1 //===- Parsing, selection, and construction of pass pipelines -------------===// 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 /// \file 9 /// 10 /// This file provides the implementation of the PassBuilder based on our 11 /// static pass registry as well as related functionality. It also provides 12 /// helpers to aid in analyzing, debugging, and testing passes and pass 13 /// pipelines. 14 /// 15 //===----------------------------------------------------------------------===// 16 17 #include "llvm/Passes/PassBuilder.h" 18 #include "llvm/ADT/StringSwitch.h" 19 #include "llvm/Analysis/AliasAnalysis.h" 20 #include "llvm/Analysis/AliasAnalysisEvaluator.h" 21 #include "llvm/Analysis/AliasSetTracker.h" 22 #include "llvm/Analysis/AssumptionCache.h" 23 #include "llvm/Analysis/BasicAliasAnalysis.h" 24 #include "llvm/Analysis/BlockFrequencyInfo.h" 25 #include "llvm/Analysis/BranchProbabilityInfo.h" 26 #include "llvm/Analysis/CFGPrinter.h" 27 #include "llvm/Analysis/CFLAndersAliasAnalysis.h" 28 #include "llvm/Analysis/CFLSteensAliasAnalysis.h" 29 #include "llvm/Analysis/CGSCCPassManager.h" 30 #include "llvm/Analysis/CallGraph.h" 31 #include "llvm/Analysis/DDG.h" 32 #include "llvm/Analysis/Delinearization.h" 33 #include "llvm/Analysis/DemandedBits.h" 34 #include "llvm/Analysis/DependenceAnalysis.h" 35 #include "llvm/Analysis/DominanceFrontier.h" 36 #include "llvm/Analysis/FunctionPropertiesAnalysis.h" 37 #include "llvm/Analysis/GlobalsModRef.h" 38 #include "llvm/Analysis/IRSimilarityIdentifier.h" 39 #include "llvm/Analysis/IVUsers.h" 40 #include "llvm/Analysis/InlineAdvisor.h" 41 #include "llvm/Analysis/InlineSizeEstimatorAnalysis.h" 42 #include "llvm/Analysis/InstCount.h" 43 #include "llvm/Analysis/LazyCallGraph.h" 44 #include "llvm/Analysis/LazyValueInfo.h" 45 #include "llvm/Analysis/Lint.h" 46 #include "llvm/Analysis/LoopAccessAnalysis.h" 47 #include "llvm/Analysis/LoopCacheAnalysis.h" 48 #include "llvm/Analysis/LoopInfo.h" 49 #include "llvm/Analysis/LoopNestAnalysis.h" 50 #include "llvm/Analysis/MemDerefPrinter.h" 51 #include "llvm/Analysis/MemoryDependenceAnalysis.h" 52 #include "llvm/Analysis/MemorySSA.h" 53 #include "llvm/Analysis/ModuleDebugInfoPrinter.h" 54 #include "llvm/Analysis/ModuleSummaryAnalysis.h" 55 #include "llvm/Analysis/MustExecute.h" 56 #include "llvm/Analysis/ObjCARCAliasAnalysis.h" 57 #include "llvm/Analysis/OptimizationRemarkEmitter.h" 58 #include "llvm/Analysis/PhiValues.h" 59 #include "llvm/Analysis/PostDominators.h" 60 #include "llvm/Analysis/ProfileSummaryInfo.h" 61 #include "llvm/Analysis/RegionInfo.h" 62 #include "llvm/Analysis/ScalarEvolution.h" 63 #include "llvm/Analysis/ScalarEvolutionAliasAnalysis.h" 64 #include "llvm/Analysis/ScopedNoAliasAA.h" 65 #include "llvm/Analysis/StackLifetime.h" 66 #include "llvm/Analysis/StackSafetyAnalysis.h" 67 #include "llvm/Analysis/TargetLibraryInfo.h" 68 #include "llvm/Analysis/TargetTransformInfo.h" 69 #include "llvm/Analysis/TypeBasedAliasAnalysis.h" 70 #include "llvm/IR/Dominators.h" 71 #include "llvm/IR/IRPrintingPasses.h" 72 #include "llvm/IR/PassManager.h" 73 #include "llvm/IR/SafepointIRVerifier.h" 74 #include "llvm/IR/Verifier.h" 75 #include "llvm/Support/CommandLine.h" 76 #include "llvm/Support/Debug.h" 77 #include "llvm/Support/FormatVariadic.h" 78 #include "llvm/Support/Regex.h" 79 #include "llvm/Target/TargetMachine.h" 80 #include "llvm/Transforms/AggressiveInstCombine/AggressiveInstCombine.h" 81 #include "llvm/Transforms/Coroutines/CoroCleanup.h" 82 #include "llvm/Transforms/Coroutines/CoroEarly.h" 83 #include "llvm/Transforms/Coroutines/CoroElide.h" 84 #include "llvm/Transforms/Coroutines/CoroSplit.h" 85 #include "llvm/Transforms/HelloNew/HelloWorld.h" 86 #include "llvm/Transforms/IPO/AlwaysInliner.h" 87 #include "llvm/Transforms/IPO/Annotation2Metadata.h" 88 #include "llvm/Transforms/IPO/ArgumentPromotion.h" 89 #include "llvm/Transforms/IPO/Attributor.h" 90 #include "llvm/Transforms/IPO/BlockExtractor.h" 91 #include "llvm/Transforms/IPO/CalledValuePropagation.h" 92 #include "llvm/Transforms/IPO/ConstantMerge.h" 93 #include "llvm/Transforms/IPO/CrossDSOCFI.h" 94 #include "llvm/Transforms/IPO/DeadArgumentElimination.h" 95 #include "llvm/Transforms/IPO/ElimAvailExtern.h" 96 #include "llvm/Transforms/IPO/ForceFunctionAttrs.h" 97 #include "llvm/Transforms/IPO/FunctionAttrs.h" 98 #include "llvm/Transforms/IPO/FunctionImport.h" 99 #include "llvm/Transforms/IPO/GlobalDCE.h" 100 #include "llvm/Transforms/IPO/GlobalOpt.h" 101 #include "llvm/Transforms/IPO/GlobalSplit.h" 102 #include "llvm/Transforms/IPO/HotColdSplitting.h" 103 #include "llvm/Transforms/IPO/InferFunctionAttrs.h" 104 #include "llvm/Transforms/IPO/Inliner.h" 105 #include "llvm/Transforms/IPO/Internalize.h" 106 #include "llvm/Transforms/IPO/LoopExtractor.h" 107 #include "llvm/Transforms/IPO/LowerTypeTests.h" 108 #include "llvm/Transforms/IPO/MergeFunctions.h" 109 #include "llvm/Transforms/IPO/OpenMPOpt.h" 110 #include "llvm/Transforms/IPO/PartialInlining.h" 111 #include "llvm/Transforms/IPO/SCCP.h" 112 #include "llvm/Transforms/IPO/SampleProfile.h" 113 #include "llvm/Transforms/IPO/StripDeadPrototypes.h" 114 #include "llvm/Transforms/IPO/StripSymbols.h" 115 #include "llvm/Transforms/IPO/SyntheticCountsPropagation.h" 116 #include "llvm/Transforms/IPO/WholeProgramDevirt.h" 117 #include "llvm/Transforms/InstCombine/InstCombine.h" 118 #include "llvm/Transforms/Instrumentation.h" 119 #include "llvm/Transforms/Instrumentation/AddressSanitizer.h" 120 #include "llvm/Transforms/Instrumentation/BoundsChecking.h" 121 #include "llvm/Transforms/Instrumentation/CGProfile.h" 122 #include "llvm/Transforms/Instrumentation/ControlHeightReduction.h" 123 #include "llvm/Transforms/Instrumentation/DataFlowSanitizer.h" 124 #include "llvm/Transforms/Instrumentation/GCOVProfiler.h" 125 #include "llvm/Transforms/Instrumentation/HWAddressSanitizer.h" 126 #include "llvm/Transforms/Instrumentation/InstrOrderFile.h" 127 #include "llvm/Transforms/Instrumentation/InstrProfiling.h" 128 #include "llvm/Transforms/Instrumentation/MemProfiler.h" 129 #include "llvm/Transforms/Instrumentation/MemorySanitizer.h" 130 #include "llvm/Transforms/Instrumentation/PGOInstrumentation.h" 131 #include "llvm/Transforms/Instrumentation/PoisonChecking.h" 132 #include "llvm/Transforms/Instrumentation/SanitizerCoverage.h" 133 #include "llvm/Transforms/Instrumentation/ThreadSanitizer.h" 134 #include "llvm/Transforms/ObjCARC.h" 135 #include "llvm/Transforms/Scalar/ADCE.h" 136 #include "llvm/Transforms/Scalar/AlignmentFromAssumptions.h" 137 #include "llvm/Transforms/Scalar/AnnotationRemarks.h" 138 #include "llvm/Transforms/Scalar/BDCE.h" 139 #include "llvm/Transforms/Scalar/CallSiteSplitting.h" 140 #include "llvm/Transforms/Scalar/ConstantHoisting.h" 141 #include "llvm/Transforms/Scalar/ConstraintElimination.h" 142 #include "llvm/Transforms/Scalar/CorrelatedValuePropagation.h" 143 #include "llvm/Transforms/Scalar/DCE.h" 144 #include "llvm/Transforms/Scalar/DeadStoreElimination.h" 145 #include "llvm/Transforms/Scalar/DivRemPairs.h" 146 #include "llvm/Transforms/Scalar/EarlyCSE.h" 147 #include "llvm/Transforms/Scalar/Float2Int.h" 148 #include "llvm/Transforms/Scalar/GVN.h" 149 #include "llvm/Transforms/Scalar/GuardWidening.h" 150 #include "llvm/Transforms/Scalar/IVUsersPrinter.h" 151 #include "llvm/Transforms/Scalar/IndVarSimplify.h" 152 #include "llvm/Transforms/Scalar/InductiveRangeCheckElimination.h" 153 #include "llvm/Transforms/Scalar/InstSimplifyPass.h" 154 #include "llvm/Transforms/Scalar/JumpThreading.h" 155 #include "llvm/Transforms/Scalar/LICM.h" 156 #include "llvm/Transforms/Scalar/LoopAccessAnalysisPrinter.h" 157 #include "llvm/Transforms/Scalar/LoopDataPrefetch.h" 158 #include "llvm/Transforms/Scalar/LoopDeletion.h" 159 #include "llvm/Transforms/Scalar/LoopDistribute.h" 160 #include "llvm/Transforms/Scalar/LoopFlatten.h" 161 #include "llvm/Transforms/Scalar/LoopFuse.h" 162 #include "llvm/Transforms/Scalar/LoopIdiomRecognize.h" 163 #include "llvm/Transforms/Scalar/LoopInstSimplify.h" 164 #include "llvm/Transforms/Scalar/LoopInterchange.h" 165 #include "llvm/Transforms/Scalar/LoopLoadElimination.h" 166 #include "llvm/Transforms/Scalar/LoopPassManager.h" 167 #include "llvm/Transforms/Scalar/LoopPredication.h" 168 #include "llvm/Transforms/Scalar/LoopReroll.h" 169 #include "llvm/Transforms/Scalar/LoopRotation.h" 170 #include "llvm/Transforms/Scalar/LoopSimplifyCFG.h" 171 #include "llvm/Transforms/Scalar/LoopSink.h" 172 #include "llvm/Transforms/Scalar/LoopStrengthReduce.h" 173 #include "llvm/Transforms/Scalar/LoopUnrollAndJamPass.h" 174 #include "llvm/Transforms/Scalar/LoopUnrollPass.h" 175 #include "llvm/Transforms/Scalar/LoopVersioningLICM.h" 176 #include "llvm/Transforms/Scalar/LowerAtomic.h" 177 #include "llvm/Transforms/Scalar/LowerConstantIntrinsics.h" 178 #include "llvm/Transforms/Scalar/LowerExpectIntrinsic.h" 179 #include "llvm/Transforms/Scalar/LowerGuardIntrinsic.h" 180 #include "llvm/Transforms/Scalar/LowerMatrixIntrinsics.h" 181 #include "llvm/Transforms/Scalar/LowerWidenableCondition.h" 182 #include "llvm/Transforms/Scalar/MakeGuardsExplicit.h" 183 #include "llvm/Transforms/Scalar/MemCpyOptimizer.h" 184 #include "llvm/Transforms/Scalar/MergeICmps.h" 185 #include "llvm/Transforms/Scalar/MergedLoadStoreMotion.h" 186 #include "llvm/Transforms/Scalar/NaryReassociate.h" 187 #include "llvm/Transforms/Scalar/NewGVN.h" 188 #include "llvm/Transforms/Scalar/PartiallyInlineLibCalls.h" 189 #include "llvm/Transforms/Scalar/Reassociate.h" 190 #include "llvm/Transforms/Scalar/Reg2Mem.h" 191 #include "llvm/Transforms/Scalar/RewriteStatepointsForGC.h" 192 #include "llvm/Transforms/Scalar/SCCP.h" 193 #include "llvm/Transforms/Scalar/SROA.h" 194 #include "llvm/Transforms/Scalar/Scalarizer.h" 195 #include "llvm/Transforms/Scalar/SeparateConstOffsetFromGEP.h" 196 #include "llvm/Transforms/Scalar/SimpleLoopUnswitch.h" 197 #include "llvm/Transforms/Scalar/SimplifyCFG.h" 198 #include "llvm/Transforms/Scalar/Sink.h" 199 #include "llvm/Transforms/Scalar/SpeculateAroundPHIs.h" 200 #include "llvm/Transforms/Scalar/SpeculativeExecution.h" 201 #include "llvm/Transforms/Scalar/StraightLineStrengthReduce.h" 202 #include "llvm/Transforms/Scalar/StructurizeCFG.h" 203 #include "llvm/Transforms/Scalar/TailRecursionElimination.h" 204 #include "llvm/Transforms/Scalar/WarnMissedTransforms.h" 205 #include "llvm/Transforms/Utils/AddDiscriminators.h" 206 #include "llvm/Transforms/Utils/AssumeBundleBuilder.h" 207 #include "llvm/Transforms/Utils/BreakCriticalEdges.h" 208 #include "llvm/Transforms/Utils/CanonicalizeAliases.h" 