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/ConstantFolding.h" 26 #include "llvm/Analysis/IndirectCallPromotionAnalysis.h" 27 #include "llvm/Analysis/LoopInfo.h" 28 #include "llvm/Analysis/MemoryProfileInfo.h" 29 #include "llvm/Analysis/ProfileSummaryInfo.h" 30 #include "llvm/Analysis/StackSafetyAnalysis.h" 31 #include "llvm/Analysis/TypeMetadataUtils.h" 32 #include "llvm/IR/Attributes.h" 33 #include "llvm/IR/BasicBlock.h" 34 #include "llvm/IR/Constant.h" 35 #include "llvm/IR/Constants.h" 36 #include "llvm/IR/Dominators.h" 37 #include "llvm/IR/Function.h" 38 #include "llvm/IR/GlobalAlias.h" 39 #include "llvm/IR/GlobalValue.h" 40 #include "llvm/IR/GlobalVariable.h" 41 #include "llvm/IR/Instructions.h" 42 #include "llvm/IR/IntrinsicInst.h" 43 #include "llvm/IR/Metadata.h" 44 #include "llvm/IR/Module.h" 45 #include "llvm/IR/ModuleSummaryIndex.h" 46 #include "llvm/IR/Use.h" 47 #include "llvm/IR/User.h" 48 #include "llvm/InitializePasses.h" 49 #include "llvm/Object/ModuleSymbolTable.h" 50 #include "llvm/Object/SymbolicFile.h" 51 #include "llvm/Pass.h" 52 #include "llvm/Support/Casting.h" 53 #include "llvm/Support/CommandLine.h" 54 #include "llvm/Support/FileSystem.h" 55 #include <cassert> 56 #include <cstdint> 57 #include <vector> 58 59 using namespace llvm; 60 using namespace llvm::memprof; 61 62 #define DEBUG_TYPE "module-summary-analysis" 63 64 // Option to force edges cold which will block importing when the 65 // -import-cold-multiplier is set to 0. Useful for debugging. 66 namespace llvm { 67 FunctionSummary::ForceSummaryHotnessType ForceSummaryEdgesCold = 68 FunctionSummary::FSHT_None; 69 } // namespace llvm 70 71 static cl::opt<FunctionSummary::ForceSummaryHotnessType, true> FSEC( 72 "force-summary-edges-cold", cl::Hidden, cl::location(ForceSummaryEdgesCold), 73 cl::desc("Force all edges in the function summary to cold"), 74 cl::values(clEnumValN(FunctionSummary::FSHT_None, "none", "None."), 75 clEnumValN(FunctionSummary::FSHT_AllNonCritical, 76 "all-non-critical", "All non-critical edges."), 77 clEnumValN(FunctionSummary::FSHT_All, "all", "All edges."))); 78 79 static cl::opt<std::string> ModuleSummaryDotFile( 80 "module-summary-dot-file", cl::Hidden, cl::value_desc("filename"), 81 cl::desc("File to emit dot graph of new summary into")); 82 83 static cl::opt<bool> EnableMemProfIndirectCallSupport( 84 "enable-memprof-indirect-call-support", cl::init(false), cl::Hidden, 85 cl::desc( 86 "Enable MemProf support for summarizing and cloning indirect calls")); 87 88 extern cl::opt<bool> ScalePartialSampleProfileWorkingSetSize; 89 90 extern cl::opt<unsigned> MaxNumVTableAnnotations; 91 92 extern cl::opt<bool> MemProfReportHintedSizes; 93 94 // Walk through the operands of a given User via worklist iteration and populate 95 // the set of GlobalValue references encountered. Invoked either on an 96 // Instruction or a GlobalVariable (which walks its initializer). 97 // Return true if any of the operands contains blockaddress. This is important 98 // to know when computing summary for global var, because if global variable 99 // references basic block address we can't import it separately from function 100 // containing that basic block. For simplicity we currently don't import such 101 // global vars at all. When importing function we aren't interested if any 102 // instruction in it takes an address of any basic block, because instruction 103 // can only take an address of basic block located in the same function. 104 // Set `RefLocalLinkageIFunc` to true if the analyzed value references a 105 // local-linkage ifunc. 106 static bool 107 findRefEdges(ModuleSummaryIndex &Index, const User *CurUser, 108 SetVector<ValueInfo, SmallVector<ValueInfo, 0>> &RefEdges, 109 SmallPtrSet<const User *, 8> &Visited, 110 bool &RefLocalLinkageIFunc) { 111 bool HasBlockAddress = false; 112 SmallVector<const User *, 32> Worklist; 113 if (Visited.insert(CurUser).second) 114 Worklist.push_back(CurUser); 115 116 while (!Worklist.empty()) { 117 const User *U = Worklist.pop_back_val(); 118 const auto *CB = dyn_cast<CallBase>(U); 119 120 for (const auto &OI : U->operands()) { 121 const User *Operand = dyn_cast<User>(OI); 122 if (!Operand) 123 continue; 124 if (isa<BlockAddress>(Operand)) { 125 HasBlockAddress = true; 126 continue; 127 } 128 if (auto *GV = dyn_cast<GlobalValue>(Operand)) { 129 // We have a reference to a global value. This should be added to 130 // the reference set unless it is a callee. Callees are handled 131 // specially by WriteFunction and are added to a separate list. 132 if (!(CB && CB->isCallee(&OI))) { 133 // If an ifunc has local linkage, do not add it into ref edges, and 134 // sets `RefLocalLinkageIFunc` to true. The referencer is not eligible 135 // for import. An ifunc doesn't have summary and ThinLTO cannot 136 // promote it; importing the referencer may cause linkage errors. 137 if (auto *GI = dyn_cast_if_present<GlobalIFunc>(GV); 138 GI && GI->hasLocalLinkage()) { 139 RefLocalLinkageIFunc = true; 140 continue; 141 } 142 RefEdges.insert(Index.getOrInsertValueInfo(GV)); 143 } 144 continue; 145 } 146 if (Visited.insert(Operand).second) 147 Worklist.push_back(Operand); 148 } 149 } 150 151 const Instruction *I = dyn_cast<Instruction>(CurUser); 152 if (I) { 153 uint64_t TotalCount = 0; 154 // MaxNumVTableAnnotations is the maximum number of vtables annotated on 155 // the instruction. 156 auto ValueDataArray = getValueProfDataFromInst( 157 *I, IPVK_VTableTarget, MaxNumVTableAnnotations, TotalCount); 158 159 for (const auto &V : ValueDataArray) 160 RefEdges.insert(Index.getOrInsertValueInfo(/* VTableGUID = */ 161 V.Value)); 162 } 163 return HasBlockAddress; 164 } 165 166 static CalleeInfo::HotnessType getHotness(uint64_t ProfileCount, 167 ProfileSummaryInfo *PSI) { 168 if (!PSI) 169 return CalleeInfo::HotnessType::Unknown; 170 if (PSI->isHotCount(ProfileCount)) 171 return CalleeInfo::HotnessType::Hot; 172 if (PSI->isColdCount(ProfileCount)) 173 return CalleeInfo::HotnessType::Cold; 174 return CalleeInfo::HotnessType::None; 175 } 176 177 static bool isNonRenamableLocal(const GlobalValue &GV) { 178 return GV.hasSection() && GV.hasLocalLinkage(); 179 } 180 181 /// Determine whether this call has all constant integer arguments (excluding 182 /// "this") and summarize it to VCalls or ConstVCalls as appropriate. 183 static void addVCallToSet( 184 DevirtCallSite Call, GlobalValue::GUID Guid, 185 SetVector<FunctionSummary::VFuncId, std::vector<FunctionSummary::VFuncId>> 186 &VCalls, 187 SetVector<FunctionSummary::ConstVCall, 188 std::vector<FunctionSummary::ConstVCall>> &ConstVCalls) { 189 std::vector<uint64_t> Args; 190 // Start from the second argument to skip the "this" pointer. 191 for (auto &Arg : drop_begin(Call.CB.args())) { 192 auto *CI = dyn_cast<ConstantInt>(Arg); 193 if (!CI || CI->getBitWidth() > 64) { 194 VCalls.insert({Guid, Call.Offset}); 195 return; 196 } 197 Args.push_back(CI->getZExtValue()); 198 } 199 ConstVCalls.insert({{Guid, Call.Offset}, std::move(Args)}); 200 } 201 202 /// If this intrinsic call requires that we add information to the function 203 /// summary, do so via the non-constant reference arguments. 204 static void addIntrinsicToSummary( 205 const CallInst *CI, 206 SetVector<GlobalValue::GUID, std::vector<GlobalValue::GUID>> &TypeTests, 207 SetVector<FunctionSummary::VFuncId, std::vector<FunctionSummary::VFuncId>> 208 &TypeTestAssumeVCalls, 209 SetVector<FunctionSummary::VFuncId, std::vector<FunctionSummary::VFuncId>> 210 &TypeCheckedLoadVCalls, 211 SetVector<FunctionSummary::ConstVCall, 212 std::vector<FunctionSummary::ConstVCall>> 213 &TypeTestAssumeConstVCalls, 214 SetVector<FunctionSummary::ConstVCall, 215 std::vector<FunctionSummary::ConstVCall>> 216 &TypeCheckedLoadConstVCalls, 217 DominatorTree &DT) { 218 switch (CI->getCalledFunction()->getIntrinsicID()) { 219 case Intrinsic::type_test: 220 case Intrinsic::public_type_test: { 221 auto *TypeMDVal = cast<MetadataAsValue>(CI->getArgOperand(1)); 222 auto *TypeId = dyn_cast<MDString>(TypeMDVal->getMetadata()); 223 if (!TypeId) 224 break; 225 GlobalValue::GUID Guid = GlobalValue::getGUID(TypeId->getString()); 226 227 // Produce a summary from type.test intrinsics. We only summarize type.test 228 // intrinsics that are used other than by an llvm.assume intrinsic. 229 // Intrinsics that are assumed are relevant only to the devirtualization 230 // pass, not the type test lowering pass. 231 bool HasNonAssumeUses = llvm::any_of(CI->uses(), [](const Use &CIU) { 232 return !isa<AssumeInst>(CIU.getUser()); 233 }); 234 if (HasNonAssumeUses) 235 TypeTests.insert(Guid); 236 237 SmallVector<DevirtCallSite, 4> DevirtCalls; 238 SmallVector<CallInst *, 4> Assumes; 239 findDevirtualizableCallsForTypeTest(DevirtCalls, Assumes, CI, DT); 240 for (auto &Call : DevirtCalls) 241 addVCallToSet(Call, Guid, TypeTestAssumeVCalls, 242 TypeTestAssumeConstVCalls); 243 244 break; 245 } 246 247 case Intrinsic::type_checked_load_relative: 248 case Intrinsic::type_checked_load: { 249 auto *TypeMDVal = cast<MetadataAsValue>(CI->getArgOperand(2)); 250 auto *TypeId = dyn_cast<MDString>(TypeMDVal->getMetadata()); 251 if (!TypeId) 252 break; 253 GlobalValue::GUID Guid = GlobalValue::getGUID(TypeId->getString()); 254 255 SmallVector<DevirtCallSite, 4> DevirtCalls; 256 SmallVector<Instruction *, 4> LoadedPtrs; 257 SmallVector<Instruction *, 4> Preds; 258 bool HasNonCallUses = false; 259 findDevirtualizableCallsForTypeCheckedLoad(DevirtCalls, LoadedPtrs, Preds, 260 HasNonCallUses, CI, DT); 261 // Any non-call uses of the result of llvm.type.checked.load will 262 // prevent us from optimizing away the llvm.type.test. 263 if (HasNonCallUses) 264 TypeTests.insert(Guid); 265 for (auto &Call : DevirtCalls) 266 addVCallToSet(Call, Guid, TypeCheckedLoadVCalls, 267 TypeCheckedLoadConstVCalls); 268 269 break; 270 } 271 default: 272 break; 273 } 274 } 275 276 static bool isNonVolatileLoad(const Instruction *I) { 277 if (const auto *LI = dyn_cast<LoadInst>(I)) 278 return !LI->isVolatile(); 279 280 return false; 281 } 282 283 static bool isNonVolatileStore(const Instruction *I) { 284 if (const auto *SI = dyn_cast<StoreInst>(I)) 285 return !SI->isVolatile(); 286 287 return false; 288 } 289 290 // Returns true if the function definition must be unreachable. 291 // 292 // Note if this helper function returns true, `F` is guaranteed 293 // to be unreachable; if it returns false, `F` might still 294 // be unreachable but not covered by this helper function. 295 static bool mustBeUnreachableFunction(const Function &F) { 296 // A function must be unreachable if its entry block ends with an 297 // 'unreachable'. 298 assert(!F.isDeclaration()); 299 return isa<UnreachableInst>(F.getEntryBlock().getTerminator()); 300 } 301 302 static void computeFunctionSummary( 303 ModuleSummaryIndex &Index, const Module &M, const Function &F, 304 BlockFrequencyInfo *BFI, ProfileSummaryInfo *PSI, DominatorTree &DT, 305 bool HasLocalsInUsedOrAsm, DenseSet<GlobalValue::GUID> &CantBePromoted, 306 bool IsThinLTO, 307 std::function<const StackSafetyInfo *(const Function &F)> GetSSICallback) { 308 // Summary not currently supported for anonymous functions, they should 309 // have been named. 310 assert(F.hasName()); 311 312 unsigned NumInsts = 0; 313 // Map from callee ValueId to profile count. Used to accumulate profile 314 // counts for all static calls to a given callee. 315 MapVector<ValueInfo, CalleeInfo, DenseMap<ValueInfo, unsigned>, 316 SmallVector<FunctionSummary::EdgeTy, 0>> 317 CallGraphEdges; 318 SetVector<ValueInfo, SmallVector<ValueInfo, 0>> RefEdges, LoadRefEdges, 319 StoreRefEdges; 320 SetVector<GlobalValue::GUID, std::vector<GlobalValue::GUID>> TypeTests; 321 SetVector<FunctionSummary::VFuncId, std::vector<FunctionSummary::VFuncId>> 322 TypeTestAssumeVCalls, TypeCheckedLoadVCalls; 323 SetVector<FunctionSummary::ConstVCall, 324 std::vector<FunctionSummary::ConstVCall>> 325 TypeTestAssumeConstVCalls, TypeCheckedLoadConstVCalls; 326 ICallPromotionAnalysis ICallAnalysis; 327 SmallPtrSet<const User *, 8> Visited; 328 329 // Add personality function, prefix data and prologue data to function's ref 330 // list. 331 bool HasLocalIFuncCallOrRef = false; 332 findRefEdges(Index, &F, RefEdges, Visited, HasLocalIFuncCallOrRef); 333 std::vector<const Instruction *> NonVolatileLoads; 334 std::vector<const Instruction *> NonVolatileStores; 335 336 std::vector<CallsiteInfo> Callsites; 337 std::vector<AllocInfo> Allocs; 338 339 #ifndef NDEBUG 340 DenseSet<const CallBase *> CallsThatMayHaveMemprofSummary; 341 #endif 342 343 bool HasInlineAsmMaybeReferencingInternal = false; 344 bool HasIndirBranchToBlockAddress = false; 345 bool HasUnknownCall = false; 346 bool MayThrow = false; 347 for (const BasicBlock &BB : F) { 348 // We don't allow inlining of function with indirect branch to blockaddress. 349 // If the blockaddress escapes the function, e.g., via a global variable, 350 // inlining may lead to an invalid cross-function reference. So we shouldn't 351 // import such function either. 352 if (BB.hasAddressTaken()) { 353 for (User *U : BlockAddress::get(const_cast<BasicBlock *>(&BB))->users()) 354 if (!isa<CallBrInst>(*U)) { 355 HasIndirBranchToBlockAddress = true; 356 break; 357 } 358 } 359 360 for (const Instruction &I : BB) { 361 if (I.isDebugOrPseudoInst()) 362 continue; 363 ++NumInsts; 364 365 // Regular LTO module doesn't participate in ThinLTO import, 366 // so no reference from it can be read/writeonly, since this 367 // would require importing variable as local copy 368 if (IsThinLTO) { 369 if (isNonVolatileLoad(&I)) { 370 // Postpone processing of non-volatile load instructions 371 // See comments below 372 Visited.insert(&I); 373 NonVolatileLoads.push_back(&I); 374 continue; 375 } else if (isNonVolatileStore(&I)) { 376 Visited.insert(&I); 377 NonVolatileStores.push_back(&I); 378 // All references from second operand of store (destination address) 379 // can be considered write-only if they're not referenced by any 380 // non-store instruction. References from first operand of store 381 // (stored value) can't be treated either as read- or as write-only 382 // so we add them to RefEdges as we do with all other instructions 383 // except non-volatile load. 384 Value *Stored = I.getOperand(0); 385 if (auto *GV = dyn_cast<GlobalValue>(Stored)) 386 // findRefEdges will try to examine GV operands, so instead 387 // of calling it we should add GV to RefEdges directly. 388 RefEdges.insert(Index.getOrInsertValueInfo(GV)); 389 else if (auto *U = dyn_cast<User>(Stored)) 390 findRefEdges(Index, U, RefEdges, Visited, HasLocalIFuncCallOrRef); 391 continue; 392 } 393 } 394 findRefEdges(Index, &I, RefEdges, Visited, HasLocalIFuncCallOrRef); 395 const auto *CB = dyn_cast<CallBase>(&I); 396 if (!CB) { 397 if (I.mayThrow()) 398 MayThrow = true; 399 continue; 400 } 401 402 const auto *CI = dyn_cast<CallInst>(&I); 403 // Since we don't know exactly which local values are referenced in inline 404 // assembly, conservatively mark the function as possibly referencing 405 // a local value from inline assembly to ensure we don't export a 406 // reference (which would require renaming and promotion of the 407 // referenced value). 408 if (HasLocalsInUsedOrAsm && CI && CI->isInlineAsm()) 409 HasInlineAsmMaybeReferencingInternal = true; 410 411 // Compute this once per indirect call. 412 uint32_t NumCandidates = 0; 413 uint64_t TotalCount = 0; 414 MutableArrayRef<InstrProfValueData> CandidateProfileData; 415 416 auto *CalledValue = CB->getCalledOperand(); 417 auto *CalledFunction = CB->getCalledFunction(); 418 if (CalledValue && !CalledFunction) { 419 CalledValue = CalledValue->stripPointerCasts(); 420 // Stripping pointer casts can reveal a called function. 421 CalledFunction = dyn_cast<Function>(CalledValue); 422 } 423 // Check if this is an alias to a function. If so, get the 424 // called aliasee for the checks below. 425 if (auto *GA = dyn_cast<GlobalAlias>(CalledValue)) { 426 assert(!CalledFunction && "Expected null called function in callsite for alias"); 427 CalledFunction = dyn_cast<Function>(GA->getAliaseeObject()); 428 } 429 // Check if this is a direct call to a known function or a known 430 // intrinsic, or an indirect call with profile data. 431 if (CalledFunction) { 432 if (CI && CalledFunction->isIntrinsic()) { 433 addIntrinsicToSummary( 434 CI, TypeTests, TypeTestAssumeVCalls, TypeCheckedLoadVCalls, 435 TypeTestAssumeConstVCalls, TypeCheckedLoadConstVCalls, DT); 436 continue; 437 } 438 // We should have named any anonymous globals 439 assert(CalledFunction->hasName()); 440 auto ScaledCount = PSI->getProfileCount(*CB, BFI); 441 auto Hotness = ScaledCount ? getHotness(*ScaledCount, PSI) 442 : CalleeInfo::HotnessType::Unknown; 443 if (ForceSummaryEdgesCold != FunctionSummary::FSHT_None) 444 Hotness = CalleeInfo::HotnessType::Cold; 445 446 // Use the original CalledValue, in case it was an alias. We want 447 // to record the call edge to the alias in that case. Eventually 448 // an alias summary will be created to associate the alias and 449 // aliasee. 450 auto &ValueInfo = CallGraphEdges[Index.getOrInsertValueInfo( 451 cast<GlobalValue>(CalledValue))]; 452 ValueInfo.updateHotness(Hotness); 453 if (CB->isTailCall()) 454 ValueInfo.setHasTailCall(true); 455 // Add the relative block frequency to CalleeInfo if there is no profile 456 // information. 457 if (BFI != nullptr && Hotness == CalleeInfo::HotnessType::Unknown) { 458 uint64_t BBFreq = BFI->getBlockFreq(&BB).getFrequency(); 459 uint64_t EntryFreq = BFI->getEntryFreq().getFrequency(); 460 ValueInfo.updateRelBlockFreq(BBFreq, EntryFreq); 461 } 462 } else { 463 HasUnknownCall = true; 464 // If F is imported, a local linkage ifunc (e.g. target_clones on a 465 // static function) called by F will be cloned. Since summaries don't 466 // track ifunc, we do not know implementation functions referenced by 467 // the ifunc resolver need to be promoted in the exporter, and we will 468 // get linker errors due to cloned declarations for implementation 469 // functions. As a simple fix, just mark F as not eligible for import. 470 // Non-local ifunc is not cloned and does not have the issue. 471 if (auto *GI = dyn_cast_if_present<GlobalIFunc>(CalledValue)) 472 if (GI->hasLocalLinkage()) 473 HasLocalIFuncCallOrRef = true; 474 // Skip inline assembly calls. 475 if (CI && CI->isInlineAsm()) 476 continue; 477 // Skip direct calls. 478 if (!CalledValue || isa<Constant>(CalledValue)) 479 continue; 480 481 // Check if the instruction has a callees metadata. If so, add callees 482 // to CallGraphEdges to reflect the references from the metadata, and 483 // to enable importing for subsequent indirect call promotion and 484 // inlining. 485 if (auto *MD = I.getMetadata(LLVMContext::MD_callees)) { 486 for (const auto &Op : MD->operands()) { 487 Function *Callee = mdconst::extract_or_null<Function>(Op); 488 if (Callee) 489 CallGraphEdges[Index.getOrInsertValueInfo(Callee)]; 490 } 491 } 492 493 CandidateProfileData = 494 ICallAnalysis.getPromotionCandidatesForInstruction(&I, TotalCount, 495 NumCandidates); 496 for (const auto &Candidate : CandidateProfileData) 497 CallGraphEdges[Index.getOrInsertValueInfo(Candidate.Value)] 498 .updateHotness(getHotness(Candidate.Count, PSI)); 499 } 500 501 // Summarize memprof related metadata. This is only needed for ThinLTO. 502 if (!IsThinLTO) 503 continue; 504 505 // Skip indirect calls if we haven't enabled memprof ICP. 506 if (!CalledFunction && !EnableMemProfIndirectCallSupport) 507 continue; 508 509 // Ensure we keep this analysis in sync with the handling in the ThinLTO 510 // backend (see MemProfContextDisambiguation::applyImport). Save this call 511 // so that we can skip it in checking the reverse case later. 512 assert(mayHaveMemprofSummary(CB)); 513 #ifndef NDEBUG 514 CallsThatMayHaveMemprofSummary.insert(CB); 515 #endif 516 517 // Compute the list of stack ids first (so we can trim them from the stack 518 // ids on any MIBs). 519 CallStack<MDNode, MDNode::op_iterator> InstCallsite( 520 I.getMetadata(LLVMContext::MD_callsite)); 521 auto *MemProfMD = I.getMetadata(LLVMContext::MD_memprof); 522 if (MemProfMD) { 523 std::vector<MIBInfo> MIBs; 524 std::vector<uint64_t> TotalSizes; 525 for (auto &MDOp : MemProfMD->operands()) { 526 auto *MIBMD = cast<const MDNode>(MDOp); 527 MDNode *StackNode = getMIBStackNode(MIBMD); 528 assert(StackNode); 529 SmallVector<unsigned> StackIdIndices; 530 CallStack<MDNode, MDNode::op_iterator> StackContext(StackNode); 531 // Collapse out any on the allocation call (inlining). 532 for (auto ContextIter = 533 StackContext.beginAfterSharedPrefix(InstCallsite); 534 ContextIter != StackContext.end(); ++ContextIter) { 535 unsigned StackIdIdx = Index.addOrGetStackIdIndex(*ContextIter); 536 // If this is a direct recursion, simply skip the duplicate 537 // entries. If this is mutual recursion, handling is left to 538 // the LTO link analysis client. 539 if (StackIdIndices.empty() || StackIdIndices.back() != StackIdIdx) 540 StackIdIndices.push_back(StackIdIdx); 541 } 542 MIBs.push_back( 543 MIBInfo(getMIBAllocType(MIBMD), std::move(StackIdIndices))); 544 if (MemProfReportHintedSizes) { 545 auto TotalSize = getMIBTotalSize(MIBMD); 546 assert(TotalSize); 547 TotalSizes.push_back(TotalSize); 548 } 549 } 550 Allocs.push_back(AllocInfo(std::move(MIBs))); 551 if (MemProfReportHintedSizes) { 552 assert(Allocs.back().MIBs.size() == TotalSizes.size()); 553 Allocs.back().TotalSizes = std::move(TotalSizes); 554 } 555 } else if (!InstCallsite.empty()) { 556 SmallVector<unsigned> StackIdIndices; 557 for (auto StackId : InstCallsite) 558 StackIdIndices.push_back(Index.addOrGetStackIdIndex(StackId)); 559 if (CalledFunction) { 560 // Use the original CalledValue, in case it was an alias. We want 561 // to record the call edge to the alias in that case. Eventually 562 // an alias summary will be created to associate the alias and 563 // aliasee. 564 auto CalleeValueInfo = 565 Index.getOrInsertValueInfo(cast<GlobalValue>(CalledValue)); 566 Callsites.push_back({CalleeValueInfo, StackIdIndices}); 567 } else { 568 assert(EnableMemProfIndirectCallSupport); 569 // For indirect callsites, create multiple Callsites, one per target. 570 // This enables having a different set of clone versions per target, 571 // and we will apply the cloning decisions while speculatively 572 // devirtualizing in the ThinLTO backends. 573 for (const auto &Candidate : CandidateProfileData) { 574 auto CalleeValueInfo = Index.getOrInsertValueInfo(Candidate.Value); 575 Callsites.push_back({CalleeValueInfo, StackIdIndices}); 576 } 577 } 578 } 579 } 580 } 581 582 if (PSI->hasPartialSampleProfile() && ScalePartialSampleProfileWorkingSetSize) 583 Index.addBlockCount(F.size()); 584 585 SmallVector<ValueInfo, 0> Refs; 586 if (IsThinLTO) { 587 auto AddRefEdges = 588 [&](const std::vector<const Instruction *> &Instrs, 589 SetVector<ValueInfo, SmallVector<ValueInfo, 0>> &Edges, 590 SmallPtrSet<const User *, 8> &Cache) { 591 for (const auto *I : Instrs) { 592 Cache.erase(I); 593 findRefEdges(Index, I, Edges, Cache, HasLocalIFuncCallOrRef); 594 } 595 }; 596 597 // By now we processed all instructions in a function, except 598 // non-volatile loads and non-volatile value stores. Let's find 599 // ref edges for both of instruction sets 600 AddRefEdges(NonVolatileLoads, LoadRefEdges, Visited); 601 // We can add some values to the Visited set when processing load 602 // instructions which are also used by stores in NonVolatileStores. 603 // For example this can happen if we have following code: 604 // 605 // store %Derived* @foo, %Derived** bitcast (%Base** @bar to %Derived**) 606 // %42 = load %Derived*, %Derived** bitcast (%Base** @bar to %Derived**) 607 // 608 // After processing loads we'll add bitcast to the Visited set, and if 609 // we use the same set while processing stores, we'll never see store 610 // to @bar and @bar will be mistakenly treated as readonly. 611 SmallPtrSet<const llvm::User *, 8> StoreCache; 612 AddRefEdges(NonVolatileStores, StoreRefEdges, StoreCache); 613 614 // If both load and store instruction reference the same variable 615 // we won't be able to optimize it. Add all such reference edges 616 // to RefEdges set. 617 for (const auto &VI : StoreRefEdges) 618 if (LoadRefEdges.remove(VI)) 619 RefEdges.insert(VI); 620 621 unsigned RefCnt = RefEdges.size(); 622 // All new reference edges inserted in two loops below are either 623 // read or write only. They will be grouped in the end of RefEdges 624 // vector, so we can use a single integer value to identify them. 625 for (const auto &VI : LoadRefEdges) 626 RefEdges.insert(VI); 627 628 unsigned FirstWORef = RefEdges.size(); 629 for (const auto &VI : StoreRefEdges) 630 RefEdges.insert(VI); 631 632 Refs = RefEdges.takeVector(); 633 for (; RefCnt < FirstWORef; ++RefCnt) 634 Refs[RefCnt].setReadOnly(); 635 636 for (; RefCnt < Refs.size(); ++RefCnt) 637 Refs[RefCnt].setWriteOnly(); 638 } else { 639 Refs = RefEdges.takeVector(); 640 } 641 // Explicit add hot edges to enforce importing for designated GUIDs for 642 // sample PGO, to enable the same inlines as the profiled optimized binary. 643 for (auto &I : F.getImportGUIDs()) 644 CallGraphEdges[Index.getOrInsertValueInfo(I)].updateHotness( 645 ForceSummaryEdgesCold == FunctionSummary::FSHT_All 646 ? CalleeInfo::HotnessType::Cold 647 : CalleeInfo::HotnessType::Critical); 648 649 #ifndef NDEBUG 650 // Make sure that all calls we decided could not have memprof summaries get a 651 // false value for mayHaveMemprofSummary, to ensure that this handling remains 652 // in sync with the ThinLTO backend handling. 653 if (IsThinLTO) { 654 for (const BasicBlock &BB : F) { 655 for (const Instruction &I : BB) { 656 const auto *CB = dyn_cast<CallBase>(&I); 657 if (!CB) 658 continue; 659 // We already checked these above. 660 if (CallsThatMayHaveMemprofSummary.count(CB)) 661 continue; 662 assert(!mayHaveMemprofSummary(CB)); 663 } 664 } 665 } 666 #endif 667 668 bool NonRenamableLocal = isNonRenamableLocal(F); 669 bool NotEligibleForImport = 670 NonRenamableLocal || HasInlineAsmMaybeReferencingInternal || 671 HasIndirBranchToBlockAddress || HasLocalIFuncCallOrRef; 672 GlobalValueSummary::GVFlags Flags( 673 F.getLinkage(), F.getVisibility(), NotEligibleForImport, 674 /* Live = */ false, F.isDSOLocal(), F.canBeOmittedFromSymbolTable(), 675 GlobalValueSummary::ImportKind::Definition); 676 FunctionSummary::FFlags FunFlags{ 677 F.doesNotAccessMemory(), F.onlyReadsMemory() && !F.doesNotAccessMemory(), 678 F.hasFnAttribute(Attribute::NoRecurse), F.returnDoesNotAlias(), 679 // FIXME: refactor this to use the same code that inliner is using. 680 // Don't try to import functions with noinline attribute. 681 F.getAttributes().hasFnAttr(Attribute::NoInline), 682 F.hasFnAttribute(Attribute::AlwaysInline), 683 F.hasFnAttribute(Attribute::NoUnwind), MayThrow, HasUnknownCall, 684 mustBeUnreachableFunction(F)}; 685 std::vector<FunctionSummary::ParamAccess> ParamAccesses; 686 if (auto *SSI = GetSSICallback(F)) 687 ParamAccesses = SSI->getParamAccesses(Index); 688 auto FuncSummary = std::make_unique<FunctionSummary>( 689 Flags, NumInsts, FunFlags, std::move(Refs), CallGraphEdges.takeVector(), 690 TypeTests.takeVector(), TypeTestAssumeVCalls.takeVector(), 691 TypeCheckedLoadVCalls.takeVector(), 692 TypeTestAssumeConstVCalls.takeVector(), 693 TypeCheckedLoadConstVCalls.takeVector(), std::move(ParamAccesses), 694 std::move(Callsites), std::move(Allocs)); 695 if (NonRenamableLocal) 696 CantBePromoted.insert(F.getGUID()); 697 Index.addGlobalValueSummary(F, std::move(FuncSummary)); 698 } 699 700 /// Find function pointers referenced within the given vtable initializer 701 /// (or subset of an initializer) \p I. The starting offset of \p I within 702 /// the vtable initializer is \p StartingOffset. Any discovered function 703 /// pointers are added to \p VTableFuncs along with their cumulative offset 704 /// within the initializer. 705 static void findFuncPointers(const Constant *I, uint64_t StartingOffset, 706 const Module &M, ModuleSummaryIndex &Index, 707 VTableFuncList &VTableFuncs, 708 const GlobalVariable &OrigGV) { 709 // First check if this is a function pointer. 710 if (I->getType()->isPointerTy()) { 711 auto C = I->stripPointerCasts(); 712 auto A = dyn_cast<GlobalAlias>(C); 713 if (isa<Function>(C) || (A && isa<Function>(A->getAliasee()))) { 714 auto GV = dyn_cast<GlobalValue>(C); 715 assert(GV); 716 // We can disregard __cxa_pure_virtual as a possible call target, as 717 // calls to pure virtuals are UB. 718 if (GV && GV->getName() != "__cxa_pure_virtual") 719 VTableFuncs.push_back({Index.getOrInsertValueInfo(GV), StartingOffset}); 720 return; 721 } 722 } 723 724 // Walk through the elements in the constant struct or array and recursively 725 // look for virtual function pointers. 726 const DataLayout &DL = M.getDataLayout(); 727 if (auto *C = dyn_cast<ConstantStruct>(I)) { 728 StructType *STy = dyn_cast<StructType>(C->getType()); 729 assert(STy); 730 const StructLayout *SL = DL.getStructLayout(C->getType()); 731 732 for (auto EI : llvm::enumerate(STy->elements())) { 733 auto Offset = SL->getElementOffset(EI.index()); 734 unsigned Op = SL->getElementContainingOffset(Offset); 735 findFuncPointers(cast<Constant>(I->getOperand(Op)), 736 StartingOffset + Offset, M, Index, VTableFuncs, OrigGV); 737 } 738 } else if (auto *C = dyn_cast<ConstantArray>(I)) { 739 ArrayType *ATy = C->getType(); 740 Type *EltTy = ATy->getElementType(); 741 uint64_t EltSize = DL.getTypeAllocSize(EltTy); 742 for (unsigned i = 0, e = ATy->getNumElements(); i != e; ++i) { 743 findFuncPointers(cast<Constant>(I->getOperand(i)), 744 StartingOffset + i * EltSize, M, Index, VTableFuncs, 745 OrigGV); 746 } 747 } else if (const auto *CE = dyn_cast<ConstantExpr>(I)) { 748 // For relative vtables, the next sub-component should be a trunc. 749 if (CE->getOpcode() != Instruction::Trunc || 750 !(CE = dyn_cast<ConstantExpr>(CE->getOperand(0)))) 751 return; 752 753 // If this constant can be reduced to the offset between a function and a 754 // global, then we know this is a valid virtual function if the RHS is the 755 // original vtable we're scanning through. 756 if (CE->getOpcode() == Instruction::Sub) { 757 GlobalValue *LHS, *RHS; 758 APSInt LHSOffset, RHSOffset; 759 if (IsConstantOffsetFromGlobal(CE->getOperand(0), LHS, LHSOffset, DL) && 760 IsConstantOffsetFromGlobal(CE->getOperand(1), RHS, RHSOffset, DL) && 761 RHS == &OrigGV && 762 763 // For relative vtables, this component should point to the callable 764 // function without any offsets. 765 LHSOffset == 0 && 766 767 // Also, the RHS should always point to somewhere within the vtable. 768 RHSOffset <= 769 static_cast<uint64_t>(DL.getTypeAllocSize(OrigGV.getInitializer()->getType()))) { 770 findFuncPointers(LHS, StartingOffset, M, Index, VTableFuncs, OrigGV); 771 } 772 } 773 } 774 } 775 776 // Identify the function pointers referenced by vtable definition \p V. 777 static void computeVTableFuncs(ModuleSummaryIndex &Index, 778 const GlobalVariable &V, const Module &M, 779 VTableFuncList &VTableFuncs) { 780 if (!V.isConstant()) 781 return; 782 783 findFuncPointers(V.getInitializer(), /*StartingOffset=*/0, M, Index, 784 VTableFuncs, V); 785 786 #ifndef NDEBUG 787 // Validate that the VTableFuncs list is ordered by offset. 788 uint64_t PrevOffset = 0; 789 for (auto &P : VTableFuncs) { 790 // The findVFuncPointers traversal should have encountered the 791 // functions in offset order. We need to use ">=" since PrevOffset 792 // starts at 0. 793 assert(P.VTableOffset >= PrevOffset); 794 PrevOffset = P.VTableOffset; 795 } 796 #endif 797 } 798 799 /// Record vtable definition \p V for each type metadata it references. 800 static void 801 recordTypeIdCompatibleVtableReferences(ModuleSummaryIndex &Index, 802 const GlobalVariable &V, 803 SmallVectorImpl<MDNode *> &Types) { 804 for (MDNode *Type : Types) { 805 auto TypeID = Type->getOperand(1).get(); 806 807 uint64_t Offset = 808 cast<ConstantInt>( 809 cast<ConstantAsMetadata>(Type->getOperand(0))->getValue()) 810 ->getZExtValue(); 811 812 if (auto *TypeId = dyn_cast<MDString>(TypeID)) 813 Index.getOrInsertTypeIdCompatibleVtableSummary(TypeId->getString()) 814 .push_back({Offset, Index.getOrInsertValueInfo(&V)}); 815 } 816 } 817 818 static void computeVariableSummary(ModuleSummaryIndex &Index, 819 const GlobalVariable &V, 820 DenseSet<GlobalValue::GUID> &CantBePromoted, 821 const Module &M, 822 SmallVectorImpl<MDNode *> &Types) { 823 SetVector<ValueInfo, SmallVector<ValueInfo, 0>> RefEdges; 824 SmallPtrSet<const User *, 8> Visited; 825 bool RefLocalIFunc = false; 826 bool HasBlockAddress = 827 findRefEdges(Index, &V, RefEdges, Visited, RefLocalIFunc); 828 const bool NotEligibleForImport = (HasBlockAddress || RefLocalIFunc); 829 bool NonRenamableLocal = isNonRenamableLocal(V); 830 GlobalValueSummary::GVFlags Flags( 831 V.getLinkage(), V.getVisibility(), NonRenamableLocal, 832 /* Live = */ false, V.isDSOLocal(), V.canBeOmittedFromSymbolTable(), 833 GlobalValueSummary::Definition); 834 835 VTableFuncList VTableFuncs; 836 // If splitting is not enabled, then we compute the summary information 837 // necessary for index-based whole program devirtualization. 838 if (!Index.enableSplitLTOUnit()) { 839 Types.clear(); 840 V.getMetadata(LLVMContext::MD_type, Types); 841 if (!Types.empty()) { 842 // Identify the function pointers referenced by this vtable definition. 843 computeVTableFuncs(Index, V, M, VTableFuncs); 844 845 // Record this vtable definition for each type metadata it references. 846 recordTypeIdCompatibleVtableReferences(Index, V, Types); 847 } 848 } 849 850 // Don't mark variables we won't be able to internalize as read/write-only. 851 bool CanBeInternalized = 852 !V.hasComdat() && !V.hasAppendingLinkage() && !V.isInterposable() && 853 !V.hasAvailableExternallyLinkage() && !V.hasDLLExportStorageClass(); 854 bool Constant = V.isConstant(); 855 GlobalVarSummary::GVarFlags VarFlags(CanBeInternalized, 856 Constant ? false : CanBeInternalized, 857 Constant, V.getVCallVisibility()); 858 auto GVarSummary = std::make_unique<GlobalVarSummary>(Flags, VarFlags, 859 RefEdges.takeVector()); 860 if (NonRenamableLocal) 861 CantBePromoted.insert(V.getGUID()); 862 if (NotEligibleForImport) 863 GVarSummary->setNotEligibleToImport(); 864 if (!VTableFuncs.empty()) 865 GVarSummary->setVTableFuncs(VTableFuncs); 866 Index.addGlobalValueSummary(V, std::move(GVarSummary)); 867 } 868 869 static void computeAliasSummary(ModuleSummaryIndex &Index, const GlobalAlias &A, 870 DenseSet<GlobalValue::GUID> &CantBePromoted) { 871 // Skip summary for indirect function aliases as summary for aliasee will not 872 // be emitted. 873 const GlobalObject *Aliasee = A.getAliaseeObject(); 874 if (isa<GlobalIFunc>(Aliasee)) 875 return; 876 bool NonRenamableLocal = isNonRenamableLocal(A); 877 GlobalValueSummary::GVFlags Flags( 878 A.getLinkage(), A.getVisibility(), NonRenamableLocal, 879 /* Live = */ false, A.isDSOLocal(), A.canBeOmittedFromSymbolTable(), 880 GlobalValueSummary::Definition); 881 auto AS = std::make_unique<AliasSummary>(Flags); 882 auto AliaseeVI = Index.getValueInfo(Aliasee->getGUID()); 883 assert(AliaseeVI && "Alias expects aliasee summary to be available"); 884 assert(AliaseeVI.getSummaryList().size() == 1 && 885 "Expected a single entry per aliasee in per-module index"); 886 AS->setAliasee(AliaseeVI, AliaseeVI.getSummaryList()[0].get()); 887 if (NonRenamableLocal) 888 CantBePromoted.insert(A.getGUID()); 889 Index.addGlobalValueSummary(A, std::move(AS)); 890 } 891 892 // Set LiveRoot flag on entries matching the given value name. 893 static void setLiveRoot(ModuleSummaryIndex &Index, StringRef Name) { 894 if (ValueInfo VI = Index.getValueInfo(GlobalValue::getGUID(Name))) 895 for (const auto &Summary : VI.getSummaryList()) 896 Summary->setLive(true); 897 } 898 899 ModuleSummaryIndex llvm::buildModuleSummaryIndex( 900 const Module &M, 901 std::function<BlockFrequencyInfo *(const Function &F)> GetBFICallback, 902 ProfileSummaryInfo *PSI, 903 std::function<const StackSafetyInfo *(const Function &F)> GetSSICallback) { 904 assert(PSI); 905 bool EnableSplitLTOUnit = false; 906 bool UnifiedLTO = false; 907 if (auto *MD = mdconst::extract_or_null<ConstantInt>( 908 M.getModuleFlag("EnableSplitLTOUnit"))) 909 EnableSplitLTOUnit = MD->getZExtValue(); 910 if (auto *MD = 911 mdconst::extract_or_null<ConstantInt>(M.getModuleFlag("UnifiedLTO"))) 912 UnifiedLTO = MD->getZExtValue(); 913 ModuleSummaryIndex Index(/*HaveGVs=*/true, EnableSplitLTOUnit, UnifiedLTO); 914 915 // Identify the local values in the llvm.used and llvm.compiler.used sets, 916 // which should not be exported as they would then require renaming and 917 // promotion, but we may have opaque uses e.g. in inline asm. We collect them 918 // here because we use this information to mark functions containing inline 919 // assembly calls as not importable. 920 SmallPtrSet<GlobalValue *, 4> LocalsUsed; 921 SmallVector<GlobalValue *, 4> Used; 922 // First collect those in the llvm.used set. 923 collectUsedGlobalVariables(M, Used, /*CompilerUsed=*/false); 924 // Next collect those in the llvm.compiler.used set. 925 collectUsedGlobalVariables(M, Used, /*CompilerUsed=*/true); 926 DenseSet<GlobalValue::GUID> CantBePromoted; 927 for (auto *V : Used) { 928 if (V->hasLocalLinkage()) { 929 LocalsUsed.insert(V); 930 CantBePromoted.insert(V->getGUID()); 931 } 932 } 933 934 bool HasLocalInlineAsmSymbol = false; 935 if (!M.getModuleInlineAsm().empty()) { 936 // Collect the local values defined by module level asm, and set up 937 // summaries for these symbols so that they can be marked as NoRename, 938 // to prevent export of any use of them in regular IR that would require 939 // renaming within the module level asm. Note we don't need to create a 940 // summary for weak or global defs, as they don't need to be flagged as 941 // NoRename, and defs in module level asm can't be imported anyway. 942 // Also, any values used but not defined within module level asm should 943 // be listed on the llvm.used or llvm.compiler.used global and marked as 944 // referenced from there. 945 ModuleSymbolTable::CollectAsmSymbols( 946 M, [&](StringRef Name, object::BasicSymbolRef::Flags Flags) { 947 // Symbols not marked as Weak or Global are local definitions. 948 if (Flags & (object::BasicSymbolRef::SF_Weak | 949 object::BasicSymbolRef::SF_Global)) 950 return; 951 HasLocalInlineAsmSymbol = true; 952 GlobalValue *GV = M.getNamedValue(Name); 953 if (!GV) 954 return; 955 assert(GV->isDeclaration() && "Def in module asm already has definition"); 956 GlobalValueSummary::GVFlags GVFlags( 957 GlobalValue::InternalLinkage, GlobalValue::DefaultVisibility, 958 /* NotEligibleToImport = */ true, 959 /* Live = */ true, 960 /* Local */ GV->isDSOLocal(), GV->canBeOmittedFromSymbolTable(), 961 GlobalValueSummary::Definition); 962 CantBePromoted.insert(GV->getGUID()); 963 // Create the appropriate summary type. 964 if (Function *F = dyn_cast<Function>(GV)) { 965 std::unique_ptr<FunctionSummary> Summary = 966 std::make_unique<FunctionSummary>( 967 GVFlags, /*InstCount=*/0, 968 FunctionSummary::FFlags{ 969 F->hasFnAttribute(Attribute::ReadNone), 970 F->hasFnAttribute(Attribute::ReadOnly), 971 F->hasFnAttribute(Attribute::NoRecurse), 972 F->returnDoesNotAlias(), 973 /* NoInline = */ false, 974 F->hasFnAttribute(Attribute::AlwaysInline), 975 F->hasFnAttribute(Attribute::NoUnwind), 976 /* MayThrow */ true, 977 /* HasUnknownCall */ true, 978 /* MustBeUnreachable */ false}, 979 SmallVector<ValueInfo, 0>{}, 980 SmallVector<FunctionSummary::EdgeTy, 0>{}, 981 ArrayRef<GlobalValue::GUID>{}, 982 ArrayRef<FunctionSummary::VFuncId>{}, 983 ArrayRef<FunctionSummary::VFuncId>{}, 984 ArrayRef<FunctionSummary::ConstVCall>{}, 985 ArrayRef<FunctionSummary::ConstVCall>{}, 986 ArrayRef<FunctionSummary::ParamAccess>{}, 987 ArrayRef<CallsiteInfo>{}, ArrayRef<AllocInfo>{}); 988 Index.addGlobalValueSummary(*GV, std::move(Summary)); 989 } else { 990 std::unique_ptr<GlobalVarSummary> Summary = 991 std::make_unique<GlobalVarSummary>( 992 GVFlags, 993 GlobalVarSummary::GVarFlags( 994 false, false, cast<GlobalVariable>(GV)->isConstant(), 995 GlobalObject::VCallVisibilityPublic), 996 SmallVector<ValueInfo, 0>{}); 997 Index.addGlobalValueSummary(*GV, std::move(Summary)); 998 } 999 }); 1000 } 1001 1002 bool IsThinLTO = true; 1003 if (auto *MD = 1004 mdconst::extract_or_null<ConstantInt>(M.getModuleFlag("ThinLTO"))) 1005 IsThinLTO = MD->getZExtValue(); 1006 1007 // Compute summaries for all functions defined in module, and save in the 1008 // index. 1009 for (const auto &F : M) { 1010 if (F.isDeclaration()) 1011 continue; 1012 1013 DominatorTree DT(const_cast<Function &>(F)); 1014 BlockFrequencyInfo *BFI = nullptr; 1015 std::unique_ptr<BlockFrequencyInfo> BFIPtr; 1016 if (GetBFICallback) 1017 BFI = GetBFICallback(F); 1018 else if (F.hasProfileData()) { 1019 LoopInfo LI{DT}; 1020 BranchProbabilityInfo BPI{F, LI}; 1021 BFIPtr = std::make_unique<BlockFrequencyInfo>(F, BPI, LI); 1022 BFI = BFIPtr.get(); 1023 } 1024 1025 computeFunctionSummary(Index, M, F, BFI, PSI, DT, 1026 !LocalsUsed.empty() || HasLocalInlineAsmSymbol, 1027 CantBePromoted, IsThinLTO, GetSSICallback); 1028 } 1029 1030 // Compute summaries for all variables defined in module, and save in the 1031 // index. 1032 SmallVector<MDNode *, 2> Types; 1033 for (const GlobalVariable &G : M.globals()) { 1034 if (G.isDeclaration()) 1035 continue; 1036 computeVariableSummary(Index, G, CantBePromoted, M, Types); 1037 } 1038 1039 // Compute summaries for all aliases defined in module, and save in the 1040 // index. 1041 for (const GlobalAlias &A : M.aliases()) 1042 computeAliasSummary(Index, A, CantBePromoted); 1043 1044 // Iterate through ifuncs, set their resolvers all alive. 1045 for (const GlobalIFunc &I : M.ifuncs()) { 1046 I.applyAlongResolverPath([&Index](const GlobalValue &GV) { 1047 Index.getGlobalValueSummary(GV)->setLive(true); 1048 }); 1049 } 1050 1051 for (auto *V : LocalsUsed) { 1052 auto *Summary = Index.getGlobalValueSummary(*V); 1053 assert(Summary && "Missing summary for global value"); 1054 Summary->setNotEligibleToImport(); 1055 } 1056 1057 // The linker doesn't know about these LLVM produced values, so we need 1058 // to flag them as live in the index to ensure index-based dead value 1059 // analysis treats them as live roots of the analysis. 1060 setLiveRoot(Index, "llvm.used"); 1061 setLiveRoot(Index, "llvm.compiler.used"); 1062 setLiveRoot(Index, "llvm.global_ctors"); 1063 setLiveRoot(Index, "llvm.global_dtors"); 1064 setLiveRoot(Index, "llvm.global.annotations"); 1065 1066 for (auto &GlobalList : Index) { 1067 // Ignore entries for references that are undefined in the current module. 1068 if (GlobalList.second.SummaryList.empty()) 1069 continue; 1070 1071 assert(GlobalList.second.SummaryList.size() == 1 && 1072 "Expected module's index to have one summary per GUID"); 1073 auto &Summary = GlobalList.second.SummaryList[0]; 1074 if (!IsThinLTO) { 1075 Summary->setNotEligibleToImport(); 1076 continue; 1077 } 1078 1079 bool AllRefsCanBeExternallyReferenced = 1080 llvm::all_of(Summary->refs(), [&](const ValueInfo &VI) { 1081 return !CantBePromoted.count(VI.getGUID()); 1082 }); 1083 if (!AllRefsCanBeExternallyReferenced) { 1084 Summary->setNotEligibleToImport(); 1085 continue; 1086 } 1087 1088 if (auto *FuncSummary = dyn_cast<FunctionSummary>(Summary.get())) { 1089 bool AllCallsCanBeExternallyReferenced = llvm::all_of( 1090 FuncSummary->calls(), [&](const FunctionSummary::EdgeTy &Edge) { 1091 return !CantBePromoted.count(Edge.first.getGUID()); 1092 }); 1093 if (!AllCallsCanBeExternallyReferenced) 1094 Summary->setNotEligibleToImport(); 1095 } 1096 } 1097 1098 if (!ModuleSummaryDotFile.empty()) { 1099 std::error_code EC; 1100 raw_fd_ostream OSDot(ModuleSummaryDotFile, EC, sys::fs::OpenFlags::OF_Text); 1101 if (EC) 1102 report_fatal_error(Twine("Failed to open dot file ") + 1103 ModuleSummaryDotFile + ": " + EC.message() + "\n"); 1104 Index.exportToDot(OSDot, {}); 1105 } 1106 1107 return Index; 1108 } 1109 1110 AnalysisKey ModuleSummaryIndexAnalysis::Key; 1111 1112 ModuleSummaryIndex 1113 ModuleSummaryIndexAnalysis::run(Module &M, ModuleAnalysisManager &AM) { 1114 ProfileSummaryInfo &PSI = AM.getResult<ProfileSummaryAnalysis>(M); 1115 auto &FAM = AM.getResult<FunctionAnalysisManagerModuleProxy>(M).getManager(); 1116 bool NeedSSI = needsParamAccessSummary(M); 1117 return buildModuleSummaryIndex( 1118 M, 1119 [&FAM](const Function &F) { 1120 return &FAM.getResult<BlockFrequencyAnalysis>( 1121 *const_cast<Function *>(&F)); 1122 }, 1123 &PSI, 1124 [&FAM, NeedSSI](const Function &F) -> const StackSafetyInfo * { 1125 return NeedSSI ? &FAM.getResult<StackSafetyAnalysis>( 1126 const_cast<Function &>(F)) 1127 : nullptr; 1128 }); 1129 } 1130 1131 char ModuleSummaryIndexWrapperPass::ID = 0; 1132 1133 INITIALIZE_PASS_BEGIN(ModuleSummaryIndexWrapperPass, "module-summary-analysis", 1134 "Module Summary Analysis", false, true) 1135 INITIALIZE_PASS_DEPENDENCY(BlockFrequencyInfoWrapperPass) 1136 INITIALIZE_PASS_DEPENDENCY(ProfileSummaryInfoWrapperPass) 1137 INITIALIZE_PASS_DEPENDENCY(StackSafetyInfoWrapperPass) 1138 INITIALIZE_PASS_END(ModuleSummaryIndexWrapperPass, "module-summary-analysis", 1139 "Module Summary Analysis", false, true) 1140 1141 ModulePass *llvm::createModuleSummaryIndexWrapperPass() { 1142 return new ModuleSummaryIndexWrapperPass(); 1143 } 1144 1145 ModuleSummaryIndexWrapperPass::ModuleSummaryIndexWrapperPass() 1146 : ModulePass(ID) { 1147 initializeModuleSummaryIndexWrapperPassPass(*PassRegistry::getPassRegistry()); 1148 } 1149 1150 bool ModuleSummaryIndexWrapperPass::runOnModule(Module &M) { 1151 auto *PSI = &getAnalysis<ProfileSummaryInfoWrapperPass>().getPSI(); 1152 bool NeedSSI = needsParamAccessSummary(M); 1153 Index.emplace(buildModuleSummaryIndex( 1154 M, 1155 [this](const Function &F) { 1156 return &(this->getAnalysis<BlockFrequencyInfoWrapperPass>( 1157 *const_cast<Function *>(&F)) 1158 .getBFI()); 1159 }, 1160 PSI, 1161 [&](const Function &F) -> const StackSafetyInfo * { 1162 return NeedSSI ? &getAnalysis<StackSafetyInfoWrapperPass>( 1163 const_cast<Function &>(F)) 1164 .getResult() 1165 : nullptr; 1166 })); 1167 return false; 1168 } 1169 1170 bool ModuleSummaryIndexWrapperPass::doFinalization(Module &M) { 1171 Index.reset(); 1172 return false; 1173 } 1174 1175 void ModuleSummaryIndexWrapperPass::getAnalysisUsage(AnalysisUsage &AU) const { 1176 AU.setPreservesAll(); 1177 AU.addRequired<BlockFrequencyInfoWrapperPass>(); 1178 AU.addRequired<ProfileSummaryInfoWrapperPass>(); 1179 AU.addRequired<StackSafetyInfoWrapperPass>(); 1180 } 1181 1182 char ImmutableModuleSummaryIndexWrapperPass::ID = 0; 1183 1184 ImmutableModuleSummaryIndexWrapperPass::ImmutableModuleSummaryIndexWrapperPass( 1185 const ModuleSummaryIndex *Index) 1186 : ImmutablePass(ID), Index(Index) { 1187 initializeImmutableModuleSummaryIndexWrapperPassPass( 1188 *PassRegistry::getPassRegistry()); 1189 } 1190 1191 void ImmutableModuleSummaryIndexWrapperPass::getAnalysisUsage( 1192 AnalysisUsage &AU) const { 1193 AU.setPreservesAll(); 1194 } 1195 1196 ImmutablePass *llvm::createImmutableModuleSummaryIndexWrapperPass( 1197 const ModuleSummaryIndex *Index) { 1198 return new ImmutableModuleSummaryIndexWrapperPass(Index); 1199 } 1200 1201 INITIALIZE_PASS(ImmutableModuleSummaryIndexWrapperPass, "module-summary-info", 1202 "Module summary info", false, true) 1203 1204 bool llvm::mayHaveMemprofSummary(const CallBase *CB) { 1205 if (!CB) 1206 return false; 1207 if (CB->isDebugOrPseudoInst()) 1208 return false; 1209 auto *CI = dyn_cast<CallInst>(CB); 1210 auto *CalledValue = CB->getCalledOperand(); 1211 auto *CalledFunction = CB->getCalledFunction(); 1212 if (CalledValue && !CalledFunction) { 1213 CalledValue = CalledValue->stripPointerCasts(); 1214 // Stripping pointer casts can reveal a called function. 1215 CalledFunction = dyn_cast<Function>(CalledValue); 1216 } 1217 // Check if this is an alias to a function. If so, get the 1218 // called aliasee for the checks below. 1219 if (auto *GA = dyn_cast<GlobalAlias>(CalledValue)) { 1220 assert(!CalledFunction && 1221 "Expected null called function in callsite for alias"); 1222 CalledFunction = dyn_cast<Function>(GA->getAliaseeObject()); 1223 } 1224 // Check if this is a direct call to a known function or a known 1225 // intrinsic, or an indirect call with profile data. 1226 if (CalledFunction) { 1227 if (CI && CalledFunction->isIntrinsic()) 1228 return false; 1229 } else { 1230 // Skip indirect calls if we haven't enabled memprof ICP. 1231 if (!EnableMemProfIndirectCallSupport) 1232 return false; 1233 // Skip inline assembly calls. 1234 if (CI && CI->isInlineAsm()) 1235 return false; 1236 // Skip direct calls via Constant. 1237 if (!CalledValue || isa<Constant>(CalledValue)) 1238 return false; 1239 return true; 1240 } 1241 return true; 1242 } 1243