1 //===---- MachineOutliner.cpp - Outline instructions -----------*- C++ -*-===// 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 /// \file 10 /// Replaces repeated sequences of instructions with function calls. 11 /// 12 /// This works by placing every instruction from every basic block in a 13 /// suffix tree, and repeatedly querying that tree for repeated sequences of 14 /// instructions. If a sequence of instructions appears often, then it ought 15 /// to be beneficial to pull out into a function. 16 /// 17 /// The MachineOutliner communicates with a given target using hooks defined in 18 /// TargetInstrInfo.h. The target supplies the outliner with information on how 19 /// a specific sequence of instructions should be outlined. This information 20 /// is used to deduce the number of instructions necessary to 21 /// 22 /// * Create an outlined function 23 /// * Call that outlined function 24 /// 25 /// Targets must implement 26 /// * getOutliningCandidateInfo 27 /// * buildOutlinedFrame 28 /// * insertOutlinedCall 29 /// * isFunctionSafeToOutlineFrom 30 /// 31 /// in order to make use of the MachineOutliner. 32 /// 33 /// This was originally presented at the 2016 LLVM Developers' Meeting in the 34 /// talk "Reducing Code Size Using Outlining". For a high-level overview of 35 /// how this pass works, the talk is available on YouTube at 36 /// 37 /// https://www.youtube.com/watch?v=yorld-WSOeU 38 /// 39 /// The slides for the talk are available at 40 /// 41 /// http://www.llvm.org/devmtg/2016-11/Slides/Paquette-Outliner.pdf 42 /// 43 /// The talk provides an overview of how the outliner finds candidates and 44 /// ultimately outlines them. It describes how the main data structure for this 45 /// pass, the suffix tree, is queried and purged for candidates. It also gives 46 /// a simplified suffix tree construction algorithm for suffix trees based off 47 /// of the algorithm actually used here, Ukkonen's algorithm. 48 /// 49 /// For the original RFC for this pass, please see 50 /// 51 /// http://lists.llvm.org/pipermail/llvm-dev/2016-August/104170.html 52 /// 53 /// For more information on the suffix tree data structure, please see 54 /// https://www.cs.helsinki.fi/u/ukkonen/SuffixT1withFigs.pdf 55 /// 56 //===----------------------------------------------------------------------===// 57 #include "llvm/CodeGen/MachineOutliner.h" 58 #include "llvm/ADT/DenseMap.h" 59 #include "llvm/ADT/SmallSet.h" 60 #include "llvm/ADT/Statistic.h" 61 #include "llvm/ADT/Twine.h" 62 #include "llvm/Analysis/OptimizationRemarkEmitter.h" 63 #include "llvm/CodeGen/LivePhysRegs.h" 64 #include "llvm/CodeGen/MachineModuleInfo.h" 65 #include "llvm/CodeGen/MachineOptimizationRemarkEmitter.h" 66 #include "llvm/CodeGen/Passes.h" 67 #include "llvm/CodeGen/TargetInstrInfo.h" 68 #include "llvm/CodeGen/TargetSubtargetInfo.h" 69 #include "llvm/IR/DIBuilder.h" 70 #include "llvm/IR/IRBuilder.h" 71 #include "llvm/IR/Mangler.h" 72 #include "llvm/InitializePasses.h" 73 #include "llvm/Support/CommandLine.h" 74 #include "llvm/Support/Debug.h" 75 #include "llvm/Support/SuffixTree.h" 76 #include "llvm/Support/raw_ostream.h" 77 #include <functional> 78 #include <tuple> 79 #include <vector> 80 81 #define DEBUG_TYPE "machine-outliner" 82 83 using namespace llvm; 84 using namespace ore; 85 using namespace outliner; 86 87 // Statistics for outlined functions. 88 STATISTIC(NumOutlined, "Number of candidates outlined"); 89 STATISTIC(FunctionsCreated, "Number of functions created"); 90 91 // Statistics for instruction mapping. 92 STATISTIC(NumLegalInUnsignedVec, "Outlinable instructions mapped"); 93 STATISTIC(NumIllegalInUnsignedVec, 94 "Unoutlinable instructions mapped + number of sentinel values"); 95 STATISTIC(NumSentinels, "Sentinel values inserted during mapping"); 96 STATISTIC(NumInvisible, 97 "Invisible instructions skipped during mapping"); 98 STATISTIC(UnsignedVecSize, 99 "Total number of instructions mapped and saved to mapping vector"); 100 101 // Set to true if the user wants the outliner to run on linkonceodr linkage 102 // functions. This is false by default because the linker can dedupe linkonceodr 103 // functions. Since the outliner is confined to a single module (modulo LTO), 104 // this is off by default. It should, however, be the default behaviour in 105 // LTO. 106 static cl::opt<bool> EnableLinkOnceODROutlining( 107 "enable-linkonceodr-outlining", cl::Hidden, 108 cl::desc("Enable the machine outliner on linkonceodr functions"), 109 cl::init(false)); 110 111 /// Number of times to re-run the outliner. This is not the total number of runs 112 /// as the outliner will run at least one time. The default value is set to 0, 113 /// meaning the outliner will run one time and rerun zero times after that. 114 static cl::opt<unsigned> OutlinerReruns( 115 "machine-outliner-reruns", cl::init(0), cl::Hidden, 116 cl::desc( 117 "Number of times to rerun the outliner after the initial outline")); 118 119 static cl::opt<unsigned> OutlinerBenefitThreshold( 120 "outliner-benefit-threshold", cl::init(1), cl::Hidden, 121 cl::desc( 122 "The minimum size in bytes before an outlining candidate is accepted")); 123 124 namespace { 125 126 /// Maps \p MachineInstrs to unsigned integers and stores the mappings. 127 struct InstructionMapper { 128 129 /// The next available integer to assign to a \p MachineInstr that 130 /// cannot be outlined. 131 /// 132 /// Set to -3 for compatability with \p DenseMapInfo<unsigned>. 133 unsigned IllegalInstrNumber = -3; 134 135 /// The next available integer to assign to a \p MachineInstr that can 136 /// be outlined. 137 unsigned LegalInstrNumber = 0; 138 139 /// Correspondence from \p MachineInstrs to unsigned integers. 140 DenseMap<MachineInstr *, unsigned, MachineInstrExpressionTrait> 141 InstructionIntegerMap; 142 143 /// Correspondence between \p MachineBasicBlocks and target-defined flags. 144 DenseMap<MachineBasicBlock *, unsigned> MBBFlagsMap; 145 146 /// The vector of unsigned integers that the module is mapped to. 147 SmallVector<unsigned> UnsignedVec; 148 149 /// Stores the location of the instruction associated with the integer 150 /// at index i in \p UnsignedVec for each index i. 151 SmallVector<MachineBasicBlock::iterator> InstrList; 152 153 // Set if we added an illegal number in the previous step. 154 // Since each illegal number is unique, we only need one of them between 155 // each range of legal numbers. This lets us make sure we don't add more 156 // than one illegal number per range. 157 bool AddedIllegalLastTime = false; 158 159 /// Maps \p *It to a legal integer. 160 /// 161 /// Updates \p CanOutlineWithPrevInstr, \p HaveLegalRange, \p InstrListForMBB, 162 /// \p UnsignedVecForMBB, \p InstructionIntegerMap, and \p LegalInstrNumber. 163 /// 164 /// \returns The integer that \p *It was mapped to. 165 unsigned mapToLegalUnsigned( 166 MachineBasicBlock::iterator &It, bool &CanOutlineWithPrevInstr, 167 bool &HaveLegalRange, unsigned &NumLegalInBlock, 168 SmallVector<unsigned> &UnsignedVecForMBB, 169 SmallVector<MachineBasicBlock::iterator> &InstrListForMBB) { 170 // We added something legal, so we should unset the AddedLegalLastTime 171 // flag. 172 AddedIllegalLastTime = false; 173 174 // If we have at least two adjacent legal instructions (which may have 175 // invisible instructions in between), remember that. 176 if (CanOutlineWithPrevInstr) 177 HaveLegalRange = true; 178 CanOutlineWithPrevInstr = true; 179 180 // Keep track of the number of legal instructions we insert. 181 NumLegalInBlock++; 182 183 // Get the integer for this instruction or give it the current 184 // LegalInstrNumber. 185 InstrListForMBB.push_back(It); 186 MachineInstr &MI = *It; 187 bool WasInserted; 188 DenseMap<MachineInstr *, unsigned, MachineInstrExpressionTrait>::iterator 189 ResultIt; 190 std::tie(ResultIt, WasInserted) = 191 InstructionIntegerMap.insert(std::make_pair(&MI, LegalInstrNumber)); 192 unsigned MINumber = ResultIt->second; 193 194 // There was an insertion. 195 if (WasInserted) 196 LegalInstrNumber++; 197 198 UnsignedVecForMBB.push_back(MINumber); 199 200 // Make sure we don't overflow or use any integers reserved by the DenseMap. 201 if (LegalInstrNumber >= IllegalInstrNumber) 202 report_fatal_error("Instruction mapping overflow!"); 203 204 assert(LegalInstrNumber != DenseMapInfo<unsigned>::getEmptyKey() && 205 "Tried to assign DenseMap tombstone or empty key to instruction."); 206 assert(LegalInstrNumber != DenseMapInfo<unsigned>::getTombstoneKey() && 207 "Tried to assign DenseMap tombstone or empty key to instruction."); 208 209 // Statistics. 210 ++NumLegalInUnsignedVec; 211 return MINumber; 212 } 213 214 /// Maps \p *It to an illegal integer. 215 /// 216 /// Updates \p InstrListForMBB, \p UnsignedVecForMBB, and \p 217 /// IllegalInstrNumber. 218 /// 219 /// \returns The integer that \p *It was mapped to. 220 unsigned mapToIllegalUnsigned( 221 MachineBasicBlock::iterator &It, bool &CanOutlineWithPrevInstr, 222 SmallVector<unsigned> &UnsignedVecForMBB, 223 SmallVector<MachineBasicBlock::iterator> &InstrListForMBB) { 224 // Can't outline an illegal instruction. Set the flag. 225 CanOutlineWithPrevInstr = false; 226 227 // Only add one illegal number per range of legal numbers. 228 if (AddedIllegalLastTime) 229 return IllegalInstrNumber; 230 231 // Remember that we added an illegal number last time. 232 AddedIllegalLastTime = true; 233 unsigned MINumber = IllegalInstrNumber; 234 235 InstrListForMBB.push_back(It); 236 UnsignedVecForMBB.push_back(IllegalInstrNumber); 237 IllegalInstrNumber--; 238 // Statistics. 239 ++NumIllegalInUnsignedVec; 240 241 assert(LegalInstrNumber < IllegalInstrNumber && 242 "Instruction mapping overflow!"); 243 244 assert(IllegalInstrNumber != DenseMapInfo<unsigned>::getEmptyKey() && 245 "IllegalInstrNumber cannot be DenseMap tombstone or empty key!"); 246 247 assert(IllegalInstrNumber != DenseMapInfo<unsigned>::getTombstoneKey() && 248 "IllegalInstrNumber cannot be DenseMap tombstone or empty key!"); 249 250 return MINumber; 251 } 252 253 /// Transforms a \p MachineBasicBlock into a \p vector of \p unsigneds 254 /// and appends it to \p UnsignedVec and \p InstrList. 255 /// 256 /// Two instructions are assigned the same integer if they are identical. 257 /// If an instruction is deemed unsafe to outline, then it will be assigned an 258 /// unique integer. The resulting mapping is placed into a suffix tree and 259 /// queried for candidates. 260 /// 261 /// \param MBB The \p MachineBasicBlock to be translated into integers. 262 /// \param TII \p TargetInstrInfo for the function. 263 void convertToUnsignedVec(MachineBasicBlock &MBB, 264 const TargetInstrInfo &TII) { 265 LLVM_DEBUG(dbgs() << "*** Converting MBB '" << MBB.getName() 266 << "' to unsigned vector ***\n"); 267 unsigned Flags = 0; 268 269 // Don't even map in this case. 270 if (!TII.isMBBSafeToOutlineFrom(MBB, Flags)) 271 return; 272 273 auto OutlinableRanges = TII.getOutlinableRanges(MBB, Flags); 274 LLVM_DEBUG(dbgs() << MBB.getName() << ": " << OutlinableRanges.size() 275 << " outlinable range(s)\n"); 276 if (OutlinableRanges.empty()) 277 return; 278 279 // Store info for the MBB for later outlining. 280 MBBFlagsMap[&MBB] = Flags; 281 282 MachineBasicBlock::iterator It = MBB.begin(); 283 284 // The number of instructions in this block that will be considered for 285 // outlining. 286 unsigned NumLegalInBlock = 0; 287 288 // True if we have at least two legal instructions which aren't separated 289 // by an illegal instruction. 290 bool HaveLegalRange = false; 291 292 // True if we can perform outlining given the last mapped (non-invisible) 293 // instruction. This lets us know if we have a legal range. 294 bool CanOutlineWithPrevInstr = false; 295 296 // FIXME: Should this all just be handled in the target, rather than using 297 // repeated calls to getOutliningType? 