209 #include "llvm/Transforms/Utils/CanonicalizeFreezeInLoops.h" 210 #include "llvm/Transforms/Utils/EntryExitInstrumenter.h" 211 #include "llvm/Transforms/Utils/FixIrreducible.h" 212 #include "llvm/Transforms/Utils/InjectTLIMappings.h" 213 #include "llvm/Transforms/Utils/InstructionNamer.h" 214 #include "llvm/Transforms/Utils/LCSSA.h" 215 #include "llvm/Transforms/Utils/LibCallsShrinkWrap.h" 216 #include "llvm/Transforms/Utils/LoopSimplify.h" 217 #include "llvm/Transforms/Utils/LoopVersioning.h" 218 #include "llvm/Transforms/Utils/LowerInvoke.h" 219 #include "llvm/Transforms/Utils/LowerSwitch.h" 220 #include "llvm/Transforms/Utils/Mem2Reg.h" 221 #include "llvm/Transforms/Utils/MetaRenamer.h" 222 #include "llvm/Transforms/Utils/NameAnonGlobals.h" 223 #include "llvm/Transforms/Utils/StripGCRelocates.h" 224 #include "llvm/Transforms/Utils/StripNonLineTableDebugInfo.h" 225 #include "llvm/Transforms/Utils/SymbolRewriter.h" 226 #include "llvm/Transforms/Utils/UnifyFunctionExitNodes.h" 227 #include "llvm/Transforms/Utils/UnifyLoopExits.h" 228 #include "llvm/Transforms/Utils/UniqueInternalLinkageNames.h" 229 #include "llvm/Transforms/Vectorize/LoadStoreVectorizer.h" 230 #include "llvm/Transforms/Vectorize/LoopVectorize.h" 231 #include "llvm/Transforms/Vectorize/SLPVectorizer.h" 232 #include "llvm/Transforms/Vectorize/VectorCombine.h" 233 234 using namespace llvm; 235 236 extern cl::opt<unsigned> MaxDevirtIterations; 237 238 static cl::opt<InliningAdvisorMode> UseInlineAdvisor( 239 "enable-ml-inliner", cl::init(InliningAdvisorMode::Default), cl::Hidden, 240 cl::desc("Enable ML policy for inliner. Currently trained for -Oz only"), 241 cl::values(clEnumValN(InliningAdvisorMode::Default, "default", 242 "Heuristics-based inliner version."), 243 clEnumValN(InliningAdvisorMode::Development, "development", 244 "Use development mode (runtime-loadable model)."), 245 clEnumValN(InliningAdvisorMode::Release, "release", 246 "Use release mode (AOT-compiled model)."))); 247 248 static cl::opt<bool> EnableSyntheticCounts( 249 "enable-npm-synthetic-counts", cl::init(false), cl::Hidden, cl::ZeroOrMore, 250 cl::desc("Run synthetic function entry count generation " 251 "pass")); 252 253 static const Regex DefaultAliasRegex( 254 "^(default|thinlto-pre-link|thinlto|lto-pre-link|lto)<(O[0123sz])>$"); 255 256 /// Flag to enable inline deferral during PGO. 257 static cl::opt<bool> 258 EnablePGOInlineDeferral("enable-npm-pgo-inline-deferral", cl::init(true), 259 cl::Hidden, 260 cl::desc("Enable inline deferral during PGO")); 261 262 static cl::opt<bool> EnableMemProfiler("enable-mem-prof", cl::init(false), 263 cl::Hidden, cl::ZeroOrMore, 264 cl::desc("Enable memory profiler")); 265 266 PipelineTuningOptions::PipelineTuningOptions() { 267 LoopInterleaving = true; 268 LoopVectorization = true; 269 SLPVectorization = false; 270 LoopUnrolling = true; 271 ForgetAllSCEVInLoopUnroll = ForgetSCEVInLoopUnroll; 272 Coroutines = false; 273 LicmMssaOptCap = SetLicmMssaOptCap; 274 LicmMssaNoAccForPromotionCap = SetLicmMssaNoAccForPromotionCap; 275 CallGraphProfile = true; 276 } 277 278 extern cl::opt<bool> EnableConstraintElimination; 279 extern cl::opt<bool> EnableGVNHoist; 280 extern cl::opt<bool> EnableGVNSink; 281 extern cl::opt<bool> EnableHotColdSplit; 282 extern cl::opt<bool> EnableOrderFileInstrumentation; 283 extern cl::opt<bool> EnableCHR; 284 extern cl::opt<bool> EnableUnrollAndJam; 285 extern cl::opt<bool> EnableLoopFlatten; 286 extern cl::opt<bool> RunNewGVN; 287 extern cl::opt<bool> RunPartialInlining; 288 289 extern cl::opt<bool> FlattenedProfileUsed; 290 291 extern cl::opt<AttributorRunOption> AttributorRun; 292 extern cl::opt<bool> EnableKnowledgeRetention; 293 294 extern cl::opt<bool> EnableMatrix; 295 296 extern cl::opt<bool> DisablePreInliner; 297 extern cl::opt<int> PreInlineThreshold; 298 299 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O0 = { 300 /*SpeedLevel*/ 0, 301 /*SizeLevel*/ 0}; 302 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O1 = { 303 /*SpeedLevel*/ 1, 304 /*SizeLevel*/ 0}; 305 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O2 = { 306 /*SpeedLevel*/ 2, 307 /*SizeLevel*/ 0}; 308 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O3 = { 309 /*SpeedLevel*/ 3, 310 /*SizeLevel*/ 0}; 311 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::Os = { 312 /*SpeedLevel*/ 2, 313 /*SizeLevel*/ 1}; 314 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::Oz = { 315 /*SpeedLevel*/ 2, 316 /*SizeLevel*/ 2}; 317 318 namespace { 319 320 // The following passes/analyses have custom names, otherwise their name will 321 // include `(anonymous namespace)`. These are special since they are only for 322 // testing purposes and don't live in a header file. 323 324 /// No-op module pass which does nothing. 325 struct NoOpModulePass : PassInfoMixin<NoOpModulePass> { 326 PreservedAnalyses run(Module &M, ModuleAnalysisManager &) { 327 return PreservedAnalyses::all(); 328 } 329 330 static StringRef name() { return "NoOpModulePass"; } 331 }; 332 333 /// No-op module analysis. 334 class NoOpModuleAnalysis : public AnalysisInfoMixin<NoOpModuleAnalysis> { 335 friend AnalysisInfoMixin<NoOpModuleAnalysis>; 336 static AnalysisKey Key; 337 338 public: 339 struct Result {}; 340 Result run(Module &, ModuleAnalysisManager &) { return Result(); } 341 static StringRef name() { return "NoOpModuleAnalysis"; } 342 }; 343 344 /// No-op CGSCC pass which does nothing. 345 struct NoOpCGSCCPass : PassInfoMixin<NoOpCGSCCPass> { 346 PreservedAnalyses run(LazyCallGraph::SCC &C, CGSCCAnalysisManager &, 347 LazyCallGraph &, CGSCCUpdateResult &UR) { 348 return PreservedAnalyses::all(); 349 } 350 static StringRef name() { return "NoOpCGSCCPass"; } 351 }; 352 353 /// No-op CGSCC analysis. 354 class NoOpCGSCCAnalysis : public AnalysisInfoMixin<NoOpCGSCCAnalysis> { 355 friend AnalysisInfoMixin<NoOpCGSCCAnalysis>; 356 static AnalysisKey Key; 357 358 public: 359 struct Result {}; 360 Result run(LazyCallGraph::SCC &, CGSCCAnalysisManager &, LazyCallGraph &G) { 361 return Result(); 362 } 363 static StringRef name() { return "NoOpCGSCCAnalysis"; } 364 }; 365 366 /// No-op function pass which does nothing. 367 struct NoOpFunctionPass : PassInfoMixin<NoOpFunctionPass> { 368 PreservedAnalyses run(Function &F, FunctionAnalysisManager &) { 369 return PreservedAnalyses::all(); 370 } 371 static StringRef name() { return "NoOpFunctionPass"; } 372 }; 373 374 /// No-op function analysis. 375 class NoOpFunctionAnalysis : public AnalysisInfoMixin<NoOpFunctionAnalysis> { 376 friend AnalysisInfoMixin<NoOpFunctionAnalysis>; 377 static AnalysisKey Key; 378 379 public: 380 struct Result {}; 381 Result run(Function &, FunctionAnalysisManager &) { return Result(); } 382 static StringRef name() { return "NoOpFunctionAnalysis"; } 383 }; 384 385 /// No-op loop pass which does nothing. 386 struct NoOpLoopPass : PassInfoMixin<NoOpLoopPass> { 387 PreservedAnalyses run(Loop &L, LoopAnalysisManager &, 388 LoopStandardAnalysisResults &, LPMUpdater &) { 389 return PreservedAnalyses::all(); 390 } 391 static StringRef name() { return "NoOpLoopPass"; } 392 }; 393 394 /// No-op loop analysis. 395 class NoOpLoopAnalysis : public AnalysisInfoMixin<NoOpLoopAnalysis> { 396 friend AnalysisInfoMixin<NoOpLoopAnalysis>; 397 static AnalysisKey Key; 398 399 public: 400 struct Result {}; 401 Result run(Loop &, LoopAnalysisManager &, LoopStandardAnalysisResults &) { 402 return Result(); 403 } 404 static StringRef name() { return "NoOpLoopAnalysis"; } 405 }; 406 407 AnalysisKey NoOpModuleAnalysis::Key; 408 AnalysisKey NoOpCGSCCAnalysis::Key; 409 AnalysisKey NoOpFunctionAnalysis::Key; 410 AnalysisKey NoOpLoopAnalysis::Key; 411 412 } // namespace 413 414 PassBuilder::PassBuilder(bool DebugLogging, TargetMachine *TM, 415 PipelineTuningOptions PTO, Optional<PGOOptions> PGOOpt, 416 PassInstrumentationCallbacks *PIC) 417 : DebugLogging(DebugLogging), TM(TM), PTO(PTO), PGOOpt(PGOOpt), PIC(PIC) { 418 if (TM) 419 TM->registerPassBuilderCallbacks(*this, DebugLogging); 420 } 421 422 void PassBuilder::invokePeepholeEPCallbacks( 423 FunctionPassManager &FPM, PassBuilder::OptimizationLevel Level) { 424 for (auto &C : PeepholeEPCallbacks) 425 C(FPM, Level); 426 } 427 428 void PassBuilder::registerModuleAnalyses(ModuleAnalysisManager &MAM) { 429 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 430 MAM.registerPass([&] { return CREATE_PASS; }); 431 #include "PassRegistry.def" 432 433 for (auto &C : ModuleAnalysisRegistrationCallbacks) 434 C(MAM); 435 } 436 437 void PassBuilder::registerCGSCCAnalyses(CGSCCAnalysisManager &CGAM) { 438 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 439 CGAM.registerPass([&] { return CREATE_PASS; }); 440 #include "PassRegistry.def" 441 442 for (auto &C : CGSCCAnalysisRegistrationCallbacks) 443 C(CGAM); 444 } 445 446 void PassBuilder::registerFunctionAnalyses(FunctionAnalysisManager &FAM) { 447 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 448 FAM.registerPass([&] { return CREATE_PASS; }); 449 #include "PassRegistry.def" 450 451 for (auto &C : FunctionAnalysisRegistrationCallbacks) 452 C(FAM); 453 } 454 455 void PassBuilder::registerLoopAnalyses(LoopAnalysisManager &LAM) { 456 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 457 LAM.registerPass([&] { return CREATE_PASS; }); 458 #include "PassRegistry.def" 459 460 for (auto &C : LoopAnalysisRegistrationCallbacks) 461 C(LAM); 462 } 463 464 // Helper to add AnnotationRemarksPass. 465 static void addAnnotationRemarksPass(ModulePassManager &MPM) { 466 FunctionPassManager FPM; 467 FPM.addPass(AnnotationRemarksPass()); 468 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 469 } 470 471 // TODO: Investigate the cost/benefit of tail call elimination on debugging. 472 FunctionPassManager 473 PassBuilder::buildO1FunctionSimplificationPipeline(OptimizationLevel Level, 474 ThinLTOPhase Phase) { 475 476 FunctionPassManager FPM(DebugLogging); 477 478 // Form SSA out of local memory accesses after breaking apart aggregates into 479 // scalars. 480 FPM.addPass(SROA()); 481 482 // Catch trivial redundancies 483 FPM.addPass(EarlyCSEPass(true /* Enable mem-ssa. */)); 484 485 // Hoisting of scalars and load expressions. 486 FPM.addPass(SimplifyCFGPass()); 487 FPM.addPass(InstCombinePass()); 488 489 FPM.addPass(LibCallsShrinkWrapPass()); 490 491 invokePeepholeEPCallbacks(FPM, Level); 492 493 FPM.addPass(SimplifyCFGPass()); 494 495 // Form canonically associated expression trees, and simplify the trees using 496 // basic mathematical properties. For example, this will form (nearly) 497 // minimal multiplication trees. 