) return AnyChange; if (OptimizeNoopCopyExpression(CI, *TLI, *DL)) return true; if ((isa(I) || isa(I) || isa(I) || isa(I)) && TLI->optimizeExtendOrTruncateConversion( I, LI->getLoopFor(I->getParent()), *TTI)) return true; if (isa(I) || isa(I)) { /// Sink a zext or sext into its user blocks if the target type doesn't /// fit in one register if (TLI->getTypeAction(CI->getContext(), TLI->getValueType(*DL, CI->getType())) == TargetLowering::TypeExpandInteger) { return SinkCast(CI); } else { if (TLI->optimizeExtendOrTruncateConversion( I, LI->getLoopFor(I->getParent()), *TTI)) return true; bool MadeChange = optimizeExt(I); return MadeChange | optimizeExtUses(I); } } return AnyChange; } if (auto *Cmp = dyn_cast(I)) if (optimizeCmp(Cmp, ModifiedDT)) return true; if (LoadInst *LI = dyn_cast(I)) { LI->setMetadata(LLVMContext::MD_invariant_group, nullptr); bool Modified = optimizeLoadExt(LI); unsigned AS = LI->getPointerAddressSpace(); Modified |= optimizeMemoryInst(I, I->getOperand(0), LI->getType(), AS); return Modified; } if (StoreInst *SI = dyn_cast(I)) { if (splitMergedValStore(*SI, *DL, *TLI)) return true; SI->setMetadata(LLVMContext::MD_invariant_group, nullptr); unsigned AS = SI->getPointerAddressSpace(); return optimizeMemoryInst(I, SI->getOperand(1), SI->getOperand(0)->getType(), AS); } if (AtomicRMWInst *RMW = dyn_cast(I)) { unsigned AS = RMW->getPointerAddressSpace(); return optimizeMemoryInst(I, RMW->getPointerOperand(), RMW->getType(), AS); } if (AtomicCmpXchgInst *CmpX = dyn_cast(I)) { unsigned AS = CmpX->getPointerAddressSpace(); return optimizeMemoryInst(I, CmpX->getPointerOperand(), CmpX->getCompareOperand()->getType(), AS); } BinaryOperator *BinOp = dyn_cast(I); if (BinOp && BinOp->getOpcode() == Instruction::And && EnableAndCmpSinking && sinkAndCmp0Expression(BinOp, *TLI, InsertedInsts)) return true; // TODO: Move this into the switch on opcode - it handles shifts already. if (BinOp && (BinOp->getOpcode() == Instruction::AShr || BinOp->getOpcode() == Instruction::LShr)) { ConstantInt *CI = dyn_cast(BinOp->getOperand(1)); if (CI && TLI->hasExtractBitsInsn()) if (OptimizeExtractBits(BinOp, CI, *TLI, *DL)) return true; } if (GetElementPtrInst *GEPI = dyn_cast(I)) { if (GEPI->hasAllZeroIndices()) { /// The GEP operand must be a pointer, so must its result -> BitCast Instruction *NC = new BitCastInst(GEPI->getOperand(0), GEPI->getType(), GEPI->getName(), GEPI->getIterator()); NC->setDebugLoc(GEPI->getDebugLoc()); replaceAllUsesWith(GEPI, NC, FreshBBs, IsHugeFunc); RecursivelyDeleteTriviallyDeadInstructions( GEPI, TLInfo, nullptr, [&](Value *V) { removeAllAssertingVHReferences(V); }); ++NumGEPsElim; optimizeInst(NC, ModifiedDT); return true; } if (tryUnmergingGEPsAcrossIndirectBr(GEPI, TTI)) { return true; } } if (FreezeInst *FI = dyn_cast(I)) { // freeze(icmp a, const)) -> icmp (freeze a), const // This helps generate efficient conditional jumps. Instruction *CmpI = nullptr; if (ICmpInst *II = dyn_cast(FI->getOperand(0))) CmpI = II; else if (FCmpInst *F = dyn_cast(FI->getOperand(0))) CmpI = F->getFastMathFlags().none() ? F : nullptr; if (CmpI && CmpI->hasOneUse()) { auto Op0 = CmpI->getOperand(0), Op1 = CmpI->getOperand(1); bool Const0 = isa(Op0) || isa(Op0) || isa(Op0); bool Const1 = isa(Op1) || isa(Op1) || isa(Op1); if (Const0 || Const1) { if (!Const0 || !Const1) { auto *F = new FreezeInst(Const0 ? Op1 : Op0, "", CmpI->getIterator()); F->takeName(FI); CmpI->setOperand(Const0 ? 