| File: | src/gnu/usr.bin/clang/libLLVM/../../../llvm/llvm/lib/Transforms/IPO/MergeFunctions.cpp |
| Warning: | line 705, column 23 Called C++ object pointer is null |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
| 1 | //===- MergeFunctions.cpp - Merge identical functions ---------------------===// | |||
| 2 | // | |||
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. | |||
| 4 | // See https://llvm.org/LICENSE.txt for license information. | |||
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception | |||
| 6 | // | |||
| 7 | //===----------------------------------------------------------------------===// | |||
| 8 | // | |||
| 9 | // This pass looks for equivalent functions that are mergable and folds them. | |||
| 10 | // | |||
| 11 | // Order relation is defined on set of functions. It was made through | |||
| 12 | // special function comparison procedure that returns | |||
| 13 | // 0 when functions are equal, | |||
| 14 | // -1 when Left function is less than right function, and | |||
| 15 | // 1 for opposite case. We need total-ordering, so we need to maintain | |||
| 16 | // four properties on the functions set: | |||
| 17 | // a <= a (reflexivity) | |||
| 18 | // if a <= b and b <= a then a = b (antisymmetry) | |||
| 19 | // if a <= b and b <= c then a <= c (transitivity). | |||
| 20 | // for all a and b: a <= b or b <= a (totality). | |||
| 21 | // | |||
| 22 | // Comparison iterates through each instruction in each basic block. | |||
| 23 | // Functions are kept on binary tree. For each new function F we perform | |||
| 24 | // lookup in binary tree. | |||
| 25 | // In practice it works the following way: | |||
| 26 | // -- We define Function* container class with custom "operator<" (FunctionPtr). | |||
| 27 | // -- "FunctionPtr" instances are stored in std::set collection, so every | |||
| 28 | // std::set::insert operation will give you result in log(N) time. | |||
| 29 | // | |||
| 30 | // As an optimization, a hash of the function structure is calculated first, and | |||
| 31 | // two functions are only compared if they have the same hash. This hash is | |||
| 32 | // cheap to compute, and has the property that if function F == G according to | |||
| 33 | // the comparison function, then hash(F) == hash(G). This consistency property | |||
| 34 | // is critical to ensuring all possible merging opportunities are exploited. | |||
| 35 | // Collisions in the hash affect the speed of the pass but not the correctness | |||
| 36 | // or determinism of the resulting transformation. | |||
| 37 | // | |||
| 38 | // When a match is found the functions are folded. If both functions are | |||
| 39 | // overridable, we move the functionality into a new internal function and | |||
| 40 | // leave two overridable thunks to it. | |||
| 41 | // | |||
| 42 | //===----------------------------------------------------------------------===// | |||
| 43 | // | |||
| 44 | // Future work: | |||
| 45 | // | |||
| 46 | // * virtual functions. | |||
| 47 | // | |||
| 48 | // Many functions have their address taken by the virtual function table for | |||
| 49 | // the object they belong to. However, as long as it's only used for a lookup | |||
| 50 | // and call, this is irrelevant, and we'd like to fold such functions. | |||
| 51 | // | |||
| 52 | // * be smarter about bitcasts. | |||
| 53 | // | |||
| 54 | // In order to fold functions, we will sometimes add either bitcast instructions | |||
| 55 | // or bitcast constant expressions. Unfortunately, this can confound further | |||
| 56 | // analysis since the two functions differ where one has a bitcast and the | |||
| 57 | // other doesn't. We should learn to look through bitcasts. | |||
| 58 | // | |||
| 59 | // * Compare complex types with pointer types inside. | |||
| 60 | // * Compare cross-reference cases. | |||
| 61 | // * Compare complex expressions. | |||
| 62 | // | |||
| 63 | // All the three issues above could be described as ability to prove that | |||
| 64 | // fA == fB == fC == fE == fF == fG in example below: | |||
| 65 | // | |||
| 66 | // void fA() { | |||
| 67 | // fB(); | |||
| 68 | // } | |||
| 69 | // void fB() { | |||
| 70 | // fA(); | |||
| 71 | // } | |||
| 72 | // | |||
| 73 | // void fE() { | |||
| 74 | // fF(); | |||
| 75 | // } | |||
| 76 | // void fF() { | |||
| 77 | // fG(); | |||
| 78 | // } | |||
| 79 | // void fG() { | |||
| 80 | // fE(); | |||
| 81 | // } | |||
| 82 | // | |||
| 83 | // Simplest cross-reference case (fA <--> fB) was implemented in previous | |||
| 84 | // versions of MergeFunctions, though it presented only in two function pairs | |||
| 85 | // in test-suite (that counts >50k functions) | |||
| 86 | // Though possibility to detect complex cross-referencing (e.g.: A->B->C->D->A) | |||
| 87 | // could cover much more cases. | |||
| 88 | // | |||
| 89 | //===----------------------------------------------------------------------===// | |||
| 90 | ||||
| 91 | #include "llvm/ADT/ArrayRef.h" | |||
| 92 | #include "llvm/ADT/SmallPtrSet.h" | |||
| 93 | #include "llvm/ADT/SmallVector.h" | |||
| 94 | #include "llvm/ADT/Statistic.h" | |||
| 95 | #include "llvm/IR/Argument.h" | |||
| 96 | #include "llvm/IR/Attributes.h" | |||
| 97 | #include "llvm/IR/BasicBlock.h" | |||
| 98 | #include "llvm/IR/Constant.h" | |||
| 99 | #include "llvm/IR/Constants.h" | |||
| 100 | #include "llvm/IR/DebugInfoMetadata.h" | |||
| 101 | #include "llvm/IR/DebugLoc.h" | |||
| 102 | #include "llvm/IR/DerivedTypes.h" | |||
| 103 | #include "llvm/IR/Function.h" | |||
| 104 | #include "llvm/IR/GlobalValue.h" | |||
| 105 | #include "llvm/IR/IRBuilder.h" | |||
| 106 | #include "llvm/IR/InstrTypes.h" | |||
| 107 | #include "llvm/IR/Instruction.h" | |||
| 108 | #include "llvm/IR/Instructions.h" | |||
| 109 | #include "llvm/IR/IntrinsicInst.h" | |||
| 110 | #include "llvm/IR/Module.h" | |||
| 111 | #include "llvm/IR/Type.h" | |||
| 112 | #include "llvm/IR/Use.h" | |||
| 113 | #include "llvm/IR/User.h" | |||
| 114 | #include "llvm/IR/Value.h" | |||