498 FPM.addPass(ReassociatePass()); 499 500 // Add the primary loop simplification pipeline. 501 // FIXME: Currently this is split into two loop pass pipelines because we run 502 // some function passes in between them. These can and should be removed 503 // and/or replaced by scheduling the loop pass equivalents in the correct 504 // positions. But those equivalent passes aren't powerful enough yet. 505 // Specifically, `SimplifyCFGPass` and `InstCombinePass` are currently still 506 // used. We have `LoopSimplifyCFGPass` which isn't yet powerful enough yet to 507 // fully replace `SimplifyCFGPass`, and the closest to the other we have is 508 // `LoopInstSimplify`. 509 LoopPassManager LPM1(DebugLogging), LPM2(DebugLogging); 510 511 // Simplify the loop body. We do this initially to clean up after other loop 512 // passes run, either when iterating on a loop or on inner loops with 513 // implications on the outer loop. 514 LPM1.addPass(LoopInstSimplifyPass()); 515 LPM1.addPass(LoopSimplifyCFGPass()); 516 517 LPM1.addPass(LoopRotatePass(/* Disable header duplication */ true)); 518 // TODO: Investigate promotion cap for O1. 519 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 520 LPM1.addPass(SimpleLoopUnswitchPass()); 521 522 if (EnableLoopFlatten) 523 FPM.addPass(LoopFlattenPass()); 524 LPM2.addPass(LoopIdiomRecognizePass()); 525 LPM2.addPass(IndVarSimplifyPass()); 526 527 for (auto &C : LateLoopOptimizationsEPCallbacks) 528 C(LPM2, Level); 529 530 LPM2.addPass(LoopDeletionPass()); 531 // Do not enable unrolling in PreLinkThinLTO phase during sample PGO 532 // because it changes IR to makes profile annotation in back compile 533 // inaccurate. The normal unroller doesn't pay attention to forced full unroll 534 // attributes so we need to make sure and allow the full unroll pass to pay 535 // attention to it. 536 if (Phase != ThinLTOPhase::PreLink || !PGOOpt || 537 PGOOpt->Action != PGOOptions::SampleUse) 538 LPM2.addPass(LoopFullUnrollPass(Level.getSpeedupLevel(), 539 /* OnlyWhenForced= */ !PTO.LoopUnrolling, 540 PTO.ForgetAllSCEVInLoopUnroll)); 541 542 for (auto &C : LoopOptimizerEndEPCallbacks) 543 C(LPM2, Level); 544 545 // We provide the opt remark emitter pass for LICM to use. We only need to do 546 // this once as it is immutable. 547 FPM.addPass( 548 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 549 FPM.addPass(createFunctionToLoopPassAdaptor( 550 std::move(LPM1), EnableMSSALoopDependency, /*UseBlockFrequencyInfo=*/true, 551 DebugLogging)); 552 FPM.addPass(SimplifyCFGPass()); 553 FPM.addPass(InstCombinePass()); 554 // The loop passes in LPM2 (LoopFullUnrollPass) do not preserve MemorySSA. 555 // *All* loop passes must preserve it, in order to be able to use it. 556 FPM.addPass(createFunctionToLoopPassAdaptor( 557 std::move(LPM2), /*UseMemorySSA=*/false, /*UseBlockFrequencyInfo=*/false, 558 DebugLogging)); 559 560 // Delete small array after loop unroll. 561 FPM.addPass(SROA()); 562 563 // Specially optimize memory movement as it doesn't look like dataflow in SSA. 564 FPM.addPass(MemCpyOptPass()); 565 566 // Sparse conditional constant propagation. 567 // FIXME: It isn't clear why we do this *after* loop passes rather than 568 // before... 569 FPM.addPass(SCCPPass()); 570 571 // Delete dead bit computations (instcombine runs after to fold away the dead 572 // computations, and then ADCE will run later to exploit any new DCE 573 // opportunities that creates). 574 FPM.addPass(BDCEPass()); 575 576 // Run instcombine after redundancy and dead bit elimination to exploit 577 // opportunities opened up by them. 578 FPM.addPass(InstCombinePass()); 579 invokePeepholeEPCallbacks(FPM, Level); 580 581 if (PTO.Coroutines) 582 FPM.addPass(CoroElidePass()); 583 584 for (auto &C : ScalarOptimizerLateEPCallbacks) 585 C(FPM, Level); 586 587 // Finally, do an expensive DCE pass to catch all the dead code exposed by 588 // the simplifications and basic cleanup after all the simplifications. 589 // TODO: Investigate if this is too expensive. 590 FPM.addPass(ADCEPass()); 591 FPM.addPass(SimplifyCFGPass()); 592 FPM.addPass(InstCombinePass()); 593 invokePeepholeEPCallbacks(FPM, Level); 594 595 return FPM; 596 } 597 598 FunctionPassManager 599 PassBuilder::buildFunctionSimplificationPipeline(OptimizationLevel Level, 600 ThinLTOPhase Phase) { 601 assert(Level != OptimizationLevel::O0 && "Must request optimizations!"); 602 603 // The O1 pipeline has a separate pipeline creation function to simplify 604 // construction readability. 605 if (Level.getSpeedupLevel() == 1) 606 return buildO1FunctionSimplificationPipeline(Level, Phase); 607 608 FunctionPassManager FPM(DebugLogging); 609 610 // Form SSA out of local memory accesses after breaking apart aggregates into 611 // scalars. 612 FPM.addPass(SROA()); 613 614 // Catch trivial redundancies 615 FPM.addPass(EarlyCSEPass(true /* Enable mem-ssa. */)); 616 if (EnableKnowledgeRetention) 617 FPM.addPass(AssumeSimplifyPass()); 618 619 // Hoisting of scalars and load expressions. 620 if (EnableGVNHoist) 621 FPM.addPass(GVNHoistPass()); 622 623 // Global value numbering based sinking. 624 if (EnableGVNSink) { 625 FPM.addPass(GVNSinkPass()); 626 FPM.addPass(SimplifyCFGPass()); 627 } 628 629 if (EnableConstraintElimination) 630 FPM.addPass(ConstraintEliminationPass()); 631 632 // Speculative execution if the target has divergent branches; otherwise nop. 633 FPM.addPass(SpeculativeExecutionPass(/* OnlyIfDivergentTarget =*/true)); 634 635 // Optimize based on known information about branches, and cleanup afterward. 636 FPM.addPass(JumpThreadingPass()); 637 FPM.addPass(CorrelatedValuePropagationPass()); 638 639 FPM.addPass(SimplifyCFGPass()); 640 if (Level == OptimizationLevel::O3) 641 FPM.addPass(AggressiveInstCombinePass()); 642 FPM.addPass(InstCombinePass()); 643 644 if (!Level.isOptimizingForSize()) 645 FPM.addPass(LibCallsShrinkWrapPass()); 646 647 invokePeepholeEPCallbacks(FPM, Level); 648 649 // For PGO use pipeline, try to optimize memory intrinsics such as memcpy 650 // using the size value profile. Don't perform this when optimizing for size. 651 if (PGOOpt && PGOOpt->Action == PGOOptions::IRUse && 652 !Level.isOptimizingForSize()) 653 FPM.addPass(PGOMemOPSizeOpt()); 654 655 FPM.addPass(TailCallElimPass()); 656 FPM.addPass(SimplifyCFGPass()); 657 658 // Form canonically associated expression trees, and simplify the trees using 659 // basic mathematical properties. For example, this will form (nearly) 660 // minimal multiplication trees. 661 FPM.addPass(ReassociatePass()); 662 663 // Add the primary loop simplification pipeline. 664 // FIXME: Currently this is split into two loop pass pipelines because we run 665 // some function passes in between them. These can and should be removed 666 // and/or replaced by scheduling the loop pass equivalents in the correct 667 // positions. But those equivalent passes aren't powerful enough yet. 668 // Specifically, `SimplifyCFGPass` and `InstCombinePass` are currently still 669 // used. We have `LoopSimplifyCFGPass` which isn't yet powerful enough yet to 670 // fully replace `SimplifyCFGPass`, and the closest to the other we have is 671 // `LoopInstSimplify`. 672 LoopPassManager LPM1(DebugLogging), LPM2(DebugLogging); 673 674 // Simplify the loop body. We do this initially to clean up after other loop 675 // passes run, either when iterating on a loop or on inner loops with 676 // implications on the outer loop. 677 LPM1.addPass(LoopInstSimplifyPass()); 678 LPM1.addPass(LoopSimplifyCFGPass()); 679 680 // Disable header duplication in loop rotation at -Oz. 681 LPM1.addPass(LoopRotatePass(Level != OptimizationLevel::Oz)); 682 // TODO: Investigate promotion cap for O1. 683 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 684 LPM1.addPass(SimpleLoopUnswitchPass()); 685 LPM2.addPass(LoopIdiomRecognizePass()); 686 LPM2.addPass(IndVarSimplifyPass()); 687 688 for (auto &C : LateLoopOptimizationsEPCallbacks) 689 C(LPM2, Level); 690 691 LPM2.addPass(LoopDeletionPass()); 692 // Do not enable unrolling in PreLinkThinLTO phase during sample PGO 693 // because it changes IR to makes profile annotation in back compile 694 // inaccurate. The normal unroller doesn't pay attention to forced full unroll 695 // attributes so we need to make sure and allow the full unroll pass to pay 696 // attention to it. 697 if (Phase != ThinLTOPhase::PreLink || !PGOOpt || 698 PGOOpt->Action != PGOOptions::SampleUse) 699 LPM2.addPass(LoopFullUnrollPass(Level.getSpeedupLevel(), 700 /* OnlyWhenForced= */ !PTO.LoopUnrolling, 701 PTO.ForgetAllSCEVInLoopUnroll)); 702 703 for (auto &C : LoopOptimizerEndEPCallbacks) 704 C(LPM2, Level); 705 706 // We provide the opt remark emitter pass for LICM to use. We only need to do 707 // this once as it is immutable. 708 FPM.addPass( 709 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 710 FPM.addPass(createFunctionToLoopPassAdaptor( 711 std::move(LPM1), EnableMSSALoopDependency, /*UseBlockFrequencyInfo=*/true, 712 DebugLogging)); 713 FPM.addPass(SimplifyCFGPass()); 714 FPM.addPass(InstCombinePass()); 715 // The loop passes in LPM2 (LoopIdiomRecognizePass, IndVarSimplifyPass, 716 // LoopDeletionPass and LoopFullUnrollPass) do not preserve MemorySSA. 717 // *All* loop passes must preserve it, in order to be able to use it. 718 FPM.addPass(createFunctionToLoopPassAdaptor( 719 std::move(LPM2), /*UseMemorySSA=*/false, /*UseBlockFrequencyInfo=*/false, 720 DebugLogging)); 721 722 // Delete small array after loop unroll. 723 FPM.addPass(SROA()); 724 725 // Eliminate redundancies. 726 FPM.addPass(MergedLoadStoreMotionPass()); 727 if (RunNewGVN) 728 FPM.addPass(NewGVNPass()); 729 else 730 FPM.addPass(GVN()); 731 732 // Specially optimize memory movement as it doesn't look like dataflow in SSA. 733 FPM.addPass(MemCpyOptPass()); 734 735 // Sparse conditional constant propagation. 736 // FIXME: It isn't clear why we do this *after* loop passes rather than 737 // before... 738 FPM.addPass(SCCPPass()); 739 740 // Delete dead bit computations (instcombine runs after to fold away the dead 741 // computations, and then ADCE will run later to exploit any new DCE 742 // opportunities that creates). 743 FPM.addPass(BDCEPass()); 744 745 // Run instcombine after redundancy and dead bit elimination to exploit 746 // opportunities opened up by them. 747 FPM.addPass(InstCombinePass()); 748 invokePeepholeEPCallbacks(FPM, Level); 749 750 // Re-consider control flow based optimizations after redundancy elimination, 751 // redo DCE, etc. 752 FPM.addPass(JumpThreadingPass()); 753 FPM.addPass(CorrelatedValuePropagationPass()); 754 755 // Finally, do an expensive DCE pass to catch all the dead code exposed by 756 // the simplifications and basic cleanup after all the simplifications. 