1 : 0, F); } replaceAllUsesWith(FI, CmpI, FreshBBs, IsHugeFunc); FI->eraseFromParent(); return true; } } return AnyChange; } if (tryToSinkFreeOperands(I)) return true; switch (I->getOpcode()) { case Instruction::Shl: case Instruction::LShr: case Instruction::AShr: return optimizeShiftInst(cast(I)); case Instruction::Call: return optimizeCallInst(cast(I), ModifiedDT); case Instruction::Select: return optimizeSelectInst(cast(I)); case Instruction::ShuffleVector: return optimizeShuffleVectorInst(cast(I)); case Instruction::Switch: return optimizeSwitchInst(cast(I)); case Instruction::ExtractElement: return optimizeExtractElementInst(cast(I)); case Instruction::Br: return optimizeBranch(cast(I), *TLI, FreshBBs, IsHugeFunc); } return AnyChange; } /// Given an OR instruction, check to see if this is a bitreverse /// idiom. If so, insert the new intrinsic and return true. bool CodeGenPrepare::makeBitReverse(Instruction &I) { if (!I.getType()->isIntegerTy() || !TLI->isOperationLegalOrCustom(ISD::BITREVERSE, TLI->getValueType(*DL, I.getType(), true))) return false; SmallVector Insts; if (!recognizeBSwapOrBitReverseIdiom(&I, false, true, Insts)) return false; Instruction *LastInst = Insts.back(); replaceAllUsesWith(&I, LastInst, FreshBBs, IsHugeFunc); RecursivelyDeleteTriviallyDeadInstructions( &I, TLInfo, nullptr, [&](Value *V) { removeAllAssertingVHReferences(V); }); return true; } // In this pass we look for GEP and cast instructions that are used // across basic blocks and rewrite them to improve basic-block-at-a-time // selection. bool CodeGenPrepare::optimizeBlock(BasicBlock &BB, ModifyDT &ModifiedDT) { SunkAddrs.clear(); bool MadeChange = false; do { CurInstIterator = BB.begin(); ModifiedDT = ModifyDT::NotModifyDT; while (CurInstIterator != BB.end()) { MadeChange |= optimizeInst(&*CurInstIterator++, ModifiedDT); if (ModifiedDT != ModifyDT::NotModifyDT) { // For huge function we tend to quickly go though the inner optmization // opportunities in the BB. So we go back to the BB head to re-optimize // each instruction instead of go back to the function head. if (IsHugeFunc) { DT.reset(); getDT(*BB.getParent()); break; } else { return true; } } } } while (ModifiedDT == ModifyDT::ModifyInstDT); bool MadeBitReverse = true; while (MadeBitReverse) { MadeBitReverse = false; for (auto &I : reverse(BB)) { if (makeBitReverse(I)) { MadeBitReverse = MadeChange = true; break; } } } MadeChange |= dupRetToEnableTailCallOpts(&BB, ModifiedDT); return MadeChange; } // Some CGP optimizations may move or alter what's computed in a block. Check // whether a dbg.value intrinsic could be pointed at a more appropriate operand. bool CodeGenPrepare::fixupDbgValue(Instruction *I) { assert(isa(I)); DbgValueInst &DVI = *cast(I); // Does this dbg.value refer to a sunk address calculation? bool AnyChange = false; SmallDenseSet LocationOps(DVI.location_ops().begin(), DVI.location_ops().end()); for (Value *Location : LocationOps) { WeakTrackingVH SunkAddrVH = SunkAddrs[Location]; Value *SunkAddr = SunkAddrVH.pointsToAliveValue() ? SunkAddrVH : nullptr; if (SunkAddr) { // Point dbg.value at locally computed address, which should give the best // opportunity to be accurately lowered. This update may change the type // of pointer being referred to; however this makes no difference to // debugging information, and we can't generate bitcasts that may affect // codegen. DVI.replaceVariableLocationOp(Location, SunkAddr); AnyChange = true; } } return AnyChange; } bool CodeGenPrepare::fixupDbgVariableRecordsOnInst(Instruction &I) { bool AnyChange = false; for (DbgVariableRecord &DVR : filterDbgVars(I.getDbgRecordRange())) AnyChange |= fixupDbgVariableRecord(DVR); return AnyChange; } // FIXME: should updating debug-info really cause the "changed" flag to fire, // which can cause a function to be reprocessed? bool