1 //===- ModuleSummaryAnalysis.cpp - Module summary index builder -----------===// 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 builds a ModuleSummaryIndex object for the module, to be written 10 // to bitcode or LLVM assembly. 11 // 12 //===----------------------------------------------------------------------===// 13 14 #include "llvm/Analysis/ModuleSummaryAnalysis.h" 15 #include "llvm/ADT/ArrayRef.h" 16 #include "llvm/ADT/DenseSet.h" 17 #include "llvm/ADT/MapVector.h" 18 #include "llvm/ADT/STLExtras.h" 19 #include "llvm/ADT/SetVector.h" 20 #include "llvm/ADT/SmallPtrSet.h" 21 #include "llvm/ADT/SmallVector.h" 22 #include "llvm/ADT/StringRef.h" 23 #include "llvm/Analysis/BlockFrequencyInfo.h" 24 #include "llvm/Analysis/BranchProbabilityInfo.h" 25 #include "llvm/Analysis/IndirectCallPromotionAnalysis.h" 26 #include "llvm/Analysis/LoopInfo.h" 27 #include "llvm/Analysis/ProfileSummaryInfo.h" 28 #include "llvm/Analysis/TypeMetadataUtils.h" 29 #include "llvm/IR/Attributes.h" 30 #include "llvm/IR/BasicBlock.h" 31 #include "llvm/IR/Constant.h" 32 #include "llvm/IR/Constants.h" 33 #include "llvm/IR/Dominators.h" 34 #include "llvm/IR/Function.h" 35 #include "llvm/IR/GlobalAlias.h" 36 #include "llvm/IR/GlobalValue.h" 37 #include "llvm/IR/GlobalVariable.h" 38 #include "llvm/IR/Instructions.h" 39 #include "llvm/IR/IntrinsicInst.h" 40 #include "llvm/IR/Intrinsics.h" 41 #include "llvm/IR/Metadata.h" 42 #include "llvm/IR/Module.h" 43 #include "llvm/IR/ModuleSummaryIndex.h" 44 #include "llvm/IR/Use.h" 45 #include "llvm/IR/User.h" 46 #include "llvm/InitializePasses.h" 47 #include "llvm/Object/ModuleSymbolTable.h" 48 #include "llvm/Object/SymbolicFile.h" 49 #include "llvm/Pass.h" 50 #include "llvm/Support/Casting.h" 51 #include "llvm/Support/CommandLine.h" 52 #include <algorithm> 53 #include <cassert> 54 #include <cstdint> 55 #include <vector> 56 57 using namespace llvm; 58 59 #define DEBUG_TYPE "module-summary-analysis" 60 61 // Option to force edges cold which will block importing when the 62 // -import-cold-multiplier is set to 0. Useful for debugging. 63 FunctionSummary::ForceSummaryHotnessType ForceSummaryEdgesCold = 64 FunctionSummary::FSHT_None; 65 cl::opt<FunctionSummary::ForceSummaryHotnessType, true> FSEC( 66 "force-summary-edges-cold", cl::Hidden, cl::location(ForceSummaryEdgesCold), 67 cl::desc("Force all edges in the function summary to cold"), 68 cl::values(clEnumValN(FunctionSummary::FSHT_None, "none", "None."), 69 clEnumValN(FunctionSummary::FSHT_AllNonCritical, 70 "all-non-critical", "All non-critical edges."), 71 clEnumValN(FunctionSummary::FSHT_All, "all", "All edges."))); 72 73 cl::opt<std::string> ModuleSummaryDotFile( 74 "module-summary-dot-file", cl::init(""), cl::Hidden, 75 cl::value_desc("filename"), 76 cl::desc("File to emit dot graph of new summary into.")); 77 78 // Walk through the operands of a given User via worklist iteration and populate 79 // the set of GlobalValue references encountered. Invoked either on an 80 // Instruction or a GlobalVariable (which walks its initializer). 81 // Return true if any of the operands contains blockaddress. This is important 82 // to know when computing summary for global var, because if global variable 83 // references basic block address we can't import it separately from function 84 // containing that basic block. For simplicity we currently don't import such 85 // global vars at all. When importing function we aren't interested if any 86 // instruction in it takes an address of any basic block, because instruction 87 // can only take an address of basic block located in the same function. 88 static bool findRefEdges(ModuleSummaryIndex &Index, const User *CurUser, 89 SetVector<ValueInfo> &RefEdges, 90 SmallPtrSet<const User *, 8> &Visited) { 91 bool HasBlockAddress = false; 92 SmallVector<const User *, 32> Worklist; 93 Worklist.push_back(CurUser); 94 95 while (!Worklist.empty()) { 96 const User *U = Worklist.pop_back_val(); 97 98 if (!Visited.insert(U).second) 99 continue; 100 101 const auto *CB = dyn_cast<CallBase>(U); 102 103 for (const auto &OI : U->operands()) { 104 const User *Operand = dyn_cast<User>(OI); 105 if (!Operand) 106 continue; 107 if (isa<BlockAddress>(Operand)) { 108 HasBlockAddress = true; 109 continue; 110 } 111 if (auto *GV = dyn_cast<GlobalValue>(Operand)) { 112 // We have a reference to a global value. This should be added to 113 // the reference set unless it is a callee. Callees are handled 114 // specially by WriteFunction and are added to a separate list. 115 if (!