757 // TODO: Investigate if this is too expensive. 758 FPM.addPass(ADCEPass()); 759 760 FPM.addPass(DSEPass()); 761 FPM.addPass(createFunctionToLoopPassAdaptor( 762 LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap), 763 EnableMSSALoopDependency, /*UseBlockFrequencyInfo=*/true, DebugLogging)); 764 765 if (PTO.Coroutines) 766 FPM.addPass(CoroElidePass()); 767 768 for (auto &C : ScalarOptimizerLateEPCallbacks) 769 C(FPM, Level); 770 771 FPM.addPass(SimplifyCFGPass()); 772 FPM.addPass(InstCombinePass()); 773 invokePeepholeEPCallbacks(FPM, Level); 774 775 if (EnableCHR && Level == OptimizationLevel::O3 && PGOOpt && 776 (PGOOpt->Action == PGOOptions::IRUse || 777 PGOOpt->Action == PGOOptions::SampleUse)) 778 FPM.addPass(ControlHeightReductionPass()); 779 780 return FPM; 781 } 782 783 void PassBuilder::addRequiredLTOPreLinkPasses(ModulePassManager &MPM) { 784 MPM.addPass(CanonicalizeAliasesPass()); 785 MPM.addPass(NameAnonGlobalPass()); 786 } 787 788 void PassBuilder::addPGOInstrPasses(ModulePassManager &MPM, 789 PassBuilder::OptimizationLevel Level, 790 bool RunProfileGen, bool IsCS, 791 std::string ProfileFile, 792 std::string ProfileRemappingFile) { 793 assert(Level != OptimizationLevel::O0 && "Not expecting O0 here!"); 794 // Generally running simplification passes and the inliner with an high 795 // threshold results in smaller executables, but there may be cases where 796 // the size grows, so let's be conservative here and skip this simplification 797 // at -Os/Oz. We will not do this inline for context sensistive PGO (when 798 // IsCS is true). 799 if (!Level.isOptimizingForSize() && !IsCS && !DisablePreInliner) { 800 InlineParams IP; 801 802 IP.DefaultThreshold = PreInlineThreshold; 803 804 // FIXME: The hint threshold has the same value used by the regular inliner. 805 // This should probably be lowered after performance testing. 806 // FIXME: this comment is cargo culted from the old pass manager, revisit). 807 IP.HintThreshold = 325; 808 ModuleInlinerWrapperPass MIWP(IP, DebugLogging); 809 CGSCCPassManager &CGPipeline = MIWP.getPM(); 810 811 FunctionPassManager FPM; 812 FPM.addPass(SROA()); 813 FPM.addPass(EarlyCSEPass()); // Catch trivial redundancies. 814 FPM.addPass(SimplifyCFGPass()); // Merge & remove basic blocks. 815 FPM.addPass(InstCombinePass()); // Combine silly sequences. 816 invokePeepholeEPCallbacks(FPM, Level); 817 818 CGPipeline.addPass(createCGSCCToFunctionPassAdaptor(std::move(FPM))); 819 820 MPM.addPass(std::move(MIWP)); 821 822 // Delete anything that is now dead to make sure that we don't instrument 823 // dead code. Instrumentation can end up keeping dead code around and 824 // dramatically increase code size. 825 MPM.addPass(GlobalDCEPass()); 826 } 827 828 if (!RunProfileGen) { 829 assert(!ProfileFile.empty() && "Profile use expecting a profile file!"); 830 MPM.addPass(PGOInstrumentationUse(ProfileFile, ProfileRemappingFile, IsCS)); 831 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 832 // RequireAnalysisPass for PSI before subsequent non-module passes. 833 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 834 return; 835 } 836 837 // Perform PGO instrumentation. 838 MPM.addPass(PGOInstrumentationGen(IsCS)); 839 840 FunctionPassManager FPM; 841 // Disable header duplication in loop rotation at -Oz. 842 FPM.addPass(createFunctionToLoopPassAdaptor( 843 LoopRotatePass(Level != OptimizationLevel::Oz), EnableMSSALoopDependency, 844 /*UseBlockFrequencyInfo=*/false, DebugLogging)); 845 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 846 847 // Add the profile lowering pass. 848 InstrProfOptions Options; 849 if (!ProfileFile.empty()) 850 Options.InstrProfileOutput = ProfileFile; 851 // Do counter promotion at Level greater than O0. 852 Options.DoCounterPromotion = true; 853 Options.UseBFIInPromotion = IsCS; 854 MPM.addPass(InstrProfiling(Options, IsCS)); 855 } 856 857 void PassBuilder::addPGOInstrPassesForO0(ModulePassManager &MPM, 858 bool RunProfileGen, bool IsCS, 859 std::string ProfileFile, 860 std::string ProfileRemappingFile) { 861 if (!RunProfileGen) { 862 assert(!ProfileFile.empty() && "Profile use expecting a profile file!"); 863 MPM.addPass(PGOInstrumentationUse(ProfileFile, ProfileRemappingFile, IsCS)); 864 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 865 // RequireAnalysisPass for PSI before subsequent non-module passes. 866 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 867 return; 868 } 869 870 // Perform PGO instrumentation. 871 MPM.addPass(PGOInstrumentationGen(IsCS)); 872 // Add the profile lowering pass. 873 InstrProfOptions Options; 874 if (!ProfileFile.empty()) 875 Options.InstrProfileOutput = ProfileFile; 876 // Do not do counter promotion at O0. 877 Options.DoCounterPromotion = false; 878 Options.UseBFIInPromotion = IsCS; 879 MPM.addPass(InstrProfiling(Options, IsCS)); 880 } 881 882 static InlineParams 883 getInlineParamsFromOptLevel(PassBuilder::OptimizationLevel Level) { 884 return getInlineParams(Level.getSpeedupLevel(), Level.getSizeLevel()); 885 } 886 887 ModuleInlinerWrapperPass 888 PassBuilder::buildInlinerPipeline(OptimizationLevel Level, ThinLTOPhase Phase) { 889 InlineParams IP = getInlineParamsFromOptLevel(Level); 890 if (Phase == PassBuilder::ThinLTOPhase::PreLink && PGOOpt && 891 PGOOpt->Action == PGOOptions::SampleUse) 892 IP.HotCallSiteThreshold = 0; 893 894 if (PGOOpt) 895 IP.EnableDeferral = EnablePGOInlineDeferral; 896 897 ModuleInlinerWrapperPass MIWP(IP, DebugLogging, UseInlineAdvisor, 898 MaxDevirtIterations); 899 900 // Require the GlobalsAA analysis for the module so we can query it within 901 // the CGSCC pipeline. 902 MIWP.addRequiredModuleAnalysis<GlobalsAA>(); 903 904 // Require the ProfileSummaryAnalysis for the module so we can query it within 905 // the inliner pass. 906 MIWP.addRequiredModuleAnalysis<ProfileSummaryAnalysis>(); 907 908 // Now begin the main postorder CGSCC pipeline. 909 // FIXME: The current CGSCC pipeline has its origins in the legacy pass 910 // manager and trying to emulate its precise behavior. Much of this doesn't 911 // make a lot of sense and we should revisit the core CGSCC structure. 912 CGSCCPassManager &MainCGPipeline = MIWP.getPM(); 913 914 // Note: historically, the PruneEH pass was run first to deduce nounwind and 915 // generally clean up exception handling overhead. It isn't clear this is 916 // valuable as the inliner doesn't currently care whether it is inlining an 917 // invoke or a call. 918 919 if (AttributorRun & AttributorRunOption::CGSCC) 920 MainCGPipeline.addPass(AttributorCGSCCPass()); 921 922 if (PTO.Coroutines) 923 MainCGPipeline.addPass(CoroSplitPass(Level != OptimizationLevel::O0)); 924 925 // Now deduce any function attributes based in the current code. 926 MainCGPipeline.addPass(PostOrderFunctionAttrsPass()); 927 928 // When at O3 add argument promotion to the pass pipeline. 929 // FIXME: It isn't at all clear why this should be limited to O3. 930 if (Level == OptimizationLevel::O3) 931 MainCGPipeline.addPass(ArgumentPromotionPass()); 932 933 // Try to perform OpenMP specific optimizations. This is a (quick!) no-op if 934 // there are no OpenMP runtime calls present in the module. 935 if (Level == OptimizationLevel::O2 || Level == OptimizationLevel::O3) 936 MainCGPipeline.addPass(OpenMPOptPass()); 937 938 // Lastly, add the core function simplification pipeline nested inside the 939 // CGSCC walk. 940 MainCGPipeline.addPass(createCGSCCToFunctionPassAdaptor( 941 buildFunctionSimplificationPipeline(Level, Phase))); 942 943 for (auto &C : CGSCCOptimizerLateEPCallbacks) 944 C(MainCGPipeline, Level); 945 946 return MIWP; 947 } 948 949 ModulePassManager 950 PassBuilder::buildModuleSimplificationPipeline(OptimizationLevel Level, 951 ThinLTOPhase Phase) { 952 ModulePassManager MPM(DebugLogging); 953 954 bool HasSampleProfile = PGOOpt && (PGOOpt->Action == PGOOptions::SampleUse); 955 956 // In ThinLTO mode, when flattened profile is used, all the available 957 // profile information will be annotated in PreLink phase so there is 958 // no need to load the profile again in PostLink. 959 bool LoadSampleProfile = 960 HasSampleProfile && 961 !(FlattenedProfileUsed && Phase == ThinLTOPhase::PostLink); 962 963 // During the ThinLTO backend phase we perform early indirect call promotion 964 // here, before globalopt. Otherwise imported available_externally functions 965 // look unreferenced and are removed. If we are going to load the sample 966 // profile then defer until later. 967 // TODO: See if we can move later and consolidate with the location where 968 // we perform ICP when we are loading a sample profile. 969 // TODO: We pass HasSampleProfile (whether there was a sample profile file 970 // passed to the compile) to the SamplePGO flag of ICP. This is used to 971 // determine whether the new direct calls are annotated with prof metadata. 972 // Ideally this should be determined from whether the IR is annotated with 973 // sample profile, and not whether the a sample profile was provided on the 974 // command line. E.g. for flattened profiles where we will not be reloading 975 // the sample profile in the ThinLTO backend, we ideally shouldn't have to 976 // provide the sample profile file. 977 if (Phase == ThinLTOPhase::PostLink && !LoadSampleProfile) 978 MPM.addPass(PGOIndirectCallPromotion(true /* InLTO */, HasSampleProfile)); 979 980 // Do basic inference of function attributes from known properties of system 981 // libraries and other oracles. 982 MPM.addPass(InferFunctionAttrsPass()); 983 984 // Create an early function pass manager to cleanup the output of the 985 // frontend. 986 FunctionPassManager EarlyFPM(DebugLogging); 987 EarlyFPM.addPass(SimplifyCFGPass()); 988 EarlyFPM.addPass(SROA()); 989 EarlyFPM.addPass(EarlyCSEPass()); 990 EarlyFPM.addPass(LowerExpectIntrinsicPass()); 991 if (PTO.Coroutines) 992 EarlyFPM.addPass(CoroEarlyPass()); 993 if (Level == OptimizationLevel::O3) 994 EarlyFPM.addPass(CallSiteSplittingPass()); 995 996 // In SamplePGO ThinLTO backend, we need instcombine before profile annotation 997 // to convert bitcast to direct calls so that they can be inlined during the 998 // profile annotation prepration step. 999 // More details about SamplePGO design can be found in: 1000 // https://research.google.com/pubs/pub45290.html 1001 // FIXME: revisit how SampleProfileLoad/Inliner/ICP is structured. 