| 115 | #include "llvm/IR/ValueHandle.h" | |||
| 116 | #include "llvm/IR/ValueMap.h" | |||
| 117 | #include "llvm/InitializePasses.h" | |||
| 118 | #include "llvm/Pass.h" | |||
| 119 | #include "llvm/Support/Casting.h" | |||
| 120 | #include "llvm/Support/CommandLine.h" | |||
| 121 | #include "llvm/Support/Debug.h" | |||
| 122 | #include "llvm/Support/raw_ostream.h" | |||
| 123 | #include "llvm/Transforms/IPO.h" | |||
| 124 | #include "llvm/Transforms/IPO/MergeFunctions.h" | |||
| 125 | #include "llvm/Transforms/Utils/FunctionComparator.h" | |||
| 126 | #include <algorithm> | |||
| 127 | #include <cassert> | |||
| 128 | #include <iterator> | |||
| 129 | #include <set> | |||
| 130 | #include <utility> | |||
| 131 | #include <vector> | |||
| 132 | ||||
| 133 | using namespace llvm; | |||
| 134 | ||||
| 135 | #define DEBUG_TYPE"mergefunc" "mergefunc" | |||
| 136 | ||||
| 137 | STATISTIC(NumFunctionsMerged, "Number of functions merged")static llvm::Statistic NumFunctionsMerged = {"mergefunc", "NumFunctionsMerged" , "Number of functions merged"}; | |||
| 138 | STATISTIC(NumThunksWritten, "Number of thunks generated")static llvm::Statistic NumThunksWritten = {"mergefunc", "NumThunksWritten" , "Number of thunks generated"}; | |||
| 139 | STATISTIC(NumAliasesWritten, "Number of aliases generated")static llvm::Statistic NumAliasesWritten = {"mergefunc", "NumAliasesWritten" , "Number of aliases generated"}; | |||
| 140 | STATISTIC(NumDoubleWeak, "Number of new functions created")static llvm::Statistic NumDoubleWeak = {"mergefunc", "NumDoubleWeak" , "Number of new functions created"}; | |||
| 141 | ||||
| 142 | static cl::opt<unsigned> NumFunctionsForSanityCheck( | |||
| 143 | "mergefunc-sanity", | |||
| 144 | cl::desc("How many functions in module could be used for " | |||
| 145 | "MergeFunctions pass sanity check. " | |||
| 146 | "'0' disables this check. Works only with '-debug' key."), | |||
| 147 | cl::init(0), cl::Hidden); | |||
| 148 | ||||
| 149 | // Under option -mergefunc-preserve-debug-info we: | |||
| 150 | // - Do not create a new function for a thunk. | |||
| 151 | // - Retain the debug info for a thunk's parameters (and associated | |||
| 152 | // instructions for the debug info) from the entry block. | |||
| 153 | // Note: -debug will display the algorithm at work. | |||
| 154 | // - Create debug-info for the call (to the shared implementation) made by | |||
| 155 | // a thunk and its return value. | |||
| 156 | // - Erase the rest of the function, retaining the (minimally sized) entry | |||
| 157 | // block to create a thunk. | |||
| 158 | // - Preserve a thunk's call site to point to the thunk even when both occur | |||
| 159 | // within the same translation unit, to aid debugability. Note that this | |||
| 160 | // behaviour differs from the underlying -mergefunc implementation which | |||
| 161 | // modifies the thunk's call site to point to the shared implementation | |||
| 162 | // when both occur within the same translation unit. | |||
| 163 | static cl::opt<bool> | |||
| 164 | MergeFunctionsPDI("mergefunc-preserve-debug-info", cl::Hidden, | |||
| 165 | cl::init(false), | |||
| 166 | cl::desc("Preserve debug info in thunk when mergefunc " | |||
| 167 | "transformations are made.")); | |||
| 168 | ||||
| 169 | static cl::opt<bool> | |||
| 170 | MergeFunctionsAliases("mergefunc-use-aliases", cl::Hidden, | |||
| 171 | cl::init(false), | |||
| 172 | cl::desc("Allow mergefunc to create aliases")); | |||
| 173 | ||||
| 174 | namespace { | |||
| 175 | ||||
| 176 | class FunctionNode { | |||
| 177 | mutable AssertingVH<Function> F; | |||
| 178 | FunctionComparator::FunctionHash Hash; | |||
| 179 | ||||
| 180 | public: | |||
| 181 | // Note the hash is recalculated potentially multiple times, but it is cheap. | |||
| 182 | FunctionNode(Function *F) | |||
| 183 | : F(F), Hash(FunctionComparator::functionHash(*F)) {} | |||
| 184 | ||||
| 185 | Function *getFunc() const { return F; } | |||
| 186 | FunctionComparator::FunctionHash getHash() const { return Hash; } | |||
| 187 | ||||
| 188 | /// Replace the reference to the function F by the function G, assuming their | |||
| 189 | /// implementations are equal. | |||
| 190 | void replaceBy(Function *G) const { | |||
| 191 | F = G; | |||
| 192 | } | |||
| 193 | }; | |||
| 194 | ||||
| 195 | /// MergeFunctions finds functions which will generate identical machine code, | |||
| 196 | /// by considering all pointer types to be equivalent. Once identified, | |||
| 197 | /// MergeFunctions will fold them by replacing a call to one to a call to a | |||
| 198 | /// bitcast of the other. | |||
| 199 | class MergeFunctions { | |||
| 200 | public: | |||
| 201 | MergeFunctions() : FnTree(FunctionNodeCmp(&GlobalNumbers)) { | |||
| 202 | } | |||
| 203 | ||||
| 204 | bool runOnModule(Module &M); | |||
| 205 | ||||
| 206 | private: | |||
| 207 | // The function comparison operator is provided here so that FunctionNodes do | |||
| 208 | // not need to become larger with another pointer. | |||
| 209 | class FunctionNodeCmp { | |||
| 210 | GlobalNumberState* GlobalNumbers; | |||
| 211 | ||||
| 212 | public: | |||
| 213 | FunctionNodeCmp(GlobalNumberState* GN) : GlobalNumbers(GN) {} | |||
| 214 | ||||
| 215 | bool operator()(const FunctionNode &LHS, const FunctionNode &RHS) const { | |||
| 216 | // Order first by hashes, then full function comparison. | |||
| 217 | if (LHS.getHash() != RHS.getHash()) | |||
| 218 | return LHS.getHash() < RHS.getHash(); | |||
| 219 | FunctionComparator FCmp(LHS.getFunc(), RHS.getFunc(), GlobalNumbers); | |||
| 220 | return FCmp.compare() == -1; | |||
| 221 | } | |||
| 222 | }; | |||
| 223 | using FnTreeType = std::set<FunctionNode, FunctionNodeCmp>; | |||
| 224 | ||||
| 225 | GlobalNumberState GlobalNumbers; | |||
| 226 | ||||
| 227 | /// A work queue of functions that may have been modified and should be | |||
| 228 | /// analyzed again. | |||
| 229 | std::vector<WeakTrackingVH> Deferred; | |||
| 230 | ||||
| 231 | #ifndef NDEBUG1 | |||
