//===- FunctionSpecialization.h - Function Specialization -----------------===// // // 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 // //===----------------------------------------------------------------------===// // // Overview: // --------- // Function Specialization is a transformation which propagates the constant // parameters of a function call from the caller to the callee. It is part of // the Inter-Procedural Sparse Conditional Constant Propagation (IPSCCP) pass. // The transformation runs iteratively a number of times which is controlled // by the option `funcspec-max-iters`. Running it multiple times is needed // for specializing recursive functions, but also exposes new opportunities // arising from specializations which return constant values or contain calls // which can be specialized. // // Function Specialization supports propagating constant parameters like // function pointers, literal constants and addresses of global variables. // By propagating function pointers, indirect calls become direct calls. This // exposes inlining opportunities which we would have otherwise missed. That's // why function specialization is run before the inliner in the optimization // pipeline; that is by design. // // Cost Model: // ----------- // The cost model facilitates a utility for estimating the specialization bonus // from propagating a constant argument. This is the InstCostVisitor, a class // that inherits from the InstVisitor. The bonus itself is expressed as codesize // and latency savings. Codesize savings means the amount of code that becomes // dead in the specialization from propagating the constant, whereas latency // savings represents the cycles we are saving from replacing instructions with // constant values. The InstCostVisitor overrides a set of `visit*` methods to // be able to handle different types of instructions. These attempt to constant- // fold the instruction in which case a constant is returned and propagated // further. // // Function pointers are not handled by the InstCostVisitor. They are treated // separately as they could expose inlining opportunities via indirect call // promotion. The inlining bonus contributes to the total specialization score. // // For a specialization to be profitable its bonus needs to exceed a minimum // threshold. There are three options for controlling the threshold which are // expressed as percentages of the original function size: // * funcspec-min-codesize-savings // * funcspec-min-latency-savings // * funcspec-min-inlining-bonus // There's also an option for controlling the codesize growth from recursive // specializations. That is `funcspec-max-codesize-growth`. // // Once we have all the potential specializations with their score we need to // choose the best ones, which fit in the module specialization budget. That // is controlled by the option `funcspec-max-clones`. To find the best `NSpec` // specializations we use a max-heap. For more details refer to D139346. // // Ideas: // ------ // - With a function specialization attribute for arguments, we could have // a direct way to steer function specialization, avoiding the cost-model, // and thus control compile-times / code-size. // // - Perhaps a post-inlining function specialization pass could be more // aggressive on literal constants. // // References: // ----------- // 2021 LLVM Dev Mtg “Introducing function specialisation, and can we enable // it by default?”, https://www.youtube.com/watch?v=zJiCjeXgV5Q // //===----------------------------------------------------------------------===// #ifndef LLVM_TRANSFORMS_IPO_FUNCTIONSPECIALIZATION_H #define LLVM_TRANSFORMS_IPO_FUNCTIONSPECIALIZATION_H #include "llvm/Analysis/BlockFrequencyInfo.h" #include "llvm/Analysis/CodeMetrics.h" #include "llvm/Analysis/InlineCost.h" #include "llvm/Analysis/TargetTransformInfo.h" #include "llvm/IR/InstVisitor.h" #include "llvm/Transforms/Scalar/SCCP.h" #include "llvm/Transforms/Utils/Cloning.h" #include "llvm/Transforms/Utils/SCCPSolver.h" #include "llvm/Transforms/Utils/SizeOpts.h" namespace llvm { // Map of potential specializations for each function. The FunctionSpecializer // keeps the discovered specialisation opportunities for the module in a single // vector, where the specialisations of each function form a contiguous range. // This map's value is the beginning and the end of that range. using SpecMap = DenseMap>; // Just a shorter abbreviation to improve indentation. using Cost = InstructionCost; // Map of known constants found during the specialization bonus estimation. using ConstMap = DenseMap; // Specialization signature, used to uniquely designate a specialization within // a function. struct SpecSig { // Hashing support, used to distinguish between ordinary, empty, or tombstone // keys. unsigned Key = 0; SmallVector Args; bool operator==(const SpecSig &Other) const { if (Key != Other.Key) return false; return Args == Other.Args; } friend hash_code hash_value(const SpecSig &S) { return hash_combine(hash_value(S.Key), hash_combine_range(S.Args.begin(), S.Args.end())); } }; // Specialization instance. struct Spec { // Original function. Function *F; // Cloned function, a specialized version of the original one. Function *Clone = nullptr; // Specialization signature. SpecSig Sig; // Profitability of the specialization. unsigned Score; // List of call sites, matching this specialization. SmallVector CallSites; Spec(Function *F, const SpecSig &S, unsigned Score) : F(F), Sig(S), Score(Score) {} Spec(Function *F, const SpecSig &&S, unsigned Score) : F(F), Sig(S), Score(Score) {} }; struct Bonus { unsigned CodeSize = 0; unsigned Latency = 0; Bonus() = default; Bonus(Cost CodeSize, Cost Latency) { int64_t Sz = *CodeSize.getValue(); int64_t Ltc = *Latency.getValue(); assert(Sz >= 0 && Ltc >= 0 && "CodeSize and Latency cannot be negative"); // It is safe to down cast since we know the arguments // cannot be negative and Cost is of type int64_t. this->CodeSize = static_cast(Sz); this->Latency = static_cast(Ltc); } Bonus &operator+=(const Bonus RHS) { CodeSize += RHS.CodeSize; Latency += RHS.Latency; return *this; } Bonus operator+(const Bonus RHS) const { return Bonus(CodeSize + RHS.CodeSize, Latency + RHS.Latency); } bool operator==(const Bonus RHS) const { return CodeSize == RHS.CodeSize && Latency == RHS.Latency; } }; class InstCostVisitor : public InstVisitor { const DataLayout &DL; BlockFrequencyInfo &BFI; TargetTransformInfo &TTI; SCCPSolver &Solver; ConstMap KnownConstants; // Basic blocks known to be unreachable after constant propagation. DenseSet DeadBlocks; // PHI nodes we have visited before. DenseSet VisitedPHIs; // PHI nodes we have visited once without successfully constant folding them. // Once the InstCostVisitor has processed all the specialization arguments, // it should be possible to determine whether those PHIs can be folded // (some of their incoming values may have become constant or dead). SmallVector PendingPHIs; ConstMap::iterator LastVisited; public: InstCostVisitor(const DataLayout &DL, BlockFrequencyInfo &BFI, TargetTransformInfo &TTI, SCCPSolver &Solver) : DL(DL), BFI(BFI), TTI(TTI), Solver(Solver) {} bool isBlockExecutable(BasicBlock *BB) { return Solver.isBlockExecutable(BB) && !DeadBlocks.contains(BB); } Bonus getSpecializationBonus(Argument *A, Constant *C); Bonus getBonusFromPendingPHIs(); private: friend class InstVisitor; static bool canEliminateSuccessor(BasicBlock *BB, BasicBlock *Succ, DenseSet &DeadBlocks); Bonus getUserBonus(Instruction *User, Value *Use = nullptr, Constant *C = nullptr); Cost estimateBasicBlocks(SmallVectorImpl &WorkList); Cost estimateSwitchInst(SwitchInst &I); Cost estimateBranchInst(BranchInst &I); // Transitively Incoming Values (TIV) is a set of Values that can "feed" a // value to the initial PHI-node. It is defined like this: // // * the initial PHI-node belongs to TIV. // // * for every PHI-node in TIV, its operands belong to TIV // // If TIV for the initial PHI-node (P) contains more than one constant or a // value that is not a PHI-node, then P cannot be folded to a constant. // // As soon as we detect these cases, we bail, without constructing the // full TIV. // Otherwise P can be folded to the one constant in TIV. bool discoverTransitivelyIncomingValues(Constant *Const, PHINode *Root, DenseSet &TransitivePHIs); Constant *visitInstruction(Instruction &I) { return nullptr; } Constant *visitPHINode(PHINode &I); Constant *visitFreezeInst(FreezeInst &I); Constant *visitCallBase(CallBase &I); Constant *visitLoadInst(LoadInst &I); Constant *visitGetElementPtrInst(GetElementPtrInst &I); Constant *visitSelectInst(SelectInst &I); Constant *visitCastInst(CastInst &I); Constant *visitCmpInst(CmpInst &I); Constant *visitUnaryOperator(UnaryOperator &I); Constant *visitBinaryOperator(BinaryOperator &I); }; class FunctionSpecializer { /// The IPSCCP Solver. SCCPSolver &Solver; Module &M; /// Analysis manager, needed to invalidate analyses. FunctionAnalysisManager *FAM; /// Analyses used to help determine if a function should be specialized. std::function GetBFI; std::function GetTLI; std::function GetTTI; std::function GetAC; SmallPtrSet Specializations; SmallPtrSet FullySpecialized; DenseMap FunctionMetrics; DenseMap FunctionGrowth; unsigned NGlobals = 0; public: FunctionSpecializer( SCCPSolver &Solver, Module &M, FunctionAnalysisManager *FAM, std::function GetBFI, std::function GetTLI, std::function GetTTI, std::function GetAC) : Solver(Solver), M(M), FAM(FAM), GetBFI(GetBFI), GetTLI(GetTLI), GetTTI(GetTTI), GetAC(GetAC) {} ~FunctionSpecializer(); bool run(); InstCostVisitor getInstCostVisitorFor(Function *F) { auto &BFI = GetBFI(*F); auto &TTI = GetTTI(*F); return InstCostVisitor(M.getDataLayout(), BFI, TTI, Solver); } private: Constant *getPromotableAlloca(AllocaInst *Alloca, CallInst *Call); /// A constant stack value is an AllocaInst that has a single constant /// value stored to it. Return this constant if such an alloca stack value /// is a function argument. Constant *getConstantStackValue(CallInst *Call, Value *Val); /// See if there are any new constant values for the callers of \p F via /// stack variables and promote them to global variables. void promoteConstantStackValues(Function *F); /// Clean up fully specialized functions. void removeDeadFunctions(); /// Remove any ssa_copy intrinsics that may have been introduced. void cleanUpSSA(); /// @brief Find potential specialization opportunities. /// @param F Function to specialize /// @param FuncSize Cost of specializing a function. /// @param AllSpecs A vector to add potential specializations to. /// @param SM A map for a function's specialisation range /// @return True, if any potential specializations were found bool findSpecializations(Function *F, unsigned FuncSize, SmallVectorImpl &AllSpecs, SpecMap &SM); /// Compute the inlining bonus for replacing argument \p A with constant \p C. unsigned getInliningBonus(Argument *A, Constant *C); bool isCandidateFunction(Function *F); /// @brief Create a specialization of \p F and prime the SCCPSolver /// @param F Function to specialize /// @param S Which specialization to create /// @return The new, cloned function Function *createSpecialization(Function *F, const SpecSig &S); /// Determine if it is possible to specialise the function for constant values /// of the formal parameter \p A. bool isArgumentInteresting(Argument *A); /// Check if the value \p V (an actual argument) is a constant or can only /// have a constant value. Return that constant. Constant *getCandidateConstant(Value *V); /// @brief Find and update calls to \p F, which match a specialization /// @param F Orginal function /// @param Begin Start of a range of possibly matching specialisations /// @param End End of a range (exclusive) of possibly matching specialisations void updateCallSites(Function *F, const Spec *Begin, const Spec *End); }; } // namespace llvm #endif // LLVM_TRANSFORMS_IPO_FUNCTIONSPECIALIZATION_H