(CB && CB->isCallee(&OI))) 116 RefEdges.insert(Index.getOrInsertValueInfo(GV)); 117 continue; 118 } 119 Worklist.push_back(Operand); 120 } 121 } 122 return HasBlockAddress; 123 } 124 125 static CalleeInfo::HotnessType getHotness(uint64_t ProfileCount, 126 ProfileSummaryInfo *PSI) { 127 if (!PSI) 128 return CalleeInfo::HotnessType::Unknown; 129 if (PSI->isHotCount(ProfileCount)) 130 return CalleeInfo::HotnessType::Hot; 131 if (PSI->isColdCount(ProfileCount)) 132 return CalleeInfo::HotnessType::Cold; 133 return CalleeInfo::HotnessType::None; 134 } 135 136 static bool isNonRenamableLocal(const GlobalValue &GV) { 137 return GV.hasSection() && GV.hasLocalLinkage(); 138 } 139 140 /// Determine whether this call has all constant integer arguments (excluding 141 /// "this") and summarize it to VCalls or ConstVCalls as appropriate. 142 static void addVCallToSet(DevirtCallSite Call, GlobalValue::GUID Guid, 143 SetVector<FunctionSummary::VFuncId> &VCalls, 144 SetVector<FunctionSummary::ConstVCall> &ConstVCalls) { 145 std::vector<uint64_t> Args; 146 // Start from the second argument to skip the "this" pointer. 147 for (auto &Arg : make_range(Call.CB.arg_begin() + 1, Call.CB.arg_end())) { 148 auto *CI = dyn_cast<ConstantInt>(Arg); 149 if (!CI || CI->getBitWidth() > 64) { 150 VCalls.insert({Guid, Call.Offset}); 151 return; 152 } 153 Args.push_back(CI->getZExtValue()); 154 } 155 ConstVCalls.insert({{Guid, Call.Offset}, std::move(Args)}); 156 } 157 158 /// If this intrinsic call requires that we add information to the function 159 /// summary, do so via the non-constant reference arguments. 160 static void addIntrinsicToSummary( 161 const CallInst *CI, SetVector<GlobalValue::GUID> &TypeTests, 162 SetVector<FunctionSummary::VFuncId> &TypeTestAssumeVCalls, 163 SetVector<FunctionSummary::VFuncId> &TypeCheckedLoadVCalls, 164 SetVector<FunctionSummary::ConstVCall> &TypeTestAssumeConstVCalls, 165 SetVector<FunctionSummary::ConstVCall> &TypeCheckedLoadConstVCalls, 166 DominatorTree &DT) { 167 switch (CI->getCalledFunction()->getIntrinsicID()) { 168 case Intrinsic::type_test: { 169 auto *TypeMDVal = cast<MetadataAsValue>(CI->getArgOperand(1)); 170 auto *TypeId = dyn_cast<MDString>(TypeMDVal->getMetadata()); 171 if (!TypeId) 172 break; 173 GlobalValue::GUID Guid = GlobalValue::getGUID(TypeId->getString()); 174 175 // Produce a summary from type.test intrinsics. We only summarize type.test 176 // intrinsics that are used other than by an llvm.assume intrinsic. 177 // Intrinsics that are assumed are relevant only to the devirtualization 178 // pass, not the type test lowering pass. 179 bool HasNonAssumeUses = llvm::any_of(CI->uses(), [](const Use &CIU) { 180 auto *AssumeCI = dyn_cast<CallInst>(CIU.getUser()); 181 if (!AssumeCI) 182 return true; 183 Function *F = AssumeCI->getCalledFunction(); 184 return !F || F->getIntrinsicID() != Intrinsic::assume; 185 }); 186 if (HasNonAssumeUses) 187 TypeTests.insert(Guid); 188 189 SmallVector<DevirtCallSite, 4> DevirtCalls; 190 SmallVector<CallInst *, 4> Assumes; 191 findDevirtualizableCallsForTypeTest(DevirtCalls, Assumes, CI, DT); 192 for (auto &Call : DevirtCalls) 193 addVCallToSet(Call, Guid, TypeTestAssumeVCalls, 194 TypeTestAssumeConstVCalls); 195 196 break; 197 } 198 199 case Intrinsic::type_checked_load: { 200 auto *TypeMDVal = cast<MetadataAsValue>(CI->getArgOperand(2)); 201 auto *TypeId = dyn_cast<MDString>(TypeMDVal->getMetadata()); 202 if (!TypeId) 203 break; 204 GlobalValue::GUID Guid = GlobalValue::getGUID(TypeId->getString()); 205 206 SmallVector<DevirtCallSite, 4> DevirtCalls; 207 SmallVector<Instruction *, 4> LoadedPtrs; 208 SmallVector<Instruction *, 4> Preds; 209 bool HasNonCallUses = false; 210 findDevirtualizableCallsForTypeCheckedLoad(DevirtCalls, LoadedPtrs, Preds, 211 HasNonCallUses, CI, DT); 212 // Any non-call uses of the result of llvm.type.checked.load will 213 // prevent us from optimizing away the llvm.type.test. 214 if (HasNonCallUses) 215 TypeTests.insert(Guid); 216 for (auto &Call : DevirtCalls) 217 addVCallToSet(Call, Guid, TypeCheckedLoadVCalls, 218 TypeCheckedLoadConstVCalls); 219 220 break; 221 } 222 default: 223 break; 224 } 225 } 226 227 static bool isNonVolatileLoad(const Instruction *I) { 228 if (const auto *LI = dyn_cast<LoadInst>(I)) 229 return !LI->isVolatile(); 230 231 return false; 232 } 233 234 static bool isNonVolatileStore(const Instruction *I) { 235 if (const auto *SI = dyn_cast<StoreInst>(I)) 236 return !SI->isVolatile(); 237 238 return false; 239 } 240 241 static void computeFunctionSummary(ModuleSummaryIndex &Index, const Module &M, 242 const Function &F, BlockFrequencyInfo *BFI, 243 ProfileSummaryInfo *PSI, DominatorTree &DT, 244 bool HasLocalsInUsedOrAsm, 245 DenseSet<GlobalValue::GUID> &CantBePromoted, 246 bool IsThinLTO) { 247 // Summary not currently supported for anonymous functions, they should 248 // have been named. 249 assert(F.hasName()); 250 251 unsigned NumInsts = 0; 252 // Map from callee ValueId to profile count. Used to accumulate profile 253 // counts for all static calls to a given callee. 254 MapVector<ValueInfo, CalleeInfo> CallGraphEdges; 255 SetVector<ValueInfo> RefEdges, LoadRefEdges, StoreRefEdges; 256 SetVector<GlobalValue::GUID> TypeTests; 257 SetVector<FunctionSummary::VFuncId> TypeTestAssumeVCalls, 258 TypeCheckedLoadVCalls; 259 SetVector<FunctionSummary::ConstVCall> TypeTestAssumeConstVCalls, 260 TypeCheckedLoadConstVCalls; 261 ICallPromotionAnalysis ICallAnalysis; 262 SmallPtrSet<const User *, 8> Visited; 263 264 // Add personality function, prefix data and prologue data to function's ref 265 // list. 266 findRefEdges(Index, &F, RefEdges, Visited); 267 std::vector<const Instruction *> NonVolatileLoads; 268 std::vector<const Instruction *> NonVolatileStores; 269 270 bool HasInlineAsmMaybeReferencingInternal = false; 271 for (const BasicBlock &BB : F) 272 for (const Instruction &I : BB) { 273 if (isa<DbgInfoIntrinsic>(I)) 274 continue; 275 ++NumInsts; 276 // Regular LTO module doesn't participate in ThinLTO import, 277 // so no reference from it can be read/writeonly, since this 278 // would require importing variable as local copy 279 if (IsThinLTO) { 280 if (isNonVolatileLoad(&I)) { 281 // Postpone processing of non-volatile load instructions 282 // See comments below 283 Visited.insert(&I); 284 NonVolatileLoads.push_back(&I); 285 continue; 286 } else if (isNonVolatileStore(&I)) { 287 Visited.insert(&I); 288 NonVolatileStores.push_back(&I); 289 // All references from second operand of store (destination address) 290 // can be considered write-only if they're not referenced by any 291 // non-store instruction. References from first operand of store 292 // (stored value) can't be treated either as read- or as write-only 293 // so we add them to RefEdges as we do with all other instructions 294 // except non-volatile load. 295 Value *Stored = I.getOperand(0); 296 if (auto *GV = dyn_cast<GlobalValue>(Stored)) 297 // findRefEdges will try to examine GV operands, so instead 298 // of calling it we should add GV to RefEdges directly. 299 RefEdges.insert(Index.getOrInsertValueInfo(GV)); 300 else if (auto *U = dyn_cast<User>(Stored)) 301 findRefEdges(Index, U, RefEdges, Visited); 302 continue; 303 } 304 } 305 findRefEdges(Index, &I, RefEdges, Visited); 306 const auto *CB = dyn_cast<CallBase>(&I); 307 if (!CB) 308 continue; 309 310 const auto *CI = dyn_cast<CallInst>(&I); 311 // Since we don't know exactly which local values are referenced in inline 312 // assembly, conservatively mark the function as possibly referencing 313 // a local value from inline assembly to ensure we don't export a 314 // reference (which would require renaming and promotion of the 315 // referenced value). 316 if (HasLocalsInUsedOrAsm && CI && CI->isInlineAsm()) 317 HasInlineAsmMaybeReferencingInternal = true; 318 319 auto *CalledValue = CB->getCalledOperand(); 320 auto *CalledFunction = CB->getCalledFunction(); 321 if (CalledValue && !CalledFunction) { 322 CalledValue = CalledValue->stripPointerCasts(); 323 // Stripping pointer casts can reveal a called function. 324 CalledFunction = dyn_cast<Function>(CalledValue); 325 } 326 // Check if this is an alias to a function. If so, get the 327 // called aliasee for the checks below. 328 if (auto *GA = dyn_cast<GlobalAlias>(CalledValue)) { 329 assert(!CalledFunction && "Expected null called function in callsite for alias"); 330 CalledFunction = dyn_cast<Function>(GA->getBaseObject()); 331 } 332 // Check if this is a direct call to a known function or a known 333 // intrinsic, or an indirect call with profile data. 334 if (CalledFunction) { 335 if (CI && CalledFunction->isIntrinsic()) { 336 addIntrinsicToSummary( 337 CI, TypeTests, TypeTestAssumeVCalls, TypeCheckedLoadVCalls, 338 TypeTestAssumeConstVCalls, TypeCheckedLoadConstVCalls, DT); 339 continue; 340 } 341 // We should have named any anonymous globals 342 assert(CalledFunction->hasName()); 343 auto ScaledCount = PSI->getProfileCount(*CB, BFI); 344 auto Hotness = ScaledCount ? getHotness(ScaledCount.getValue(), PSI) 345 : CalleeInfo::HotnessType::Unknown; 346 if (ForceSummaryEdgesCold != FunctionSummary::FSHT_None) 347 Hotness = CalleeInfo::HotnessType::Cold; 348 349 // Use the original CalledValue, in case it was an alias. We want 350 // to record the call edge to the alias in that case. Eventually 351 // an alias summary will be created to associate the alias and 352 // aliasee. 353 auto &ValueInfo = CallGraphEdges[Index.getOrInsertValueInfo( 354 cast<GlobalValue>(CalledValue))]; 355 ValueInfo.updateHotness(Hotness); 356 // Add the relative block frequency to CalleeInfo if there is no profile 357 // information. 358 if (BFI != nullptr && Hotness == CalleeInfo::HotnessType::Unknown) { 359 uint64_t BBFreq = BFI->getBlockFreq(&BB).getFrequency(); 360 uint64_t EntryFreq = BFI->getEntryFreq(); 361 ValueInfo.updateRelBlockFreq(BBFreq, EntryFreq); 362 } 363 } else { 364 // Skip inline assembly calls. 365 if (CI && CI->isInlineAsm()) 366 continue; 367 // Skip direct calls. 368 if (!CalledValue || isa<Constant>(CalledValue)) 369 continue; 370 371 // Check if the instruction has a callees metadata. If so, add callees 372 // to CallGraphEdges to reflect the references from the metadata, and 373 // to enable importing for subsequent indirect call promotion and 374 // inlining. 375 if (auto *MD = I.getMetadata(LLVMContext::MD_callees)) { 376 for (auto &Op : MD->operands()) { 377 Function *Callee = mdconst::extract_or_null<Function>(Op); 378 if (Callee) 379 CallGraphEdges[Index.getOrInsertValueInfo(Callee)]; 380 } 381 } 382 383 uint32_t NumVals, NumCandidates; 384 uint64_t TotalCount; 385 auto CandidateProfileData = 386 ICallAnalysis.getPromotionCandidatesForInstruction( 387 &I, NumVals, TotalCount, NumCandidates); 388 for (auto &Candidate : CandidateProfileData) 389 CallGraphEdges[Index.getOrInsertValueInfo(Candidate.Value)] 390 .updateHotness(getHotness(Candidate.Count, PSI)); 391 } 392 } 393 Index.addBlockCount(F.size()); 394 395 std::vector<ValueInfo> Refs; 396 if (IsThinLTO) { 397 auto AddRefEdges = [&](const std::vector<const Instruction *> &Instrs, 398 SetVector<ValueInfo> &Edges, 399 SmallPtrSet<const User *, 8> &Cache) { 400 for (const auto *I : Instrs) { 401 Cache.erase(I); 402 findRefEdges(Index, I, Edges, Cache); 403 } 404 }; 405 406 // By now we processed all instructions in a function, except 407 // non-volatile loads and non-volatile value stores. Let's find 408 // ref edges for both of instruction sets 409 AddRefEdges(NonVolatileLoads, LoadRefEdges, Visited); 410 // We can add some values to the Visited set when processing load 411 // instructions which are also used by stores in NonVolatileStores. 412 // For example this can happen if we have following code: 413 // 414 // store %Derived* @foo, %Derived** bitcast (%Base** @bar to %Derived**) 415 // %42 = load %Derived*, %Derived** bitcast (%Base** @bar to %Derived**) 416 // 417 // After processing loads we'll add bitcast to the Visited set, and if 418 // we use the same set while processing stores, we'll never see store 419 // to @bar and @bar will be mistakenly treated as readonly. 420 SmallPtrSet<const llvm::User *, 8> StoreCache; 421 AddRefEdges(NonVolatileStores, StoreRefEdges, StoreCache); 422 423 // If both load and store instruction reference the same variable 424 // we won't be able to optimize it. Add all such reference edges 425 // to RefEdges set. 426 for (auto &VI : StoreRefEdges) 427 if (LoadRefEdges.remove(VI)) 428 RefEdges.insert(VI); 429 430 unsigned RefCnt = RefEdges.size(); 431 // All new reference edges inserted in two loops below are either 432 // read or write only. They will be grouped in the end of RefEdges 433 // vector, so we can use a single integer value to identify them. 434 for (auto &VI : LoadRefEdges) 435 RefEdges.insert(VI); 436 437 unsigned FirstWORef = RefEdges.size(); 438 for (auto &VI : StoreRefEdges) 439 RefEdges.insert(VI); 440 441 Refs = RefEdges.takeVector(); 442 for (; RefCnt < FirstWORef; ++RefCnt) 443 Refs[RefCnt].setReadOnly(); 444 445 for (; RefCnt < Refs.size(); ++RefCnt) 446 Refs[RefCnt].setWriteOnly(); 447 } else { 448 Refs = RefEdges.takeVector(); 449 } 450 // Explicit add hot edges to enforce importing for designated GUIDs for 451 // sample PGO, to enable the same inlines as the profiled optimized binary. 452 for (auto &I : F.getImportGUIDs()) 453 CallGraphEdges[Index.getOrInsertValueInfo(I)].updateHotness( 454 ForceSummaryEdgesCold == FunctionSummary::FSHT_All 455 ? CalleeInfo::HotnessType::Cold 456 : CalleeInfo::HotnessType::Critical); 457 458 bool NonRenamableLocal = isNonRenamableLocal(F); 459 bool NotEligibleForImport = 460 NonRenamableLocal || HasInlineAsmMaybeReferencingInternal; 461 GlobalValueSummary::GVFlags Flags(F.getLinkage(), NotEligibleForImport, 462 /* Live = */ false, F.isDSOLocal(), 463 F.hasLinkOnceODRLinkage() && F.hasGlobalUnnamedAddr()); 464 FunctionSummary::FFlags FunFlags{ 465 F.hasFnAttribute(Attribute::ReadNone), 466 F.hasFnAttribute(Attribute::ReadOnly), 467 F.hasFnAttribute(Attribute::NoRecurse), F.returnDoesNotAlias(), 468 // FIXME: refactor this to use the same code that inliner is using. 469 // Don't try to import functions with noinline attribute. 470 F.getAttributes().hasFnAttribute(Attribute::NoInline), 471 F.hasFnAttribute(Attribute::AlwaysInline)}; 472 auto FuncSummary = std::make_unique<FunctionSummary>( 473 Flags, NumInsts, FunFlags, /*EntryCount=*/0, std::move(Refs), 474 CallGraphEdges.takeVector(), TypeTests.takeVector(), 475 TypeTestAssumeVCalls.takeVector(), TypeCheckedLoadVCalls.takeVector(), 476 TypeTestAssumeConstVCalls.takeVector(), 477 TypeCheckedLoadConstVCalls.takeVector()); 478 if (NonRenamableLocal) 479 CantBePromoted.insert(F.getGUID()); 480 Index.addGlobalValueSummary(F, std::move(FuncSummary)); 481 } 482 483 /// Find function pointers referenced within the given vtable initializer 484 /// (or subset of an initializer) \p I. The starting offset of \p I within 485 /// the vtable initializer is \p StartingOffset. Any discovered function 486 /// pointers are added to \p VTableFuncs along with their cumulative offset 487 /// within the initializer. 488 static void findFuncPointers(const Constant *I, uint64_t StartingOffset, 489 const Module &M, ModuleSummaryIndex &Index, 490 VTableFuncList &VTableFuncs) { 491 // First check if this is a function pointer. 492 if (I->getType()->isPointerTy()) { 493 auto Fn = dyn_cast<Function>(I->stripPointerCasts()); 494 // We can disregard __cxa_pure_virtual as a possible call target, as 495 // calls to pure virtuals are UB. 496 if (Fn && Fn->getName() != "__cxa_pure_virtual") 497 VTableFuncs.push_back({Index.getOrInsertValueInfo(Fn), StartingOffset}); 498 return; 499 } 500 501 // Walk through the elements in the constant struct or array and recursively 502 // look for virtual function pointers. 503 const DataLayout &DL = M.getDataLayout(); 504 if (auto *C = dyn_cast<ConstantStruct>(I)) { 505 StructType *STy = dyn_cast<StructType>(C->getType()); 506 assert(STy); 507 const StructLayout *SL = DL.getStructLayout(C->getType()); 508 509 for (StructType::element_iterator EB = STy->element_begin(), EI = EB, 510 EE = STy->element_end(); 511 EI != EE; ++EI) { 512 auto Offset = SL->getElementOffset(EI - EB); 513 unsigned Op = SL->getElementContainingOffset(Offset); 514 findFuncPointers(cast<Constant>(I->getOperand(Op)), 515 StartingOffset + Offset, M, Index, VTableFuncs); 516 } 517 } else if (auto *C = dyn_cast<ConstantArray>(I)) { 518 ArrayType *ATy = C->getType(); 519 Type *EltTy = ATy->getElementType(); 520 uint64_t EltSize = DL.getTypeAllocSize(EltTy); 521 for (unsigned i = 0, e = ATy->getNumElements(); i != e; ++i) { 522 findFuncPointers(cast<Constant>(I->getOperand(i)), 523 StartingOffset + i * EltSize, M, Index, VTableFuncs); 524 } 525 } 526 } 527 528 // Identify the function pointers referenced by vtable definition \p V. 529 static void computeVTableFuncs(ModuleSummaryIndex &Index, 530 const GlobalVariable &V, const Module &M, 531 VTableFuncList &VTableFuncs) { 532 if (!V.isConstant()) 533 return; 534 535 findFuncPointers(V.getInitializer(), /*StartingOffset=*/0, M, Index, 536 VTableFuncs); 537 538 #ifndef NDEBUG 539 // Validate that the VTableFuncs list is ordered by offset. 540 uint64_t PrevOffset = 0; 541 for (auto &P : VTableFuncs) { 542 // The findVFuncPointers traversal should have encountered the 543 // functions in offset order. We need to use ">=" since PrevOffset 544 // starts at 0. 545 assert(P.VTableOffset >= PrevOffset); 546 PrevOffset = P.VTableOffset; 547 } 548 #endif 549 } 550 551 /// Record vtable definition \p V for each type metadata it references. 552 static void 553 recordTypeIdCompatibleVtableReferences(ModuleSummaryIndex &Index, 554 const GlobalVariable &V, 555 SmallVectorImpl<MDNode *> &Types) { 556 for (MDNode *Type : Types) { 557 auto TypeID = Type->getOperand(1).get(); 558 559 uint64_t Offset = 560 cast<ConstantInt>( 561 cast<ConstantAsMetadata>(Type->getOperand(0))->getValue()) 562 ->getZExtValue(); 563 564 if (auto *TypeId = dyn_cast<MDString>(TypeID)) 565 Index.getOrInsertTypeIdCompatibleVtableSummary(TypeId->getString()) 566 .push_back({Offset, Index.getOrInsertValueInfo(&V)}); 567 } 568 } 569 570 static void computeVariableSummary(ModuleSummaryIndex &Index, 571 const GlobalVariable &V, 572 DenseSet<GlobalValue::GUID> &CantBePromoted, 573 const Module &M, 574 SmallVectorImpl<MDNode *> &Types) { 575 SetVector<ValueInfo> RefEdges; 576 SmallPtrSet<const User *, 8> Visited; 577 bool HasBlockAddress = findRefEdges(Index, &V, RefEdges, Visited); 578 bool NonRenamableLocal = isNonRenamableLocal(V); 579 GlobalValueSummary::GVFlags Flags(V.getLinkage(), NonRenamableLocal, 580 /* Live = */ false, V.isDSOLocal(), 581 V.hasLinkOnceODRLinkage() && V.hasGlobalUnnamedAddr()); 582 583 VTableFuncList VTableFuncs; 584 // If splitting is not enabled, then we compute the summary information 585 // necessary for index-based whole program devirtualization. 586 if (!Index.enableSplitLTOUnit()) { 587 Types.clear(); 588 V.getMetadata(LLVMContext::MD_type, Types); 589 if (!Types.empty()) { 590 // Identify the function pointers referenced by this vtable definition. 591 computeVTableFuncs(Index, V, M, VTableFuncs); 592 593 // Record this vtable definition for each type metadata it references. 594 recordTypeIdCompatibleVtableReferences(Index, V, Types); 595 } 596 } 597 598 // Don't mark variables we won't be able to internalize as read/write-only. 599 bool CanBeInternalized = 600 !V.hasComdat() && !V.hasAppendingLinkage() && !V.isInterposable() && 601 !V.hasAvailableExternallyLinkage() && !V.hasDLLExportStorageClass(); 602 bool Constant = V.isConstant(); 603 GlobalVarSummary::GVarFlags VarFlags(CanBeInternalized, 604 Constant ? false : CanBeInternalized, 605 Constant, V.getVCallVisibility()); 606 auto GVarSummary = std::make_unique<GlobalVarSummary>(Flags, VarFlags, 607 RefEdges.takeVector()); 608 if (NonRenamableLocal) 609 CantBePromoted.insert(V.getGUID()); 610 if (HasBlockAddress) 611 GVarSummary->setNotEligibleToImport(); 612 if (!VTableFuncs.empty()) 613 GVarSummary->setVTableFuncs(VTableFuncs); 614 Index.addGlobalValueSummary(V, std::move(GVarSummary)); 615 } 616 617 static void 618 computeAliasSummary(ModuleSummaryIndex &Index, const GlobalAlias &A, 619 DenseSet<GlobalValue::GUID> &CantBePromoted) { 620 bool NonRenamableLocal = isNonRenamableLocal(A); 621 GlobalValueSummary::GVFlags Flags(A.getLinkage(), NonRenamableLocal, 622 /* Live = */ false, A.isDSOLocal(), 623 A.hasLinkOnceODRLinkage() && A.hasGlobalUnnamedAddr()); 624 auto AS = std::make_unique<AliasSummary>(Flags); 625 auto *Aliasee = A.getBaseObject(); 626 auto AliaseeVI = Index.getValueInfo(Aliasee->getGUID()); 627 assert(AliaseeVI && "Alias expects aliasee summary to be available"); 628 assert(AliaseeVI.getSummaryList().size() == 1 && 629 "Expected a single entry per aliasee in per-module index"); 630 AS->setAliasee(AliaseeVI, AliaseeVI.getSummaryList()[0].get()); 631 if (NonRenamableLocal) 632 CantBePromoted.insert(A.getGUID()); 633 Index.addGlobalValueSummary(A, std::move(AS)); 634 } 635 636 // Set LiveRoot flag on entries matching the given value name. 637 static void setLiveRoot(ModuleSummaryIndex &Index, StringRef Name) { 638 if (ValueInfo VI = Index.getValueInfo(GlobalValue::getGUID(Name))) 639 for (auto &Summary : VI.getSummaryList()) 640 Summary->setLive(true); 641 } 642 643 ModuleSummaryIndex llvm::buildModuleSummaryIndex( 644 const Module &M, 645 std::function<BlockFrequencyInfo *(const Function &F)> GetBFICallback, 646 ProfileSummaryInfo *PSI) { 647 assert(PSI); 648 bool EnableSplitLTOUnit = false; 649 if (auto *MD = mdconst::extract_or_null<ConstantInt>( 650 M.getModuleFlag("EnableSplitLTOUnit"))) 651 EnableSplitLTOUnit = MD->getZExtValue(); 652 ModuleSummaryIndex Index(/*HaveGVs=*/true, EnableSplitLTOUnit); 653 654 // Identify the local values in the llvm.used and llvm.compiler.used sets, 655 // which should not be exported as they would then require renaming and 656 // promotion, but we may have opaque uses e.g. in inline asm. We collect them 657 // here because we use this information to mark functions containing inline 658 // assembly calls as not importable. 659 SmallPtrSet<GlobalValue *, 8> LocalsUsed; 660 SmallPtrSet<GlobalValue *, 8> Used; 661 // First collect those in the llvm.used set. 662 collectUsedGlobalVariables(M, Used, /*CompilerUsed*/ false); 663 // Next collect those in the llvm.compiler.used set. 664 collectUsedGlobalVariables(M, Used, /*CompilerUsed*/ true); 665 DenseSet<GlobalValue::GUID> CantBePromoted; 666 for (auto *V : Used) { 667 if (V->hasLocalLinkage()) { 668 LocalsUsed.insert(V); 669 CantBePromoted.insert(V->getGUID()); 670 } 671 } 672 673 bool HasLocalInlineAsmSymbol = false; 674 if (!M.getModuleInlineAsm().empty()) { 675 // Collect the local values defined by module level asm, and set up 676 // summaries for these symbols so that they can be marked as NoRename, 677 // to prevent export of any use of them in regular IR that would require 678 // renaming within the module level asm. Note we don't need to create a 679 // summary for weak or global defs, as they don't need to be flagged as 680 // NoRename, and defs in module level asm can't be imported anyway. 681 // Also, any values used but not defined within module level asm should 682 // be listed on the llvm.used or llvm.compiler.used global and marked as 683 // referenced from there. 684 ModuleSymbolTable::CollectAsmSymbols( 685 M, [&](StringRef Name, object::BasicSymbolRef::Flags Flags) { 686 // Symbols not marked as Weak or Global are local definitions. 687 if (Flags & (object::BasicSymbolRef::SF_Weak | 688 object::BasicSymbolRef::SF_Global)) 689 return; 690 HasLocalInlineAsmSymbol = true; 691 GlobalValue *GV = M.getNamedValue(Name); 692 if (!GV) 693 return; 694 assert(GV->isDeclaration() && "Def in module asm already has definition"); 695 GlobalValueSummary::GVFlags GVFlags(GlobalValue::InternalLinkage, 696 /* NotEligibleToImport = */ true, 697 /* Live = */ true, 698 /* Local */ GV->isDSOLocal(), 699 GV->hasLinkOnceODRLinkage() && GV->hasGlobalUnnamedAddr()); 700 CantBePromoted.insert(GV->getGUID()); 701 // Create the appropriate summary type. 702 if (Function *F = dyn_cast<Function>(GV)) { 703 std::unique_ptr<FunctionSummary> Summary = 704 std::make_unique<FunctionSummary>( 705 GVFlags, /*InstCount=*/0, 706 FunctionSummary::FFlags{ 707 F->hasFnAttribute(Attribute::ReadNone), 708 F->hasFnAttribute(Attribute::ReadOnly), 709 F->hasFnAttribute(Attribute::NoRecurse), 710 F->returnDoesNotAlias(), 711 /* NoInline = */ false, 712 F->hasFnAttribute(Attribute::AlwaysInline)}, 713 /*EntryCount=*/0, ArrayRef<ValueInfo>{}, 714 ArrayRef<FunctionSummary::EdgeTy>{}, 715 ArrayRef<GlobalValue::GUID>{}, 716 ArrayRef<FunctionSummary::VFuncId>{}, 717 ArrayRef<FunctionSummary::VFuncId>{}, 718 ArrayRef<FunctionSummary::ConstVCall>{}, 719 ArrayRef<FunctionSummary::ConstVCall>{}); 720 Index.addGlobalValueSummary(*GV, std::move(Summary)); 721 } else { 722 std::unique_ptr<GlobalVarSummary> Summary = 723 std::make_unique<GlobalVarSummary>( 724 GVFlags, 725 GlobalVarSummary::GVarFlags( 726 false, false, cast<GlobalVariable>(GV)->isConstant(), 727 GlobalObject::VCallVisibilityPublic), 728 ArrayRef<ValueInfo>{}); 729 Index.addGlobalValueSummary(*GV, std::move(Summary)); 730 } 731 }); 732 } 733 734 bool IsThinLTO = true; 735 if (auto *MD = 736 mdconst::extract_or_null<ConstantInt>(M.getModuleFlag("ThinLTO"))) 737 IsThinLTO = MD->getZExtValue(); 738 739 // Compute summaries for all functions defined in module, and save in the 740 // index. 741 for (auto &F : M) { 742 if (F.isDeclaration()) 743 continue; 744 745 DominatorTree DT(const_cast<Function &>(F)); 746 BlockFrequencyInfo *BFI = nullptr; 747 std::unique_ptr<BlockFrequencyInfo> BFIPtr; 748 if (GetBFICallback) 749 BFI = GetBFICallback(F); 750 else if (F.hasProfileData()) { 751 LoopInfo LI{DT}; 752 BranchProbabilityInfo BPI{F, LI}; 753 BFIPtr = std::make_unique<BlockFrequencyInfo>(F, BPI, LI); 754 BFI = BFIPtr.get(); 755 } 756 757 computeFunctionSummary(Index, M, F, BFI, PSI, DT, 758 !LocalsUsed.empty() || HasLocalInlineAsmSymbol, 759 CantBePromoted, IsThinLTO); 760 } 761 762 // Compute summaries for all variables defined in module, and save in the 763 // index. 764 SmallVector<MDNode *, 2> Types; 765 for (const GlobalVariable &G : M.globals()) { 766 if (G.isDeclaration()) 767 continue; 768 computeVariableSummary(Index, G, CantBePromoted, M, Types); 769 } 770 771 // Compute summaries for all aliases defined in module, and save in the 772 // index. 773 for (const GlobalAlias &A : M.aliases()) 774 computeAliasSummary(Index, A, CantBePromoted); 775 776 for (auto *V : LocalsUsed) { 777 auto *Summary = Index.getGlobalValueSummary(*V); 778 assert(Summary && "Missing summary for global value"); 779 Summary->setNotEligibleToImport(); 780 } 781 782 // The linker doesn't know about these LLVM produced values, so we need 783 // to flag them as live in the index to ensure index-based dead value 784 // analysis treats them as live roots of the analysis. 785 setLiveRoot(Index, "llvm.used"); 786 setLiveRoot(Index, "llvm.compiler.used"); 787 setLiveRoot(Index, "llvm.global_ctors"); 788 setLiveRoot(Index, "llvm.global_dtors"); 789 setLiveRoot(Index, "llvm.global.annotations"); 790 791 for (auto &GlobalList : Index) { 792 // Ignore entries for references that are undefined in the current module. 793 if (GlobalList.second.SummaryList.empty()) 794 continue; 795 796 assert(GlobalList.second.SummaryList.size() == 1 && 797 "Expected module's index to have one summary per GUID"); 798 auto &Summary = GlobalList.second.SummaryList[0]; 799 if (!IsThinLTO) { 800 Summary->setNotEligibleToImport(); 801 continue; 802 } 803 804 bool AllRefsCanBeExternallyReferenced = 805 llvm::all_of(Summary->refs(), [&](const ValueInfo &VI) { 806 return !CantBePromoted.count(VI.getGUID()); 807 }); 808 if (!AllRefsCanBeExternallyReferenced) { 809 Summary->setNotEligibleToImport(); 810 continue; 811 } 812 813 if (auto *FuncSummary = dyn_cast<FunctionSummary>(Summary.get())) { 814 bool AllCallsCanBeExternallyReferenced = llvm::all_of( 815 FuncSummary->calls(), [&](const FunctionSummary::EdgeTy &Edge) { 816 return !CantBePromoted.count(Edge.first.getGUID()); 817 }); 818 if (!AllCallsCanBeExternallyReferenced) 819 Summary->setNotEligibleToImport(); 820 } 821 } 822 823 if (!ModuleSummaryDotFile.empty()) { 824 std::error_code EC; 825 raw_fd_ostream OSDot(ModuleSummaryDotFile, EC, sys::fs::OpenFlags::OF_None); 826 if (EC) 827 report_fatal_error(Twine("Failed to open dot file ") + 828 ModuleSummaryDotFile + ": " + EC.message() + "\n"); 829 Index.exportToDot(OSDot, {}); 830 } 831 832 return Index; 833 } 834 835 AnalysisKey ModuleSummaryIndexAnalysis::Key; 836 837 ModuleSummaryIndex 838 ModuleSummaryIndexAnalysis::run(Module &M, ModuleAnalysisManager &AM) { 839 ProfileSummaryInfo &PSI = AM.getResult<ProfileSummaryAnalysis>(M); 840 auto &FAM = AM.getResult<FunctionAnalysisManagerModuleProxy>(M).getManager(); 841 return buildModuleSummaryIndex( 842 M, 843 [&FAM](const Function &F) { 844 return &FAM.getResult<BlockFrequencyAnalysis>( 845 *const_cast<Function *>(&F)); 846 }, 847 &PSI); 848 } 849 850 char ModuleSummaryIndexWrapperPass::ID = 0; 851 852 INITIALIZE_PASS_BEGIN(ModuleSummaryIndexWrapperPass, "module-summary-analysis", 853 "Module Summary Analysis", false, true) 854 INITIALIZE_PASS_DEPENDENCY(BlockFrequencyInfoWrapperPass) 855 INITIALIZE_PASS_DEPENDENCY(ProfileSummaryInfoWrapperPass) 856 INITIALIZE_PASS_END(ModuleSummaryIndexWrapperPass, "module-summary-analysis", 857 "Module Summary Analysis", false, true) 858 859 ModulePass *llvm::createModuleSummaryIndexWrapperPass() { 860 return new ModuleSummaryIndexWrapperPass(); 861 } 862 863 ModuleSummaryIndexWrapperPass::ModuleSummaryIndexWrapperPass() 864 : ModulePass(ID) { 865 initializeModuleSummaryIndexWrapperPassPass(*PassRegistry::getPassRegistry()); 866 } 867 868 bool ModuleSummaryIndexWrapperPass::runOnModule(Module &M) { 869 auto *PSI = &getAnalysis<ProfileSummaryInfoWrapperPass>().getPSI(); 870 Index.emplace(buildModuleSummaryIndex( 871 M, 872 [this](const Function &F) { 873 return &(this->getAnalysis<BlockFrequencyInfoWrapperPass>( 874 *const_cast<Function *>(&F)) 875 .getBFI()); 876 }, 877 PSI)); 878 return false; 879 } 880 881 bool ModuleSummaryIndexWrapperPass::doFinalization(Module &M) { 882 Index.reset(); 883 return false; 884 } 885 886 void ModuleSummaryIndexWrapperPass::getAnalysisUsage(AnalysisUsage &AU) const { 887 AU.setPreservesAll(); 888 AU.addRequired<BlockFrequencyInfoWrapperPass>(); 889 AU.addRequired<ProfileSummaryInfoWrapperPass>(); 890 } 891