| 232 | /// Checks the rules of order relation introduced among functions set. | |||
| 233 | /// Returns true, if sanity check has been passed, and false if failed. | |||
| 234 | bool doSanityCheck(std::vector<WeakTrackingVH> &Worklist); | |||
| 235 | #endif | |||
| 236 | ||||
| 237 | /// Insert a ComparableFunction into the FnTree, or merge it away if it's | |||
| 238 | /// equal to one that's already present. | |||
| 239 | bool insert(Function *NewFunction); | |||
| 240 | ||||
| 241 | /// Remove a Function from the FnTree and queue it up for a second sweep of | |||
| 242 | /// analysis. | |||
| 243 | void remove(Function *F); | |||
| 244 | ||||
| 245 | /// Find the functions that use this Value and remove them from FnTree and | |||
| 246 | /// queue the functions. | |||
| 247 | void removeUsers(Value *V); | |||
| 248 | ||||
| 249 | /// Replace all direct calls of Old with calls of New. Will bitcast New if | |||
| 250 | /// necessary to make types match. | |||
| 251 | void replaceDirectCallers(Function *Old, Function *New); | |||
| 252 | ||||
| 253 | /// Merge two equivalent functions. Upon completion, G may be deleted, or may | |||
| 254 | /// be converted into a thunk. In either case, it should never be visited | |||
| 255 | /// again. | |||
| 256 | void mergeTwoFunctions(Function *F, Function *G); | |||
| 257 | ||||
| 258 | /// Fill PDIUnrelatedWL with instructions from the entry block that are | |||
| 259 | /// unrelated to parameter related debug info. | |||
| 260 | void filterInstsUnrelatedToPDI(BasicBlock *GEntryBlock, | |||
| 261 | std::vector<Instruction *> &PDIUnrelatedWL); | |||
| 262 | ||||
| 263 | /// Erase the rest of the CFG (i.e. barring the entry block). | |||
| 264 | void eraseTail(Function *G); | |||
| 265 | ||||
| 266 | /// Erase the instructions in PDIUnrelatedWL as they are unrelated to the | |||
| 267 | /// parameter debug info, from the entry block. | |||
| 268 | void eraseInstsUnrelatedToPDI(std::vector<Instruction *> &PDIUnrelatedWL); | |||
| 269 | ||||
| 270 | /// Replace G with a simple tail call to bitcast(F). Also (unless | |||
| 271 | /// MergeFunctionsPDI holds) replace direct uses of G with bitcast(F), | |||
| 272 | /// delete G. | |||
| 273 | void writeThunk(Function *F, Function *G); | |||
| 274 | ||||
| 275 | // Replace G with an alias to F (deleting function G) | |||
| 276 | void writeAlias(Function *F, Function *G); | |||
| 277 | ||||
| 278 | // Replace G with an alias to F if possible, or a thunk to F if possible. | |||
| 279 | // Returns false if neither is the case. | |||
| 280 | bool writeThunkOrAlias(Function *F, Function *G); | |||
| 281 | ||||
| 282 | /// Replace function F with function G in the function tree. | |||
| 283 | void replaceFunctionInTree(const FunctionNode &FN, Function *G); | |||
| 284 | ||||
| 285 | /// The set of all distinct functions. Use the insert() and remove() methods | |||
| 286 | /// to modify it. The map allows efficient lookup and deferring of Functions. | |||
| 287 | FnTreeType FnTree; | |||
| 288 | ||||
| 289 | // Map functions to the iterators of the FunctionNode which contains them | |||
| 290 | // in the FnTree. This must be updated carefully whenever the FnTree is | |||
| 291 | // modified, i.e. in insert(), remove(), and replaceFunctionInTree(), to avoid | |||
| 292 | // dangling iterators into FnTree. The invariant that preserves this is that | |||
| 293 | // there is exactly one mapping F -> FN for each FunctionNode FN in FnTree. | |||
| 294 | DenseMap<AssertingVH<Function>, FnTreeType::iterator> FNodesInTree; | |||
| 295 | }; | |||
| 296 | ||||
| 297 | class MergeFunctionsLegacyPass : public ModulePass { | |||
| 298 | public: | |||
| 299 | static char ID; | |||
| 300 | ||||
| 301 | MergeFunctionsLegacyPass(): ModulePass(ID) { | |||
| 302 | initializeMergeFunctionsLegacyPassPass(*PassRegistry::getPassRegistry()); | |||
| 303 | } | |||
| 304 | ||||
| 305 | bool runOnModule(Module &M) override { | |||
| 306 | if (skipModule(M)) | |||
| 307 | return false; | |||
| 308 | ||||
| 309 | MergeFunctions MF; | |||
| 310 | return MF.runOnModule(M); | |||
| 311 | } | |||
| 312 | }; | |||
| 313 | ||||
| 314 | } // end anonymous namespace | |||
| 315 | ||||
| 316 | char MergeFunctionsLegacyPass::ID = 0; | |||
| 317 | INITIALIZE_PASS(MergeFunctionsLegacyPass, "mergefunc",static void *initializeMergeFunctionsLegacyPassPassOnce(PassRegistry &Registry) { PassInfo *PI = new PassInfo( "Merge Functions" , "mergefunc", &MergeFunctionsLegacyPass::ID, PassInfo::NormalCtor_t (callDefaultCtor<MergeFunctionsLegacyPass>), false, false ); Registry.registerPass(*PI, true); return PI; } static llvm ::once_flag InitializeMergeFunctionsLegacyPassPassFlag; void llvm ::initializeMergeFunctionsLegacyPassPass(PassRegistry &Registry ) { llvm::call_once(InitializeMergeFunctionsLegacyPassPassFlag , initializeMergeFunctionsLegacyPassPassOnce, std::ref(Registry )); } | |||
| 318 | "Merge Functions", false, false)static void *initializeMergeFunctionsLegacyPassPassOnce(PassRegistry &Registry) { PassInfo *PI = new PassInfo( "Merge Functions" , "mergefunc", &MergeFunctionsLegacyPass::ID, PassInfo::NormalCtor_t (callDefaultCtor<MergeFunctionsLegacyPass>), false, false ); Registry.registerPass(*PI, true); return PI; } static llvm ::once_flag InitializeMergeFunctionsLegacyPassPassFlag; void llvm ::initializeMergeFunctionsLegacyPassPass(PassRegistry &Registry ) { llvm::call_once(InitializeMergeFunctionsLegacyPassPassFlag , initializeMergeFunctionsLegacyPassPassOnce, std::ref(Registry )); } | |||
| 319 | ||||
| 320 | ModulePass *llvm::createMergeFunctionsPass() { | |||
| 321 | return new MergeFunctionsLegacyPass(); | |||
| 322 | } | |||
| 323 | ||||
| 324 | PreservedAnalyses MergeFunctionsPass::run(Module &M, | |||
| 325 | ModuleAnalysisManager &AM) { | |||
| 326 | MergeFunctions MF; | |||
| 327 | if (!MF.runOnModule(M)) | |||
| 328 | return PreservedAnalyses::all(); | |||
| 329 | return PreservedAnalyses::none(); | |||
| 330 | } | |||
| 331 | ||||
| 332 | #ifndef NDEBUG1 | |||