CodeGenPrepare::fixupDbgVariableRecord(DbgVariableRecord &DVR) { if (DVR.Type != DbgVariableRecord::LocationType::Value && DVR.Type != DbgVariableRecord::LocationType::Assign) return false; // Does this DbgVariableRecord refer to a sunk address calculation? bool AnyChange = false; SmallDenseSet LocationOps(DVR.location_ops().begin(), DVR.location_ops().end()); for (Value *Location : LocationOps) { WeakTrackingVH SunkAddrVH = SunkAddrs[Location]; Value *SunkAddr = SunkAddrVH.pointsToAliveValue() ? SunkAddrVH : nullptr; if (SunkAddr) { // Point dbg.value at locally computed address, which should give the best // opportunity to be accurately lowered. This update may change the type // of pointer being referred to; however this makes no difference to // debugging information, and we can't generate bitcasts that may affect // codegen. DVR.replaceVariableLocationOp(Location, SunkAddr); AnyChange = true; } } return AnyChange; } static void DbgInserterHelper(DbgValueInst *DVI, Instruction *VI) { DVI->removeFromParent(); if (isa(VI)) DVI->insertBefore(&*VI->getParent()->getFirstInsertionPt()); else DVI->insertAfter(VI); } static void DbgInserterHelper(DbgVariableRecord *DVR, Instruction *VI) { DVR->removeFromParent(); BasicBlock *VIBB = VI->getParent(); if (isa(VI)) VIBB->insertDbgRecordBefore(DVR, VIBB->getFirstInsertionPt()); else VIBB->insertDbgRecordAfter(DVR, VI); } // A llvm.dbg.value may be using a value before its definition, due to // optimizations in this pass and others. Scan for such dbg.values, and rescue // them by moving the dbg.value to immediately after the value definition. // FIXME: Ideally this should never be necessary, and this has the potential // to re-order dbg.value intrinsics. bool CodeGenPrepare::placeDbgValues(Function &F) { bool MadeChange = false; DominatorTree DT(F); auto DbgProcessor = [&](auto *DbgItem, Instruction *Position) { SmallVector VIs; for (Value *V : DbgItem->location_ops()) if (Instruction *VI = dyn_cast_or_null(V)) VIs.push_back(VI); // This item may depend on multiple instructions, complicating any // potential sink. This block takes the defensive approach, opting to // "undef" the item if it has more than one instruction and any of them do // not dominate iem. for (Instruction *VI : VIs) { if (VI->isTerminator()) continue; // If VI is a phi in a block with an EHPad terminator, we can't insert // after it. if (isa(VI) && VI->getParent()->getTerminator()->isEHPad()) continue; // If the defining instruction dominates the dbg.value, we do not need // to move the dbg.value. if (DT.dominates(VI, Position)) continue; // If we depend on multiple instructions and any of them doesn't // dominate this DVI, we probably can't salvage it: moving it to // after any of the instructions could cause us to lose the others. if (VIs.size() > 1) { LLVM_DEBUG( dbgs() << "Unable to find valid location for Debug Value, undefing:\n" << *DbgItem); DbgItem->setKillLocation(); break; } LLVM_DEBUG(dbgs() << "Moving Debug Value before :\n" << *DbgItem << ' ' << *VI); DbgInserterHelper(DbgItem, VI); MadeChange = true; ++NumDbgValueMoved; } }; for (BasicBlock &BB : F) { for (Instruction &Insn : llvm::make_early_inc_range(BB)) { // Process dbg.value intrinsics. DbgValueInst *DVI = dyn_cast(&Insn); if (DVI) { DbgProcessor(DVI, DVI); continue; } // If this isn't a dbg.value, process any attached DbgVariableRecord // records attached to this instruction. for (DbgVariableRecord &DVR : llvm::make_early_inc_range( filterDbgVars(Insn.getDbgRecordRange()))) { if (DVR.Type != DbgVariableRecord::LocationType::Value) continue; DbgProcessor(&DVR, &Insn); } } } return MadeChange; } // Group scattered pseudo probes in a block to favor SelectionDAG. Scattered // probes can be chained dependencies of other regular DAG nodes and block DAG // combine optimizations. bool CodeGenPrepare::placePseudoProbes(Function &F) { bool MadeChange = false; for (auto &Block : F) { // Move the rest probes to the beginning of the block. auto FirstInst = Block.getFirstInsertionPt(); while (FirstInst != Block.end() && FirstInst->isDebugOrPseudoInst()) ++FirstInst; BasicBlock::iterator I(FirstInst); I++; while (I != Block.end()) { if (auto *II = dyn_cast(I++)) { II->moveBefore(&*FirstInst); MadeChange = true; } } } return MadeChange; } /// Scale down both weights to fit into uint32_t. static void scaleWeights(uint64_t &NewTrue, uint64_t &NewFalse) { uint64_t NewMax = (NewTrue > NewFalse) ? NewTrue : NewFalse; uint32_t Scale = (NewMax / std::numeric_limits::max()) + 1; NewTrue = NewTrue / Scale; NewFalse = NewFalse / Scale; } /// Some targets prefer to split a conditional branch like: /// \code /// %0 = icmp ne i32 %a, 0 /// %1 = icmp ne i32 %b, 0 /// %or.cond = or i1 %0, %1 /// br i1 %or.cond, label %TrueBB, label %FalseBB /// \endcode /// into multiple branch instructions like: /// \code /// bb1: /// %0 = icmp ne i32 %a, 0 /// br i1 %0, label %TrueBB, label %bb2 /// bb2: /// %1 = icmp ne i32 %b, 0 /// br i1 %1, label %TrueBB, label %FalseBB /// \endcode /// This usually allows instruction selection to do even further optimizations /// and combine the compare with the branch instruction. Currently this is /// applied for targets which have "cheap" jump instructions. /// /// FIXME: Remove the (equivalent?) implementation in SelectionDAG. /// bool CodeGenPrepare::splitBranchCondition(Function &F, ModifyDT &ModifiedDT) { if (!TM->Options.EnableFastISel || TLI->isJumpExpensive()) return false; bool MadeChange = false; for (auto &BB : F) { // Does this BB end with the following? // %cond1 = icmp|fcmp|binary instruction ... // %cond2 = icmp|fcmp|binary instruction ... // %cond.or = or|and i1 %cond1, cond2 // br i1 %cond.or label %dest1, label %dest2" Instruction *LogicOp; BasicBlock *TBB, *FBB; if (!match(BB.getTerminator(), m_Br(m_OneUse(m_Instruction(LogicOp)), TBB, FBB))) continue; auto *Br1 = cast(BB.getTerminator()); if (Br1->getMetadata(LLVMContext::MD_unpredictable)) continue; // The merging of mostly empty BB can cause a degenerate branch. if (TBB == FBB) continue; unsigned Opc; Value *Cond1, *Cond2; if (match(LogicOp, m_LogicalAnd(m_OneUse(m_Value(Cond1)), m_OneUse(m_Value(Cond2))))) Opc = Instruction::And; else if (match(LogicOp, m_LogicalOr(m_OneUse(m_Value(Cond1)), m_OneUse(m_Value(Cond2))))) Opc = Instruction::Or; else continue; auto IsGoodCond = [](Value *Cond) { return match( Cond, m_CombineOr(m_Cmp(), m_CombineOr(m_LogicalAnd(m_Value(), m_Value()), m_LogicalOr(m_Value(), m_Value())))); }; if (!IsGoodCond(Cond1) || !IsGoodCond(Cond2)) continue; LLVM_DEBUG(dbgs() << "Before branch condition splitting\n"; BB.dump()); // Create a new BB. auto *TmpBB = BasicBlock::Create(BB.getContext(), BB.getName() + ".cond.split", BB.getParent(), BB.getNextNode()); if (IsHugeFunc) FreshBBs.insert(TmpBB); // Update original basic block by using the first condition directly by the // branch instruction and removing the no longer needed and/or instruction. Br1->setCondition(Cond1); LogicOp->eraseFromParent(); // Depending on the condition we have to either replace the true or the // false successor of the original branch instruction. if (Opc == Instruction::And) Br1->setSuccessor(0, TmpBB); else Br1->setSuccessor(1, TmpBB); // Fill in the new basic block. auto *Br2 = IRBuilder<>(TmpBB).CreateCondBr(Cond2, TBB, FBB); if (auto *I = dyn_cast(Cond2)) { I->removeFromParent(); I->insertBefore(Br2); } // Update PHI nodes in both successors. The original BB needs to be // replaced in one successor's PHI nodes, because the branch comes now from // the newly generated BB (NewBB). In the other successor we need to add one // incoming edge to the PHI nodes, because both branch instructions target // now the same successor. Depending on the original branch condition // (and/or) we have to swap the successors (TrueDest, FalseDest), so that // we perform the correct update for the PHI nodes. // This doesn't change the successor order of the just created branch // instruction (or any other instruction). if (Opc == Instruction::Or) std::swap(TBB, FBB); // Replace the old BB with the new BB. TBB->replacePhiUsesWith(&BB, TmpBB); // Add another incoming edge from the new BB. for (PHINode &PN : FBB->phis()) { auto *Val = PN.getIncomingValueForBlock(&BB); PN.addIncoming(Val, TmpBB); } // Update the branch weights (from SelectionDAGBuilder:: // FindMergedConditions). if (Opc == Instruction::Or) { // Codegen X | Y as: // BB1: // jmp_if_X TBB // jmp TmpBB // TmpBB: // jmp_if_Y TBB // jmp FBB // // We have flexibility in setting Prob for BB1 and Prob for NewBB. // The requirement is that // TrueProb for BB1 + (FalseProb for BB1 * TrueProb for TmpBB) // = TrueProb for original BB. // Assuming the original weights are A and B, one choice is to set BB1's // weights to A and A+2B, and set TmpBB's weights to A and 2B. This choice // assumes that // TrueProb for BB1 == FalseProb for BB1 * TrueProb for TmpBB. // Another choice is to assume TrueProb for BB1 equals to TrueProb for // TmpBB, but the math is more complicated. uint64_t TrueWeight, FalseWeight; if (extractBranchWeights(*Br1, TrueWeight, FalseWeight)) { uint64_t NewTrueWeight = TrueWeight; uint64_t NewFalseWeight = TrueWeight + 2 * FalseWeight; scaleWeights(NewTrueWeight, NewFalseWeight); Br1->setMetadata(LLVMContext::MD_prof, MDBuilder(Br1->getContext()) .createBranchWeights(TrueWeight, FalseWeight, hasBranchWeightOrigin(*Br1))); NewTrueWeight = TrueWeight; NewFalseWeight = 2 * FalseWeight; scaleWeights(NewTrueWeight, NewFalseWeight); Br2->setMetadata(LLVMContext::MD_prof, MDBuilder(Br2->getContext()) .createBranchWeights(TrueWeight, FalseWeight)); } } else { // Codegen X & Y as: // BB1: // jmp_if_X TmpBB // jmp FBB // TmpBB: // jmp_if_Y TBB // jmp FBB // // This requires creation of TmpBB after CurBB. // We have flexibility in setting Prob for BB1 and Prob for TmpBB. // The requirement is that // FalseProb for BB1 + (TrueProb for BB1 * FalseProb for TmpBB) // = FalseProb for original BB. // Assuming the original weights are A and B, one choice is to set BB1's // weights to 2A+B and B, and set TmpBB's weights to 2A and B. This choice // assumes that // FalseProb for BB1 == TrueProb for BB1 * FalseProb for TmpBB. uint64_t TrueWeight, FalseWeight; if (extractBranchWeights(*Br1, TrueWeight, FalseWeight)) { uint64_t NewTrueWeight = 2 * TrueWeight + FalseWeight; uint64_t NewFalseWeight = FalseWeight; scaleWeights(NewTrueWeight, NewFalseWeight); Br1->setMetadata(LLVMContext::MD_prof, MDBuilder(Br1->getContext()) .createBranchWeights(TrueWeight, FalseWeight)); NewTrueWeight = 2 * TrueWeight; NewFalseWeight = FalseWeight; scaleWeights(NewTrueWeight, NewFalseWeight); Br2->setMetadata(LLVMContext::MD_prof, MDBuilder(Br2->getContext()) .createBranchWeights(TrueWeight, FalseWeight)); } } ModifiedDT = ModifyDT::ModifyBBDT; MadeChange = true; LLVM_DEBUG(dbgs() << "After branch condition splitting\n"; BB.dump(); TmpBB->dump()); } return MadeChange; }