298 SmallVector<unsigned> UnsignedVecForMBB; 299 SmallVector<MachineBasicBlock::iterator> InstrListForMBB; 300 301 LLVM_DEBUG(dbgs() << "*** Mapping outlinable ranges ***\n"); 302 for (auto &OutlinableRange : OutlinableRanges) { 303 auto OutlinableRangeBegin = OutlinableRange.first; 304 auto OutlinableRangeEnd = OutlinableRange.second; 305 #ifndef NDEBUG 306 LLVM_DEBUG( 307 dbgs() << "Mapping " 308 << std::distance(OutlinableRangeBegin, OutlinableRangeEnd) 309 << " instruction range\n"); 310 // Everything outside of an outlinable range is illegal. 311 unsigned NumSkippedInRange = 0; 312 #endif 313 for (; It != OutlinableRangeBegin; ++It) { 314 #ifndef NDEBUG 315 ++NumSkippedInRange; 316 #endif 317 mapToIllegalUnsigned(It, CanOutlineWithPrevInstr, UnsignedVecForMBB, 318 InstrListForMBB); 319 } 320 #ifndef NDEBUG 321 LLVM_DEBUG(dbgs() << "Skipped " << NumSkippedInRange 322 << " instructions outside outlinable range\n"); 323 #endif 324 assert(It != MBB.end() && "Should still have instructions?"); 325 // `It` is now positioned at the beginning of a range of instructions 326 // which may be outlinable. Check if each instruction is known to be safe. 327 for (; It != OutlinableRangeEnd; ++It) { 328 // Keep track of where this instruction is in the module. 329 switch (TII.getOutliningType(It, Flags)) { 330 case InstrType::Illegal: 331 mapToIllegalUnsigned(It, CanOutlineWithPrevInstr, UnsignedVecForMBB, 332 InstrListForMBB); 333 break; 334 335 case InstrType::Legal: 336 mapToLegalUnsigned(It, CanOutlineWithPrevInstr, HaveLegalRange, 337 NumLegalInBlock, UnsignedVecForMBB, 338 InstrListForMBB); 339 break; 340 341 case InstrType::LegalTerminator: 342 mapToLegalUnsigned(It, CanOutlineWithPrevInstr, HaveLegalRange, 343 NumLegalInBlock, UnsignedVecForMBB, 344 InstrListForMBB); 345 // The instruction also acts as a terminator, so we have to record 346 // that in the string. 347 mapToIllegalUnsigned(It, CanOutlineWithPrevInstr, UnsignedVecForMBB, 348 InstrListForMBB); 349 break; 350 351 case InstrType::Invisible: 352 // Normally this is set by mapTo(Blah)Unsigned, but we just want to 353 // skip this instruction. So, unset the flag here. 354 ++NumInvisible; 355 AddedIllegalLastTime = false; 356 break; 357 } 358 } 359 } 360 361 LLVM_DEBUG(dbgs() << "HaveLegalRange = " << HaveLegalRange << "\n"); 362 363 // Are there enough legal instructions in the block for outlining to be 364 // possible? 365 if (HaveLegalRange) { 366 // After we're done every insertion, uniquely terminate this part of the 367 // "string". This makes sure we won't match across basic block or function 368 // boundaries since the "end" is encoded uniquely and thus appears in no 369 // repeated substring. 370 mapToIllegalUnsigned(It, CanOutlineWithPrevInstr, UnsignedVecForMBB, 371 InstrListForMBB); 372 ++NumSentinels; 373 append_range(InstrList, InstrListForMBB); 374 append_range(UnsignedVec, UnsignedVecForMBB); 375 } 376 } 377 378 InstructionMapper() { 379 // Make sure that the implementation of DenseMapInfo<unsigned> hasn't 380 // changed. 381 assert(DenseMapInfo<unsigned>::getEmptyKey() == (unsigned)-1 && 382 "DenseMapInfo<unsigned>'s empty key isn't -1!"); 383 assert(DenseMapInfo<unsigned>::getTombstoneKey() == (unsigned)-2 && 384 "DenseMapInfo<unsigned>'s tombstone key isn't -2!"); 385 } 386 }; 387 388 /// An interprocedural pass which finds repeated sequences of 389 /// instructions and replaces them with calls to functions. 390 /// 391 /// Each instruction is mapped to an unsigned integer and placed in a string. 392 /// The resulting mapping is then placed in a \p SuffixTree. The \p SuffixTree 393 /// is then repeatedly queried for repeated sequences of instructions. Each 394 /// non-overlapping repeated sequence is then placed in its own 395 /// \p MachineFunction and each instance is then replaced with a call to that 396 /// function. 397 struct MachineOutliner : public ModulePass { 398 399 static char ID; 400 401 /// Set to true if the outliner should consider functions with 402 /// linkonceodr linkage. 403 bool OutlineFromLinkOnceODRs = false; 404 405 /// The current repeat number of machine outlining. 406 unsigned OutlineRepeatedNum = 0; 407 408 /// Set to true if the outliner should run on all functions in the module 409 /// considered safe for outlining. 410 /// Set to true by default for compatibility with llc's -run-pass option. 411 /// Set when the pass is constructed in TargetPassConfig. 412 bool RunOnAllFunctions = true; 413 414 StringRef getPassName() const override { return "Machine Outliner"; } 415 416 void getAnalysisUsage(AnalysisUsage &AU) const override { 417 AU.addRequired<MachineModuleInfoWrapperPass>(); 418 AU.addPreserved<MachineModuleInfoWrapperPass>(); 419 AU.setPreservesAll(); 420 ModulePass::getAnalysisUsage(AU); 421 } 422 423 MachineOutliner() : ModulePass(ID) { 424 initializeMachineOutlinerPass(*PassRegistry::getPassRegistry()); 425 } 426 427 /// Remark output explaining that not outlining a set of candidates would be 428 /// better than outlining that set. 429 void emitNotOutliningCheaperRemark( 430 unsigned StringLen, std::vector<Candidate> &CandidatesForRepeatedSeq, 431 OutlinedFunction &OF); 432 433 /// Remark output explaining that a function was outlined. 434 void emitOutlinedFunctionRemark(OutlinedFunction &OF); 435 436 /// Find all repeated substrings that satisfy the outlining cost model by 437 /// constructing a suffix tree. 438 /// 439 /// If a substring appears at least twice, then it must be represented by 440 /// an internal node which appears in at least two suffixes. Each suffix 441 /// is represented by a leaf node. To do this, we visit each internal node 442 /// in the tree, using the leaf children of each internal node. If an 443 /// internal node represents a beneficial substring, then we use each of 444 /// its leaf children to find the locations of its substring. 445 /// 446 /// \param Mapper Contains outlining mapping information. 447 /// \param[out] FunctionList Filled with a list of \p OutlinedFunctions 448 /// each type of candidate. 449 void findCandidates(InstructionMapper &Mapper, 450 std::vector<OutlinedFunction> &FunctionList); 451 452 /// Replace the sequences of instructions represented by \p OutlinedFunctions 453 /// with calls to functions. 454 /// 455 /// \param M The module we are outlining from. 456 /// \param FunctionList A list of functions to be inserted into the module. 457 /// \param Mapper Contains the instruction mappings for the module. 458 bool outline(Module &M, std::vector<OutlinedFunction> &FunctionList, 459 InstructionMapper &Mapper, unsigned &OutlinedFunctionNum); 460 461 /// Creates a function for \p OF and inserts it into the module. 462 MachineFunction *createOutlinedFunction(Module &M, OutlinedFunction &OF, 463 InstructionMapper &Mapper, 464 unsigned Name); 465 466 /// Calls 'doOutline()' 1 + OutlinerReruns times. 467 bool runOnModule(Module &M) override; 468 469 /// Construct a suffix tree on the instructions in \p M and outline repeated 470 /// strings from that tree. 471 bool doOutline(Module &M, unsigned &OutlinedFunctionNum); 472 473 /// Return a DISubprogram for OF if one exists, and null otherwise. Helper 474 /// function for remark emission. 475 DISubprogram *getSubprogramOrNull(const OutlinedFunction &OF) { 476 for (const Candidate &C : OF.Candidates) 477 if (MachineFunction *MF = C.getMF()) 478 if (DISubprogram *SP = MF->getFunction().getSubprogram()) 479 return SP; 480 return nullptr; 481 } 482 483 /// Populate and \p InstructionMapper with instruction-to-integer mappings. 484 /// These are used to construct a suffix tree. 485 void populateMapper(InstructionMapper &Mapper, Module &M, 486 MachineModuleInfo &MMI); 487 488 /// Initialize information necessary to output a size remark. 489 /// FIXME: This should be handled by the pass manager, not the outliner. 490 /// FIXME: This is nearly identical to the initSizeRemarkInfo in the legacy 491 /// pass manager. 492 void initSizeRemarkInfo(const Module &M, const MachineModuleInfo &MMI, 493 StringMap<unsigned> &FunctionToInstrCount); 494 495 /// Emit the remark. 496 // FIXME: This should be handled by the pass manager, not the outliner. 497 void 498 emitInstrCountChangedRemark(const Module &M, const MachineModuleInfo &MMI, 499 const StringMap<unsigned> &FunctionToInstrCount); 500 }; 501 } // Anonymous namespace. 502 503 char MachineOutliner::ID = 0; 504 505 namespace llvm { 506 ModulePass *createMachineOutlinerPass(bool RunOnAllFunctions) { 507 MachineOutliner *OL = new MachineOutliner(); 508 OL->RunOnAllFunctions = RunOnAllFunctions; 509 return OL; 510 } 511 512 } // namespace llvm 513 514 INITIALIZE_PASS(MachineOutliner, DEBUG_TYPE, "Machine Function Outliner", false, 515 false) 516 517 void MachineOutliner::emitNotOutliningCheaperRemark( 518 unsigned StringLen, std::vector<Candidate> &CandidatesForRepeatedSeq, 519 OutlinedFunction &OF) { 520 // FIXME: Right now, we arbitrarily choose some Candidate from the 521 // OutlinedFunction. This isn't necessarily fixed, nor does it have to be. 522 // We should probably sort these by function name or something to make sure 523 // the remarks are stable. 524 Candidate &C = CandidatesForRepeatedSeq.front(); 525 MachineOptimizationRemarkEmitter MORE(*(C.getMF()), nullptr); 526 MORE.emit([&]() { 527 MachineOptimizationRemarkMissed R(DEBUG_TYPE, "NotOutliningCheaper", 528 C.front().getDebugLoc(), C.getMBB()); 529 R << "Did not outline " << NV("Length", StringLen) << " instructions" 530 << " from " << NV("NumOccurrences", CandidatesForRepeatedSeq.size()) 531 << " locations." 532 << " Bytes from outlining all occurrences (" 533 << NV("OutliningCost", OF.getOutliningCost()) << ")" 534 << " >= Unoutlined instruction bytes (" 535 << NV("NotOutliningCost", OF.getNotOutlinedCost()) << ")" 536 << " (Also found at: "; 537 538 // Tell the user the other places the candidate was found. 539 for (unsigned i = 1, e = CandidatesForRepeatedSeq.size(); i < e; i++) { 540 R << NV((Twine("OtherStartLoc") + Twine(i)).str(), 541 CandidatesForRepeatedSeq[i].front().getDebugLoc()); 542 if (i != e - 1) 543 R << ", "; 544 } 545 546 R << ")"; 547 return R; 548 }); 549 } 550 551 void MachineOutliner::emitOutlinedFunctionRemark(OutlinedFunction &OF) { 552 MachineBasicBlock *MBB = &*OF.MF->begin(); 553 MachineOptimizationRemarkEmitter MORE(*OF.MF, nullptr); 554 MachineOptimizationRemark R(DEBUG_TYPE, "OutlinedFunction", 555 MBB->findDebugLoc(MBB->begin()), MBB); 556 R << "Saved " << NV("OutliningBenefit", OF.getBenefit()) << " bytes by " 557 << "outlining " << NV("Length", OF.getNumInstrs()) << " instructions " 558 << "from " << NV("NumOccurrences", OF.getOccurrenceCount()) 559 << " locations. " 560 << "(Found at: "; 561 562 // Tell the user the other places the candidate was found. 563 for (size_t i = 0, e = OF.Candidates.size(); i < e; i++) { 564 565 R << NV((Twine("StartLoc") + Twine(i)).str(), 566 OF.Candidates[i].front().getDebugLoc()); 567 if (i != e - 1) 568 R << ", "; 569 } 570 571 R << ")"; 572 573 MORE.emit(R); 574 } 575 576 void MachineOutliner::findCandidates( 577 InstructionMapper &Mapper, std::vector<OutlinedFunction> &FunctionList) { 578 FunctionList.clear(); 579 SuffixTree ST(Mapper.UnsignedVec); 580 581 // First, find all of the repeated substrings in the tree of minimum length 582 // 2. 583 std::vector<Candidate> CandidatesForRepeatedSeq; 584 LLVM_DEBUG(dbgs() << "*** Discarding overlapping candidates *** \n"); 585 LLVM_DEBUG( 586 dbgs() << "Searching for overlaps in all repeated sequences...\n"); 587 for (const SuffixTree::RepeatedSubstring &RS : ST) { 588 CandidatesForRepeatedSeq.clear(); 589 unsigned StringLen = RS.Length; 590 LLVM_DEBUG(dbgs() << " Sequence length: " << StringLen << "\n"); 591 // Debug code to keep track of how many candidates we removed. 592 #ifndef NDEBUG 593 unsigned NumDiscarded = 0; 594 unsigned NumKept = 0; 595 #endif 596 for (const unsigned &StartIdx : RS.StartIndices) { 597 // Trick: Discard some candidates that would be incompatible with the 598 // ones we've already found for this sequence. This will save us some 599 // work in candidate selection. 600 // 601 // If two candidates overlap, then we can't outline them both. This 602 // happens when we have candidates that look like, say 603 // 604 // AA (where each "A" is an instruction). 605 // 606 // We might have some portion of the module that looks like this: 607 // AAAAAA (6 A's) 608 // 609 // In this case, there are 5 different copies of "AA" in this range, but 610 // at most 3 can be outlined. If only outlining 3 of these is going to 611 // be unbeneficial, then we ought to not bother. 612 // 613 // Note that two things DON'T overlap when they look like this: 614 // start1...end1 .... start2...end2 615 // That is, one must either 616 // * End before the other starts 617 // * Start after the other ends 618 unsigned EndIdx = StartIdx + StringLen - 1; 619 auto FirstOverlap = find_if( 620 CandidatesForRepeatedSeq, [StartIdx, EndIdx](const Candidate &C) { 621 return EndIdx >= C.getStartIdx() && StartIdx <= C.getEndIdx(); 622 }); 623 if (FirstOverlap != CandidatesForRepeatedSeq.end()) { 624 #ifndef NDEBUG 625 ++NumDiscarded; 626 LLVM_DEBUG(dbgs() << " .. DISCARD candidate @ [" << StartIdx 627 << ", " << EndIdx << "]; overlaps with candidate @ [" 628 << FirstOverlap->getStartIdx() << ", " 629 << FirstOverlap->getEndIdx() << "]\n"); 630 #endif 631 continue; 632 } 633 // It doesn't overlap with anything, so we can outline it. 634 // Each sequence is over [StartIt, EndIt]. 635 // Save the candidate and its location. 636 #ifndef NDEBUG 637 ++NumKept; 638 #endif 639 MachineBasicBlock::iterator StartIt = Mapper.InstrList[StartIdx]; 640 MachineBasicBlock::iterator EndIt = Mapper.InstrList[EndIdx]; 641 MachineBasicBlock *MBB = StartIt->getParent(); 642 CandidatesForRepeatedSeq.emplace_back(StartIdx, StringLen, StartIt, EndIt, 643 MBB, FunctionList.size(), 644 Mapper.MBBFlagsMap[MBB]); 645 } 646 #ifndef NDEBUG 647 LLVM_DEBUG(dbgs() << " Candidates discarded: " << NumDiscarded 648 << "\n"); 649 LLVM_DEBUG(dbgs() << " Candidates kept: " << NumKept << "\n\n"); 650 #endif 651 652 // We've found something we might want to outline. 653 // Create an OutlinedFunction to store it and check if it'd be beneficial 654 // to outline. 655 if (CandidatesForRepeatedSeq.size() < 2) 656 continue; 657 658 // Arbitrarily choose a TII from the first candidate. 659 // FIXME: Should getOutliningCandidateInfo move to TargetMachine? 660 const TargetInstrInfo *TII = 661 CandidatesForRepeatedSeq[0].getMF()->getSubtarget().getInstrInfo(); 662 663 std::optional<OutlinedFunction> OF = 664 TII->getOutliningCandidateInfo(CandidatesForRepeatedSeq); 665 666 // If we deleted too many candidates, then there's nothing worth outlining. 667 // FIXME: This should take target-specified instruction sizes into account. 668 if (!OF || OF->Candidates.size() < 2) 669 continue; 670 671 // Is it better to outline this candidate than not? 672 if (OF->getBenefit() < OutlinerBenefitThreshold) { 673 emitNotOutliningCheaperRemark(StringLen, CandidatesForRepeatedSeq, *OF); 674 continue; 675 } 676 677 FunctionList.push_back(*OF); 678 } 679 } 680 681 MachineFunction *MachineOutliner::createOutlinedFunction( 682 Module &M, OutlinedFunction &OF, InstructionMapper &Mapper, unsigned Name) { 683 684 // Create the function name. This should be unique. 685 // FIXME: We should have a better naming scheme. This should be stable, 686 // regardless of changes to the outliner's cost model/traversal order. 687 std::string FunctionName = "OUTLINED_FUNCTION_"; 688 if (OutlineRepeatedNum > 0) 689 FunctionName += std::to_string(OutlineRepeatedNum + 1) + "_"; 690 FunctionName += std::to_string(Name); 691 LLVM_DEBUG(dbgs() << "NEW FUNCTION: " << FunctionName << "\n"); 692 693 // Create the function using an IR-level function. 694 LLVMContext &C = M.getContext(); 695 Function *F = Function::Create(FunctionType::get(Type::getVoidTy(C), false), 696 Function::ExternalLinkage, FunctionName, M); 697 698 // NOTE: If this is linkonceodr, then we can take advantage of linker deduping 699 // which gives us better results when we outline from linkonceodr functions. 700 F->setLinkage(GlobalValue::InternalLinkage); 701 F->setUnnamedAddr(GlobalValue::UnnamedAddr::Global); 702 703 // Set optsize/minsize, so we don't insert padding between outlined 704 // functions. 705 F->addFnAttr(Attribute::OptimizeForSize); 706 F->addFnAttr(Attribute::MinSize); 707 708 Candidate &FirstCand = OF.Candidates.front(); 709 const TargetInstrInfo &TII = 710 *FirstCand.getMF()->getSubtarget().getInstrInfo(); 711 712 TII.mergeOutliningCandidateAttributes(*F, OF.Candidates); 713 714 // Set uwtable, so we generate eh_frame. 715 UWTableKind UW = std::accumulate( 716 OF.Candidates.cbegin(), OF.Candidates.cend(), UWTableKind::None, 717 [](UWTableKind K, const outliner::Candidate &C) { 718 return std::max(K, C.getMF()->getFunction().getUWTableKind()); 719 }); 720 F->setUWTableKind(UW); 721 722 BasicBlock *EntryBB = BasicBlock::Create(C, "entry", F); 723 IRBuilder<> Builder(EntryBB); 724 Builder.CreateRetVoid(); 725 726 MachineModuleInfo &MMI = getAnalysis<MachineModuleInfoWrapperPass>().getMMI(); 727 MachineFunction &MF = MMI.getOrCreateMachineFunction(*F); 728 MF.setIsOutlined(true); 729 MachineBasicBlock &MBB = *MF.CreateMachineBasicBlock(); 730 731 // Insert the new function into the module. 732 MF.insert(MF.begin(), &MBB); 733 734 MachineFunction *OriginalMF = FirstCand.front().getMF(); 735 const std::vector<MCCFIInstruction> &Instrs = 736 OriginalMF->getFrameInstructions(); 737 for (auto &MI : FirstCand) { 738 if (MI.isDebugInstr()) 739 continue; 740 741 // Don't keep debug information for outlined instructions. 742 auto DL = DebugLoc(); 743 if (MI.isCFIInstruction()) { 744 unsigned CFIIndex = MI.getOperand(0).getCFIIndex(); 745 MCCFIInstruction CFI = Instrs[CFIIndex]; 746 BuildMI(MBB, MBB.end(), DL, TII.get(TargetOpcode::CFI_INSTRUCTION)) 747 .addCFIIndex(MF.addFrameInst(CFI)); 748 } else { 749 MachineInstr *NewMI = MF.CloneMachineInstr(&MI); 750 NewMI->dropMemRefs(MF); 751 NewMI->setDebugLoc(DL); 752 MBB.insert(MBB.end(), NewMI); 753 } 754 } 755 756 // Set normal properties for a late MachineFunction. 757 MF.getProperties().reset(MachineFunctionProperties::Property::IsSSA); 758 MF.getProperties().set(MachineFunctionProperties::Property::NoPHIs); 759 MF.getProperties().set(MachineFunctionProperties::Property::NoVRegs); 760 MF.getProperties().set(MachineFunctionProperties::Property::TracksLiveness); 761 MF.getRegInfo().freezeReservedRegs(); 762 763 // Compute live-in set for outlined fn 764 const MachineRegisterInfo &MRI = MF.getRegInfo(); 765 const TargetRegisterInfo &TRI = *MRI.getTargetRegisterInfo(); 766 LivePhysRegs LiveIns(TRI); 767 for (auto &Cand : OF.Candidates) { 768 // Figure out live-ins at the first instruction. 769 MachineBasicBlock &OutlineBB = *Cand.front().getParent(); 770 LivePhysRegs CandLiveIns(TRI); 771 CandLiveIns.addLiveOuts(OutlineBB); 772 for (const MachineInstr &MI : 773 reverse(make_range(Cand.begin(), OutlineBB.end()))) 774 CandLiveIns.stepBackward(MI); 775 776 // The live-in set for the outlined function is the union of the live-ins 777 // from all the outlining points. 778 for (MCPhysReg Reg : CandLiveIns) 779 LiveIns.addReg(Reg); 780 } 781 addLiveIns(MBB, LiveIns); 782 783 TII.buildOutlinedFrame(MBB, MF, OF); 784 785 // If there's a DISubprogram associated with this outlined function, then 786 // emit debug info for the outlined function. 787 if (DISubprogram *SP = getSubprogramOrNull(OF)) { 788 // We have a DISubprogram. Get its DICompileUnit. 789 DICompileUnit *CU = SP->getUnit(); 790 DIBuilder DB(M, true, CU); 791 DIFile *Unit = SP->getFile(); 792 Mangler Mg; 793 // Get the mangled name of the function for the linkage name. 794 std::string Dummy; 795 raw_string_ostream MangledNameStream(Dummy); 796 Mg.getNameWithPrefix(MangledNameStream, F, false); 797 798 DISubprogram *OutlinedSP = DB.createFunction( 799 Unit /* Context */, F->getName(), StringRef(MangledNameStream.str()), 800 Unit /* File */, 801 0 /* Line 0 is reserved for compiler-generated code. */, 802 DB.createSubroutineType( 803 DB.getOrCreateTypeArray(std::nullopt)), /* void type */ 804 0, /* Line 0 is reserved for compiler-generated code. */ 805 DINode::DIFlags::FlagArtificial /* Compiler-generated code. */, 806 /* Outlined code is optimized code by definition. */ 807 DISubprogram::SPFlagDefinition | DISubprogram::SPFlagOptimized); 808 809 // Don't add any new variables to the subprogram. 810 DB.finalizeSubprogram(OutlinedSP); 811 812 // Attach subprogram to the function. 813 F->setSubprogram(OutlinedSP); 814 // We're done with the DIBuilder. 815 DB.finalize(); 816 } 817 818 return &MF; 819 } 820 821 bool MachineOutliner::outline(Module &M, 822 std::vector<OutlinedFunction> &FunctionList, 823 InstructionMapper &Mapper, 824 unsigned &OutlinedFunctionNum) { 825 LLVM_DEBUG(dbgs() << "*** Outlining ***\n"); 826 LLVM_DEBUG(dbgs() << "NUMBER OF POTENTIAL FUNCTIONS: " << FunctionList.size() 827 << "\n"); 828 bool OutlinedSomething = false; 829 830 // Sort by benefit. The most beneficial functions should be outlined first. 831 stable_sort(FunctionList, 832 [](const OutlinedFunction &LHS, const OutlinedFunction &RHS) { 833 return LHS.getBenefit() > RHS.getBenefit(); 834 }); 835 836 // Walk over each function, outlining them as we go along. Functions are 837 // outlined greedily, based off the sort above. 838 auto *UnsignedVecBegin = Mapper.UnsignedVec.begin(); 839 LLVM_DEBUG(dbgs() << "WALKING FUNCTION LIST\n"); 840 for (OutlinedFunction &OF : FunctionList) { 841 #ifndef NDEBUG 842 auto NumCandidatesBefore = OF.Candidates.size(); 843 #endif 844 // If we outlined something that overlapped with a candidate in a previous 845 // step, then we can't outline from it. 846 erase_if(OF.Candidates, [&UnsignedVecBegin](Candidate &C) { 847 return std::any_of(UnsignedVecBegin + C.getStartIdx(), 848 UnsignedVecBegin + C.getEndIdx() + 1, [](unsigned I) { 849 return I == static_cast<unsigned>(-1); 850 }); 851 }); 852 853 #ifndef NDEBUG 854 auto NumCandidatesAfter = OF.Candidates.size(); 855 LLVM_DEBUG(dbgs() << "PRUNED: " << NumCandidatesBefore - NumCandidatesAfter 856 << "/" << NumCandidatesBefore << " candidates\n"); 857 #endif 858 859 // If we made it unbeneficial to outline this function, skip it. 860 if (OF.getBenefit() < OutlinerBenefitThreshold) { 861 LLVM_DEBUG(dbgs() << "SKIP: Expected benefit (" << OF.getBenefit() 862 << " B) < threshold (" << OutlinerBenefitThreshold 863 << " B)\n"); 864 continue; 865 } 866 867 LLVM_DEBUG(dbgs() << "OUTLINE: Expected benefit (" << OF.getBenefit() 868 << " B) > threshold (" << OutlinerBenefitThreshold 869 << " B)\n"); 870 871 // It's beneficial. Create the function and outline its sequence's 872 // occurrences. 873 OF.MF = createOutlinedFunction(M, OF, Mapper, OutlinedFunctionNum); 874 emitOutlinedFunctionRemark(OF); 875 FunctionsCreated++; 876 OutlinedFunctionNum++; // Created a function, move to the next name. 877 MachineFunction *MF = OF.MF; 878 const TargetSubtargetInfo &STI = MF->getSubtarget(); 879 const TargetInstrInfo &TII = *STI.getInstrInfo(); 880 881 // Replace occurrences of the sequence with calls to the new function. 882 LLVM_DEBUG(dbgs() << "CREATE OUTLINED CALLS\n"); 883 for (Candidate &C : OF.Candidates) { 884 MachineBasicBlock &MBB = *C.getMBB(); 885 MachineBasicBlock::iterator StartIt = C.begin(); 886 MachineBasicBlock::iterator EndIt = std::prev(C.end()); 887 888 // Insert the call. 889 auto CallInst = TII.insertOutlinedCall(M, MBB, StartIt, *MF, C); 890 // Insert the call. 891 #ifndef NDEBUG 892 auto MBBBeingOutlinedFromName = 893 MBB.getName().empty() ? "<unknown>" : MBB.getName().str(); 894 auto MFBeingOutlinedFromName = MBB.getParent()->getName().empty() 895 ? "<unknown>" 896 : MBB.getParent()->getName().str(); 897 LLVM_DEBUG(dbgs() << " CALL: " << MF->getName() << " in " 898 << MFBeingOutlinedFromName << ":" 899 << MBBBeingOutlinedFromName << "\n"); 900 LLVM_DEBUG(dbgs() << " .. " << *CallInst); 901 #endif 902 903 // If the caller tracks liveness, then we need to make sure that 904 // anything we outline doesn't break liveness assumptions. The outlined 905 // functions themselves currently don't track liveness, but we should 906 // make sure that the ranges we yank things out of aren't wrong. 907 if (MBB.getParent()->getProperties().hasProperty( 908 MachineFunctionProperties::Property::TracksLiveness)) { 909 // The following code is to add implicit def operands to the call 910 // instruction. It also updates call site information for moved 911 // code. 912 SmallSet<Register, 2> UseRegs, DefRegs; 913 // Copy over the defs in the outlined range. 914 // First inst in outlined range <-- Anything that's defined in this 915 // ... .. range has to be added as an 916 // implicit Last inst in outlined range <-- def to the call 917 // instruction. Also remove call site information for outlined block 918 // of code. The exposed uses need to be copied in the outlined range. 919 for (MachineBasicBlock::reverse_iterator 920 Iter = EndIt.getReverse(), 921 Last = std::next(CallInst.getReverse()); 922 Iter != Last; Iter++) { 923 MachineInstr *MI = &*Iter; 924 SmallSet<Register, 2> InstrUseRegs; 925 for (MachineOperand &MOP : MI->operands()) { 926 // Skip over anything that isn't a register. 927 if (!MOP.isReg()) 928 continue; 929 930 if (MOP.isDef()) { 931 // Introduce DefRegs set to skip the redundant register. 932 DefRegs.insert(MOP.getReg()); 933 if (UseRegs.count(MOP.getReg()) && 934 !InstrUseRegs.count(MOP.getReg())) 935 // Since the regiester is modeled as defined, 936 // it is not necessary to be put in use register set. 937 UseRegs.erase(MOP.getReg()); 938 } else if (!MOP.isUndef()) { 939 // Any register which is not undefined should 940 // be put in the use register set. 941 UseRegs.insert(MOP.getReg()); 942 InstrUseRegs.insert(MOP.getReg()); 943 } 944 } 945 if (MI->isCandidateForCallSiteEntry()) 946 MI->getMF()->eraseCallSiteInfo(MI); 947 } 948 949 for (const Register &I : DefRegs) 950 // If it's a def, add it to the call instruction. 951 CallInst->addOperand( 952 MachineOperand::CreateReg(I, true, /* isDef = true */ 953 true /* isImp = true */)); 954 955 for (const Register &I : UseRegs) 956 // If it's a exposed use, add it to the call instruction. 957 CallInst->addOperand( 958 MachineOperand::CreateReg(I, false, /* isDef = false */ 959 true /* isImp = true */)); 960 } 961 962 // Erase from the point after where the call was inserted up to, and 963 // including, the final instruction in the sequence. 964 // Erase needs one past the end, so we need std::next there too. 965 MBB.erase(std::next(StartIt), std::next(EndIt)); 966 967 // Keep track of what we removed by marking them all as -1. 968 for (unsigned &I : make_range(UnsignedVecBegin + C.getStartIdx(), 969 UnsignedVecBegin + C.getEndIdx() + 1)) 970 I = static_cast<unsigned>(-1); 971 OutlinedSomething = true; 972 973 // Statistics. 974 NumOutlined++; 975 } 976 } 977 978 LLVM_DEBUG(dbgs() << "OutlinedSomething = " << OutlinedSomething << "\n";); 979 return OutlinedSomething; 980 } 981 982 void MachineOutliner::populateMapper(InstructionMapper &Mapper, Module &M, 983 MachineModuleInfo &MMI) { 984 // Build instruction mappings for each function in the module. Start by 985 // iterating over each Function in M. 986 LLVM_DEBUG(dbgs() << "*** Populating mapper ***\n"); 987 for (Function &F : M) { 988 LLVM_DEBUG(dbgs() << "MAPPING FUNCTION: " << F.getName() << "\n"); 989 990 if (F.hasFnAttribute("nooutline")) { 991 LLVM_DEBUG(dbgs() << "SKIP: Function has nooutline attribute\n"); 992 continue; 993 } 994 995 // There's something in F. Check if it has a MachineFunction associated with 996 // it. 997 MachineFunction *MF = MMI.getMachineFunction(F); 998 999 // If it doesn't, then there's nothing to outline from. Move to the next 1000 // Function. 1001 if (!MF) { 1002 LLVM_DEBUG(dbgs() << "SKIP: Function does not have a MachineFunction\n"); 1003 continue; 1004 } 1005 1006 const TargetInstrInfo *TII = MF->getSubtarget().getInstrInfo(); 1007 if (!RunOnAllFunctions && !TII->shouldOutlineFromFunctionByDefault(*MF)) { 1008 LLVM_DEBUG(dbgs() << "SKIP: Target does not want to outline from " 1009 "function by default\n"); 1010 continue; 1011 } 1012 1013 // We have a MachineFunction. Ask the target if it's suitable for outlining. 1014 // If it isn't, then move on to the next Function in the module. 1015 if (!TII->isFunctionSafeToOutlineFrom(*MF, OutlineFromLinkOnceODRs)) { 1016 LLVM_DEBUG(dbgs() << "SKIP: " << MF->getName() 1017 << ": unsafe to outline from\n"); 1018 continue; 1019 } 1020 1021 // We have a function suitable for outlining. Iterate over every 1022 // MachineBasicBlock in MF and try to map its instructions to a list of 1023 // unsigned integers. 1024 const unsigned MinMBBSize = 2; 1025 1026 for (MachineBasicBlock &MBB : *MF) { 1027 LLVM_DEBUG(dbgs() << " MAPPING MBB: '" << MBB.getName() << "'\n"); 1028 // If there isn't anything in MBB, then there's no point in outlining from 1029 // it. 1030 // If there are fewer than 2 instructions in the MBB, then it can't ever 1031 // contain something worth outlining. 1032 // FIXME: This should be based off of the maximum size in B of an outlined 1033 // call versus the size in B of the MBB. 1034 if (MBB.size() < MinMBBSize) { 1035 LLVM_DEBUG(dbgs() << " SKIP: MBB size less than minimum size of " 1036 << MinMBBSize << "\n"); 1037 continue; 1038 } 1039 1040 // Check if MBB could be the target of an indirect branch. If it is, then 1041 // we don't want to outline from it. 1042 if (MBB.hasAddressTaken()) { 1043 LLVM_DEBUG(dbgs() << " SKIP: MBB's address is taken\n"); 1044 continue; 1045 } 1046 1047 // MBB is suitable for outlining. Map it to a list of unsigneds. 1048 Mapper.convertToUnsignedVec(MBB, *TII); 1049 } 1050 } 1051 // Statistics. 1052 UnsignedVecSize = Mapper.UnsignedVec.size(); 1053 } 1054 1055 void MachineOutliner::initSizeRemarkInfo( 1056 const Module &M, const MachineModuleInfo &MMI, 1057 StringMap<unsigned> &FunctionToInstrCount) { 1058 // Collect instruction counts for every function. We'll use this to emit 1059 // per-function size remarks later. 1060 for (const Function &F : M) { 1061 MachineFunction *MF = MMI.getMachineFunction(F); 1062 1063 // We only care about MI counts here. If there's no MachineFunction at this 1064 // point, then there won't be after the outliner runs, so let's move on. 1065 if (!MF) 1066 continue; 1067 FunctionToInstrCount[F.getName().str()] = MF->getInstructionCount(); 1068 } 1069 } 1070 1071 void MachineOutliner::emitInstrCountChangedRemark( 1072 const Module &M, const MachineModuleInfo &MMI, 1073 const StringMap<unsigned> &FunctionToInstrCount) { 1074 // Iterate over each function in the module and emit remarks. 1075 // Note that we won't miss anything by doing this, because the outliner never 1076 // deletes functions. 1077 for (const Function &F : M) { 1078 MachineFunction *MF = MMI.getMachineFunction(F); 1079 1080 // The outliner never deletes functions. If we don't have a MF here, then we 1081 // didn't have one prior to outlining either. 1082 if (!MF) 1083 continue; 1084 1085 std::string Fname = std::string(F.getName()); 1086 unsigned FnCountAfter = MF->getInstructionCount(); 1087 unsigned FnCountBefore = 0; 1088 1089 // Check if the function was recorded before. 1090 auto It = FunctionToInstrCount.find(Fname); 1091 1092 // Did we have a previously-recorded size? If yes, then set FnCountBefore 1093 // to that. 1094 if (It != FunctionToInstrCount.end()) 1095 FnCountBefore = It->second; 1096 1097 // Compute the delta and emit a remark if there was a change. 1098 int64_t FnDelta = static_cast<int64_t>(FnCountAfter) - 1099 static_cast<int64_t>(FnCountBefore); 1100 if (FnDelta == 0) 1101 continue; 1102 1103 MachineOptimizationRemarkEmitter MORE(*MF, nullptr); 1104 MORE.emit([&]() { 1105 MachineOptimizationRemarkAnalysis R("size-info", "FunctionMISizeChange", 1106 DiagnosticLocation(), &MF->front()); 1107 R << DiagnosticInfoOptimizationBase::Argument("Pass", "Machine Outliner") 1108 << ": Function: " 1109 << DiagnosticInfoOptimizationBase::Argument("Function", F.getName()) 1110 << ": MI instruction count changed from " 1111 << DiagnosticInfoOptimizationBase::Argument("MIInstrsBefore", 1112 FnCountBefore) 1113 << " to " 1114 << DiagnosticInfoOptimizationBase::Argument("MIInstrsAfter", 1115 FnCountAfter) 1116 << "; Delta: " 1117 << DiagnosticInfoOptimizationBase::Argument("Delta", FnDelta); 1118 return R; 1119 }); 1120 } 1121 } 1122 1123 bool MachineOutliner::runOnModule(Module &M) { 1124 // Check if there's anything in the module. If it's empty, then there's 1125 // nothing to outline. 1126 if (M.empty()) 1127 return false; 1128 1129 // Number to append to the current outlined function. 1130 unsigned OutlinedFunctionNum = 0; 1131 1132 OutlineRepeatedNum = 0; 1133 if (!doOutline(M, OutlinedFunctionNum)) 1134 return false; 1135 1136 for (unsigned I = 0; I < OutlinerReruns; ++I) { 1137 OutlinedFunctionNum = 0; 1138 OutlineRepeatedNum++; 1139 if (!doOutline(M, OutlinedFunctionNum)) { 1140 LLVM_DEBUG({ 1141 dbgs() << "Did not outline on iteration " << I + 2 << " out of " 1142 << OutlinerReruns + 1 << "\n"; 1143 }); 1144 break; 1145 } 1146 } 1147 1148 return true; 1149 } 1150 1151 bool MachineOutliner::doOutline(Module &M, unsigned &OutlinedFunctionNum) { 1152 MachineModuleInfo &MMI = getAnalysis<MachineModuleInfoWrapperPass>().getMMI(); 1153 1154 // If the user passed -enable-machine-outliner=always or 1155 // -enable-machine-outliner, the pass will run on all functions in the module. 1156 // Otherwise, if the target supports default outlining, it will run on all 1157 // functions deemed by the target to be worth outlining from by default. Tell 1158 // the user how the outliner is running. 1159 LLVM_DEBUG({ 1160 dbgs() << "Machine Outliner: Running on "; 1161 if (RunOnAllFunctions) 1162 dbgs() << "all functions"; 1163 else 1164 dbgs() << "target-default functions"; 1165 dbgs() << "\n"; 1166 }); 1167 1168 // If the user specifies that they want to outline from linkonceodrs, set 1169 // it here. 1170 OutlineFromLinkOnceODRs = EnableLinkOnceODROutlining; 1171 InstructionMapper Mapper; 1172 1173 // Prepare instruction mappings for the suffix tree. 1174 populateMapper(Mapper, M, MMI); 1175 std::vector<OutlinedFunction> FunctionList; 1176 1177 // Find all of the outlining candidates. 1178 findCandidates(Mapper, FunctionList); 1179 1180 // If we've requested size remarks, then collect the MI counts of every 1181 // function before outlining, and the MI counts after outlining. 1182 // FIXME: This shouldn't be in the outliner at all; it should ultimately be 1183 // the pass manager's responsibility. 1184 // This could pretty easily be placed in outline instead, but because we 1185 // really ultimately *don't* want this here, it's done like this for now 1186 // instead. 1187 1188 // Check if we want size remarks. 1189 bool ShouldEmitSizeRemarks = M.shouldEmitInstrCountChangedRemark(); 1190 StringMap<unsigned> FunctionToInstrCount; 1191 if (ShouldEmitSizeRemarks) 1192 initSizeRemarkInfo(M, MMI, FunctionToInstrCount); 1193 1194 // Outline each of the candidates and return true if something was outlined. 1195 bool OutlinedSomething = 1196 outline(M, FunctionList, Mapper, OutlinedFunctionNum); 1197 1198 // If we outlined something, we definitely changed the MI count of the 1199 // module. If we've asked for size remarks, then output them. 1200 // FIXME: This should be in the pass manager. 1201 if (ShouldEmitSizeRemarks && OutlinedSomething) 1202 emitInstrCountChangedRemark(M, MMI, FunctionToInstrCount); 1203 1204 LLVM_DEBUG({ 1205 if (!OutlinedSomething) 1206 dbgs() << "Stopped outlining at iteration " << OutlineRepeatedNum 1207 << " because no changes were found.\n"; 1208 }); 1209 1210 return OutlinedSomething; 1211 } 1212