| 333 | bool MergeFunctions::doSanityCheck(std::vector<WeakTrackingVH> &Worklist) { | |||
| 334 | if (const unsigned Max = NumFunctionsForSanityCheck) { | |||
| 335 | unsigned TripleNumber = 0; | |||
| 336 | bool Valid = true; | |||
| 337 | ||||
| 338 | dbgs() << "MERGEFUNC-SANITY: Started for first " << Max << " functions.\n"; | |||
| 339 | ||||
| 340 | unsigned i = 0; | |||
| 341 | for (std::vector<WeakTrackingVH>::iterator I = Worklist.begin(), | |||
| 342 | E = Worklist.end(); | |||
| 343 | I != E && i < Max; ++I, ++i) { | |||
| 344 | unsigned j = i; | |||
| 345 | for (std::vector<WeakTrackingVH>::iterator J = I; J != E && j < Max; | |||
| 346 | ++J, ++j) { | |||
| 347 | Function *F1 = cast<Function>(*I); | |||
| 348 | Function *F2 = cast<Function>(*J); | |||
| 349 | int Res1 = FunctionComparator(F1, F2, &GlobalNumbers).compare(); | |||
| 350 | int Res2 = FunctionComparator(F2, F1, &GlobalNumbers).compare(); | |||
| 351 | ||||
| 352 | // If F1 <= F2, then F2 >= F1, otherwise report failure. | |||
| 353 | if (Res1 != -Res2) { | |||
| 354 | dbgs() << "MERGEFUNC-SANITY: Non-symmetric; triple: " << TripleNumber | |||
| 355 | << "\n"; | |||
| 356 | dbgs() << *F1 << '\n' << *F2 << '\n'; | |||
| 357 | Valid = false; | |||
| 358 | } | |||
| 359 | ||||
| 360 | if (Res1 == 0) | |||
| 361 | continue; | |||
| 362 | ||||
| 363 | unsigned k = j; | |||
| 364 | for (std::vector<WeakTrackingVH>::iterator K = J; K != E && k < Max; | |||
| 365 | ++k, ++K, ++TripleNumber) { | |||
| 366 | if (K == J) | |||
| 367 | continue; | |||
| 368 | ||||
| 369 | Function *F3 = cast<Function>(*K); | |||
| 370 | int Res3 = FunctionComparator(F1, F3, &GlobalNumbers).compare(); | |||
| 371 | int Res4 = FunctionComparator(F2, F3, &GlobalNumbers).compare(); | |||
| 372 | ||||
| 373 | bool Transitive = true; | |||
| 374 | ||||
| 375 | if (Res1 != 0 && Res1 == Res4) { | |||
| 376 | // F1 > F2, F2 > F3 => F1 > F3 | |||
| 377 | Transitive = Res3 == Res1; | |||
| 378 | } else if (Res3 != 0 && Res3 == -Res4) { | |||
| 379 | // F1 > F3, F3 > F2 => F1 > F2 | |||
| 380 | Transitive = Res3 == Res1; | |||
| 381 | } else if (Res4 != 0 && -Res3 == Res4) { | |||
| 382 | // F2 > F3, F3 > F1 => F2 > F1 | |||
| 383 | Transitive = Res4 == -Res1; | |||
| 384 | } | |||
| 385 | ||||
| 386 | if (!Transitive) { | |||
| 387 | dbgs() << "MERGEFUNC-SANITY: Non-transitive; triple: " | |||
| 388 | << TripleNumber << "\n"; | |||
| 389 | dbgs() << "Res1, Res3, Res4: " << Res1 << ", " << Res3 << ", " | |||
| 390 | << Res4 << "\n"; | |||
| 391 | dbgs() << *F1 << '\n' << *F2 << '\n' << *F3 << '\n'; | |||
| 392 | Valid = false; | |||
| 393 | } | |||
| 394 | } | |||
| 395 | } | |||
| 396 | } | |||
| 397 | ||||
| 398 | dbgs() << "MERGEFUNC-SANITY: " << (Valid ? "Passed." : "Failed.") << "\n"; | |||
| 399 | return Valid; | |||
| 400 | } | |||
| 401 | return true; | |||
| 402 | } | |||
| 403 | #endif | |||
| 404 | ||||
| 405 | /// Check whether \p F is eligible for function merging. | |||
| 406 | static bool isEligibleForMerging(Function &F) { | |||
| 407 | return !F.isDeclaration() && !F.hasAvailableExternallyLinkage(); | |||
| 408 | } | |||
| 409 | ||||
| 410 | bool MergeFunctions::runOnModule(Module &M) { | |||
| 411 | bool Changed = false; | |||
| 412 | ||||
| 413 | // All functions in the module, ordered by hash. Functions with a unique | |||
| 414 | // hash value are easily eliminated. | |||
| 415 | std::vector<std::pair<FunctionComparator::FunctionHash, Function *>> | |||
| 416 | HashedFuncs; | |||
| 417 | for (Function &Func : M) { | |||
| 418 | if (isEligibleForMerging(Func)) { | |||
| 419 | HashedFuncs.push_back({FunctionComparator::functionHash(Func), &Func}); | |||
| 420 | } | |||
| 421 | } | |||
| 422 | ||||
| 423 | llvm::stable_sort(HashedFuncs, less_first()); | |||
| 424 | ||||
| 425 | auto S = HashedFuncs.begin(); | |||
| 426 | for (auto I = HashedFuncs.begin(), IE = HashedFuncs.end(); I != IE; ++I) { | |||
| 427 | // If the hash value matches the previous value or the next one, we must | |||
| 428 | // consider merging it. Otherwise it is dropped and never considered again. | |||
| 429 | if ((I != S && std::prev(I)->first == I->first) || | |||
| 430 | (std::next(I) != IE && std::next(I)->first == I->first) ) { | |||
| 431 | Deferred.push_back(WeakTrackingVH(I->second)); | |||
| 432 | } | |||
| 433 | } | |||
| 434 | ||||
| 435 | do { | |||
| 436 | std::vector<WeakTrackingVH> Worklist; | |||
| 437 | Deferred.swap(Worklist); | |||
| 438 | ||||
| 439 | LLVM_DEBUG(doSanityCheck(Worklist))do { } while (false); | |||
| 440 | ||||
| 441 | LLVM_DEBUG(dbgs() << "size of module: " << M.size() << '\n')do { } while (false); | |||
| 442 | LLVM_DEBUG(dbgs() << "size of worklist: " << Worklist.size() << '\n')do { } while (false); | |||
| 443 | ||||
| 444 | // Insert functions and merge them. | |||
| 445 | for (WeakTrackingVH &I : Worklist) { | |||
| 446 | if (!I) | |||
| 447 | continue; | |||
| 448 | Function *F = cast<Function>(I); | |||
| 449 | if (!F->isDeclaration() && !F->hasAvailableExternallyLinkage()) { | |||
| 450 | Changed |= insert(F); | |||
| 451 | } | |||
| 452 | } | |||
| 453 | LLVM_DEBUG(dbgs() << "size of FnTree: " << FnTree.size() << '\n')do { } while (false); | |||
| 454 | } while (!Deferred.empty()); | |||
| 455 | ||||
| 456 | FnTree.clear(); | |||
| 457 | FNodesInTree.clear(); | |||
| 458 | GlobalNumbers.clear(); | |||
| 459 | ||||
| 460 | return Changed; | |||
| 461 | } | |||
| 462 | ||||
| 463 | // Replace direct callers of Old with New. | |||
| 464 | void MergeFunctions::replaceDirectCallers(Function *Old, Function *New) { | |||
| 465 | Constant *BitcastNew = ConstantExpr::getBitCast(New, Old->getType()); | |||
| 466 | for (auto UI = Old->use_begin(), UE = Old->use_end(); UI != UE;) { | |||
| 467 | Use *U = &*UI; | |||
| 468 | ++UI; | |||
| 469 | CallBase *CB = dyn_cast<CallBase>(U->getUser()); | |||
| 470 | if (CB && CB->isCallee(U)) { | |||
| 471 | // Do not copy attributes from the called function to the call-site. | |||
| 472 | // Function comparison ensures that the attributes are the same up to | |||
| 473 | // type congruences in byval(), in which case we need to keep the byval | |||
| 474 | // type of the call-site, not the callee function. | |||
| 475 | remove(CB->getFunction()); | |||
| 476 | U->set(BitcastNew); | |||
| 477 | } | |||
| 478 | } | |||
| 479 | } | |||
| 480 | ||||
| 481 | // Helper for writeThunk, | |||
| 482 | // Selects proper bitcast operation, | |||
| 483 | // but a bit simpler then CastInst::getCastOpcode. | |||
| 484 | static Value *createCast(IRBuilder<> &Builder, Value *V, Type *DestTy) { | |||
| 485 | Type *SrcTy = V->getType(); | |||
| 486 | if (SrcTy->isStructTy()) { | |||
| 487 | assert(DestTy->isStructTy())((void)0); | |||
| 488 | assert(SrcTy->getStructNumElements() == DestTy->getStructNumElements())((void)0); | |||
| 489 | Value *Result = UndefValue::get(DestTy); | |||
| 490 | for (unsigned int I = 0, E = SrcTy->getStructNumElements(); I < E; ++I) { | |||
| 491 | Value *Element = createCast( | |||
| 492 | Builder, Builder.CreateExtractValue(V, makeArrayRef(I)), | |||
| 493 | DestTy->getStructElementType(I)); | |||
| 494 | ||||
| 495 | Result = | |||
| 496 | Builder.CreateInsertValue(Result, Element, makeArrayRef(I)); | |||
| 497 | } | |||
| 498 | return Result; | |||
| 499 | } | |||
| 500 | assert(!DestTy->isStructTy())((void)0); | |||
| 501 | if (SrcTy->isIntegerTy() && DestTy->isPointerTy()) | |||
| 502 | return Builder.CreateIntToPtr(V, DestTy); | |||
| 503 | else if (SrcTy->isPointerTy() && DestTy->isIntegerTy()) | |||
| 504 | return Builder.CreatePtrToInt(V, DestTy); | |||
| 505 | else | |||
| 506 | return Builder.CreateBitCast(V, DestTy); | |||
| 507 | } | |||
| 508 | ||||
| 509 | // Erase the instructions in PDIUnrelatedWL as they are unrelated to the | |||
| 510 | // parameter debug info, from the entry block. | |||
| 511 | void MergeFunctions::eraseInstsUnrelatedToPDI( | |||
| 512 | std::vector<Instruction *> &PDIUnrelatedWL) { | |||
| 513 | LLVM_DEBUG(do { } while (false) | |||
| 514 | dbgs() << " Erasing instructions (in reverse order of appearance in "do { } while (false) | |||
| 515 | "entry block) unrelated to parameter debug info from entry "do { } while (false) | |||
| 516 | "block: {\n")do { } while (false); | |||
| 517 | while (!PDIUnrelatedWL.empty()) { | |||
| 518 | Instruction *I = PDIUnrelatedWL.back(); | |||
| 519 | LLVM_DEBUG(dbgs() << " Deleting Instruction: ")do { } while (false); | |||
| 520 | LLVM_DEBUG(I->print(dbgs()))do { } while (false); | |||
| 521 | LLVM_DEBUG(dbgs() << "\n")do { } while (false); | |||
| 522 | I->eraseFromParent(); | |||
| 523 | PDIUnrelatedWL.pop_back(); | |||
| 524 | } | |||
| 525 | LLVM_DEBUG(dbgs() << " } // Done erasing instructions unrelated to parameter "do { } while (false) | |||
| 526 | "debug info from entry block. \n")do { } while (false); | |||
| 527 | } | |||
| 528 | ||||
| 529 | // Reduce G to its entry block. | |||
| 530 | void MergeFunctions::eraseTail(Function *G) { | |||
| 531 | std::vector<BasicBlock *> WorklistBB; | |||
| 532 | for (BasicBlock &BB : drop_begin(*G)) { | |||
| 533 | BB.dropAllReferences(); | |||
| 534 | WorklistBB.push_back(&BB); | |||
| 535 | } | |||
| 536 | while (!WorklistBB.empty()) { | |||
| 537 | BasicBlock *BB = WorklistBB.back(); | |||
| 538 | BB->eraseFromParent(); | |||
| 539 | WorklistBB.pop_back(); | |||
| 540 | } | |||
| 541 | } | |||
| 542 | ||||
| 543 | // We are interested in the following instructions from the entry block as being | |||
| 544 | // related to parameter debug info: | |||
| 545 | // - @llvm.dbg.declare | |||
| 546 | // - stores from the incoming parameters to locations on the stack-frame | |||
| 547 | // - allocas that create these locations on the stack-frame | |||
| 548 | // - @llvm.dbg.value | |||
| 549 | // - the entry block's terminator | |||
| 550 | // The rest are unrelated to debug info for the parameters; fill up | |||
| 551 | // PDIUnrelatedWL with such instructions. | |||
| 552 | void MergeFunctions::filterInstsUnrelatedToPDI( | |||
| 553 | BasicBlock *GEntryBlock, std::vector<Instruction *> &PDIUnrelatedWL) { | |||
| 554 | std::set<Instruction *> PDIRelated; | |||
| 555 | for (BasicBlock::iterator BI = GEntryBlock->begin(), BIE = GEntryBlock->end(); | |||
| 556 | BI != BIE; ++BI) { | |||
| 557 | if (auto *DVI = dyn_cast<DbgValueInst>(&*BI)) { | |||
| 558 | LLVM_DEBUG(dbgs() << " Deciding: ")do { } while (false); | |||
| 559 | LLVM_DEBUG(BI->print(dbgs()))do { } while (false); | |||
| 560 | LLVM_DEBUG(dbgs() << "\n")do { } while (false); | |||
| 561 | DILocalVariable *DILocVar = DVI->getVariable(); | |||
| 562 | if (DILocVar->isParameter()) { | |||
| 563 | LLVM_DEBUG(dbgs() << " Include (parameter): ")do { } while (false); | |||
| 564 | LLVM_DEBUG(BI->print(dbgs()))do { } while (false); | |||
| 565 | LLVM_DEBUG(dbgs() << "\n")do { } while (false); | |||
| 566 | PDIRelated.insert(&*BI); | |||
| 567 | } else { | |||
| 568 | LLVM_DEBUG(dbgs() << " Delete (!parameter): ")do { } while (false); | |||
| 569 | LLVM_DEBUG(BI->print(dbgs()))do { } while (false); | |||
| 570 | LLVM_DEBUG(dbgs() << "\n")do { } while (false); | |||
| 571 | } | |||
| 572 | } else if (auto *DDI = dyn_cast<DbgDeclareInst>(&*BI)) { | |||
| 573 | LLVM_DEBUG(dbgs() << " Deciding: ")do { } while (false); | |||
| 574 | LLVM_DEBUG(BI->print(dbgs()))do { } while (false); | |||
| 575 | LLVM_DEBUG(dbgs() << "\n")do { } while (false); | |||
| 576 | DILocalVariable *DILocVar = DDI->getVariable(); | |||
| 577 | if (DILocVar->isParameter()) { | |||
| 578 | LLVM_DEBUG(dbgs() << " Parameter: ")do { } while (false); | |||
| 579 | LLVM_DEBUG(DILocVar->print(dbgs()))do { } while (false); | |||
| 580 | AllocaInst *AI = dyn_cast_or_null<AllocaInst>(DDI->getAddress()); | |||
| 581 | if (AI) { | |||
| 582 | LLVM_DEBUG(dbgs() << " Processing alloca users: ")do { } while (false); | |||
| 583 | LLVM_DEBUG(dbgs() << "\n")do { } while (false); | |||
| 584 | for (User *U : AI->users()) { | |||
| 585 | if (StoreInst *SI = dyn_cast<StoreInst>(U)) { | |||
| 586 | if (Value *Arg = SI->getValueOperand()) { | |||
| 587 | if (isa<Argument>(Arg)) { | |||
| 588 | LLVM_DEBUG(dbgs() << " Include: ")do { } while (false); | |||
| 589 | LLVM_DEBUG(AI->print(dbgs()))do { } while (false); | |||
| 590 | LLVM_DEBUG(dbgs() << "\n")do { } while (false); | |||
| 591 | PDIRelated.insert(AI); | |||
| 592 | LLVM_DEBUG(dbgs() << " Include (parameter): ")do { } while (false); | |||
| 593 | LLVM_DEBUG(SI->print(dbgs()))do { } while (false); | |||
| 594 | LLVM_DEBUG(dbgs() << "\n")do { } while (false); | |||
| 595 | PDIRelated.insert(SI); | |||
| 596 | LLVM_DEBUG(dbgs() << " Include: ")do { } while (false); | |||
| 597 | LLVM_DEBUG(BI->print(dbgs()))do { } while (false); | |||
| 598 | LLVM_DEBUG(dbgs() << "\n")do { } while (false); | |||
| 599 | PDIRelated.insert(&*BI); | |||
| 600 | } else { | |||
| 601 | LLVM_DEBUG(dbgs() << " Delete (!parameter): ")do { } while (false); | |||
| 602 | LLVM_DEBUG(SI->print(dbgs()))do { } while (false); | |||
| 603 | LLVM_DEBUG(dbgs() << "\n")do { } while (false); | |||
| 604 | } | |||
| 605 | } | |||
| 606 | } else { | |||
| 607 | LLVM_DEBUG(dbgs() << " Defer: ")do { } while (false); | |||
| 608 | LLVM_DEBUG(U->print(dbgs()))do { } while (false); | |||
| 609 | LLVM_DEBUG(dbgs() << "\n")do { } while (false); | |||
| 610 | } | |||
| 611 | } | |||
| 612 | } else { | |||
| 613 | LLVM_DEBUG(dbgs() << " Delete (alloca NULL): ")do { } while (false); | |||
| 614 | LLVM_DEBUG(BI->print(dbgs()))do { } while (false); | |||
| 615 | LLVM_DEBUG(dbgs() << "\n")do { } while (false); | |||
| 616 | } | |||
| 617 | } else { | |||
| 618 | LLVM_DEBUG(dbgs() << " Delete (!parameter): ")do { } while (false); | |||
| 619 | LLVM_DEBUG(BI->print(dbgs()))do { } while (false); | |||
| 620 | LLVM_DEBUG(dbgs() << "\n")do { } while (false); | |||
| 621 | } | |||
| 622 | } else if (BI->isTerminator() && &*BI == GEntryBlock->getTerminator()) { | |||
| 623 | LLVM_DEBUG(dbgs() << " Will Include Terminator: ")do { } while (false); | |||
| 624 | LLVM_DEBUG(BI->print(dbgs()))do { } while (false); | |||
| 625 | LLVM_DEBUG(dbgs() << "\n")do { } while (false); | |||
| 626 | PDIRelated.insert(&*BI); | |||
| 627 | } else { | |||
| 628 | LLVM_DEBUG(dbgs() << " Defer: ")do { } while (false); | |||
| 629 | LLVM_DEBUG(BI->print(dbgs()))do { } while (false); | |||
| 630 | LLVM_DEBUG(dbgs() << "\n")do { } while (false); | |||
| 631 | } | |||
| 632 | } | |||
| 633 | LLVM_DEBUG(do { } while (false) | |||
| 634 | dbgs()do { } while (false) | |||
| 635 | << " Report parameter debug info related/related instructions: {\n")do { } while (false); | |||
| 636 | for (Instruction &I : *GEntryBlock) { | |||
| 637 | if (PDIRelated.find(&I) == PDIRelated.end()) { | |||
| 638 | LLVM_DEBUG(dbgs() << " !PDIRelated: ")do { } while (false); | |||
| 639 | LLVM_DEBUG(I.print(dbgs()))do { } while (false); | |||
| 640 | LLVM_DEBUG(dbgs() << "\n")do { } while (false); | |||
| 641 | PDIUnrelatedWL.push_back(&I); | |||
| 642 | } else { | |||
| 643 | LLVM_DEBUG(dbgs() << " PDIRelated: ")do { } while (false); | |||
| 644 | LLVM_DEBUG(I.print(dbgs()))do { } while (false); | |||
| 645 | LLVM_DEBUG(dbgs() << "\n")do { } while (false); | |||
| 646 | } | |||
| 647 | } | |||
| 648 | LLVM_DEBUG(dbgs() << " }\n")do { } while (false); | |||
| 649 | } | |||
| 650 | ||||
| 651 | /// Whether this function may be replaced by a forwarding thunk. | |||
| 652 | static bool canCreateThunkFor(Function *F) { | |||
| 653 | if (F->isVarArg()) | |||
| 654 | return false; | |||
| 655 | ||||
| 656 | // Don't merge tiny functions using a thunk, since it can just end up | |||
| 657 | // making the function larger. | |||
| 658 | if (F->size() == 1) { | |||
| 659 | if (F->front().size() <= 2) { | |||
| 660 | LLVM_DEBUG(dbgs() << "canCreateThunkFor: " << F->getName()do { } while (false) | |||
| 661 | << " is too small to bother creating a thunk for\n")do { } while (false); | |||
| 662 | return false; | |||
| 663 | } | |||
| 664 | } | |||
| 665 | return true; | |||
| 666 | } | |||
| 667 | ||||
| 668 | // Replace G with a simple tail call to bitcast(F). Also (unless | |||
| 669 | // MergeFunctionsPDI holds) replace direct uses of G with bitcast(F), | |||
| 670 | // delete G. Under MergeFunctionsPDI, we use G itself for creating | |||
| 671 | // the thunk as we preserve the debug info (and associated instructions) | |||
| 672 | // from G's entry block pertaining to G's incoming arguments which are | |||
| 673 | // passed on as corresponding arguments in the call that G makes to F. | |||
| 674 | // For better debugability, under MergeFunctionsPDI, we do not modify G's | |||
| 675 | // call sites to point to F even when within the same translation unit. | |||
| 676 | void MergeFunctions::writeThunk(Function *F, Function *G) { | |||
| 677 | BasicBlock *GEntryBlock = nullptr; | |||
| 678 | std::vector<Instruction *> PDIUnrelatedWL; | |||
| 679 | BasicBlock *BB = nullptr; | |||
| 680 | Function *NewG = nullptr; | |||
| ||||
| 681 | if (MergeFunctionsPDI) { | |||
| 682 | LLVM_DEBUG(dbgs() << "writeThunk: (MergeFunctionsPDI) Do not create a new "do { } while (false) | |||
| 683 | "function as thunk; retain original: "do { } while (false) | |||
| 684 | << G->getName() << "()\n")do { } while (false); | |||
| 685 | GEntryBlock = &G->getEntryBlock(); | |||
| 686 | LLVM_DEBUG(do { } while (false) | |||
| 687 | dbgs() << "writeThunk: (MergeFunctionsPDI) filter parameter related "do { } while (false) | |||
| 688 | "debug info for "do { } while (false) | |||
| 689 | << G->getName() << "() {\n")do { } while (false); | |||
| 690 | filterInstsUnrelatedToPDI(GEntryBlock, PDIUnrelatedWL); | |||
| 691 | GEntryBlock->getTerminator()->eraseFromParent(); | |||
| 692 | BB = GEntryBlock; | |||
| 693 | } else { | |||
| 694 | NewG = Function::Create(G->getFunctionType(), G->getLinkage(), | |||
| 695 | G->getAddressSpace(), "", G->getParent()); | |||
| 696 | NewG->setComdat(G->getComdat()); | |||
| 697 | BB = BasicBlock::Create(F->getContext(), "", NewG); | |||
| 698 | } | |||
| 699 | ||||
| 700 | IRBuilder<> Builder(BB); | |||
| 701 | Function *H = MergeFunctionsPDI ? G : NewG; | |||
| 702 | SmallVector<Value *, 16> Args; | |||
| 703 | unsigned i = 0; | |||
| 704 | FunctionType *FFTy = F->getFunctionType(); | |||
| 705 | for (Argument &AI : H->args()) { | |||
| ||||
| 706 | Args.push_back(createCast(Builder, &AI, FFTy->getParamType(i))); | |||
| 707 | ++i; | |||
| 708 | } | |||
| 709 | ||||
| 710 | CallInst *CI = Builder.CreateCall(F, Args); | |||
| 711 | ReturnInst *RI = nullptr; | |||
| 712 | bool isSwiftTailCall = F->getCallingConv() == CallingConv::SwiftTail && | |||
| 713 | G->getCallingConv() == CallingConv::SwiftTail; | |||
| 714 | CI->setTailCallKind(isSwiftTailCall ? llvm::CallInst::TCK_MustTail | |||
| 715 | : llvm::CallInst::TCK_Tail); | |||
| 716 | CI->setCallingConv(F->getCallingConv()); | |||
| 717 | CI->setAttributes(F->getAttributes()); | |||
| 718 | if (H->getReturnType()->isVoidTy()) { | |||
| 719 | RI = Builder.CreateRetVoid(); | |||
| 720 | } else { | |||
| 721 | RI = Builder.CreateRet(createCast(Builder, CI, H->getReturnType())); | |||
| 722 | } | |||
| 723 | ||||
| 724 | if (MergeFunctionsPDI) { | |||
| 725 | DISubprogram *DIS = G->getSubprogram(); | |||
| 726 | if (DIS) { | |||
| 727 | DebugLoc CIDbgLoc = | |||
| 728 | DILocation::get(DIS->getContext(), DIS->getScopeLine(), 0, DIS); | |||
| 729 | DebugLoc RIDbgLoc = | |||
| 730 | DILocation::get(DIS->getContext(), DIS->getScopeLine(), 0, DIS); | |||
| 731 | CI->setDebugLoc(CIDbgLoc); | |||
| 732 | RI->setDebugLoc(RIDbgLoc); | |||
| 733 | } else { | |||
| 734 | LLVM_DEBUG(do { } while (false) | |||
| 735 | dbgs() << "writeThunk: (MergeFunctionsPDI) No DISubprogram for "do { } while (false) | |||
| 736 | << G->getName() << "()\n")do { } while (false); | |||
| 737 | } | |||
| 738 | eraseTail(G); | |||
| 739 | eraseInstsUnrelatedToPDI(PDIUnrelatedWL); | |||
| 740 | LLVM_DEBUG(do { } while (false) | |||
| 741 | dbgs() << "} // End of parameter related debug info filtering for: "do { } while (false) | |||
| 742 | << G->getName() << "()\n")do { } while (false); | |||
| 743 | } else { | |||
| 744 | NewG->copyAttributesFrom(G); | |||
| 745 | NewG->takeName(G); | |||
| 746 | removeUsers(G); | |||
| 747 | G->replaceAllUsesWith(NewG); | |||
| 748 | G->eraseFromParent(); | |||
| 749 | } | |||
| 750 | ||||
| 751 | LLVM_DEBUG(dbgs() << "writeThunk: " << H->getName() << '\n')do { } while (false); | |||
| 752 | ++NumThunksWritten; | |||
| 753 | } | |||
| 754 | ||||
| 755 | // Whether this function may be replaced by an alias | |||
| 756 | static bool canCreateAliasFor(Function *F) { | |||
| 757 | if (!MergeFunctionsAliases || !F->hasGlobalUnnamedAddr()) | |||
| 758 | return false; | |||
| 759 | ||||
| 760 | // We should only see linkages supported by aliases here | |||
| 761 | assert(F->hasLocalLinkage() || F->hasExternalLinkage()((void)0) | |||
| 762 | || F->hasWeakLinkage() || F->hasLinkOnceLinkage())((void)0); | |||
| 763 | return true; | |||
| 764 | } | |||
| 765 | ||||
| 766 | // Replace G with an alias to F (deleting function G) | |||
| 767 | void MergeFunctions::writeAlias(Function *F, Function *G) { | |||
| 768 | Constant *BitcastF = ConstantExpr::getBitCast(F, G->getType()); | |||
| 769 | PointerType *PtrType = G->getType(); | |||
| 770 | auto *GA = GlobalAlias::create(G->getValueType(), PtrType->getAddressSpace(), | |||
| 771 | G->getLinkage(), "", BitcastF, G->getParent()); | |||
| 772 | ||||
| 773 | F->setAlignment(MaybeAlign(std::max(F->getAlignment(), G->getAlignment()))); | |||
| 774 | GA->takeName(G); | |||
| 775 | GA->setVisibility(G->getVisibility()); | |||
| 776 | GA->setUnnamedAddr(GlobalValue::UnnamedAddr::Global); | |||
| 777 | ||||
| 778 | removeUsers(G); | |||
| 779 | G->replaceAllUsesWith(GA); | |||
| 780 | G->eraseFromParent(); | |||
| 781 | ||||
| 782 | LLVM_DEBUG(dbgs() << "writeAlias: " << GA->getName() << '\n')do { } while (false); | |||
| 783 | ++NumAliasesWritten; | |||
| 784 | } | |||
| 785 | ||||
| 786 | // Replace G with an alias to F if possible, or a thunk to F if | |||
| 787 | // profitable. Returns false if neither is the case. | |||
| 788 | bool MergeFunctions::writeThunkOrAlias(Function *F, Function *G) { | |||
| 789 | if (canCreateAliasFor(G)) { | |||
| 790 | writeAlias(F, G); | |||
| 791 | return true; | |||
| 792 | } | |||
| 793 | if (canCreateThunkFor(F)) { | |||
| 794 | writeThunk(F, G); | |||
| 795 | return true; | |||
| 796 | } | |||
| 797 | return false; | |||
| 798 | } | |||
| 799 | ||||
| 800 | // Merge two equivalent functions. Upon completion, Function G is deleted. | |||
| 801 | void MergeFunctions::mergeTwoFunctions(Function *F, Function *G) { | |||
| 802 | if (F->isInterposable()) { | |||
| 803 | assert(G->isInterposable())((void)0); | |||
| 804 | ||||
| 805 | // Both writeThunkOrAlias() calls below must succeed, either because we can | |||
| 806 | // create aliases for G and NewF, or because a thunk for F is profitable. | |||
| 807 | // F here has the same signature as NewF below, so that's what we check. | |||
| 808 | if (!canCreateThunkFor(F) && | |||
| 809 | (!canCreateAliasFor(F) || !canCreateAliasFor(G))) | |||
| 810 | return; | |||
| 811 | ||||
| 812 | // Make them both thunks to the same internal function. | |||
| 813 | Function *NewF = Function::Create(F->getFunctionType(), F->getLinkage(), | |||
| 814 | F->getAddressSpace(), "", F->getParent()); | |||
| 815 | NewF->copyAttributesFrom(F); | |||
| 816 | NewF->takeName(F); | |||
| 817 | removeUsers(F); | |||
| 818 | F->replaceAllUsesWith(NewF); | |||
| 819 | ||||
| 820 | MaybeAlign MaxAlignment(std::max(G->getAlignment(), NewF->getAlignment())); | |||
| 821 | ||||
| 822 | writeThunkOrAlias(F, G); | |||
| 823 | writeThunkOrAlias(F, NewF); | |||
| 824 | ||||
| 825 | F->setAlignment(MaxAlignment); | |||
| 826 | F->setLinkage(GlobalValue::PrivateLinkage); | |||
| 827 | ++NumDoubleWeak; | |||
| 828 | ++NumFunctionsMerged; | |||
| 829 | } else { | |||
| 830 | // For better debugability, under MergeFunctionsPDI, we do not modify G's | |||
| 831 | // call sites to point to F even when within the same translation unit. | |||
| 832 | if (!G->isInterposable() && !MergeFunctionsPDI) { | |||
| 833 | if (G->hasGlobalUnnamedAddr()) { | |||
| 834 | // G might have been a key in our GlobalNumberState, and it's illegal | |||
| 835 | // to replace a key in ValueMap<GlobalValue *> with a non-global. | |||
| 836 | GlobalNumbers.erase(G); | |||
| 837 | // If G's address is not significant, replace it entirely. | |||
| 838 | Constant *BitcastF = ConstantExpr::getBitCast(F, G->getType()); | |||
| 839 | removeUsers(G); | |||
| 840 | G->replaceAllUsesWith(BitcastF); | |||
| 841 | } else { | |||
| 842 | // Redirect direct callers of G to F. (See note on MergeFunctionsPDI | |||
| 843 | // above). | |||
| 844 | replaceDirectCallers(G, F); | |||
| 845 | } | |||
| 846 | } | |||
| 847 | ||||
| 848 | // If G was internal then we may have replaced all uses of G with F. If so, | |||
| 849 | // stop here and delete G. There's no need for a thunk. (See note on | |||
| 850 | // MergeFunctionsPDI above). | |||
| 851 | if (G->isDiscardableIfUnused() && G->use_empty() && !MergeFunctionsPDI) { | |||
| 852 | G->eraseFromParent(); | |||
| 853 | ++NumFunctionsMerged; | |||
| 854 | return; | |||
| 855 | } | |||
| 856 | ||||
| 857 | if (writeThunkOrAlias(F, G)) { | |||
| 858 | ++NumFunctionsMerged; | |||
| 859 | } | |||
| 860 | } | |||
| 861 | } | |||
| 862 | ||||
| 863 | /// Replace function F by function G. | |||
| 864 | void MergeFunctions::replaceFunctionInTree(const FunctionNode &FN, | |||
| 865 | Function *G) { | |||
| 866 | Function *F = FN.getFunc(); | |||
| 867 | assert(FunctionComparator(F, G, &GlobalNumbers).compare() == 0 &&((void)0) | |||
| 868 | "The two functions must be equal")((void)0); | |||
| 869 | ||||
| 870 | auto I = FNodesInTree.find(F); | |||
| 871 | assert(I != FNodesInTree.end() && "F should be in FNodesInTree")((void)0); | |||
| 872 | assert(FNodesInTree.count(G) == 0 && "FNodesInTree should not contain G")((void)0); | |||
| 873 | ||||
| 874 | FnTreeType::iterator IterToFNInFnTree = I->second; | |||
| 875 | assert(&(*IterToFNInFnTree) == &FN && "F should map to FN in FNodesInTree.")((void)0); | |||
| 876 | // Remove F -> FN and insert G -> FN | |||
| 877 | FNodesInTree.erase(I); | |||
| 878 | FNodesInTree.insert({G, IterToFNInFnTree}); | |||
| 879 | // Replace F with G in FN, which is stored inside the FnTree. | |||
| 880 | FN.replaceBy(G); | |||
| 881 | } | |||
| 882 | ||||
| 883 | // Ordering for functions that are equal under FunctionComparator | |||
| 884 | static bool isFuncOrderCorrect(const Function *F, const Function *G) { | |||
| 885 | if (F->isInterposable() != G->isInterposable()) { | |||
| 886 | // Strong before weak, because the weak function may call the strong | |||
| 887 | // one, but not the other way around. | |||
| 888 | return !F->isInterposable(); | |||
| 889 | } | |||
| 890 | if (F->hasLocalLinkage() != G->hasLocalLinkage()) { | |||
| 891 | // External before local, because we definitely have to keep the external | |||
| 892 | // function, but may be able to drop the local one. | |||
| 893 | return !F->hasLocalLinkage(); | |||
| 894 | } | |||
| 895 | // Impose a total order (by name) on the replacement of functions. This is | |||
| 896 | // important when operating on more than one module independently to prevent | |||
| 897 | // cycles of thunks calling each other when the modules are linked together. | |||
| 898 | return F->getName() <= G->getName(); | |||
| 899 | } | |||
| 900 | ||||
| 901 | // Insert a ComparableFunction into the FnTree, or merge it away if equal to one | |||
| 902 | // that was already inserted. | |||
| 903 | bool MergeFunctions::insert(Function *NewFunction) { | |||
| 904 | std::pair<FnTreeType::iterator, bool> Result = | |||
| 905 | FnTree.insert(FunctionNode(NewFunction)); | |||
| 906 | ||||
| 907 | if (Result.second) { | |||
| 908 | assert(FNodesInTree.count(NewFunction) == 0)((void)0); | |||
| 909 | FNodesInTree.insert({NewFunction, Result.first}); | |||
| 910 | LLVM_DEBUG(dbgs() << "Inserting as unique: " << NewFunction->getName()do { } while (false) | |||
| 911 | << '\n')do { } while (false); | |||
| 912 | return false; | |||
| 913 | } | |||
| 914 | ||||
| 915 | const FunctionNode &OldF = *Result.first; | |||
| 916 | ||||
| 917 | if (!isFuncOrderCorrect(OldF.getFunc(), NewFunction)) { | |||
| 918 | // Swap the two functions. | |||
| 919 | Function *F = OldF.getFunc(); | |||
| 920 | replaceFunctionInTree(*Result.first, NewFunction); | |||
| 921 | NewFunction = F; | |||
| 922 | assert(OldF.getFunc() != F && "Must have swapped the functions.")((void)0); | |||
| 923 | } | |||
| 924 | ||||
| 925 | LLVM_DEBUG(dbgs() << " " << OldF.getFunc()->getName()do { } while (false) | |||
| 926 | << " == " << NewFunction->getName() << '\n')do { } while (false); | |||
| 927 | ||||
| 928 | Function *DeleteF = NewFunction; | |||
| 929 | mergeTwoFunctions(OldF.getFunc(), DeleteF); | |||
| 930 | return true; | |||
| 931 | } | |||
| 932 | ||||
| 933 | // Remove a function from FnTree. If it was already in FnTree, add | |||
| 934 | // it to Deferred so that we'll look at it in the next round. | |||
| 935 | void MergeFunctions::remove(Function *F) { | |||
| 936 | auto I = FNodesInTree.find(F); | |||
| 937 | if (I != FNodesInTree.end()) { | |||
| 938 | LLVM_DEBUG(dbgs() << "Deferred " << F->getName() << ".\n")do { } while (false); | |||
| 939 | FnTree.erase(I->second); | |||
| 940 | // I->second has been invalidated, remove it from the FNodesInTree map to | |||
| 941 | // preserve the invariant. | |||
| 942 | FNodesInTree.erase(I); | |||
| 943 | Deferred.emplace_back(F); | |||
| 944 | } | |||
| 945 | } | |||
| 946 | ||||
| 947 | // For each instruction used by the value, remove() the function that contains | |||
| 948 | // the instruction. This should happen right before a call to RAUW. | |||
| 949 | void MergeFunctions::removeUsers(Value *V) { | |||
| 950 | for (User *U : V->users()) | |||
| 951 | if (auto *I = dyn_cast<Instruction>(U)) | |||
| 952 | remove(I->getFunction()); | |||
| 953 | } |