1002 if (LoadSampleProfile) 1003 EarlyFPM.addPass(InstCombinePass()); 1004 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(EarlyFPM))); 1005 1006 if (LoadSampleProfile) { 1007 // Annotate sample profile right after early FPM to ensure freshness of 1008 // the debug info. 1009 MPM.addPass(SampleProfileLoaderPass(PGOOpt->ProfileFile, 1010 PGOOpt->ProfileRemappingFile, 1011 Phase == ThinLTOPhase::PreLink)); 1012 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 1013 // RequireAnalysisPass for PSI before subsequent non-module passes. 1014 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 1015 // Do not invoke ICP in the ThinLTOPrelink phase as it makes it hard 1016 // for the profile annotation to be accurate in the ThinLTO backend. 1017 if (Phase != ThinLTOPhase::PreLink) 1018 // We perform early indirect call promotion here, before globalopt. 1019 // This is important for the ThinLTO backend phase because otherwise 1020 // imported available_externally functions look unreferenced and are 1021 // removed. 1022 MPM.addPass(PGOIndirectCallPromotion(Phase == ThinLTOPhase::PostLink, 1023 true /* SamplePGO */)); 1024 } 1025 1026 if (AttributorRun & AttributorRunOption::MODULE) 1027 MPM.addPass(AttributorPass()); 1028 1029 // Lower type metadata and the type.test intrinsic in the ThinLTO 1030 // post link pipeline after ICP. This is to enable usage of the type 1031 // tests in ICP sequences. 1032 if (Phase == ThinLTOPhase::PostLink) 1033 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1034 1035 for (auto &C : PipelineEarlySimplificationEPCallbacks) 1036 C(MPM, Level); 1037 1038 // Interprocedural constant propagation now that basic cleanup has occurred 1039 // and prior to optimizing globals. 1040 // FIXME: This position in the pipeline hasn't been carefully considered in 1041 // years, it should be re-analyzed. 1042 MPM.addPass(IPSCCPPass()); 1043 1044 // Attach metadata to indirect call sites indicating the set of functions 1045 // they may target at run-time. This should follow IPSCCP. 1046 MPM.addPass(CalledValuePropagationPass()); 1047 1048 // Optimize globals to try and fold them into constants. 1049 MPM.addPass(GlobalOptPass()); 1050 1051 // Promote any localized globals to SSA registers. 1052 // FIXME: Should this instead by a run of SROA? 1053 // FIXME: We should probably run instcombine and simplify-cfg afterward to 1054 // delete control flows that are dead once globals have been folded to 1055 // constants. 1056 MPM.addPass(createModuleToFunctionPassAdaptor(PromotePass())); 1057 1058 // Remove any dead arguments exposed by cleanups and constant folding 1059 // globals. 1060 MPM.addPass(DeadArgumentEliminationPass()); 1061 1062 // Create a small function pass pipeline to cleanup after all the global 1063 // optimizations. 1064 FunctionPassManager GlobalCleanupPM(DebugLogging); 1065 GlobalCleanupPM.addPass(InstCombinePass()); 1066 invokePeepholeEPCallbacks(GlobalCleanupPM, Level); 1067 1068 GlobalCleanupPM.addPass(SimplifyCFGPass()); 1069 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(GlobalCleanupPM))); 1070 1071 // Add all the requested passes for instrumentation PGO, if requested. 1072 if (PGOOpt && Phase != ThinLTOPhase::PostLink && 1073 (PGOOpt->Action == PGOOptions::IRInstr || 1074 PGOOpt->Action == PGOOptions::IRUse)) { 1075 addPGOInstrPasses(MPM, Level, 1076 /* RunProfileGen */ PGOOpt->Action == PGOOptions::IRInstr, 1077 /* IsCS */ false, PGOOpt->ProfileFile, 1078 PGOOpt->ProfileRemappingFile); 1079 MPM.addPass(PGOIndirectCallPromotion(false, false)); 1080 } 1081 if (PGOOpt && Phase != ThinLTOPhase::PostLink && 1082 PGOOpt->CSAction == PGOOptions::CSIRInstr) 1083 MPM.addPass(PGOInstrumentationGenCreateVar(PGOOpt->CSProfileGenFile)); 1084 1085 // Synthesize function entry counts for non-PGO compilation. 1086 if (EnableSyntheticCounts && !PGOOpt) 1087 MPM.addPass(SyntheticCountsPropagation()); 1088 1089 MPM.addPass(buildInlinerPipeline(Level, Phase)); 1090 1091 if (EnableMemProfiler && Phase != ThinLTOPhase::PreLink) { 1092 MPM.addPass(createModuleToFunctionPassAdaptor(MemProfilerPass())); 1093 MPM.addPass(ModuleMemProfilerPass()); 1094 } 1095 1096 return MPM; 1097 } 1098 1099 ModulePassManager 1100 PassBuilder::buildModuleOptimizationPipeline(OptimizationLevel Level, 1101 bool LTOPreLink) { 1102 ModulePassManager MPM(DebugLogging); 1103 1104 // Optimize globals now that the module is fully simplified. 1105 MPM.addPass(GlobalOptPass()); 1106 MPM.addPass(GlobalDCEPass()); 1107 1108 // Run partial inlining pass to partially inline functions that have 1109 // large bodies. 1110 if (RunPartialInlining) 1111 MPM.addPass(PartialInlinerPass()); 1112 1113 // Remove avail extern fns and globals definitions since we aren't compiling 1114 // an object file for later LTO. For LTO we want to preserve these so they 1115 // are eligible for inlining at link-time. Note if they are unreferenced they 1116 // will be removed by GlobalDCE later, so this only impacts referenced 1117 // available externally globals. Eventually they will be suppressed during 1118 // codegen, but eliminating here enables more opportunity for GlobalDCE as it 1119 // may make globals referenced by available external functions dead and saves 1120 // running remaining passes on the eliminated functions. These should be 1121 // preserved during prelinking for link-time inlining decisions. 1122 if (!LTOPreLink) 1123 MPM.addPass(EliminateAvailableExternallyPass()); 1124 1125 if (EnableOrderFileInstrumentation) 1126 MPM.addPass(InstrOrderFilePass()); 1127 1128 // Do RPO function attribute inference across the module to forward-propagate 1129 // attributes where applicable. 1130 // FIXME: Is this really an optimization rather than a canonicalization? 1131 MPM.addPass(ReversePostOrderFunctionAttrsPass()); 1132 1133 // Do a post inline PGO instrumentation and use pass. This is a context 1134 // sensitive PGO pass. We don't want to do this in LTOPreLink phrase as 1135 // cross-module inline has not been done yet. The context sensitive 1136 // instrumentation is after all the inlines are done. 1137 if (!LTOPreLink && PGOOpt) { 1138 if (PGOOpt->CSAction == PGOOptions::CSIRInstr) 1139 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ true, 1140 /* IsCS */ true, PGOOpt->CSProfileGenFile, 1141 PGOOpt->ProfileRemappingFile); 1142 else if (PGOOpt->CSAction == PGOOptions::CSIRUse) 1143 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ false, 1144 /* IsCS */ true, PGOOpt->ProfileFile, 1145 PGOOpt->ProfileRemappingFile); 1146 } 1147 1148 // Re-require GloblasAA here prior to function passes. This is particularly 1149 // useful as the above will have inlined, DCE'ed, and function-attr 1150 // propagated everything. We should at this point have a reasonably minimal 1151 // and richly annotated call graph. By computing aliasing and mod/ref 1152 // information for all local globals here, the late loop passes and notably 1153 // the vectorizer will be able to use them to help recognize vectorizable 1154 // memory operations. 1155 MPM.addPass(RequireAnalysisPass<GlobalsAA, Module>()); 1156 1157 FunctionPassManager OptimizePM(DebugLogging); 1158 OptimizePM.addPass(Float2IntPass()); 1159 OptimizePM.addPass(LowerConstantIntrinsicsPass()); 1160 1161 if (EnableMatrix) { 1162 OptimizePM.addPass(LowerMatrixIntrinsicsPass()); 1163 OptimizePM.addPass(EarlyCSEPass()); 1164 } 1165 1166 // FIXME: We need to run some loop optimizations to re-rotate loops after 1167 // simplify-cfg and others undo their rotation. 1168 1169 // Optimize the loop execution. These passes operate on entire loop nests 1170 // rather than on each loop in an inside-out manner, and so they are actually 1171 // function passes. 1172 1173 for (auto &C : VectorizerStartEPCallbacks) 1174 C(OptimizePM, Level); 1175 1176 // First rotate loops that may have been un-rotated by prior passes. 1177 // Disable header duplication at -Oz. 1178 OptimizePM.addPass(createFunctionToLoopPassAdaptor( 1179 LoopRotatePass(Level != OptimizationLevel::Oz), EnableMSSALoopDependency, 1180 /*UseBlockFrequencyInfo=*/false, DebugLogging)); 1181 1182 // Distribute loops to allow partial vectorization. I.e. isolate dependences 1183 // into separate loop that would otherwise inhibit vectorization. This is 1184 // currently only performed for loops marked with the metadata 1185 // llvm.loop.distribute=true or when -enable-loop-distribute is specified. 1186 OptimizePM.addPass(LoopDistributePass()); 1187 1188 // Populates the VFABI attribute with the scalar-to-vector mappings 1189 // from the TargetLibraryInfo. 1190 OptimizePM.addPass(InjectTLIMappings()); 1191 1192 // Now run the core loop vectorizer. 1193 OptimizePM.addPass(LoopVectorizePass( 1194 LoopVectorizeOptions(!PTO.LoopInterleaving, !PTO.LoopVectorization))); 1195 1196 // Eliminate loads by forwarding stores from the previous iteration to loads 1197 // of the current iteration. 1198 OptimizePM.addPass(LoopLoadEliminationPass()); 1199 1200 // Cleanup after the loop optimization passes. 1201 OptimizePM.addPass(InstCombinePass()); 1202 1203 // Now that we've formed fast to execute loop structures, we do further 1204 // optimizations. These are run afterward as they might block doing complex 1205 // analyses and transforms such as what are needed for loop vectorization. 1206 1207 // Cleanup after loop vectorization, etc. Simplification passes like CVP and 1208 // GVN, loop transforms, and others have already run, so it's now better to 1209 // convert to more optimized IR using more aggressive simplify CFG options. 1210 // The extra sinking transform can create larger basic blocks, so do this 1211 // before SLP vectorization. 1212 // FIXME: study whether hoisting and/or sinking of common instructions should 1213 // be delayed until after SLP vectorizer. 1214 OptimizePM.addPass(SimplifyCFGPass(SimplifyCFGOptions() 1215 .forwardSwitchCondToPhi(true) 1216 .convertSwitchToLookupTable(true) 1217 .needCanonicalLoops(false) 1218 .hoistCommonInsts(true) 1219 .sinkCommonInsts(true))); 1220 1221 // Optimize parallel scalar instruction chains into SIMD instructions. 1222 if (PTO.SLPVectorization) 1223 OptimizePM.addPass(SLPVectorizerPass()); 1224 1225 // Enhance/cleanup vector code. 1226 OptimizePM.addPass(VectorCombinePass()); 1227 OptimizePM.addPass(InstCombinePass()); 1228 1229 // Unroll small loops to hide loop backedge latency and saturate any parallel 1230 // execution resources of an out-of-order processor. We also then need to 1231 // clean up redundancies and loop invariant code. 1232 // FIXME: It would be really good to use a loop-integrated instruction 1233 // combiner for cleanup here so that the unrolling and LICM can be pipelined 1234 // across the loop nests. 1235 // We do UnrollAndJam in a separate LPM to ensure it happens before unroll 1236 if (EnableUnrollAndJam && PTO.LoopUnrolling) { 1237 OptimizePM.addPass(LoopUnrollAndJamPass(Level.getSpeedupLevel())); 1238 } 1239 OptimizePM.addPass(LoopUnrollPass(LoopUnrollOptions( 1240 Level.getSpeedupLevel(), /*OnlyWhenForced=*/!PTO.LoopUnrolling, 1241 PTO.ForgetAllSCEVInLoopUnroll))); 1242 OptimizePM.addPass(WarnMissedTransformationsPass()); 1243 OptimizePM.addPass(InstCombinePass()); 1244 OptimizePM.addPass(RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 1245 OptimizePM.addPass(createFunctionToLoopPassAdaptor( 1246 LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap), 1247 EnableMSSALoopDependency, /*UseBlockFrequencyInfo=*/true, DebugLogging)); 1248 1249 // Now that we've vectorized and unrolled loops, we may have more refined 1250 // alignment information, try to re-derive it here. 1251 OptimizePM.addPass(AlignmentFromAssumptionsPass()); 1252 1253 // Split out cold code. Splitting is done late to avoid hiding context from 1254 // other optimizations and inadvertently regressing performance. The tradeoff 1255 // is that this has a higher code size cost than splitting early. 1256 if (EnableHotColdSplit && !LTOPreLink) 1257 MPM.addPass(HotColdSplittingPass()); 1258 1259 // LoopSink pass sinks instructions hoisted by LICM, which serves as a 1260 // canonicalization pass that enables other optimizations. As a result, 1261 // LoopSink pass needs to be a very late IR pass to avoid undoing LICM 1262 // result too early. 1263 OptimizePM.addPass(LoopSinkPass()); 1264 1265 // And finally clean up LCSSA form before generating code. 1266 OptimizePM.addPass(InstSimplifyPass()); 1267 1268 // This hoists/decomposes div/rem ops. It should run after other sink/hoist 1269 // passes to avoid re-sinking, but before SimplifyCFG because it can allow 1270 // flattening of blocks. 1271 OptimizePM.addPass(DivRemPairsPass()); 1272 1273 // LoopSink (and other loop passes since the last simplifyCFG) might have 1274 // resulted in single-entry-single-exit or empty blocks. Clean up the CFG. 1275 OptimizePM.addPass(SimplifyCFGPass()); 1276 1277 // Optimize PHIs by speculating around them when profitable. Note that this 1278 // pass needs to be run after any PRE or similar pass as it is essentially 1279 // inserting redundancies into the program. This even includes SimplifyCFG. 1280 OptimizePM.addPass(SpeculateAroundPHIsPass()); 1281 1282 if (PTO.Coroutines) 1283 OptimizePM.addPass(CoroCleanupPass()); 1284 1285 // Add the core optimizing pipeline. 1286 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(OptimizePM))); 1287 1288 for (auto &C : OptimizerLastEPCallbacks) 1289 C(MPM, Level); 1290 1291 if (PTO.CallGraphProfile) 1292 MPM.addPass(CGProfilePass()); 1293 1294 // Now we need to do some global optimization transforms. 1295 // FIXME: It would seem like these should come first in the optimization 1296 // pipeline and maybe be the bottom of the canonicalization pipeline? Weird 1297 // ordering here. 1298 MPM.addPass(GlobalDCEPass()); 1299 MPM.addPass(ConstantMergePass()); 1300 1301 return MPM; 1302 } 1303 1304 ModulePassManager 1305 PassBuilder::buildPerModuleDefaultPipeline(OptimizationLevel Level, 1306 bool LTOPreLink) { 1307 assert(Level != OptimizationLevel::O0 && 1308 "Must request optimizations for the default pipeline!"); 1309 1310 ModulePassManager MPM(DebugLogging); 1311 1312 // Convert @llvm.global.annotations to !annotation metadata. 1313 MPM.addPass(Annotation2MetadataPass()); 1314 1315 // Force any function attributes we want the rest of the pipeline to observe. 1316 MPM.addPass(ForceFunctionAttrsPass()); 1317 1318 // Apply module pipeline start EP callback. 1319 for (auto &C : PipelineStartEPCallbacks) 1320 C(MPM, Level); 1321 1322 if (PGOOpt && PGOOpt->SamplePGOSupport) 1323 MPM.addPass(createModuleToFunctionPassAdaptor(AddDiscriminatorsPass())); 1324 1325 // Add the core simplification pipeline. 1326 MPM.addPass(buildModuleSimplificationPipeline(Level, ThinLTOPhase::None)); 1327 1328 // Now add the optimization pipeline. 1329 MPM.addPass(buildModuleOptimizationPipeline(Level, LTOPreLink)); 1330 1331 // Emit annotation remarks. 1332 addAnnotationRemarksPass(MPM); 1333 1334 if (LTOPreLink) 1335 addRequiredLTOPreLinkPasses(MPM); 1336 1337 return MPM; 1338 } 1339 1340 ModulePassManager 1341 PassBuilder::buildThinLTOPreLinkDefaultPipeline(OptimizationLevel Level) { 1342 assert(Level != OptimizationLevel::O0 && 1343 "Must request optimizations for the default pipeline!"); 1344 1345 ModulePassManager MPM(DebugLogging); 1346 1347 // Convert @llvm.global.annotations to !annotation metadata. 1348 MPM.addPass(Annotation2MetadataPass()); 1349 1350 // Force any function attributes we want the rest of the pipeline to observe. 1351 MPM.addPass(ForceFunctionAttrsPass()); 1352 1353 if (PGOOpt && PGOOpt->SamplePGOSupport) 1354 MPM.addPass(createModuleToFunctionPassAdaptor(AddDiscriminatorsPass())); 1355 1356 // Apply module pipeline start EP callback. 1357 for (auto &C : PipelineStartEPCallbacks) 1358 C(MPM, Level); 1359 1360 // If we are planning to perform ThinLTO later, we don't bloat the code with 1361 // unrolling/vectorization/... now. Just simplify the module as much as we 1362 // can. 1363 MPM.addPass(buildModuleSimplificationPipeline(Level, ThinLTOPhase::PreLink)); 1364 1365 // Run partial inlining pass to partially inline functions that have 1366 // large bodies. 1367 // FIXME: It isn't clear whether this is really the right place to run this 1368 // in ThinLTO. Because there is another canonicalization and simplification 1369 // phase that will run after the thin link, running this here ends up with 1370 // less information than will be available later and it may grow functions in 1371 // ways that aren't beneficial. 1372 if (RunPartialInlining) 1373 MPM.addPass(PartialInlinerPass()); 1374 1375 // Reduce the size of the IR as much as possible. 1376 MPM.addPass(GlobalOptPass()); 1377 1378 // Module simplification splits coroutines, but does not fully clean up 1379 // coroutine intrinsics. To ensure ThinLTO optimization passes don't trip up 1380 // on these, we schedule the cleanup here. 1381 if (PTO.Coroutines) 1382 MPM.addPass(createModuleToFunctionPassAdaptor(CoroCleanupPass())); 1383 1384 // Emit annotation remarks. 1385 addAnnotationRemarksPass(MPM); 1386 1387 addRequiredLTOPreLinkPasses(MPM); 1388 1389 return MPM; 1390 } 1391 1392 ModulePassManager PassBuilder::buildThinLTODefaultPipeline( 1393 OptimizationLevel Level, const ModuleSummaryIndex *ImportSummary) { 1394 ModulePassManager MPM(DebugLogging); 1395 1396 // Convert @llvm.global.annotations to !annotation metadata. 1397 MPM.addPass(Annotation2MetadataPass()); 1398 1399 if (ImportSummary) { 1400 // These passes import type identifier resolutions for whole-program 1401 // devirtualization and CFI. They must run early because other passes may 1402 // disturb the specific instruction patterns that these passes look for, 1403 // creating dependencies on resolutions that may not appear in the summary. 1404 // 1405 // For example, GVN may transform the pattern assume(type.test) appearing in 1406 // two basic blocks into assume(phi(type.test, type.test)), which would 1407 // transform a dependency on a WPD resolution into a dependency on a type 1408 // identifier resolution for CFI. 1409 // 1410 // Also, WPD has access to more precise information than ICP and can 1411 // devirtualize more effectively, so it should operate on the IR first. 1412 // 1413 // The WPD and LowerTypeTest passes need to run at -O0 to lower type 1414 // metadata and intrinsics. 1415 MPM.addPass(WholeProgramDevirtPass(nullptr, ImportSummary)); 1416 MPM.addPass(LowerTypeTestsPass(nullptr, ImportSummary)); 1417 } 1418 1419 if (Level == OptimizationLevel::O0) 1420 return MPM; 1421 1422 // Force any function attributes we want the rest of the pipeline to observe. 1423 MPM.addPass(ForceFunctionAttrsPass()); 1424 1425 // Add the core simplification pipeline. 1426 MPM.addPass(buildModuleSimplificationPipeline(Level, ThinLTOPhase::PostLink)); 1427 1428 // Now add the optimization pipeline. 1429 MPM.addPass(buildModuleOptimizationPipeline(Level)); 1430 1431 // Emit annotation remarks. 1432 addAnnotationRemarksPass(MPM); 1433 1434 return MPM; 1435 } 1436 1437 ModulePassManager 1438 PassBuilder::buildLTOPreLinkDefaultPipeline(OptimizationLevel Level) { 1439 assert(Level != OptimizationLevel::O0 && 1440 "Must request optimizations for the default pipeline!"); 1441 // FIXME: We should use a customized pre-link pipeline! 1442 return buildPerModuleDefaultPipeline(Level, 1443 /* LTOPreLink */ true); 1444 } 1445 1446 ModulePassManager 1447 PassBuilder::buildLTODefaultPipeline(OptimizationLevel Level, 1448 ModuleSummaryIndex *ExportSummary) { 1449 ModulePassManager MPM(DebugLogging); 1450 1451 // Convert @llvm.global.annotations to !annotation metadata. 1452 MPM.addPass(Annotation2MetadataPass()); 1453 1454 if (Level == OptimizationLevel::O0) { 1455 // The WPD and LowerTypeTest passes need to run at -O0 to lower type 1456 // metadata and intrinsics. 1457 MPM.addPass(WholeProgramDevirtPass(ExportSummary, nullptr)); 1458 MPM.addPass(LowerTypeTestsPass(ExportSummary, nullptr)); 1459 // Run a second time to clean up any type tests left behind by WPD for use 1460 // in ICP. 1461 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1462 1463 // Emit annotation remarks. 1464 addAnnotationRemarksPass(MPM); 1465 1466 return MPM; 1467 } 1468 1469 if (PGOOpt && PGOOpt->Action == PGOOptions::SampleUse) { 1470 // Load sample profile before running the LTO optimization pipeline. 1471 MPM.addPass(SampleProfileLoaderPass(PGOOpt->ProfileFile, 1472 PGOOpt->ProfileRemappingFile, 1473 false /* ThinLTOPhase::PreLink */)); 1474 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 1475 // RequireAnalysisPass for PSI before subsequent non-module passes. 1476 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 1477 } 1478 1479 // Remove unused virtual tables to improve the quality of code generated by 1480 // whole-program devirtualization and bitset lowering. 1481 MPM.addPass(GlobalDCEPass()); 1482 1483 // Force any function attributes we want the rest of the pipeline to observe. 1484 MPM.addPass(ForceFunctionAttrsPass()); 1485 1486 // Do basic inference of function attributes from known properties of system 1487 // libraries and other oracles. 1488 MPM.addPass(InferFunctionAttrsPass()); 1489 1490 if (Level.getSpeedupLevel() > 1) { 1491 FunctionPassManager EarlyFPM(DebugLogging); 1492 EarlyFPM.addPass(CallSiteSplittingPass()); 1493 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(EarlyFPM))); 1494 1495 // Indirect call promotion. This should promote all the targets that are 1496 // left by the earlier promotion pass that promotes intra-module targets. 1497 // This two-step promotion is to save the compile time. For LTO, it should 1498 // produce the same result as if we only do promotion here. 1499 MPM.addPass(PGOIndirectCallPromotion( 1500 true /* InLTO */, PGOOpt && PGOOpt->Action == PGOOptions::SampleUse)); 1501 // Propagate constants at call sites into the functions they call. This 1502 // opens opportunities for globalopt (and inlining) by substituting function 1503 // pointers passed as arguments to direct uses of functions. 1504 MPM.addPass(IPSCCPPass()); 1505 1506 // Attach metadata to indirect call sites indicating the set of functions 1507 // they may target at run-time. This should follow IPSCCP. 1508 MPM.addPass(CalledValuePropagationPass()); 1509 } 1510 1511 // Now deduce any function attributes based in the current code. 1512 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor( 1513 PostOrderFunctionAttrsPass())); 1514 1515 // Do RPO function attribute inference across the module to forward-propagate 1516 // attributes where applicable. 1517 // FIXME: Is this really an optimization rather than a canonicalization? 1518 MPM.addPass(ReversePostOrderFunctionAttrsPass()); 1519 1520 // Use in-range annotations on GEP indices to split globals where beneficial. 1521 MPM.addPass(GlobalSplitPass()); 1522 1523 // Run whole program optimization of virtual call when the list of callees 1524 // is fixed. 1525 MPM.addPass(WholeProgramDevirtPass(ExportSummary, nullptr)); 1526 1527 // Stop here at -O1. 1528 if (Level == OptimizationLevel::O1) { 1529 // The LowerTypeTestsPass needs to run to lower type metadata and the 1530 // type.test intrinsics. The pass does nothing if CFI is disabled. 1531 MPM.addPass(LowerTypeTestsPass(ExportSummary, nullptr)); 1532 // Run a second time to clean up any type tests left behind by WPD for use 1533 // in ICP (which is performed earlier than this in the regular LTO 1534 // pipeline). 1535 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1536 1537 // Emit annotation remarks. 1538 addAnnotationRemarksPass(MPM); 1539 1540 return MPM; 1541 } 1542 1543 // Optimize globals to try and fold them into constants. 1544 MPM.addPass(GlobalOptPass()); 1545 1546 // Promote any localized globals to SSA registers. 1547 MPM.addPass(createModuleToFunctionPassAdaptor(PromotePass())); 1548 1549 // Linking modules together can lead to duplicate global constant, only 1550 // keep one copy of each constant. 1551 MPM.addPass(ConstantMergePass()); 1552 1553 // Remove unused arguments from functions. 1554 MPM.addPass(DeadArgumentEliminationPass()); 1555 1556 // Reduce the code after globalopt and ipsccp. Both can open up significant 1557 // simplification opportunities, and both can propagate functions through 1558 // function pointers. When this happens, we often have to resolve varargs 1559 // calls, etc, so let instcombine do this. 1560 FunctionPassManager PeepholeFPM(DebugLogging); 1561 if (Level == OptimizationLevel::O3) 1562 PeepholeFPM.addPass(AggressiveInstCombinePass()); 1563 PeepholeFPM.addPass(InstCombinePass()); 1564 invokePeepholeEPCallbacks(PeepholeFPM, Level); 1565 1566 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(PeepholeFPM))); 1567 1568 // Note: historically, the PruneEH pass was run first to deduce nounwind and 1569 // generally clean up exception handling overhead. It isn't clear this is 1570 // valuable as the inliner doesn't currently care whether it is inlining an 1571 // invoke or a call. 1572 // Run the inliner now. 1573 MPM.addPass(ModuleInlinerWrapperPass(getInlineParamsFromOptLevel(Level), 1574 DebugLogging)); 1575 1576 // Optimize globals again after we ran the inliner. 1577 MPM.addPass(GlobalOptPass()); 1578 1579 // Garbage collect dead functions. 1580 // FIXME: Add ArgumentPromotion pass after once it's ported. 1581 MPM.addPass(GlobalDCEPass()); 1582 1583 FunctionPassManager FPM(DebugLogging); 1584 // The IPO Passes may leave cruft around. Clean up after them. 1585 FPM.addPass(InstCombinePass()); 1586 invokePeepholeEPCallbacks(FPM, Level); 1587 1588 FPM.addPass(JumpThreadingPass(/*InsertFreezeWhenUnfoldingSelect*/ true)); 1589 1590 // Do a post inline PGO instrumentation and use pass. This is a context 1591 // sensitive PGO pass. 1592 if (PGOOpt) { 1593 if (PGOOpt->CSAction == PGOOptions::CSIRInstr) 1594 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ true, 1595 /* IsCS */ true, PGOOpt->CSProfileGenFile, 1596 PGOOpt->ProfileRemappingFile); 1597 else if (PGOOpt->CSAction == PGOOptions::CSIRUse) 1598 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ false, 1599 /* IsCS */ true, PGOOpt->ProfileFile, 1600 PGOOpt->ProfileRemappingFile); 1601 } 1602 1603 // Break up allocas 1604 FPM.addPass(SROA()); 1605 1606 // LTO provides additional opportunities for tailcall elimination due to 1607 // link-time inlining, and visibility of nocapture attribute. 1608 FPM.addPass(TailCallElimPass()); 1609 1610 // Run a few AA driver optimizations here and now to cleanup the code. 1611 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 1612 1613 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor( 1614 PostOrderFunctionAttrsPass())); 1615 // FIXME: here we run IP alias analysis in the legacy PM. 1616 1617 FunctionPassManager MainFPM; 1618 1619 // FIXME: once we fix LoopPass Manager, add LICM here. 1620 // FIXME: once we provide support for enabling MLSM, add it here. 1621 if (RunNewGVN) 1622 MainFPM.addPass(NewGVNPass()); 1623 else 1624 MainFPM.addPass(GVN()); 1625 1626 // Remove dead memcpy()'s. 1627 MainFPM.addPass(MemCpyOptPass()); 1628 1629 // Nuke dead stores. 1630 MainFPM.addPass(DSEPass()); 1631 1632 // FIXME: at this point, we run a bunch of loop passes: 1633 // indVarSimplify, loopDeletion, loopInterchange, loopUnroll, 1634 // loopVectorize. Enable them once the remaining issue with LPM 1635 // are sorted out. 1636 1637 MainFPM.addPass(InstCombinePass()); 1638 MainFPM.addPass(SimplifyCFGPass()); 1639 MainFPM.addPass(SCCPPass()); 1640 MainFPM.addPass(InstCombinePass()); 1641 MainFPM.addPass(BDCEPass()); 1642 1643 // FIXME: We may want to run SLPVectorizer here. 1644 // After vectorization, assume intrinsics may tell us more 1645 // about pointer alignments. 1646 #if 0 1647 MainFPM.add(AlignmentFromAssumptionsPass()); 1648 #endif 1649 1650 // FIXME: Conditionally run LoadCombine here, after it's ported 1651 // (in case we still have this pass, given its questionable usefulness). 1652 1653 MainFPM.addPass(InstCombinePass()); 1654 invokePeepholeEPCallbacks(MainFPM, Level); 1655 MainFPM.addPass(JumpThreadingPass(/*InsertFreezeWhenUnfoldingSelect*/ true)); 1656 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(MainFPM))); 1657 1658 // Create a function that performs CFI checks for cross-DSO calls with 1659 // targets in the current module. 1660 MPM.addPass(CrossDSOCFIPass()); 1661 1662 // Lower type metadata and the type.test intrinsic. This pass supports 1663 // clang's control flow integrity mechanisms (-fsanitize=cfi*) and needs 1664 // to be run at link time if CFI is enabled. This pass does nothing if 1665 // CFI is disabled. 1666 MPM.addPass(LowerTypeTestsPass(ExportSummary, nullptr)); 1667 // Run a second time to clean up any type tests left behind by WPD for use 1668 // in ICP (which is performed earlier than this in the regular LTO pipeline). 1669 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1670 1671 // Enable splitting late in the FullLTO post-link pipeline. This is done in 1672 // the same stage in the old pass manager (\ref addLateLTOOptimizationPasses). 1673 if (EnableHotColdSplit) 1674 MPM.addPass(HotColdSplittingPass()); 1675 1676 // Add late LTO optimization passes. 1677 // Delete basic blocks, which optimization passes may have killed. 1678 MPM.addPass(createModuleToFunctionPassAdaptor(SimplifyCFGPass())); 1679 1680 // Drop bodies of available eternally objects to improve GlobalDCE. 1681 MPM.addPass(EliminateAvailableExternallyPass()); 1682 1683 // Now that we have optimized the program, discard unreachable functions. 1684 MPM.addPass(GlobalDCEPass()); 1685 1686 // Emit annotation remarks. 1687 addAnnotationRemarksPass(MPM); 1688 1689 // FIXME: Maybe enable MergeFuncs conditionally after it's ported. 1690 return MPM; 1691 } 1692 1693 ModulePassManager PassBuilder::buildO0DefaultPipeline(OptimizationLevel Level, 1694 bool LTOPreLink) { 1695 assert(Level == OptimizationLevel::O0 && 1696 "buildO0DefaultPipeline should only be used with O0"); 1697 1698 ModulePassManager MPM(DebugLogging); 1699 1700 if (PGOOpt && (PGOOpt->Action == PGOOptions::IRInstr || 1701 PGOOpt->Action == PGOOptions::IRUse)) 1702 addPGOInstrPassesForO0( 1703 MPM, 1704 /* RunProfileGen */ (PGOOpt->Action == PGOOptions::IRInstr), 1705 /* IsCS */ false, PGOOpt->ProfileFile, PGOOpt->ProfileRemappingFile); 1706 1707 for (auto &C : PipelineStartEPCallbacks) 1708 C(MPM, Level); 1709 for (auto &C : PipelineEarlySimplificationEPCallbacks) 1710 C(MPM, Level); 1711 1712 // Build a minimal pipeline based on the semantics required by LLVM, 1713 // which is just that always inlining occurs. Further, disable generating 1714 // lifetime intrinsics to avoid enabling further optimizations during 1715 // code generation. 1716 // However, we need to insert lifetime intrinsics to avoid invalid access 1717 // caused by multithreaded coroutines. 1718 MPM.addPass(AlwaysInlinerPass( 1719 /*InsertLifetimeIntrinsics=*/PTO.Coroutines)); 1720 1721 if (EnableMatrix) 1722 MPM.addPass( 1723 createModuleToFunctionPassAdaptor(LowerMatrixIntrinsicsPass(true))); 1724 1725 if (!LateLoopOptimizationsEPCallbacks.empty()) { 1726 LoopPassManager LPM(DebugLogging); 1727 for (auto &C : LateLoopOptimizationsEPCallbacks) 1728 C(LPM, Level); 1729 if (!LPM.isEmpty()) { 1730 MPM.addPass(createModuleToFunctionPassAdaptor( 1731 createFunctionToLoopPassAdaptor(std::move(LPM)))); 1732 } 1733 } 1734 if (!LoopOptimizerEndEPCallbacks.empty()) { 1735 LoopPassManager LPM(DebugLogging); 1736 for (auto &C : LoopOptimizerEndEPCallbacks) 1737 C(LPM, Level); 1738 if (!LPM.isEmpty()) { 1739 MPM.addPass(createModuleToFunctionPassAdaptor( 1740 createFunctionToLoopPassAdaptor(std::move(LPM)))); 1741 } 1742 } 1743 if (!ScalarOptimizerLateEPCallbacks.empty()) { 1744 FunctionPassManager FPM(DebugLogging); 1745 for (auto &C : ScalarOptimizerLateEPCallbacks) 1746 C(FPM, Level); 1747 if (!FPM.isEmpty()) 1748 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 1749 } 1750 if (!CGSCCOptimizerLateEPCallbacks.empty()) { 1751 CGSCCPassManager CGPM(DebugLogging); 1752 for (auto &C : CGSCCOptimizerLateEPCallbacks) 1753 C(CGPM, Level); 1754 if (!CGPM.isEmpty()) 1755 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPM))); 1756 } 1757 if (!VectorizerStartEPCallbacks.empty()) { 1758 FunctionPassManager FPM(DebugLogging); 1759 for (auto &C : VectorizerStartEPCallbacks) 1760 C(FPM, Level); 1761 if (!FPM.isEmpty()) 1762 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 1763 } 1764 1765 if (PTO.Coroutines) { 1766 MPM.addPass(createModuleToFunctionPassAdaptor(CoroEarlyPass())); 1767 1768 CGSCCPassManager CGPM(DebugLogging); 1769 CGPM.addPass(CoroSplitPass()); 1770 CGPM.addPass(createCGSCCToFunctionPassAdaptor(CoroElidePass())); 1771 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPM))); 1772 1773 MPM.addPass(createModuleToFunctionPassAdaptor(CoroCleanupPass())); 1774 } 1775 1776 for (auto &C : OptimizerLastEPCallbacks) 1777 C(MPM, Level); 1778 1779 if (LTOPreLink) 1780 addRequiredLTOPreLinkPasses(MPM); 1781 1782 return MPM; 1783 } 1784 1785 AAManager PassBuilder::buildDefaultAAPipeline() { 1786 AAManager AA; 1787 1788 // The order in which these are registered determines their priority when 1789 // being queried. 1790 1791 // First we register the basic alias analysis that provides the majority of 1792 // per-function local AA logic. This is a stateless, on-demand local set of 1793 // AA techniques. 1794 AA.registerFunctionAnalysis<BasicAA>(); 1795 1796 // Next we query fast, specialized alias analyses that wrap IR-embedded 1797 // information about aliasing. 1798 AA.registerFunctionAnalysis<ScopedNoAliasAA>(); 1799 AA.registerFunctionAnalysis<TypeBasedAA>(); 1800 1801 // Add support for querying global aliasing information when available. 1802 // Because the `AAManager` is a function analysis and `GlobalsAA` is a module 1803 // analysis, all that the `AAManager` can do is query for any *cached* 1804 // results from `GlobalsAA` through a readonly proxy. 1805 AA.registerModuleAnalysis<GlobalsAA>(); 1806 1807 return AA; 1808 } 1809 1810 static Optional<int> parseRepeatPassName(StringRef Name) { 1811 if (!Name.consume_front("repeat<") || !Name.consume_back(">")) 1812 return None; 1813 int Count; 1814 if (Name.getAsInteger(0, Count) || Count <= 0) 1815 return None; 1816 return Count; 1817 } 1818 1819 static Optional<int> parseDevirtPassName(StringRef Name) { 1820 if (!Name.consume_front("devirt<") || !Name.consume_back(">")) 1821 return None; 1822 int Count; 1823 if (Name.getAsInteger(0, Count) || Count < 0) 1824 return None; 1825 return Count; 1826 } 1827 1828 static bool checkParametrizedPassName(StringRef Name, StringRef PassName) { 1829 if (!Name.consume_front(PassName)) 1830 return false; 1831 // normal pass name w/o parameters == default parameters 1832 if (Name.empty()) 1833 return true; 1834 return Name.startswith("<") && Name.endswith(">"); 1835 } 1836 1837 namespace { 1838 1839 /// This performs customized parsing of pass name with parameters. 1840 /// 1841 /// We do not need parametrization of passes in textual pipeline very often, 1842 /// yet on a rare occasion ability to specify parameters right there can be 1843 /// useful. 1844 /// 1845 /// \p Name - parameterized specification of a pass from a textual pipeline 1846 /// is a string in a form of : 1847 /// PassName '<' parameter-list '>' 1848 /// 1849 /// Parameter list is being parsed by the parser callable argument, \p Parser, 1850 /// It takes a string-ref of parameters and returns either StringError or a 1851 /// parameter list in a form of a custom parameters type, all wrapped into 1852 /// Expected<> template class. 1853 /// 1854 template <typename ParametersParseCallableT> 1855 auto parsePassParameters(ParametersParseCallableT &&Parser, StringRef Name, 1856 StringRef PassName) -> decltype(Parser(StringRef{})) { 1857 using ParametersT = typename decltype(Parser(StringRef{}))::value_type; 1858 1859 StringRef Params = Name; 1860 if (!Params.consume_front(PassName)) { 1861 assert(false && 1862 "unable to strip pass name from parametrized pass specification"); 1863 } 1864 if (Params.empty()) 1865 return ParametersT{}; 1866 if (!Params.consume_front("<") || !Params.consume_back(">")) { 1867 assert(false && "invalid format for parametrized pass name"); 1868 } 1869 1870 Expected<ParametersT> Result = Parser(Params); 1871 assert((Result || Result.template errorIsA<StringError>()) && 1872 "Pass parameter parser can only return StringErrors."); 1873 return Result; 1874 } 1875 1876 /// Parser of parameters for LoopUnroll pass. 1877 Expected<LoopUnrollOptions> parseLoopUnrollOptions(StringRef Params) { 1878 LoopUnrollOptions UnrollOpts; 1879 while (!Params.empty()) { 1880 StringRef ParamName; 1881 std::tie(ParamName, Params) = Params.split(';'); 1882 int OptLevel = StringSwitch<int>(ParamName) 1883 .Case("O0", 0) 1884 .Case("O1", 1) 1885 .Case("O2", 2) 1886 .Case("O3", 3) 1887 .Default(-1); 1888 if (OptLevel >= 0) { 1889 UnrollOpts.setOptLevel(OptLevel); 1890 continue; 1891 } 1892 if (ParamName.consume_front("full-unroll-max=")) { 1893 int Count; 1894 if (ParamName.getAsInteger(0, Count)) 1895 return make_error<StringError>( 1896 formatv("invalid LoopUnrollPass parameter '{0}' ", ParamName).str(), 1897 inconvertibleErrorCode()); 1898 UnrollOpts.setFullUnrollMaxCount(Count); 1899 continue; 1900 } 1901 1902 bool Enable = !ParamName.consume_front("no-"); 1903 if (ParamName == "partial") { 1904 UnrollOpts.setPartial(Enable); 1905 } else if (ParamName == "peeling") { 1906 UnrollOpts.setPeeling(Enable); 1907 } else if (ParamName == "profile-peeling") { 1908 UnrollOpts.setProfileBasedPeeling(Enable); 1909 } else if (ParamName == "runtime") { 1910 UnrollOpts.setRuntime(Enable); 1911 } else if (ParamName == "upperbound") { 1912 UnrollOpts.setUpperBound(Enable); 1913 } else { 1914 return make_error<StringError>( 1915 formatv("invalid LoopUnrollPass parameter '{0}' ", ParamName).str(), 1916 inconvertibleErrorCode()); 1917 } 1918 } 1919 return UnrollOpts; 1920 } 1921 1922 Expected<MemorySanitizerOptions> parseMSanPassOptions(StringRef Params) { 1923 MemorySanitizerOptions Result; 1924 while (!Params.empty()) { 1925 StringRef ParamName; 1926 std::tie(ParamName, Params) = Params.split(';'); 1927 1928 if (ParamName == "recover") { 1929 Result.Recover = true; 1930 } else if (ParamName == "kernel") { 1931 Result.Kernel = true; 1932 } else if (ParamName.consume_front("track-origins=")) { 1933 if (ParamName.getAsInteger(0, Result.TrackOrigins)) 1934 return make_error<StringError>( 1935 formatv("invalid argument to MemorySanitizer pass track-origins " 1936 "parameter: '{0}' ", 1937 ParamName) 1938 .str(), 1939 inconvertibleErrorCode()); 1940 } else { 1941 return make_error<StringError>( 1942 formatv("invalid MemorySanitizer pass parameter '{0}' ", ParamName) 1943 .str(), 1944 inconvertibleErrorCode()); 1945 } 1946 } 1947 return Result; 1948 } 1949 1950 /// Parser of parameters for SimplifyCFG pass. 1951 Expected<SimplifyCFGOptions> parseSimplifyCFGOptions(StringRef Params) { 1952 SimplifyCFGOptions Result; 1953 while (!Params.empty()) { 1954 StringRef ParamName; 1955 std::tie(ParamName, Params) = Params.split(';'); 1956 1957 bool Enable = !ParamName.consume_front("no-"); 1958 if (ParamName == "forward-switch-cond") { 1959 Result.forwardSwitchCondToPhi(Enable); 1960 } else if (ParamName == "switch-to-lookup") { 1961 Result.convertSwitchToLookupTable(Enable); 1962 } else if (ParamName == "keep-loops") { 1963 Result.needCanonicalLoops(Enable); 1964 } else if (ParamName == "hoist-common-insts") { 1965 Result.hoistCommonInsts(Enable); 1966 } else if (ParamName == "sink-common-insts") { 1967 Result.sinkCommonInsts(Enable); 1968 } else if (Enable && ParamName.consume_front("bonus-inst-threshold=")) { 1969 APInt BonusInstThreshold; 1970 if (ParamName.getAsInteger(0, BonusInstThreshold)) 1971 return make_error<StringError>( 1972 formatv("invalid argument to SimplifyCFG pass bonus-threshold " 1973 "parameter: '{0}' ", 1974 ParamName).str(), 1975 inconvertibleErrorCode()); 1976 Result.bonusInstThreshold(BonusInstThreshold.getSExtValue()); 1977 } else { 1978 return make_error<StringError>( 1979 formatv("invalid SimplifyCFG pass parameter '{0}' ", ParamName).str(), 1980 inconvertibleErrorCode()); 1981 } 1982 } 1983 return Result; 1984 } 1985 1986 /// Parser of parameters for LoopVectorize pass. 1987 Expected<LoopVectorizeOptions> parseLoopVectorizeOptions(StringRef Params) { 1988 LoopVectorizeOptions Opts; 1989 while (!Params.empty()) { 1990 StringRef ParamName; 1991 std::tie(ParamName, Params) = Params.split(';'); 1992 1993 bool Enable = !ParamName.consume_front("no-"); 1994 if (ParamName == "interleave-forced-only") { 1995 Opts.setInterleaveOnlyWhenForced(Enable); 1996 } else if (ParamName == "vectorize-forced-only") { 1997 Opts.setVectorizeOnlyWhenForced(Enable); 1998 } else { 1999 return make_error<StringError>( 2000 formatv("invalid LoopVectorize parameter '{0}' ", ParamName).str(), 2001 inconvertibleErrorCode()); 2002 } 2003 } 2004 return Opts; 2005 } 2006 2007 Expected<bool> parseLoopUnswitchOptions(StringRef Params) { 2008 bool Result = false; 2009 while (!Params.empty()) { 2010 StringRef ParamName; 2011 std::tie(ParamName, Params) = Params.split(';'); 2012 2013 bool Enable = !ParamName.consume_front("no-"); 2014 if (ParamName == "nontrivial") { 2015 Result = Enable; 2016 } else { 2017 return make_error<StringError>( 2018 formatv("invalid LoopUnswitch pass parameter '{0}' ", ParamName) 2019 .str(), 2020 inconvertibleErrorCode()); 2021 } 2022 } 2023 return Result; 2024 } 2025 2026 Expected<bool> parseMergedLoadStoreMotionOptions(StringRef Params) { 2027 bool Result = false; 2028 while (!Params.empty()) { 2029 StringRef ParamName; 2030 std::tie(ParamName, Params) = Params.split(';'); 2031 2032 bool Enable = !ParamName.consume_front("no-"); 2033 if (ParamName == "split-footer-bb") { 2034 Result = Enable; 2035 } else { 2036 return make_error<StringError>( 2037 formatv("invalid MergedLoadStoreMotion pass parameter '{0}' ", 2038 ParamName) 2039 .str(), 2040 inconvertibleErrorCode()); 2041 } 2042 } 2043 return Result; 2044 } 2045 2046 Expected<GVNOptions> parseGVNOptions(StringRef Params) { 2047 GVNOptions Result; 2048 while (!Params.empty()) { 2049 StringRef ParamName; 2050 std::tie(ParamName, Params) = