1 //===- WholeProgramDevirt.cpp - Whole program virtual call optimization ---===// 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 implements whole program optimization of virtual calls in cases 10 // where we know (via !type metadata) that the list of callees is fixed. This 11 // includes the following: 12 // - Single implementation devirtualization: if a virtual call has a single 13 // possible callee, replace all calls with a direct call to that callee. 14 // - Virtual constant propagation: if the virtual function's return type is an 15 // integer <=64 bits and all possible callees are readnone, for each class and 16 // each list of constant arguments: evaluate the function, store the return 17 // value alongside the virtual table, and rewrite each virtual call as a load 18 // from the virtual table. 19 // - Uniform return value optimization: if the conditions for virtual constant 20 // propagation hold and each function returns the same constant value, replace 21 // each virtual call with that constant. 22 // - Unique return value optimization for i1 return values: if the conditions 23 // for virtual constant propagation hold and a single vtable's function 24 // returns 0, or a single vtable's function returns 1, replace each virtual 25 // call with a comparison of the vptr against that vtable's address. 26 // 27 // This pass is intended to be used during the regular and thin LTO pipelines: 28 // 29 // During regular LTO, the pass determines the best optimization for each 30 // virtual call and applies the resolutions directly to virtual calls that are 31 // eligible for virtual call optimization (i.e. calls that use either of the 32 // llvm.assume(llvm.type.test) or llvm.type.checked.load intrinsics). 33 // 34 // During hybrid Regular/ThinLTO, the pass operates in two phases: 35 // - Export phase: this is run during the thin link over a single merged module 36 // that contains all vtables with !type metadata that participate in the link. 37 // The pass computes a resolution for each virtual call and stores it in the 38 // type identifier summary. 39 // - Import phase: this is run during the thin backends over the individual 40 // modules. The pass applies the resolutions previously computed during the 41 // import phase to each eligible virtual call. 42 // 43 // During ThinLTO, the pass operates in two phases: 44 // - Export phase: this is run during the thin link over the index which 45 // contains a summary of all vtables with !type metadata that participate in 46 // the link. It computes a resolution for each virtual call and stores it in 47 // the type identifier summary. Only single implementation devirtualization 48 // is supported. 49 // - Import phase: (same as with hybrid case above). 50 // 51 //===----------------------------------------------------------------------===// 52 53 #include "llvm/Transforms/IPO/WholeProgramDevirt.h" 54 #include "llvm/ADT/ArrayRef.h" 55 #include "llvm/ADT/DenseMap.h" 56 #include "llvm/ADT/DenseMapInfo.h" 57 #include "llvm/ADT/DenseSet.h" 58 #include "llvm/ADT/MapVector.h" 59 #include "llvm/ADT/SmallVector.h" 60 #include "llvm/ADT/Triple.h" 61 #include "llvm/ADT/iterator_range.h" 62 #include "llvm/Analysis/AssumptionCache.h" 63 #include "llvm/Analysis/BasicAliasAnalysis.h" 64 #include "llvm/Analysis/OptimizationRemarkEmitter.h" 65 #include "llvm/Analysis/TypeMetadataUtils.h" 66 #include "llvm/Bitcode/BitcodeReader.h" 67 #include "llvm/Bitcode/BitcodeWriter.h" 68 #include "llvm/IR/Constants.h" 69 #include "llvm/IR/DataLayout.h" 70 #include "llvm/IR/DebugLoc.h" 71 #include "llvm/IR/DerivedTypes.h" 72 #include "llvm/IR/Dominators.h" 73 #include "llvm/IR/Function.h" 74 #include "llvm/IR/GlobalAlias.h" 75 #include "llvm/IR/GlobalVariable.h" 76 #include "llvm/IR/IRBuilder.h" 77 #include "llvm/IR/InstrTypes.h" 78 #include "llvm/IR/Instruction.h" 79 #include "llvm/IR/Instructions.h" 80 #include "llvm/IR/Intrinsics.h" 81 #include "llvm/IR/LLVMContext.h" 82 #include "llvm/IR/Metadata.h" 83 #include "llvm/IR/Module.h" 84 #include "llvm/IR/ModuleSummaryIndexYAML.h" 85 #include "llvm/InitializePasses.h" 86 #include "llvm/Pass.h" 87 #include "llvm/PassRegistry.h" 88 #include "llvm/Support/Casting.h" 89 #include "llvm/Support/CommandLine.h" 90 #include "llvm/Support/Errc.h" 91 #include "llvm/Support/Error.h" 92 #include "llvm/Support/FileSystem.h" 93 #include "llvm/Support/GlobPattern.h" 94 #include "llvm/Support/MathExtras.h" 95 #include "llvm/Transforms/IPO.h" 96 #include "llvm/Transforms/IPO/FunctionAttrs.h" 97 #include "llvm/Transforms/Utils/Evaluator.h" 98 #include <algorithm> 99 #include <cstddef> 100 #include <map> 101 #include <set> 102 #include <string> 103 104 using namespace llvm; 105 using namespace wholeprogramdevirt; 106 107 #define DEBUG_TYPE "wholeprogramdevirt" 108 109 static cl::opt<PassSummaryAction> ClSummaryAction( 110 "wholeprogramdevirt-summary-action", 111 cl::desc("What to do with the summary when running this pass"), 112 cl::values(clEnumValN(PassSummaryAction::None, "none", "Do nothing"), 113 clEnumValN(PassSummaryAction::Import, "import", 114 "Import typeid resolutions from summary and globals"), 115 clEnumValN(PassSummaryAction::Export, "export", 116 "Export typeid resolutions to summary and globals")), 117 cl::Hidden); 118 119 static cl::opt<std::string> ClReadSummary( 120 "wholeprogramdevirt-read-summary", 121 cl::desc( 122 "Read summary from given bitcode or YAML file before running pass"), 123 cl::Hidden); 124 125 static cl::opt<std::string> ClWriteSummary( 126 "wholeprogramdevirt-write-summary", 127 cl::desc("Write summary to given bitcode or YAML file after running pass. " 128 "Output file format is deduced from extension: *.bc means writing " 129 "bitcode, otherwise YAML"), 130 cl::Hidden); 131 132 static cl::opt<unsigned> 133 ClThreshold("wholeprogramdevirt-branch-funnel-threshold", cl::Hidden, 134 cl::init(10), cl::ZeroOrMore, 135 cl::desc("Maximum number of call targets per " 136 "call site to enable branch funnels")); 137 138 static cl::opt<bool> 139 PrintSummaryDevirt("wholeprogramdevirt-print-index-based", cl::Hidden, 140 cl::init(false), cl::ZeroOrMore, 141 cl::desc("Print index-based devirtualization messages")); 142 143 /// Provide a way to force enable whole program visibility in tests. 144 /// This is needed to support legacy tests that don't contain 145 /// !vcall_visibility metadata (the mere presense of type tests 146 /// previously implied hidden visibility). 147 cl::opt<bool> 148 WholeProgramVisibility("whole-program-visibility", cl::init(false), 149 cl::Hidden, cl::ZeroOrMore, 150 cl::desc("Enable whole program visibility")); 151 152 /// Provide a way to force disable whole program for debugging or workarounds, 153 /// when enabled via the linker. 154 cl::opt<bool> DisableWholeProgramVisibility( 155 "disable-whole-program-visibility", cl::init(false), cl::Hidden, 156 cl::ZeroOrMore, 157 cl::desc("Disable whole program visibility (overrides enabling options)")); 158 159 /// Provide way to prevent certain function from being devirtualized 160 cl::list<std::string> 161 SkipFunctionNames("wholeprogramdevirt-skip", 162 cl::desc("Prevent function(s) from being devirtualized"), 163 cl::Hidden, cl::ZeroOrMore, cl::CommaSeparated); 164 165 namespace { 166 struct PatternList { 167 std::vector<GlobPattern> Patterns; 168 template <class T> void init(const T &StringList) { 169 for (const auto &S : StringList) 170 if (Expected<GlobPattern> Pat = GlobPattern::create(S)) 171 Patterns.push_back(std::move(*Pat)); 172 } 173 bool match(StringRef S) { 174 for (const GlobPattern &P : Patterns) 175 if (P.match(S)) 176 return true; 177 return false; 178 } 179 }; 180 } // namespace 181 182 // Find the minimum offset that we may store a value of size Size bits at. If 183 // IsAfter is set, look for an offset before the object, otherwise look for an 184 // offset after the object. 185 uint64_t 186 wholeprogramdevirt::findLowestOffset(ArrayRef<VirtualCallTarget> Targets, 187 bool IsAfter, uint64_t Size) { 188 // Find a minimum offset taking into account only vtable sizes. 189 uint64_t MinByte = 0; 190 for (const VirtualCallTarget &Target : Targets) { 191 if (IsAfter) 192 MinByte = std::max(MinByte, Target.minAfterBytes()); 193 else 194 MinByte = std::max(MinByte, Target.minBeforeBytes()); 195 } 196 197 // Build a vector of arrays of bytes covering, for each target, a slice of the 198 // used region (see AccumBitVector::BytesUsed in 199 // llvm/Transforms/IPO/WholeProgramDevirt.h) starting at MinByte. Effectively, 200 // this aligns the used regions to start at MinByte. 201 // 202 // In this example, A, B and C are vtables, # is a byte already allocated for 203 // a virtual function pointer, AAAA... (etc.) are the used regions for the 204 // vtables and Offset(X) is the value computed for the Offset variable below 205 // for X. 206 // 207 // Offset(A) 208 // | | 209 // |MinByte 210 // A: ################AAAAAAAA|AAAAAAAA 211 // B: ########BBBBBBBBBBBBBBBB|BBBB 212 // C: ########################|CCCCCCCCCCCCCCCC 213 // | Offset(B) | 214 // 215 // This code produces the slices of A, B and C that appear after the divider 216 // at MinByte. 217 std::vector<ArrayRef<uint8_t>> Used; 218 for (const VirtualCallTarget &Target : Targets) { 219 ArrayRef<uint8_t> VTUsed = IsAfter ? Target.TM->Bits->After.BytesUsed 220 : Target.TM->Bits->Before.BytesUsed; 221 uint64_t Offset = IsAfter ? MinByte - Target.minAfterBytes() 222 : MinByte - Target.minBeforeBytes(); 223 224 // Disregard used regions that are smaller than Offset. These are 225 // effectively all-free regions that do not need to be checked. 226 if (VTUsed.size() > Offset) 227 Used.push_back(VTUsed.slice(Offset)); 228 } 229 230 if (Size == 1) { 231 // Find a free bit in each member of Used. 232 for (unsigned I = 0;; ++I) { 233 uint8_t BitsUsed = 0; 234 for (auto &&B : Used) 235 if (I < B.size()) 236 BitsUsed |= B[I]; 237 if (BitsUsed != 0xff) 238 return (MinByte + I) * 8 + 239 countTrailingZeros(uint8_t(~BitsUsed), ZB_Undefined); 240 } 241 } else { 242 // Find a free (Size/8) byte region in each member of Used. 243 // FIXME: see if alignment helps. 244 for (unsigned I = 0;; ++I) { 245 for (auto &&B : Used) { 246 unsigned Byte = 0; 247 while ((I + Byte) < B.size() && Byte < (Size / 8)) { 248 if (B[I + Byte]) 249 goto NextI; 250 ++Byte; 251 } 252 } 253 return (MinByte + I) * 8; 254 NextI:; 255 } 256 } 257 } 258 259 void wholeprogramdevirt::setBeforeReturnValues( 260 MutableArrayRef<VirtualCallTarget> Targets, uint64_t AllocBefore, 261 unsigned BitWidth, int64_t &OffsetByte, uint64_t &OffsetBit) { 262 if (BitWidth == 1) 263 OffsetByte = -(AllocBefore / 8 + 1); 264 else 265 OffsetByte = -((AllocBefore + 7) / 8 + (BitWidth + 7) / 8); 266 OffsetBit = AllocBefore % 8; 267 268 for (VirtualCallTarget &Target : Targets) { 269 if (BitWidth == 1) 270 Target.setBeforeBit(AllocBefore); 271 else 272 Target.setBeforeBytes(AllocBefore, (BitWidth + 7) / 8); 273 } 274 } 275 276 void wholeprogramdevirt::setAfterReturnValues( 277 MutableArrayRef<VirtualCallTarget> Targets, uint64_t AllocAfter, 278 unsigned BitWidth, int64_t &OffsetByte, uint64_t &OffsetBit) { 279 if (BitWidth == 1) 280 OffsetByte = AllocAfter / 8; 281 else 282 OffsetByte = (AllocAfter + 7) / 8; 283 OffsetBit = AllocAfter % 8; 284 285 for (VirtualCallTarget &Target : Targets) { 286 if (BitWidth == 1) 287 Target.setAfterBit(AllocAfter); 288 else 289 Target.setAfterBytes(AllocAfter, (BitWidth + 7) / 8); 290 } 291 } 292 293 VirtualCallTarget::VirtualCallTarget(Function *Fn, const TypeMemberInfo *TM) 294 : Fn(Fn), TM(TM), 295 IsBigEndian(Fn->getParent()->getDataLayout().isBigEndian()), WasDevirt(false) {} 296 297 namespace { 298 299 // A slot in a set of virtual tables. The TypeID identifies the set of virtual 300 // tables, and the ByteOffset is the offset in bytes from the address point to 301 // the virtual function pointer. 302 struct VTableSlot { 303 Metadata *TypeID; 304 uint64_t ByteOffset; 305 }; 306 307 } // end anonymous namespace 308 309 namespace llvm { 310 311 template <> struct DenseMapInfo<VTableSlot> { 312 static VTableSlot getEmptyKey() { 313 return {DenseMapInfo<Metadata *>::getEmptyKey(), 314 DenseMapInfo<uint64_t>::getEmptyKey()}; 315 } 316 static VTableSlot getTombstoneKey() { 317 return {DenseMapInfo<Metadata *>::getTombstoneKey(), 318 DenseMapInfo<uint64_t>::getTombstoneKey()}; 319 } 320 static unsigned getHashValue(const VTableSlot &I) { 321 return DenseMapInfo<Metadata *>::getHashValue(I.TypeID) ^ 322 DenseMapInfo<uint64_t>::getHashValue(I.ByteOffset); 323 } 324 static bool isEqual(const VTableSlot &LHS, 325 const VTableSlot &RHS) { 326 return LHS.TypeID == RHS.TypeID && LHS.ByteOffset == RHS.ByteOffset; 327 } 328 }; 329 330 template <> struct DenseMapInfo<VTableSlotSummary> { 331 static VTableSlotSummary getEmptyKey() { 332 return {DenseMapInfo<StringRef>::getEmptyKey(), 333 DenseMapInfo<uint64_t>::getEmptyKey()}; 334 } 335 static VTableSlotSummary getTombstoneKey() { 336 return {DenseMapInfo<StringRef>::getTombstoneKey(), 337 DenseMapInfo<uint64_t>::getTombstoneKey()}; 338 } 339 static unsigned getHashValue(const VTableSlotSummary &I) { 340 return DenseMapInfo<StringRef>::getHashValue(I.TypeID) ^ 341 DenseMapInfo<uint64_t>::getHashValue(I.ByteOffset); 342 } 343 static bool isEqual(const VTableSlotSummary &LHS, 344 const VTableSlotSummary &RHS) { 345 return LHS.TypeID == RHS.TypeID && LHS.ByteOffset == RHS.ByteOffset; 346 } 347 }; 348 349 } // end namespace llvm 350 351 namespace { 352 353 // A virtual call site. VTable is the loaded virtual table pointer, and CS is 354 // the indirect virtual call. 355 struct VirtualCallSite { 356 Value *VTable = nullptr; 357 CallBase &CB; 358 359 // If non-null, this field points to the associated unsafe use count stored in 360 // the DevirtModule::NumUnsafeUsesForTypeTest map below. See the description 361 // of that field for details. 362 unsigned *NumUnsafeUses = nullptr; 363 364 void 365 emitRemark(const StringRef OptName, const StringRef TargetName, 366 function_ref<OptimizationRemarkEmitter &(Function *)> OREGetter) { 367 Function *F = CB.getCaller(); 368 DebugLoc DLoc = CB.getDebugLoc(); 369 BasicBlock *Block = CB.getParent(); 370 371 using namespace ore; 372 OREGetter(F).emit(OptimizationRemark(DEBUG_TYPE, OptName, DLoc, Block) 373 << NV("Optimization", OptName) 374 << ": devirtualized a call to " 375 << NV("FunctionName", TargetName)); 376 } 377 378 void replaceAndErase( 379 const StringRef OptName, const StringRef TargetName, bool RemarksEnabled, 380 function_ref<OptimizationRemarkEmitter &(Function *)> OREGetter, 381 Value *New) { 382 if (RemarksEnabled) 383 emitRemark(OptName, TargetName, OREGetter); 384 CB.replaceAllUsesWith(New); 385 if (auto *II = dyn_cast<InvokeInst>(&CB)) { 386 BranchInst::Create(II->getNormalDest(), &CB); 387 II->getUnwindDest()->removePredecessor(II->getParent()); 388 } 389 CB.eraseFromParent(); 390 // This use is no longer unsafe. 391 if (NumUnsafeUses) 392 --*NumUnsafeUses; 393 } 394 }; 395 396 // Call site information collected for a specific VTableSlot and possibly a list 397 // of constant integer arguments. The grouping by arguments is handled by the 398 // VTableSlotInfo class. 399 struct CallSiteInfo { 400 /// The set of call sites for this slot. Used during regular LTO and the 401 /// import phase of ThinLTO (as well as the export phase of ThinLTO for any 402 /// call sites that appear in the merged module itself); in each of these 403 /// cases we are directly operating on the call sites at the IR level. 404 std::vector<VirtualCallSite> CallSites; 405 406 /// Whether all call sites represented by this CallSiteInfo, including those 407 /// in summaries, have been devirtualized. This starts off as true because a 408 /// default constructed CallSiteInfo represents no call sites. 409 bool AllCallSitesDevirted = true; 410 411 // These fields are used during the export phase of ThinLTO and reflect 412 // information collected from function summaries. 413 414 /// Whether any function summary contains an llvm.assume(llvm.type.test) for 415 /// this slot. 416 bool SummaryHasTypeTestAssumeUsers = false; 417 418 /// CFI-specific: a vector containing the list of function summaries that use 419 /// the llvm.type.checked.load intrinsic and therefore will require 420 /// resolutions for llvm.type.test in order to implement CFI checks if 421 /// devirtualization was unsuccessful. If devirtualization was successful, the 422 /// pass will clear this vector by calling markDevirt(). If at the end of the 423 /// pass the vector is non-empty, we will need to add a use of llvm.type.test 424 /// to each of the function summaries in the vector. 425 std::vector<FunctionSummary *> SummaryTypeCheckedLoadUsers; 426 std::vector<FunctionSummary *> SummaryTypeTestAssumeUsers; 427 428 bool isExported() const { 429 return SummaryHasTypeTestAssumeUsers || 430 !SummaryTypeCheckedLoadUsers.empty(); 431 } 432 433 void addSummaryTypeCheckedLoadUser(FunctionSummary *FS) { 434 SummaryTypeCheckedLoadUsers.push_back(FS); 435 AllCallSitesDevirted = false; 436 } 437 438 void addSummaryTypeTestAssumeUser(FunctionSummary *FS) { 439 SummaryTypeTestAssumeUsers.push_back(FS); 440 SummaryHasTypeTestAssumeUsers = true; 441 AllCallSitesDevirted = false; 442 } 443 444 void markDevirt() { 445 AllCallSitesDevirted = true; 446 447 // As explained in the comment for SummaryTypeCheckedLoadUsers. 448 SummaryTypeCheckedLoadUsers.clear(); 449 } 450 }; 451 452 // Call site information collected for a specific VTableSlot. 453 struct VTableSlotInfo { 454 // The set of call sites which do not have all constant integer arguments 455 // (excluding "this"). 456 CallSiteInfo CSInfo; 457 458 // The set of call sites with all constant integer arguments (excluding 459 // "this"), grouped by argument list. 460 std::map<std::vector<uint64_t>, CallSiteInfo> ConstCSInfo; 461 462 void addCallSite(Value *VTable, CallBase &CB, unsigned *NumUnsafeUses); 463 464 private: 465 CallSiteInfo &findCallSiteInfo(CallBase &CB); 466 }; 467 468 CallSiteInfo &VTableSlotInfo::findCallSiteInfo(CallBase &CB) { 469 std::vector<uint64_t> Args; 470 auto *CBType = dyn_cast<IntegerType>(CB.getType()); 471 if (!CBType || CBType->getBitWidth() > 64 || CB.arg_empty()) 472 return CSInfo; 473 for (auto &&Arg : drop_begin(CB.args())) { 474 auto *CI = dyn_cast<ConstantInt>(Arg); 475 if (!CI || CI->getBitWidth() > 64) 476 return CSInfo; 477 Args.push_back(CI->getZExtValue()); 478 } 479 return ConstCSInfo[Args]; 480 } 481 482 void VTableSlotInfo::addCallSite(Value *VTable, CallBase &CB, 483 unsigned *NumUnsafeUses) { 484 auto &CSI = findCallSiteInfo(CB); 485 CSI.AllCallSitesDevirted = false; 486 CSI.CallSites.push_back({VTable, CB, NumUnsafeUses}); 487 } 488 489 struct DevirtModule { 490 Module &M; 491 function_ref<AAResults &(Function &)> AARGetter; 492 function_ref<DominatorTree &(Function &)> LookupDomTree; 493 494 ModuleSummaryIndex *ExportSummary; 495 const ModuleSummaryIndex *ImportSummary; 496 497 IntegerType *Int8Ty; 498 PointerType *Int8PtrTy; 499 IntegerType *Int32Ty; 500 IntegerType *Int64Ty; 501 IntegerType *IntPtrTy; 502 /// Sizeless array type, used for imported vtables. This provides a signal 503 /// to analyzers that these imports may alias, as they do for example 504 /// when multiple unique return values occur in the same vtable. 505 ArrayType *Int8Arr0Ty; 506 507 bool RemarksEnabled; 508 function_ref<OptimizationRemarkEmitter &(Function *)> OREGetter; 509 510 MapVector<VTableSlot, VTableSlotInfo> CallSlots; 511 512 // This map keeps track of the number of "unsafe" uses of a loaded function 513 // pointer. The key is the associated llvm.type.test intrinsic call generated 514 // by this pass. An unsafe use is one that calls the loaded function pointer 515 // directly. Every time we eliminate an unsafe use (for example, by 516 // devirtualizing it or by applying virtual constant propagation), we 517 // decrement the value stored in this map. If a value reaches zero, we can 518 // eliminate the type check by RAUWing the associated llvm.type.test call with 519 // true. 520 std::map<CallInst *, unsigned> NumUnsafeUsesForTypeTest; 521 PatternList FunctionsToSkip; 522 523 DevirtModule(Module &M, function_ref<AAResults &(Function &)> AARGetter, 524 function_ref<OptimizationRemarkEmitter &(Function *)> OREGetter, 525 function_ref<DominatorTree &(Function &)> LookupDomTree, 526 ModuleSummaryIndex *ExportSummary, 527 const ModuleSummaryIndex *ImportSummary) 528 : M(M), AARGetter(AARGetter), LookupDomTree(LookupDomTree), 529 ExportSummary(ExportSummary), ImportSummary(ImportSummary), 530 Int8Ty(Type::getInt8Ty(M.getContext())), 531 Int8PtrTy(Type::getInt8PtrTy(M.getContext())), 532 Int32Ty(Type::getInt32Ty(M.getContext())), 533 Int64Ty(Type::getInt64Ty(M.getContext())), 534 IntPtrTy(M.getDataLayout().getIntPtrType(M.getContext(), 0)), 535 Int8Arr0Ty(ArrayType::get(Type::getInt8Ty(M.getContext()), 0)), 536 RemarksEnabled(areRemarksEnabled()), OREGetter(OREGetter) { 537 assert(!(ExportSummary && ImportSummary)); 538 FunctionsToSkip.init(SkipFunctionNames); 539 } 540 541 bool areRemarksEnabled(); 542 543 void 544 scanTypeTestUsers(Function *TypeTestFunc, 545 DenseMap<Metadata *, std::set<TypeMemberInfo>> &TypeIdMap); 546 void scanTypeCheckedLoadUsers(Function *TypeCheckedLoadFunc); 547 548 void buildTypeIdentifierMap( 549 std::vector<VTableBits> &Bits, 550 DenseMap<Metadata *, std::set<TypeMemberInfo>> &TypeIdMap); 551 bool 552 tryFindVirtualCallTargets(std::vector<VirtualCallTarget> &TargetsForSlot, 553 const std::set<TypeMemberInfo> &TypeMemberInfos, 554 uint64_t ByteOffset); 555 556 void applySingleImplDevirt(VTableSlotInfo &SlotInfo, Constant *TheFn, 557 bool &IsExported); 558 bool trySingleImplDevirt(ModuleSummaryIndex *ExportSummary, 559 MutableArrayRef<VirtualCallTarget> TargetsForSlot, 560 VTableSlotInfo &SlotInfo, 561 WholeProgramDevirtResolution *Res); 562 563 void applyICallBranchFunnel(VTableSlotInfo &SlotInfo, Constant *JT, 564 bool &IsExported); 565 void tryICallBranchFunnel(MutableArrayRef<VirtualCallTarget> TargetsForSlot, 566 VTableSlotInfo &SlotInfo, 567 WholeProgramDevirtResolution *Res, VTableSlot Slot); 568 569 bool tryEvaluateFunctionsWithArgs( 570 MutableArrayRef<VirtualCallTarget> TargetsForSlot, 571 ArrayRef<uint64_t> Args); 572 573 void applyUniformRetValOpt(CallSiteInfo &CSInfo, StringRef FnName, 574 uint64_t TheRetVal); 575 bool tryUniformRetValOpt(MutableArrayRef<VirtualCallTarget> TargetsForSlot, 576 CallSiteInfo &CSInfo, 577 WholeProgramDevirtResolution::ByArg *Res); 578 579 // Returns the global symbol name that is used to export information about the 580 // given vtable slot and list of arguments. 581 std::string getGlobalName(VTableSlot Slot, ArrayRef<uint64_t> Args, 582 StringRef Name); 583 584 bool shouldExportConstantsAsAbsoluteSymbols(); 585 586 // This function is called during the export phase to create a symbol 587 // definition containing information about the given vtable slot and list of 588 // arguments. 589 void exportGlobal(VTableSlot Slot, ArrayRef<uint64_t> Args, StringRef Name, 590 Constant *C); 591 void exportConstant(VTableSlot Slot, ArrayRef<uint64_t> Args, StringRef Name, 592 uint32_t Const, uint32_t &Storage); 593 594 // This function is called during the import phase to create a reference to 595 // the symbol definition created during the export phase. 596 Constant *importGlobal(VTableSlot Slot, ArrayRef<uint64_t> Args, 597 StringRef Name); 598 Constant *importConstant(VTableSlot Slot, ArrayRef<uint64_t> Args, 599 StringRef Name, IntegerType *IntTy, 600 uint32_t Storage); 601 602 Constant *getMemberAddr(const TypeMemberInfo *M); 603 604 void applyUniqueRetValOpt(CallSiteInfo &CSInfo, StringRef FnName, bool IsOne, 605 Constant *UniqueMemberAddr); 606 bool tryUniqueRetValOpt(unsigned BitWidth, 607 MutableArrayRef<VirtualCallTarget> TargetsForSlot, 608 CallSiteInfo &CSInfo, 609 WholeProgramDevirtResolution::ByArg *Res, 610 VTableSlot Slot, ArrayRef<uint64_t> Args); 611 612 void applyVirtualConstProp(CallSiteInfo &CSInfo, StringRef FnName, 613 Constant *Byte, Constant *Bit); 614 bool tryVirtualConstProp(MutableArrayRef<VirtualCallTarget> TargetsForSlot, 615 VTableSlotInfo &SlotInfo, 616 WholeProgramDevirtResolution *Res, VTableSlot Slot); 617 618 void rebuildGlobal(VTableBits &B); 619 620 // Apply the summary resolution for Slot to all virtual calls in SlotInfo. 621 void importResolution(VTableSlot Slot, VTableSlotInfo &SlotInfo); 622 623 // If we were able to eliminate all unsafe uses for a type checked load, 624 // eliminate the associated type tests by replacing them with true. 625 void removeRedundantTypeTests(); 626 627 bool run(); 628 629 // Lower the module using the action and summary passed as command line 630 // arguments. For testing purposes only. 631 static bool 632 runForTesting(Module &M, function_ref<AAResults &(Function &)> AARGetter, 633 function_ref<OptimizationRemarkEmitter &(Function *)> OREGetter, 634 function_ref<DominatorTree &(Function &)> LookupDomTree); 635 }; 636 637 struct DevirtIndex { 638 ModuleSummaryIndex &ExportSummary; 639 // The set in which to record GUIDs exported from their module by 640 // devirtualization, used by client to ensure they are not internalized. 641 std::set<GlobalValue::GUID> &ExportedGUIDs; 642 // A map in which to record the information necessary to locate the WPD 643 // resolution for local targets in case they are exported by cross module 644 // importing. 645 std::map<ValueInfo, std::vector<VTableSlotSummary>> &LocalWPDTargetsMap; 646 647 MapVector<VTableSlotSummary, VTableSlotInfo> CallSlots; 648 649 PatternList FunctionsToSkip; 650 651 DevirtIndex( 652 ModuleSummaryIndex &ExportSummary, 653 std::set<GlobalValue::GUID> &ExportedGUIDs, 654 std::map<ValueInfo, std::vector<VTableSlotSummary>> &LocalWPDTargetsMap) 655 : ExportSummary(ExportSummary), ExportedGUIDs(ExportedGUIDs), 656 LocalWPDTargetsMap(LocalWPDTargetsMap) { 657 FunctionsToSkip.init(SkipFunctionNames); 658 } 659 660 bool tryFindVirtualCallTargets(std::vector<ValueInfo> &TargetsForSlot, 661 const TypeIdCompatibleVtableInfo TIdInfo, 662 uint64_t ByteOffset); 663 664 bool trySingleImplDevirt(MutableArrayRef<ValueInfo> TargetsForSlot, 665 VTableSlotSummary &SlotSummary, 666 VTableSlotInfo &SlotInfo, 667 WholeProgramDevirtResolution *Res, 668 std::set<ValueInfo> &DevirtTargets); 669 670 void run(); 671 }; 672 673 struct WholeProgramDevirt : public ModulePass { 674 static char ID; 675 676 bool UseCommandLine = false; 677 678 ModuleSummaryIndex *ExportSummary = nullptr; 679 const ModuleSummaryIndex *ImportSummary = nullptr; 680 681 WholeProgramDevirt() : ModulePass(ID), UseCommandLine(true) { 682 initializeWholeProgramDevirtPass(*PassRegistry::getPassRegistry()); 683 } 684 685 WholeProgramDevirt(ModuleSummaryIndex *ExportSummary, 686 const ModuleSummaryIndex *ImportSummary) 687 : ModulePass(ID), ExportSummary(ExportSummary), 688 ImportSummary(ImportSummary) { 689 initializeWholeProgramDevirtPass(*PassRegistry::getPassRegistry()); 690 } 691 692 bool runOnModule(Module &M) override { 693 if (skipModule(M)) 694 return false; 695 696 // In the new pass manager, we can request the optimization 697 // remark emitter pass on a per-function-basis, which the 698 // OREGetter will do for us. 699 // In the old pass manager, this is harder, so we just build 700 // an optimization remark emitter on the fly, when we need it. 701 std::unique_ptr<OptimizationRemarkEmitter> ORE; 702 auto OREGetter = [&](Function *F) -> OptimizationRemarkEmitter & { 703 ORE = std::make_unique<OptimizationRemarkEmitter>(F); 704 return *ORE; 705 }; 706 707 auto LookupDomTree = [this](Function &F) -> DominatorTree & { 708 return this->getAnalysis<DominatorTreeWrapperPass>(F).getDomTree(); 709 }; 710 711 if (UseCommandLine) 712 return DevirtModule::runForTesting(M, LegacyAARGetter(*this), OREGetter, 713 LookupDomTree); 714 715 return DevirtModule(M, LegacyAARGetter(*this), OREGetter, LookupDomTree, 716 ExportSummary, ImportSummary) 717 .run(); 718 } 719 720 void getAnalysisUsage(AnalysisUsage &AU) const override { 721 AU.addRequired<AssumptionCacheTracker>(); 722 AU.addRequired<TargetLibraryInfoWrapperPass>(); 723 AU.addRequired<DominatorTreeWrapperPass>(); 724 } 725 }; 726 727 } // end anonymous namespace 728 729 INITIALIZE_PASS_BEGIN(WholeProgramDevirt, "wholeprogramdevirt", 730 "Whole program devirtualization", false, false) 731 INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker) 732 INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass) 733 INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass) 734 INITIALIZE_PASS_END(WholeProgramDevirt, "wholeprogramdevirt", 735 "Whole program devirtualization", false, false) 736 char WholeProgramDevirt::ID = 0; 737 738 ModulePass * 739 llvm::createWholeProgramDevirtPass(ModuleSummaryIndex *ExportSummary, 740 const ModuleSummaryIndex *ImportSummary) { 741 return new WholeProgramDevirt(ExportSummary, ImportSummary); 742 } 743 744 PreservedAnalyses WholeProgramDevirtPass::run(Module &M, 745 ModuleAnalysisManager &AM) { 746 auto &FAM = AM.getResult<FunctionAnalysisManagerModuleProxy>(M).getManager(); 747 auto AARGetter = [&](Function &F) -> AAResults & { 748 return FAM.getResult<AAManager>(F); 749 }; 750 auto OREGetter = [&](Function *F) -> OptimizationRemarkEmitter & { 751 return FAM.getResult<OptimizationRemarkEmitterAnalysis>(*F); 752 }; 753 auto LookupDomTree = [&FAM](Function &F) -> DominatorTree & { 754 return FAM.getResult<DominatorTreeAnalysis>(F); 755 }; 756 if (UseCommandLine) { 757 if (DevirtModule::runForTesting(M, AARGetter, OREGetter, LookupDomTree)) 758 return PreservedAnalyses::all(); 759 return PreservedAnalyses::none(); 760 } 761 if (!DevirtModule(M, AARGetter, OREGetter, LookupDomTree, ExportSummary, 762 ImportSummary) 763 .run()) 764 return PreservedAnalyses::all(); 765 return PreservedAnalyses::none(); 766 } 767 768 // Enable whole program visibility if enabled by client (e.g. linker) or 769 // internal option, and not force disabled. 770 static bool hasWholeProgramVisibility(bool WholeProgramVisibilityEnabledInLTO) { 771 return (WholeProgramVisibilityEnabledInLTO || WholeProgramVisibility) && 772 !DisableWholeProgramVisibility; 773 } 774 775 namespace llvm { 776 777 /// If whole program visibility asserted, then upgrade all public vcall 778 /// visibility metadata on vtable definitions to linkage unit visibility in 779 /// Module IR (for regular or hybrid LTO). 780 void updateVCallVisibilityInModule(Module &M, 781 bool WholeProgramVisibilityEnabledInLTO) { 782 if (!hasWholeProgramVisibility(WholeProgramVisibilityEnabledInLTO)) 783 return; 784 for (GlobalVariable &GV : M.globals()) 785 // Add linkage unit visibility to any variable with type metadata, which are 786 // the vtable definitions. We won't have an existing vcall_visibility 787 // metadata on vtable definitions with public visibility. 788 if (GV.hasMetadata(LLVMContext::MD_type) && 789 GV.getVCallVisibility() == GlobalObject::VCallVisibilityPublic) 790 GV.setVCallVisibilityMetadata(GlobalObject::VCallVisibilityLinkageUnit); 791 } 792 793 /// If whole program visibility asserted, then upgrade all public vcall 794 /// visibility metadata on vtable definition summaries to linkage unit 795 /// visibility in Module summary index (for ThinLTO). 796 void updateVCallVisibilityInIndex(ModuleSummaryIndex &Index, 797 bool WholeProgramVisibilityEnabledInLTO) { 798 if (!hasWholeProgramVisibility(WholeProgramVisibilityEnabledInLTO)) 799 return; 800 for (auto &P : Index) { 801 for (auto &S : P.second.SummaryList) { 802 auto *GVar = dyn_cast<GlobalVarSummary>(S.get()); 803 if (!GVar || GVar->vTableFuncs().empty() || 804 GVar->getVCallVisibility() != GlobalObject::VCallVisibilityPublic) 805 continue; 806 GVar->setVCallVisibility(GlobalObject::VCallVisibilityLinkageUnit); 807 } 808 } 809 } 810 811 void runWholeProgramDevirtOnIndex( 812 ModuleSummaryIndex &Summary, std::set<GlobalValue::GUID> &ExportedGUIDs, 813 std::map<ValueInfo, std::vector<VTableSlotSummary>> &LocalWPDTargetsMap) { 814 DevirtIndex(Summary, ExportedGUIDs, LocalWPDTargetsMap).run(); 815 } 816 817 void updateIndexWPDForExports( 818 ModuleSummaryIndex &Summary, 819 function_ref<bool(StringRef, ValueInfo)> isExported, 820 std::map<ValueInfo, std::vector<VTableSlotSummary>> &LocalWPDTargetsMap) { 821 for (auto &T : LocalWPDTargetsMap) { 822 auto &VI = T.first; 823 // This was enforced earlier during trySingleImplDevirt. 824 assert(VI.getSummaryList().size() == 1 && 825 "Devirt of local target has more than one copy"); 826 auto &S = VI.getSummaryList()[0]; 827 if (!isExported(S->modulePath(), VI)) 828 continue; 829 830 // It's been exported by a cross module import. 831 for (auto &SlotSummary : T.second) { 832 auto *TIdSum = Summary.getTypeIdSummary(SlotSummary.TypeID); 833 assert(TIdSum); 834 auto WPDRes = TIdSum->WPDRes.find(SlotSummary.ByteOffset); 835 assert(WPDRes != TIdSum->WPDRes.end()); 836 WPDRes->second.SingleImplName = ModuleSummaryIndex::getGlobalNameForLocal( 837 WPDRes->second.SingleImplName, 838 Summary.getModuleHash(S->modulePath())); 839 } 840 } 841 } 842 843 } // end namespace llvm 844 845 static Error checkCombinedSummaryForTesting(ModuleSummaryIndex *Summary) { 846 // Check that summary index contains regular LTO module when performing 847 // export to prevent occasional use of index from pure ThinLTO compilation 848 // (-fno-split-lto-module). This kind of summary index is passed to 849 // DevirtIndex::run, not to DevirtModule::run used by opt/runForTesting. 850 const auto &ModPaths = Summary->modulePaths(); 851 if (ClSummaryAction != PassSummaryAction::Import && 852 ModPaths.find(ModuleSummaryIndex::getRegularLTOModuleName()) == 853 ModPaths.end()) 854 return createStringError( 855 errc::invalid_argument, 856 "combined summary should contain Regular LTO module"); 857 return ErrorSuccess(); 858 } 859 860 bool DevirtModule::runForTesting( 861 Module &M, function_ref<AAResults &(Function &)> AARGetter, 862 function_ref<OptimizationRemarkEmitter &(Function *)> OREGetter, 863 function_ref<DominatorTree &(Function &)> LookupDomTree) { 864 std::unique_ptr<ModuleSummaryIndex> Summary = 865 std::make_unique<ModuleSummaryIndex>(/*HaveGVs=*/false); 866 867 // Handle the command-line summary arguments. This code is for testing 868 // purposes only, so we handle errors directly. 869 if (!ClReadSummary.empty()) { 870 ExitOnError ExitOnErr("-wholeprogramdevirt-read-summary: " + ClReadSummary + 871 ": "); 872 auto ReadSummaryFile = 873 ExitOnErr(errorOrToExpected(MemoryBuffer::getFile(ClReadSummary))); 874 if (Expected<std::unique_ptr<ModuleSummaryIndex>> SummaryOrErr = 875 getModuleSummaryIndex(*ReadSummaryFile)) { 876 Summary = std::move(*SummaryOrErr); 877 ExitOnErr(checkCombinedSummaryForTesting(Summary.get())); 878 } else { 879 // Try YAML if we've failed with bitcode. 880 consumeError(SummaryOrErr.takeError()); 881 yaml::Input In(ReadSummaryFile->getBuffer()); 882 In >> *Summary; 883 ExitOnErr(errorCodeToError(In.error())); 884 } 885 } 886 887 bool Changed = 888 DevirtModule(M, AARGetter, OREGetter, LookupDomTree, 889 ClSummaryAction == PassSummaryAction::Export ? Summary.get() 890 : nullptr, 891 ClSummaryAction == PassSummaryAction::Import ? Summary.get() 892 : nullptr) 893 .run(); 894 895 if (!ClWriteSummary.empty()) { 896 ExitOnError ExitOnErr( 897 "-wholeprogramdevirt-write-summary: " + ClWriteSummary + ": "); 898 std::error_code EC; 899 if (StringRef(ClWriteSummary).endswith(".bc")) { 900 raw_fd_ostream OS(ClWriteSummary, EC, sys::fs::OF_None); 901 ExitOnErr(errorCodeToError(EC)); 902 WriteIndexToFile(*Summary, OS); 903 } else { 904 raw_fd_ostream OS(ClWriteSummary, EC, sys::fs::OF_Text); 905 ExitOnErr(errorCodeToError(EC)); 906 yaml::Output Out(OS); 907 Out << *Summary; 908 } 909 } 910 911 return Changed; 912 } 913 914 void DevirtModule::buildTypeIdentifierMap( 915 std::vector<VTableBits> &Bits, 916 DenseMap<Metadata *, std::set<TypeMemberInfo>> &TypeIdMap) { 917 DenseMap<GlobalVariable *, VTableBits *> GVToBits; 918 Bits.reserve(M.getGlobalList().size()); 919 SmallVector<MDNode *, 2> Types; 920 for (GlobalVariable &GV : M.globals()) { 921 Types.clear(); 922 GV.getMetadata(LLVMContext::MD_type, Types); 923 if (GV.isDeclaration() || Types.empty()) 924 continue; 925 926 VTableBits *&BitsPtr = GVToBits[&GV]; 927 if (!BitsPtr) { 928 Bits.emplace_back(); 929 Bits.back().GV = &GV; 930 Bits.back().ObjectSize = 931 M.getDataLayout().getTypeAllocSize(GV.getInitializer()->getType()); 932 BitsPtr = &Bits.back(); 933 } 934 935 for (MDNode *Type : Types) { 936 auto TypeID = Type->getOperand(1).get(); 937 938 uint64_t Offset = 939 cast<ConstantInt>( 940 cast<ConstantAsMetadata>(Type->getOperand(0))->getValue()) 941 ->getZExtValue(); 942 943 TypeIdMap[TypeID].insert({BitsPtr, Offset}); 944 } 945 } 946 } 947 948 bool DevirtModule::tryFindVirtualCallTargets( 949 std::vector<VirtualCallTarget> &TargetsForSlot, 950 const std::set<TypeMemberInfo> &TypeMemberInfos, uint64_t ByteOffset) { 951 for (const TypeMemberInfo &TM : TypeMemberInfos) { 952 if (!TM.Bits->GV->isConstant()) 953 return false; 954 955 // We cannot perform whole program devirtualization analysis on a vtable 956 // with public LTO visibility. 957 if (TM.Bits->GV->getVCallVisibility() == 958 GlobalObject::VCallVisibilityPublic) 959 return false; 960 961 Constant *Ptr = getPointerAtOffset(TM.Bits->GV->getInitializer(), 962 TM.Offset + ByteOffset, M); 963 if (!Ptr) 964 return false; 965 966 auto Fn = dyn_cast<Function>(Ptr->stripPointerCasts()); 967 if (!Fn) 968 return false; 969 970 if (FunctionsToSkip.match(Fn->getName())) 971 return false; 972 973 // We can disregard __cxa_pure_virtual as a possible call target, as 974 // calls to pure virtuals are UB. 975 if (Fn->getName() == "__cxa_pure_virtual") 976 continue; 977 978 TargetsForSlot.push_back({Fn, &TM}); 979 } 980 981 // Give up if we couldn't find any targets. 982 return !TargetsForSlot.empty(); 983 } 984 985 bool DevirtIndex::tryFindVirtualCallTargets( 986 std::vector<ValueInfo> &TargetsForSlot, const TypeIdCompatibleVtableInfo TIdInfo, 987 uint64_t ByteOffset) { 988 for (const TypeIdOffsetVtableInfo &P : TIdInfo) { 989 // Find the first non-available_externally linkage vtable initializer. 990 // We can have multiple available_externally, linkonce_odr and weak_odr 991 // vtable initializers, however we want to skip available_externally as they 992 // do not have type metadata attached, and therefore the summary will not 993 // contain any vtable functions. We can also have multiple external 994 // vtable initializers in the case of comdats, which we cannot check here. 995 // The linker should give an error in this case. 996 // 997 // Also, handle the case of same-named local Vtables with the same path 998 // and therefore the same GUID. This can happen if there isn't enough 999 // distinguishing path when compiling the source file. In that case we 1000 // conservatively return false early. 1001 const GlobalVarSummary *VS = nullptr; 1002 bool LocalFound = false; 1003 for (auto &S : P.VTableVI.getSummaryList()) { 1004 if (GlobalValue::isLocalLinkage(S->linkage())) { 1005 if (LocalFound) 1006 return false; 1007 LocalFound = true; 1008 } 1009 if (!GlobalValue::isAvailableExternallyLinkage(S->linkage())) { 1010 VS = cast<GlobalVarSummary>(S->getBaseObject()); 1011 // We cannot perform whole program devirtualization analysis on a vtable 1012 // with public LTO visibility. 1013 if (VS->getVCallVisibility() == GlobalObject::VCallVisibilityPublic) 1014 return false; 1015 } 1016 } 1017 if (!VS->isLive()) 1018 continue; 1019 for (auto VTP : VS->vTableFuncs()) { 1020 if (VTP.VTableOffset != P.AddressPointOffset + ByteOffset) 1021 continue; 1022 1023 TargetsForSlot.push_back(VTP.FuncVI); 1024 } 1025 } 1026 1027 // Give up if we couldn't find any targets. 1028 return !TargetsForSlot.empty(); 1029 } 1030 1031 void DevirtModule::applySingleImplDevirt(VTableSlotInfo &SlotInfo, 1032 Constant *TheFn, bool &IsExported) { 1033 // Don't devirtualize function if we're told to skip it 1034 // in -wholeprogramdevirt-skip. 1035 if (FunctionsToSkip.match(TheFn->stripPointerCasts()->getName())) 1036 return; 1037 auto Apply = [&](CallSiteInfo &CSInfo) { 1038 for (auto &&VCallSite : CSInfo.CallSites) { 1039 if (RemarksEnabled) 1040 VCallSite.emitRemark("single-impl", 1041 TheFn->stripPointerCasts()->getName(), OREGetter); 1042 VCallSite.CB.setCalledOperand(ConstantExpr::getBitCast( 1043 TheFn, VCallSite.CB.getCalledOperand()->getType())); 1044 // This use is no longer unsafe. 1045 if (VCallSite.NumUnsafeUses) 1046 --*VCallSite.NumUnsafeUses; 1047 } 1048 if (CSInfo.isExported()) 1049 IsExported = true; 1050 CSInfo.markDevirt(); 1051 }; 1052 Apply(SlotInfo.CSInfo); 1053 for (auto &P : SlotInfo.ConstCSInfo) 1054 Apply(P.second); 1055 } 1056 1057 static bool AddCalls(VTableSlotInfo &SlotInfo, const ValueInfo &Callee) { 1058 // We can't add calls if we haven't seen a definition 1059 if (Callee.getSummaryList().empty()) 1060 return false; 1061 1062 // Insert calls into the summary index so that the devirtualized targets 1063 // are eligible for import. 1064 // FIXME: Annotate type tests with hotness. For now, mark these as hot 1065 // to better ensure we have the opportunity to inline them. 1066 bool IsExported = false; 1067 auto &S = Callee.getSummaryList()[0]; 1068 CalleeInfo CI(CalleeInfo::HotnessType::Hot, /* RelBF = */ 0); 1069 auto AddCalls = [&](CallSiteInfo &CSInfo) { 1070 for (auto *FS : CSInfo.SummaryTypeCheckedLoadUsers) { 1071 FS->addCall({Callee, CI}); 1072 IsExported |= S->modulePath() != FS->modulePath(); 1073 } 1074 for (auto *FS : CSInfo.SummaryTypeTestAssumeUsers) { 1075 FS->addCall({Callee, CI}); 1076 IsExported |= S->modulePath() != FS->modulePath(); 1077 } 1078 }; 1079 AddCalls(SlotInfo.CSInfo); 1080 for (auto &P : SlotInfo.ConstCSInfo) 1081 AddCalls(P.second); 1082 return IsExported; 1083 } 1084 1085 bool DevirtModule::trySingleImplDevirt( 1086 ModuleSummaryIndex *ExportSummary, 1087 MutableArrayRef<VirtualCallTarget> TargetsForSlot, VTableSlotInfo &SlotInfo, 1088 WholeProgramDevirtResolution *Res) { 1089 // See if the program contains a single implementation of this virtual 1090 // function. 1091 Function *TheFn = TargetsForSlot[0].Fn; 1092 for (auto &&Target : TargetsForSlot) 1093 if (TheFn != Target.Fn) 1094 return false; 1095 1096 // If so, update each call site to call that implementation directly. 1097 if (RemarksEnabled) 1098 TargetsForSlot[0].WasDevirt = true; 1099 1100 bool IsExported = false; 1101 applySingleImplDevirt(SlotInfo, TheFn, IsExported); 1102 if (!IsExported) 1103 return false; 1104 1105 // If the only implementation has local linkage, we must promote to external 1106 // to make it visible to thin LTO objects. We can only get here during the 1107 // ThinLTO export phase. 1108 if (TheFn->hasLocalLinkage()) { 1109 std::string NewName = (TheFn->getName() + "$merged").str(); 1110 1111 // Since we are renaming the function, any comdats with the same name must 1112 // also be renamed. This is required when targeting COFF, as the comdat name 1113 // must match one of the names of the symbols in the comdat. 1114 if (Comdat *C = TheFn->getComdat()) { 1115 if (C->getName() == TheFn->getName()) { 1116 Comdat *NewC = M.getOrInsertComdat(NewName); 1117 NewC->setSelectionKind(C->getSelectionKind()); 1118 for (GlobalObject &GO : M.global_objects()) 1119 if (GO.getComdat() == C) 1120 GO.setComdat(NewC); 1121 } 1122 } 1123 1124 TheFn->setLinkage(GlobalValue::ExternalLinkage); 1125 TheFn->setVisibility(GlobalValue::HiddenVisibility); 1126 TheFn->setName(NewName); 1127 } 1128 if (ValueInfo TheFnVI = ExportSummary->getValueInfo(TheFn->getGUID())) 1129 // Any needed promotion of 'TheFn' has already been done during 1130 // LTO unit split, so we can ignore return value of AddCalls. 1131 AddCalls(SlotInfo, TheFnVI); 1132 1133 Res->TheKind = WholeProgramDevirtResolution::SingleImpl; 1134 Res->SingleImplName = std::string(TheFn->getName()); 1135 1136 return true; 1137 } 1138 1139 bool DevirtIndex::trySingleImplDevirt(MutableArrayRef<ValueInfo> TargetsForSlot, 1140 VTableSlotSummary &SlotSummary, 1141 VTableSlotInfo &SlotInfo, 1142 WholeProgramDevirtResolution *Res, 1143 std::set<ValueInfo> &DevirtTargets) { 1144 // See if the program contains a single implementation of this virtual 1145 // function. 1146 auto TheFn = TargetsForSlot[0]; 1147 for (auto &&Target : TargetsForSlot) 1148 if (TheFn != Target) 1149 return false; 1150 1151 // Don't devirtualize if we don't have target definition. 1152 auto Size = TheFn.getSummaryList().size(); 1153 if (!Size) 1154 return false; 1155 1156 // Don't devirtualize function if we're told to skip it 1157 // in -wholeprogramdevirt-skip. 1158 if (FunctionsToSkip.match(TheFn.name())) 1159 return false; 1160 1161 // If the summary list contains multiple summaries where at least one is 1162 // a local, give up, as we won't know which (possibly promoted) name to use. 1163 for (auto &S : TheFn.getSummaryList()) 1164 if (GlobalValue::isLocalLinkage(S->linkage()) && Size > 1) 1165 return false; 1166 1167 // Collect functions devirtualized at least for one call site for stats. 1168 if (PrintSummaryDevirt) 1169 DevirtTargets.insert(TheFn); 1170 1171 auto &S = TheFn.getSummaryList()[0]; 1172 bool IsExported = AddCalls(SlotInfo, TheFn); 1173 if (IsExported) 1174 ExportedGUIDs.insert(TheFn.getGUID()); 1175 1176 // Record in summary for use in devirtualization during the ThinLTO import 1177 // step. 1178 Res->TheKind = WholeProgramDevirtResolution::SingleImpl; 1179 if (GlobalValue::isLocalLinkage(S->linkage())) { 1180 if (IsExported) 1181 // If target is a local function and we are exporting it by 1182 // devirtualizing a call in another module, we need to record the 1183 // promoted name. 1184 Res->SingleImplName = ModuleSummaryIndex::getGlobalNameForLocal( 1185 TheFn.name(), ExportSummary.getModuleHash(S->modulePath())); 1186 else { 1187 LocalWPDTargetsMap[TheFn].push_back(SlotSummary); 1188 Res->SingleImplName = std::string(TheFn.name()); 1189 } 1190 } else 1191 Res->SingleImplName = std::string(TheFn.name()); 1192 1193 // Name will be empty if this thin link driven off of serialized combined 1194 // index (e.g. llvm-lto). However, WPD is not supported/invoked for the 1195 // legacy LTO API anyway. 1196 assert(!Res->SingleImplName.empty()); 1197 1198 return true; 1199 } 1200 1201 void DevirtModule::tryICallBranchFunnel( 1202 MutableArrayRef<VirtualCallTarget> TargetsForSlot, VTableSlotInfo &SlotInfo, 1203 WholeProgramDevirtResolution *Res, VTableSlot Slot) { 1204 Triple T(M.getTargetTriple()); 1205 if (T.getArch() != Triple::x86_64) 1206 return; 1207 1208 if (TargetsForSlot.size() > ClThreshold) 1209 return; 1210 1211 bool HasNonDevirt = !SlotInfo.CSInfo.AllCallSitesDevirted; 1212 if (!HasNonDevirt) 1213 for (auto &P : SlotInfo.ConstCSInfo) 1214 if (!P.second.AllCallSitesDevirted) { 1215 HasNonDevirt = true; 1216 break; 1217 } 1218 1219 if (!HasNonDevirt) 1220 return; 1221 1222 FunctionType *FT = 1223 FunctionType::get(Type::getVoidTy(M.getContext()), {Int8PtrTy}, true); 1224 Function *JT; 1225 if (isa<MDString>(Slot.TypeID)) { 1226 JT = Function::Create(FT, Function::ExternalLinkage, 1227 M.getDataLayout().getProgramAddressSpace(), 1228 getGlobalName(Slot, {}, "branch_funnel"), &M); 1229 JT->setVisibility(GlobalValue::HiddenVisibility); 1230 } else { 1231 JT = Function::Create(FT, Function::InternalLinkage, 1232 M.getDataLayout().getProgramAddressSpace(), 1233 "branch_funnel", &M); 1234 } 1235 JT->addAttribute(1, Attribute::Nest); 1236 1237 std::vector<Value *> JTArgs; 1238 JTArgs.push_back(JT->arg_begin()); 1239 for (auto &T : TargetsForSlot) { 1240 JTArgs.push_back(getMemberAddr(T.TM)); 1241 JTArgs.push_back(T.Fn); 1242 } 1243 1244 BasicBlock *BB = BasicBlock::Create(M.getContext(), "", JT, nullptr); 1245 Function *Intr = 1246 Intrinsic::getDeclaration(&M, llvm::Intrinsic::icall_branch_funnel, {}); 1247 1248 auto *CI = CallInst::Create(Intr, JTArgs, "", BB); 1249 CI->setTailCallKind(CallInst::TCK_MustTail); 1250 ReturnInst::Create(M.getContext(), nullptr, BB); 1251 1252 bool IsExported = false; 1253 applyICallBranchFunnel(SlotInfo, JT, IsExported); 1254 if (IsExported) 1255 Res->TheKind = WholeProgramDevirtResolution::BranchFunnel; 1256 } 1257 1258 void DevirtModule::applyICallBranchFunnel(VTableSlotInfo &SlotInfo, 1259 Constant *JT, bool &IsExported) { 1260 auto Apply = [&](CallSiteInfo &CSInfo) { 1261 if (CSInfo.isExported()) 1262 IsExported = true; 1263 if (CSInfo.AllCallSitesDevirted) 1264 return; 1265 for (auto &&VCallSite : CSInfo.CallSites) { 1266 CallBase &CB = VCallSite.CB; 1267 1268 // Jump tables are only profitable if the retpoline mitigation is enabled. 1269 Attribute FSAttr = CB.getCaller()->getFnAttribute("target-features"); 1270 if (!FSAttr.isValid() || 1271 !FSAttr.getValueAsString().contains("+retpoline")) 1272 continue; 1273 1274 if (RemarksEnabled) 1275 VCallSite.emitRemark("branch-funnel", 1276 JT->stripPointerCasts()->getName(), OREGetter); 1277 1278 // Pass the address of the vtable in the nest register, which is r10 on 1279 // x86_64. 1280 std::vector<Type *> NewArgs; 1281 NewArgs.push_back(Int8PtrTy); 1282 append_range(NewArgs, CB.getFunctionType()->params()); 1283 FunctionType *NewFT = 1284 FunctionType::get(CB.getFunctionType()->getReturnType(), NewArgs, 1285 CB.getFunctionType()->isVarArg()); 1286 PointerType *NewFTPtr = PointerType::getUnqual(NewFT); 1287 1288 IRBuilder<> IRB(&CB); 1289 std::vector<Value *> Args; 1290 Args.push_back(IRB.CreateBitCast(VCallSite.VTable, Int8PtrTy)); 1291 llvm::append_range(Args, CB.args()); 1292 1293 CallBase *NewCS = nullptr; 1294 if (isa<CallInst>(CB)) 1295 NewCS = IRB.CreateCall(NewFT, IRB.CreateBitCast(JT, NewFTPtr), Args); 1296 else 1297 NewCS = IRB.CreateInvoke(NewFT, IRB.CreateBitCast(JT, NewFTPtr), 1298 cast<InvokeInst>(CB).getNormalDest(), 1299 cast<InvokeInst>(CB).getUnwindDest(), Args); 1300 NewCS->setCallingConv(CB.getCallingConv()); 1301 1302 AttributeList Attrs = CB.getAttributes(); 1303 std::vector<AttributeSet> NewArgAttrs; 1304 NewArgAttrs.push_back(AttributeSet::get( 1305 M.getContext(), ArrayRef<Attribute>{Attribute::get( 1306 M.getContext(), Attribute::Nest)})); 1307 for (unsigned I = 0; I + 2 < Attrs.getNumAttrSets(); ++I) 1308 NewArgAttrs.push_back(Attrs.getParamAttributes(I)); 1309 NewCS->setAttributes( 1310 AttributeList::get(M.getContext(), Attrs.getFnAttributes(), 1311 Attrs.getRetAttributes(), NewArgAttrs)); 1312 1313 CB.replaceAllUsesWith(NewCS); 1314 CB.eraseFromParent(); 1315 1316 // This use is no longer unsafe. 1317 if (VCallSite.NumUnsafeUses) 1318 --*VCallSite.NumUnsafeUses; 1319 } 1320 // Don't mark as devirtualized because there may be callers compiled without 1321 // retpoline mitigation, which would mean that they are lowered to 1322 // llvm.type.test and therefore require an llvm.type.test resolution for the 1323 // type identifier. 1324 }; 1325 Apply(SlotInfo.CSInfo); 1326 for (auto &P : SlotInfo.ConstCSInfo) 1327 Apply(P.second); 1328 } 1329 1330 bool DevirtModule::tryEvaluateFunctionsWithArgs( 1331 MutableArrayRef<VirtualCallTarget> TargetsForSlot, 1332 ArrayRef<uint64_t> Args) { 1333 // Evaluate each function and store the result in each target's RetVal 1334 // field. 1335 for (VirtualCallTarget &Target : TargetsForSlot) { 1336 if (Target.Fn->arg_size() != Args.size() + 1) 1337 return false; 1338 1339 Evaluator Eval(M.getDataLayout(), nullptr); 1340 SmallVector<Constant *, 2> EvalArgs; 1341 EvalArgs.push_back( 1342 Constant::getNullValue(Target.Fn->getFunctionType()->getParamType(0))); 1343 for (unsigned I = 0; I != Args.size(); ++I) { 1344 auto *ArgTy = dyn_cast<IntegerType>( 1345 Target.Fn->getFunctionType()->getParamType(I + 1)); 1346 if (!ArgTy) 1347 return false; 1348 EvalArgs.push_back(ConstantInt::get(ArgTy, Args[I])); 1349 } 1350 1351 Constant *RetVal; 1352 if (!Eval.EvaluateFunction(Target.Fn, RetVal, EvalArgs) || 1353 !isa<ConstantInt>(RetVal)) 1354 return false; 1355 Target.RetVal = cast<ConstantInt>(RetVal)->getZExtValue(); 1356 } 1357 return true; 1358 } 1359 1360 void DevirtModule::applyUniformRetValOpt(CallSiteInfo &CSInfo, StringRef FnName, 1361 uint64_t TheRetVal) { 1362 for (auto Call : CSInfo.CallSites) 1363 Call.replaceAndErase( 1364 "uniform-ret-val", FnName, RemarksEnabled, OREGetter, 1365 ConstantInt::get(cast<IntegerType>(Call.CB.getType()), TheRetVal)); 1366 CSInfo.markDevirt(); 1367 } 1368 1369 bool DevirtModule::tryUniformRetValOpt( 1370 MutableArrayRef<VirtualCallTarget> TargetsForSlot, CallSiteInfo &CSInfo, 1371 WholeProgramDevirtResolution::ByArg *Res) { 1372 // Uniform return value optimization. If all functions return the same 1373 // constant, replace all calls with that constant. 1374 uint64_t TheRetVal = TargetsForSlot[0].RetVal; 1375 for (const VirtualCallTarget &Target : TargetsForSlot) 1376 if (Target.RetVal != TheRetVal) 1377 return false; 1378 1379 if (CSInfo.isExported()) { 1380 Res->TheKind = WholeProgramDevirtResolution::ByArg::UniformRetVal; 1381 Res->Info = TheRetVal; 1382 } 1383 1384 applyUniformRetValOpt(CSInfo, TargetsForSlot[0].Fn->getName(), TheRetVal); 1385 if (RemarksEnabled) 1386 for (auto &&Target : TargetsForSlot) 1387 Target.WasDevirt = true; 1388 return true; 1389 } 1390 1391 std::string DevirtModule::getGlobalName(VTableSlot Slot, 1392 ArrayRef<uint64_t> Args, 1393 StringRef Name) { 1394 std::string FullName = "__typeid_"; 1395 raw_string_ostream OS(FullName); 1396 OS << cast<MDString>(Slot.TypeID)->getString() << '_' << Slot.ByteOffset; 1397 for (uint64_t Arg : Args) 1398 OS << '_' << Arg; 1399 OS << '_' << Name; 1400 return OS.str(); 1401 } 1402 1403 bool DevirtModule::shouldExportConstantsAsAbsoluteSymbols() { 1404 Triple T(M.getTargetTriple()); 1405 return T.isX86() && T.getObjectFormat() == Triple::ELF; 1406 } 1407 1408 void DevirtModule::exportGlobal(VTableSlot Slot, ArrayRef<uint64_t> Args, 1409 StringRef Name, Constant *C) { 1410 GlobalAlias *GA = GlobalAlias::create(Int8Ty, 0, GlobalValue::ExternalLinkage, 1411 getGlobalName(Slot, Args, Name), C, &M); 1412 GA->setVisibility(GlobalValue::HiddenVisibility); 1413 } 1414 1415 void DevirtModule::exportConstant(VTableSlot Slot, ArrayRef<uint64_t> Args, 1416 StringRef Name, uint32_t Const, 1417 uint32_t &Storage) { 1418 if (shouldExportConstantsAsAbsoluteSymbols()) { 1419 exportGlobal( 1420 Slot, Args, Name, 1421 ConstantExpr::getIntToPtr(ConstantInt::get(Int32Ty, Const), Int8PtrTy)); 1422 return; 1423 } 1424 1425 Storage = Const; 1426 } 1427 1428 Constant *DevirtModule::importGlobal(VTableSlot Slot, ArrayRef<uint64_t> Args, 1429 StringRef Name) { 1430 Constant *C = 1431 M.getOrInsertGlobal(getGlobalName(Slot, Args, Name), Int8Arr0Ty); 1432 auto *GV = dyn_cast<GlobalVariable>(C); 1433 if (GV) 1434 GV->setVisibility(GlobalValue::HiddenVisibility); 1435 return C; 1436 } 1437 1438 Constant *DevirtModule::importConstant(VTableSlot Slot, ArrayRef<uint64_t> Args, 1439 StringRef Name, IntegerType *IntTy, 1440 uint32_t Storage) { 1441 if (!shouldExportConstantsAsAbsoluteSymbols()) 1442 return ConstantInt::get(IntTy, Storage); 1443 1444 Constant *C = importGlobal(Slot, Args, Name); 1445 auto *GV = cast<GlobalVariable>(C->stripPointerCasts()); 1446 C = ConstantExpr::getPtrToInt(C, IntTy); 1447 1448 // We only need to set metadata if the global is newly created, in which 1449 // case it would not have hidden visibility. 1450 if (GV->hasMetadata(LLVMContext::MD_absolute_symbol)) 1451 return C; 1452 1453 auto SetAbsRange = [&](uint64_t Min, uint64_t Max) { 1454 auto *MinC = ConstantAsMetadata::get(ConstantInt::get(IntPtrTy, Min)); 1455 auto *MaxC = ConstantAsMetadata::get(ConstantInt::get(IntPtrTy, Max)); 1456 GV->setMetadata(LLVMContext::MD_absolute_symbol, 1457 MDNode::get(M.getContext(), {MinC, MaxC})); 1458 }; 1459 unsigned AbsWidth = IntTy->getBitWidth(); 1460 if (AbsWidth == IntPtrTy->getBitWidth()) 1461 SetAbsRange(~0ull, ~0ull); // Full set. 1462 else 1463 SetAbsRange(0, 1ull << AbsWidth); 1464 return C; 1465 } 1466 1467 void DevirtModule::applyUniqueRetValOpt(CallSiteInfo &CSInfo, StringRef FnName, 1468 bool IsOne, 1469 Constant *UniqueMemberAddr) { 1470 for (auto &&Call : CSInfo.CallSites) { 1471 IRBuilder<> B(&Call.CB); 1472 Value *Cmp = 1473 B.CreateICmp(IsOne ? ICmpInst::ICMP_EQ : ICmpInst::ICMP_NE, Call.VTable, 1474 B.CreateBitCast(UniqueMemberAddr, Call.VTable->getType())); 1475 Cmp = B.CreateZExt(Cmp, Call.CB.getType()); 1476 Call.replaceAndErase("unique-ret-val", FnName, RemarksEnabled, OREGetter, 1477 Cmp); 1478 } 1479 CSInfo.markDevirt(); 1480 } 1481 1482 Constant *DevirtModule::getMemberAddr(const TypeMemberInfo *M) { 1483 Constant *C = ConstantExpr::getBitCast(M->Bits->GV, Int8PtrTy); 1484 return ConstantExpr::getGetElementPtr(Int8Ty, C, 1485 ConstantInt::get(Int64Ty, M->Offset)); 1486 } 1487 1488 bool DevirtModule::tryUniqueRetValOpt( 1489 unsigned BitWidth, MutableArrayRef<VirtualCallTarget> TargetsForSlot, 1490 CallSiteInfo &CSInfo, WholeProgramDevirtResolution::ByArg *Res, 1491 VTableSlot Slot, ArrayRef<uint64_t> Args) { 1492 // IsOne controls whether we look for a 0 or a 1. 1493 auto tryUniqueRetValOptFor = [&](bool IsOne) { 1494 const TypeMemberInfo *UniqueMember = nullptr; 1495 for (const VirtualCallTarget &Target : TargetsForSlot) { 1496 if (Target.RetVal == (IsOne ? 1 : 0)) { 1497 if (UniqueMember) 1498 return false; 1499 UniqueMember = Target.TM; 1500 } 1501 } 1502 1503 // We should have found a unique member or bailed out by now. We already 1504 // checked for a uniform return value in tryUniformRetValOpt. 1505 assert(UniqueMember); 1506 1507 Constant *UniqueMemberAddr = getMemberAddr(UniqueMember); 1508 if (CSInfo.isExported()) { 1509 Res->TheKind = WholeProgramDevirtResolution::ByArg::UniqueRetVal; 1510 Res->Info = IsOne; 1511 1512 exportGlobal(Slot, Args, "unique_member", UniqueMemberAddr); 1513 } 1514 1515 // Replace each call with the comparison. 1516 applyUniqueRetValOpt(CSInfo, TargetsForSlot[0].Fn->getName(), IsOne, 1517 UniqueMemberAddr); 1518 1519 // Update devirtualization statistics for targets. 1520 if (RemarksEnabled) 1521 for (auto &&Target : TargetsForSlot) 1522 Target.WasDevirt = true; 1523 1524 return true; 1525 }; 1526 1527 if (BitWidth == 1) { 1528 if (tryUniqueRetValOptFor(true)) 1529 return true; 1530 if (tryUniqueRetValOptFor(false)) 1531 return true; 1532 } 1533 return false; 1534 } 1535 1536 void DevirtModule::applyVirtualConstProp(CallSiteInfo &CSInfo, StringRef FnName, 1537 Constant *Byte, Constant *Bit) { 1538 for (auto Call : CSInfo.CallSites) { 1539 auto *RetType = cast<IntegerType>(Call.CB.getType()); 1540 IRBuilder<> B(&Call.CB); 1541 Value *Addr = 1542 B.CreateGEP(Int8Ty, B.CreateBitCast(Call.VTable, Int8PtrTy), Byte); 1543 if (RetType->getBitWidth() == 1) { 1544 Value *Bits = B.CreateLoad(Int8Ty, Addr); 1545 Value *BitsAndBit = B.CreateAnd(Bits, Bit); 1546 auto IsBitSet = B.CreateICmpNE(BitsAndBit, ConstantInt::get(Int8Ty, 0)); 1547 Call.replaceAndErase("virtual-const-prop-1-bit", FnName, RemarksEnabled, 1548 OREGetter, IsBitSet); 1549 } else { 1550 Value *ValAddr = B.CreateBitCast(Addr, RetType->getPointerTo()); 1551 Value *Val = B.CreateLoad(RetType, ValAddr); 1552 Call.replaceAndErase("virtual-const-prop", FnName, RemarksEnabled, 1553 OREGetter, Val); 1554 } 1555 } 1556 CSInfo.markDevirt(); 1557 } 1558 1559 bool DevirtModule::tryVirtualConstProp( 1560 MutableArrayRef<VirtualCallTarget> TargetsForSlot, VTableSlotInfo &SlotInfo, 1561 WholeProgramDevirtResolution *Res, VTableSlot Slot) { 1562 // This only works if the function returns an integer. 1563 auto RetType = dyn_cast<IntegerType>(TargetsForSlot[0].Fn->getReturnType()); 1564 if (!RetType) 1565 return false; 1566 unsigned BitWidth = RetType->getBitWidth(); 1567 if (BitWidth > 64) 1568 return false; 1569 1570 // Make sure that each function is defined, does not access memory, takes at 1571 // least one argument, does not use its first argument (which we assume is 1572 // 'this'), and has the same return type. 1573 // 1574 // Note that we test whether this copy of the function is readnone, rather 1575 // than testing function attributes, which must hold for any copy of the 1576 // function, even a less optimized version substituted at link time. This is 1577 // sound because the virtual constant propagation optimizations effectively 1578 // inline all implementations of the virtual function into each call site, 1579 // rather than using function attributes to perform local optimization. 1580 for (VirtualCallTarget &Target : TargetsForSlot) { 1581 if (Target.Fn->isDeclaration() || 1582 computeFunctionBodyMemoryAccess(*Target.Fn, AARGetter(*Target.Fn)) != 1583 MAK_ReadNone || 1584 Target.Fn->arg_empty() || !Target.Fn->arg_begin()->use_empty() || 1585 Target.Fn->getReturnType() != RetType) 1586 return false; 1587 } 1588 1589 for (auto &&CSByConstantArg : SlotInfo.ConstCSInfo) { 1590 if (!tryEvaluateFunctionsWithArgs(TargetsForSlot, CSByConstantArg.first)) 1591 continue; 1592 1593 WholeProgramDevirtResolution::ByArg *ResByArg = nullptr; 1594 if (Res) 1595 ResByArg = &Res->ResByArg[CSByConstantArg.first]; 1596 1597 if (tryUniformRetValOpt(TargetsForSlot, CSByConstantArg.second, ResByArg)) 1598 continue; 1599 1600 if (tryUniqueRetValOpt(BitWidth, TargetsForSlot, CSByConstantArg.second, 1601 ResByArg, Slot, CSByConstantArg.first)) 1602 continue; 1603 1604 // Find an allocation offset in bits in all vtables associated with the 1605 // type. 1606 uint64_t AllocBefore = 1607 findLowestOffset(TargetsForSlot, /*IsAfter=*/false, BitWidth); 1608 uint64_t AllocAfter = 1609 findLowestOffset(TargetsForSlot, /*IsAfter=*/true, BitWidth); 1610 1611 // Calculate the total amount of padding needed to store a value at both 1612 // ends of the object. 1613 uint64_t TotalPaddingBefore = 0, TotalPaddingAfter = 0; 1614 for (auto &&Target : TargetsForSlot) { 1615 TotalPaddingBefore += std::max<int64_t>( 1616 (AllocBefore + 7) / 8 - Target.allocatedBeforeBytes() - 1, 0); 1617 TotalPaddingAfter += std::max<int64_t>( 1618 (AllocAfter + 7) / 8 - Target.allocatedAfterBytes() - 1, 0); 1619 } 1620 1621 // If the amount of padding is too large, give up. 1622 // FIXME: do something smarter here. 1623 if (std::min(TotalPaddingBefore, TotalPaddingAfter) > 128) 1624 continue; 1625 1626 // Calculate the offset to the value as a (possibly negative) byte offset 1627 // and (if applicable) a bit offset, and store the values in the targets. 1628 int64_t OffsetByte; 1629 uint64_t OffsetBit; 1630 if (TotalPaddingBefore <= TotalPaddingAfter) 1631 setBeforeReturnValues(TargetsForSlot, AllocBefore, BitWidth, OffsetByte, 1632 OffsetBit); 1633 else 1634 setAfterReturnValues(TargetsForSlot, AllocAfter, BitWidth, OffsetByte, 1635 OffsetBit); 1636 1637 if (RemarksEnabled) 1638 for (auto &&Target : TargetsForSlot) 1639 Target.WasDevirt = true; 1640 1641 1642 if (CSByConstantArg.second.isExported()) { 1643 ResByArg->TheKind = WholeProgramDevirtResolution::ByArg::VirtualConstProp; 1644 exportConstant(Slot, CSByConstantArg.first, "byte", OffsetByte, 1645 ResByArg->Byte); 1646 exportConstant(Slot, CSByConstantArg.first, "bit", 1ULL << OffsetBit, 1647 ResByArg->Bit); 1648 } 1649 1650 // Rewrite each call to a load from OffsetByte/OffsetBit. 1651 Constant *ByteConst = ConstantInt::get(Int32Ty, OffsetByte); 1652 Constant *BitConst = ConstantInt::get(Int8Ty, 1ULL << OffsetBit); 1653 applyVirtualConstProp(CSByConstantArg.second, 1654 TargetsForSlot[0].Fn->getName(), ByteConst, BitConst); 1655 } 1656 return true; 1657 } 1658 1659 void DevirtModule::rebuildGlobal(VTableBits &B) { 1660 if (B.Before.Bytes.empty() && B.After.Bytes.empty()) 1661 return; 1662 1663 // Align the before byte array to the global's minimum alignment so that we 1664 // don't break any alignment requirements on the global. 1665 Align Alignment = M.getDataLayout().getValueOrABITypeAlignment( 1666 B.GV->getAlign(), B.GV->getValueType()); 1667 B.Before.Bytes.resize(alignTo(B.Before.Bytes.size(), Alignment)); 1668 1669 // Before was stored in reverse order; flip it now. 1670 for (size_t I = 0, Size = B.Before.Bytes.size(); I != Size / 2; ++I) 1671 std::swap(B.Before.Bytes[I], B.Before.Bytes[Size - 1 - I]); 1672 1673 // Build an anonymous global containing the before bytes, followed by the 1674 // original initializer, followed by the after bytes. 1675 auto NewInit = ConstantStruct::getAnon( 1676 {ConstantDataArray::get(M.getContext(), B.Before.Bytes), 1677 B.GV->getInitializer(), 1678 ConstantDataArray::get(M.getContext(), B.After.Bytes)}); 1679 auto NewGV = 1680 new GlobalVariable(M, NewInit->getType(), B.GV->isConstant(), 1681 GlobalVariable::PrivateLinkage, NewInit, "", B.GV); 1682 NewGV->setSection(B.GV->getSection()); 1683 NewGV->setComdat(B.GV->getComdat()); 1684 NewGV->setAlignment(MaybeAlign(B.GV->getAlignment())); 1685 1686 // Copy the original vtable's metadata to the anonymous global, adjusting 1687 // offsets as required. 1688 NewGV->copyMetadata(B.GV, B.Before.Bytes.size()); 1689 1690 // Build an alias named after the original global, pointing at the second 1691 // element (the original initializer). 1692 auto Alias = GlobalAlias::create( 1693 B.GV->getInitializer()->getType(), 0, B.GV->getLinkage(), "", 1694 ConstantExpr::getGetElementPtr( 1695 NewInit->getType(), NewGV, 1696 ArrayRef<Constant *>{ConstantInt::get(Int32Ty, 0), 1697 ConstantInt::get(Int32Ty, 1)}), 1698 &M); 1699 Alias->setVisibility(B.GV->getVisibility()); 1700 Alias->takeName(B.GV); 1701 1702 B.GV->replaceAllUsesWith(Alias); 1703 B.GV->eraseFromParent(); 1704 } 1705 1706 bool DevirtModule::areRemarksEnabled() { 1707 const auto &FL = M.getFunctionList(); 1708 for (const Function &Fn : FL) { 1709 const auto &BBL = Fn.getBasicBlockList(); 1710 if (BBL.empty()) 1711 continue; 1712 auto DI = OptimizationRemark(DEBUG_TYPE, "", DebugLoc(), &BBL.front()); 1713 return DI.isEnabled(); 1714 } 1715 return false; 1716 } 1717 1718 void DevirtModule::scanTypeTestUsers( 1719 Function *TypeTestFunc, 1720 DenseMap<Metadata *, std::set<TypeMemberInfo>> &TypeIdMap) { 1721 // Find all virtual calls via a virtual table pointer %p under an assumption 1722 // of the form llvm.assume(llvm.type.test(%p, %md)). This indicates that %p 1723 // points to a member of the type identifier %md. Group calls by (type ID, 1724 // offset) pair (effectively the identity of the virtual function) and store 1725 // to CallSlots. 1726 for (auto I = TypeTestFunc->use_begin(), E = TypeTestFunc->use_end(); 1727 I != E;) { 1728 auto CI = dyn_cast<CallInst>(I->getUser()); 1729 ++I; 1730 if (!CI) 1731 continue; 1732 1733 // Search for virtual calls based on %p and add them to DevirtCalls. 1734 SmallVector<DevirtCallSite, 1> DevirtCalls; 1735 SmallVector<CallInst *, 1> Assumes; 1736 auto &DT = LookupDomTree(*CI->getFunction()); 1737 findDevirtualizableCallsForTypeTest(DevirtCalls, Assumes, CI, DT); 1738 1739 Metadata *TypeId = 1740 cast<MetadataAsValue>(CI->getArgOperand(1))->getMetadata(); 1741 // If we found any, add them to CallSlots. 1742 if (!Assumes.empty()) { 1743 Value *Ptr = CI->getArgOperand(0)->stripPointerCasts(); 1744 for (DevirtCallSite Call : DevirtCalls) 1745 CallSlots[{TypeId, Call.Offset}].addCallSite(Ptr, Call.CB, nullptr); 1746 } 1747 1748 auto RemoveTypeTestAssumes = [&]() { 1749 // We no longer need the assumes or the type test. 1750 for (auto Assume : Assumes) 1751 Assume->eraseFromParent(); 1752 // We can't use RecursivelyDeleteTriviallyDeadInstructions here because we 1753 // may use the vtable argument later. 1754 if (CI->use_empty()) 1755 CI->eraseFromParent(); 1756 }; 1757 1758 // At this point we could remove all type test assume sequences, as they 1759 // were originally inserted for WPD. However, we can keep these in the 1760 // code stream for later analysis (e.g. to help drive more efficient ICP 1761 // sequences). They will eventually be removed by a second LowerTypeTests 1762 // invocation that cleans them up. In order to do this correctly, the first 1763 // LowerTypeTests invocation needs to know that they have "Unknown" type 1764 // test resolution, so that they aren't treated as Unsat and lowered to 1765 // False, which will break any uses on assumes. Below we remove any type 1766 // test assumes that will not be treated as Unknown by LTT. 1767 1768 // The type test assumes will be treated by LTT as Unsat if the type id is 1769 // not used on a global (in which case it has no entry in the TypeIdMap). 1770 if (!TypeIdMap.count(TypeId)) 1771 RemoveTypeTestAssumes(); 1772 1773 // For ThinLTO importing, we need to remove the type test assumes if this is 1774 // an MDString type id without a corresponding TypeIdSummary. Any 1775 // non-MDString type ids are ignored and treated as Unknown by LTT, so their 1776 // type test assumes can be kept. If the MDString type id is missing a 1777 // TypeIdSummary (e.g. because there was no use on a vcall, preventing the 1778 // exporting phase of WPD from analyzing it), then it would be treated as 1779 // Unsat by LTT and we need to remove its type test assumes here. If not 1780 // used on a vcall we don't need them for later optimization use in any 1781 // case. 1782 else if (ImportSummary && isa<MDString>(TypeId)) { 1783 const TypeIdSummary *TidSummary = 1784 ImportSummary->getTypeIdSummary(cast<MDString>(TypeId)->getString()); 1785 if (!TidSummary) 1786 RemoveTypeTestAssumes(); 1787 else 1788 // If one was created it should not be Unsat, because if we reached here 1789 // the type id was used on a global. 1790 assert(TidSummary->TTRes.TheKind != TypeTestResolution::Unsat); 1791 } 1792 } 1793 } 1794 1795 void DevirtModule::scanTypeCheckedLoadUsers(Function *TypeCheckedLoadFunc) { 1796 Function *TypeTestFunc = Intrinsic::getDeclaration(&M, Intrinsic::type_test); 1797 1798 for (auto I = TypeCheckedLoadFunc->use_begin(), 1799 E = TypeCheckedLoadFunc->use_end(); 1800 I != E;) { 1801 auto CI = dyn_cast<CallInst>(I->getUser()); 1802 ++I; 1803 if (!CI) 1804 continue; 1805 1806 Value *Ptr = CI->getArgOperand(0); 1807 Value *Offset = CI->getArgOperand(1); 1808 Value *TypeIdValue = CI->getArgOperand(2); 1809 Metadata *TypeId = cast<MetadataAsValue>(TypeIdValue)->getMetadata(); 1810 1811 SmallVector<DevirtCallSite, 1> DevirtCalls; 1812 SmallVector<Instruction *, 1> LoadedPtrs; 1813 SmallVector<Instruction *, 1> Preds; 1814 bool HasNonCallUses = false; 1815 auto &DT = LookupDomTree(*CI->getFunction()); 1816 findDevirtualizableCallsForTypeCheckedLoad(DevirtCalls, LoadedPtrs, Preds, 1817 HasNonCallUses, CI, DT); 1818 1819 // Start by generating "pessimistic" code that explicitly loads the function 1820 // pointer from the vtable and performs the type check. If possible, we will 1821 // eliminate the load and the type check later. 1822 1823 // If possible, only generate the load at the point where it is used. 1824 // This helps avoid unnecessary spills. 1825 IRBuilder<> LoadB( 1826 (LoadedPtrs.size() == 1 && !HasNonCallUses) ? LoadedPtrs[0] : CI); 1827 Value *GEP = LoadB.CreateGEP(Int8Ty, Ptr, Offset); 1828 Value *GEPPtr = LoadB.CreateBitCast(GEP, PointerType::getUnqual(Int8PtrTy)); 1829 Value *LoadedValue = LoadB.CreateLoad(Int8PtrTy, GEPPtr); 1830 1831 for (Instruction *LoadedPtr : LoadedPtrs) { 1832 LoadedPtr->replaceAllUsesWith(LoadedValue); 1833 LoadedPtr->eraseFromParent(); 1834 } 1835 1836 // Likewise for the type test. 1837 IRBuilder<> CallB((Preds.size() == 1 && !HasNonCallUses) ? Preds[0] : CI); 1838 CallInst *TypeTestCall = CallB.CreateCall(TypeTestFunc, {Ptr, TypeIdValue}); 1839 1840 for (Instruction *Pred : Preds) { 1841 Pred->replaceAllUsesWith(TypeTestCall); 1842 Pred->eraseFromParent(); 1843 } 1844 1845 // We have already erased any extractvalue instructions that refer to the 1846 // intrinsic call, but the intrinsic may have other non-extractvalue uses 1847 // (although this is unlikely). In that case, explicitly build a pair and 1848 // RAUW it. 1849 if (!CI->use_empty()) { 1850 Value *Pair = UndefValue::get(CI->getType()); 1851 IRBuilder<> B(CI); 1852 Pair = B.CreateInsertValue(Pair, LoadedValue, {0}); 1853 Pair = B.CreateInsertValue(Pair, TypeTestCall, {1}); 1854 CI->replaceAllUsesWith(Pair); 1855 } 1856 1857 // The number of unsafe uses is initially the number of uses. 1858 auto &NumUnsafeUses = NumUnsafeUsesForTypeTest[TypeTestCall]; 1859 NumUnsafeUses = DevirtCalls.size(); 1860 1861 // If the function pointer has a non-call user, we cannot eliminate the type 1862 // check, as one of those users may eventually call the pointer. Increment 1863 // the unsafe use count to make sure it cannot reach zero. 1864 if (HasNonCallUses) 1865 ++NumUnsafeUses; 1866 for (DevirtCallSite Call : DevirtCalls) { 1867 CallSlots[{TypeId, Call.Offset}].addCallSite(Ptr, Call.CB, 1868 &NumUnsafeUses); 1869 } 1870 1871 CI->eraseFromParent(); 1872 } 1873 } 1874 1875 void DevirtModule::importResolution(VTableSlot Slot, VTableSlotInfo &SlotInfo) { 1876 auto *TypeId = dyn_cast<MDString>(Slot.TypeID); 1877 if (!TypeId) 1878 return; 1879 const TypeIdSummary *TidSummary = 1880 ImportSummary->getTypeIdSummary(TypeId->getString()); 1881 if (!TidSummary) 1882 return; 1883 auto ResI = TidSummary->WPDRes.find(Slot.ByteOffset); 1884 if (ResI == TidSummary->WPDRes.end()) 1885 return; 1886 const WholeProgramDevirtResolution &Res = ResI->second; 1887 1888 if (Res.TheKind == WholeProgramDevirtResolution::SingleImpl) { 1889 assert(!Res.SingleImplName.empty()); 1890 // The type of the function in the declaration is irrelevant because every 1891 // call site will cast it to the correct type. 1892 Constant *SingleImpl = 1893 cast<Constant>(M.getOrInsertFunction(Res.SingleImplName, 1894 Type::getVoidTy(M.getContext())) 1895 .getCallee()); 1896 1897 // This is the import phase so we should not be exporting anything. 1898 bool IsExported = false; 1899 applySingleImplDevirt(SlotInfo, SingleImpl, IsExported); 1900 assert(!IsExported); 1901 } 1902 1903 for (auto &CSByConstantArg : SlotInfo.ConstCSInfo) { 1904 auto I = Res.ResByArg.find(CSByConstantArg.first); 1905 if (I == Res.ResByArg.end()) 1906 continue; 1907 auto &ResByArg = I->second; 1908 // FIXME: We should figure out what to do about the "function name" argument 1909 // to the apply* functions, as the function names are unavailable during the 1910 // importing phase. For now we just pass the empty string. This does not 1911 // impact correctness because the function names are just used for remarks. 1912 switch (ResByArg.TheKind) { 1913 case WholeProgramDevirtResolution::ByArg::UniformRetVal: 1914 applyUniformRetValOpt(CSByConstantArg.second, "", ResByArg.Info); 1915 break; 1916 case WholeProgramDevirtResolution::ByArg::UniqueRetVal: { 1917 Constant *UniqueMemberAddr = 1918 importGlobal(Slot, CSByConstantArg.first, "unique_member"); 1919 applyUniqueRetValOpt(CSByConstantArg.second, "", ResByArg.Info, 1920 UniqueMemberAddr); 1921 break; 1922 } 1923 case WholeProgramDevirtResolution::ByArg::VirtualConstProp: { 1924 Constant *Byte = importConstant(Slot, CSByConstantArg.first, "byte", 1925 Int32Ty, ResByArg.Byte); 1926 Constant *Bit = importConstant(Slot, CSByConstantArg.first, "bit", Int8Ty, 1927 ResByArg.Bit); 1928 applyVirtualConstProp(CSByConstantArg.second, "", Byte, Bit); 1929 break; 1930 } 1931 default: 1932 break; 1933 } 1934 } 1935 1936 if (Res.TheKind == WholeProgramDevirtResolution::BranchFunnel) { 1937 // The type of the function is irrelevant, because it's bitcast at calls 1938 // anyhow. 1939 Constant *JT = cast<Constant>( 1940 M.getOrInsertFunction(getGlobalName(Slot, {}, "branch_funnel"), 1941 Type::getVoidTy(M.getContext())) 1942 .getCallee()); 1943 bool IsExported = false; 1944 applyICallBranchFunnel(SlotInfo, JT, IsExported); 1945 assert(!IsExported); 1946 } 1947 } 1948 1949 void DevirtModule::removeRedundantTypeTests() { 1950 auto True = ConstantInt::getTrue(M.getContext()); 1951 for (auto &&U : NumUnsafeUsesForTypeTest) { 1952 if (U.second == 0) { 1953 U.first->replaceAllUsesWith(True); 1954 U.first->eraseFromParent(); 1955 } 1956 } 1957 } 1958 1959 bool DevirtModule::run() { 1960 // If only some of the modules were split, we cannot correctly perform 1961 // this transformation. We already checked for the presense of type tests 1962 // with partially split modules during the thin link, and would have emitted 1963 // an error if any were found, so here we can simply return. 1964 if ((ExportSummary && ExportSummary->partiallySplitLTOUnits()) || 1965 (ImportSummary && ImportSummary->partiallySplitLTOUnits())) 1966 return false; 1967 1968 Function *TypeTestFunc = 1969 M.getFunction(Intrinsic::getName(Intrinsic::type_test)); 1970 Function *TypeCheckedLoadFunc = 1971 M.getFunction(Intrinsic::getName(Intrinsic::type_checked_load)); 1972 Function *AssumeFunc = M.getFunction(Intrinsic::getName(Intrinsic::assume)); 1973 1974 // Normally if there are no users of the devirtualization intrinsics in the 1975 // module, this pass has nothing to do. But if we are exporting, we also need 1976 // to handle any users that appear only in the function summaries. 1977 if (!ExportSummary && 1978 (!TypeTestFunc || TypeTestFunc->use_empty() || !AssumeFunc || 1979 AssumeFunc->use_empty()) && 1980 (!TypeCheckedLoadFunc || TypeCheckedLoadFunc->use_empty())) 1981 return false; 1982 1983 // Rebuild type metadata into a map for easy lookup. 1984 std::vector<VTableBits> Bits; 1985 DenseMap<Metadata *, std::set<TypeMemberInfo>> TypeIdMap; 1986 buildTypeIdentifierMap(Bits, TypeIdMap); 1987 1988 if (TypeTestFunc && AssumeFunc) 1989 scanTypeTestUsers(TypeTestFunc, TypeIdMap); 1990 1991 if (TypeCheckedLoadFunc) 1992 scanTypeCheckedLoadUsers(TypeCheckedLoadFunc); 1993 1994 if (ImportSummary) { 1995 for (auto &S : CallSlots) 1996 importResolution(S.first, S.second); 1997 1998 removeRedundantTypeTests(); 1999 2000 // We have lowered or deleted the type instrinsics, so we will no 2001 // longer have enough information to reason about the liveness of virtual 2002 // function pointers in GlobalDCE. 2003 for (GlobalVariable &GV : M.globals()) 2004 GV.eraseMetadata(LLVMContext::MD_vcall_visibility); 2005 2006 // The rest of the code is only necessary when exporting or during regular 2007 // LTO, so we are done. 2008 return true; 2009 } 2010 2011 if (TypeIdMap.empty()) 2012 return true; 2013 2014 // Collect information from summary about which calls to try to devirtualize. 2015 if (ExportSummary) { 2016 DenseMap<GlobalValue::GUID, TinyPtrVector<Metadata *>> MetadataByGUID; 2017 for (auto &P : TypeIdMap) { 2018 if (auto *TypeId = dyn_cast<MDString>(P.first)) 2019 MetadataByGUID[GlobalValue::getGUID(TypeId->getString())].push_back( 2020 TypeId); 2021 } 2022 2023 for (auto &P : *ExportSummary) { 2024 for (auto &S : P.second.SummaryList) { 2025 auto *FS = dyn_cast<FunctionSummary>(S.get()); 2026 if (!FS) 2027 continue; 2028 // FIXME: Only add live functions. 2029 for (FunctionSummary::VFuncId VF : FS->type_test_assume_vcalls()) { 2030 for (Metadata *MD : MetadataByGUID[VF.GUID]) { 2031 CallSlots[{MD, VF.Offset}].CSInfo.addSummaryTypeTestAssumeUser(FS); 2032 } 2033 } 2034 for (FunctionSummary::VFuncId VF : FS->type_checked_load_vcalls()) { 2035 for (Metadata *MD : MetadataByGUID[VF.GUID]) { 2036 CallSlots[{MD, VF.Offset}].CSInfo.addSummaryTypeCheckedLoadUser(FS); 2037 } 2038 } 2039 for (const FunctionSummary::ConstVCall &VC : 2040 FS->type_test_assume_const_vcalls()) { 2041 for (Metadata *MD : MetadataByGUID[VC.VFunc.GUID]) { 2042 CallSlots[{MD, VC.VFunc.Offset}] 2043 .ConstCSInfo[VC.Args] 2044 .addSummaryTypeTestAssumeUser(FS); 2045 } 2046 } 2047 for (const FunctionSummary::ConstVCall &VC : 2048 FS->type_checked_load_const_vcalls()) { 2049 for (Metadata *MD : MetadataByGUID[VC.VFunc.GUID]) { 2050 CallSlots[{MD, VC.VFunc.Offset}] 2051 .ConstCSInfo[VC.Args] 2052 .addSummaryTypeCheckedLoadUser(FS); 2053 } 2054 } 2055 } 2056 } 2057 } 2058 2059 // For each (type, offset) pair: 2060 bool DidVirtualConstProp = false; 2061 std::map<std::string, Function*> DevirtTargets; 2062 for (auto &S : CallSlots) { 2063 // Search each of the members of the type identifier for the virtual 2064 // function implementation at offset S.first.ByteOffset, and add to 2065 // TargetsForSlot. 2066 std::vector<VirtualCallTarget> TargetsForSlot; 2067 WholeProgramDevirtResolution *Res = nullptr; 2068 const std::set<TypeMemberInfo> &TypeMemberInfos = TypeIdMap[S.first.TypeID]; 2069 if (ExportSummary && isa<MDString>(S.first.TypeID) && 2070 TypeMemberInfos.size()) 2071 // For any type id used on a global's type metadata, create the type id 2072 // summary resolution regardless of whether we can devirtualize, so that 2073 // lower type tests knows the type id is not Unsat. If it was not used on 2074 // a global's type metadata, the TypeIdMap entry set will be empty, and 2075 // we don't want to create an entry (with the default Unknown type 2076 // resolution), which can prevent detection of the Unsat. 2077 Res = &ExportSummary 2078 ->getOrInsertTypeIdSummary( 2079 cast<MDString>(S.first.TypeID)->getString()) 2080 .WPDRes[S.first.ByteOffset]; 2081 if (tryFindVirtualCallTargets(TargetsForSlot, TypeMemberInfos, 2082 S.first.ByteOffset)) { 2083 2084 if (!trySingleImplDevirt(ExportSummary, TargetsForSlot, S.second, Res)) { 2085 DidVirtualConstProp |= 2086 tryVirtualConstProp(TargetsForSlot, S.second, Res, S.first); 2087 2088 tryICallBranchFunnel(TargetsForSlot, S.second, Res, S.first); 2089 } 2090 2091 // Collect functions devirtualized at least for one call site for stats. 2092 if (RemarksEnabled) 2093 for (const auto &T : TargetsForSlot) 2094 if (T.WasDevirt) 2095 DevirtTargets[std::string(T.Fn->getName())] = T.Fn; 2096 } 2097 2098 // CFI-specific: if we are exporting and any llvm.type.checked.load 2099 // intrinsics were *not* devirtualized, we need to add the resulting 2100 // llvm.type.test intrinsics to the function summaries so that the 2101 // LowerTypeTests pass will export them. 2102 if (ExportSummary && isa<MDString>(S.first.TypeID)) { 2103 auto GUID = 2104 GlobalValue::getGUID(cast<MDString>(S.first.TypeID)->getString()); 2105 for (auto FS : S.second.CSInfo.SummaryTypeCheckedLoadUsers) 2106 FS->addTypeTest(GUID); 2107 for (auto &CCS : S.second.ConstCSInfo) 2108 for (auto FS : CCS.second.SummaryTypeCheckedLoadUsers) 2109 FS->addTypeTest(GUID); 2110 } 2111 } 2112 2113 if (RemarksEnabled) { 2114 // Generate remarks for each devirtualized function. 2115 for (const auto &DT : DevirtTargets) { 2116 Function *F = DT.second; 2117 2118 using namespace ore; 2119 OREGetter(F).emit(OptimizationRemark(DEBUG_TYPE, "Devirtualized", F) 2120 << "devirtualized " 2121 << NV("FunctionName", DT.first)); 2122 } 2123 } 2124 2125 removeRedundantTypeTests(); 2126 2127 // Rebuild each global we touched as part of virtual constant propagation to 2128 // include the before and after bytes. 2129 if (DidVirtualConstProp) 2130 for (VTableBits &B : Bits) 2131 rebuildGlobal(B); 2132 2133 // We have lowered or deleted the type instrinsics, so we will no 2134 // longer have enough information to reason about the liveness of virtual 2135 // function pointers in GlobalDCE. 2136 for (GlobalVariable &GV : M.globals()) 2137 GV.eraseMetadata(LLVMContext::MD_vcall_visibility); 2138 2139 return true; 2140 } 2141 2142 void DevirtIndex::run() { 2143 if (ExportSummary.typeIdCompatibleVtableMap().empty()) 2144 return; 2145 2146 DenseMap<GlobalValue::GUID, std::vector<StringRef>> NameByGUID; 2147 for (auto &P : ExportSummary.typeIdCompatibleVtableMap()) { 2148 NameByGUID[GlobalValue::getGUID(P.first)].push_back(P.first); 2149 } 2150 2151 // Collect information from summary about which calls to try to devirtualize. 2152 for (auto &P : ExportSummary) { 2153 for (auto &S : P.second.SummaryList) { 2154 auto *FS = dyn_cast<FunctionSummary>(S.get()); 2155 if (!FS) 2156 continue; 2157 // FIXME: Only add live functions. 2158 for (FunctionSummary::VFuncId VF : FS->type_test_assume_vcalls()) { 2159 for (StringRef Name : NameByGUID[VF.GUID]) { 2160 CallSlots[{Name, VF.Offset}].CSInfo.addSummaryTypeTestAssumeUser(FS); 2161 } 2162 } 2163 for (FunctionSummary::VFuncId VF : FS->type_checked_load_vcalls()) { 2164 for (StringRef Name : NameByGUID[VF.GUID]) { 2165 CallSlots[{Name, VF.Offset}].CSInfo.addSummaryTypeCheckedLoadUser(FS); 2166 } 2167 } 2168 for (const FunctionSummary::ConstVCall &VC : 2169 FS->type_test_assume_const_vcalls()) { 2170 for (StringRef Name : NameByGUID[VC.VFunc.GUID]) { 2171 CallSlots[{Name, VC.VFunc.Offset}] 2172 .ConstCSInfo[VC.Args] 2173 .addSummaryTypeTestAssumeUser(FS); 2174 } 2175 } 2176 for (const FunctionSummary::ConstVCall &VC : 2177 FS->type_checked_load_const_vcalls()) { 2178 for (StringRef Name : NameByGUID[VC.VFunc.GUID]) { 2179 CallSlots[{Name, VC.VFunc.Offset}] 2180 .ConstCSInfo[VC.Args] 2181 .addSummaryTypeCheckedLoadUser(FS); 2182 } 2183 } 2184 } 2185 } 2186 2187 std::set<ValueInfo> DevirtTargets; 2188 // For each (type, offset) pair: 2189 for (auto &S : CallSlots) { 2190 // Search each of the members of the type identifier for the virtual 2191 // function implementation at offset S.first.ByteOffset, and add to 2192 // TargetsForSlot. 2193 std::vector<ValueInfo> TargetsForSlot; 2194 auto TidSummary = ExportSummary.getTypeIdCompatibleVtableSummary(S.first.TypeID); 2195 assert(TidSummary); 2196 // Create the type id summary resolution regardlness of whether we can 2197 // devirtualize, so that lower type tests knows the type id is used on 2198 // a global and not Unsat. 2199 WholeProgramDevirtResolution *Res = 2200 &ExportSummary.getOrInsertTypeIdSummary(S.first.TypeID) 2201 .WPDRes[S.first.ByteOffset]; 2202 if (tryFindVirtualCallTargets(TargetsForSlot, *TidSummary, 2203 S.first.ByteOffset)) { 2204 2205 if (!trySingleImplDevirt(TargetsForSlot, S.first, S.second, Res, 2206 DevirtTargets)) 2207 continue; 2208 } 2209 } 2210 2211 // Optionally have the thin link print message for each devirtualized 2212 // function. 2213 if (PrintSummaryDevirt) 2214 for (const auto &DT : DevirtTargets) 2215 errs() << "Devirtualized call to " << DT << "\n"; 2216 } 2217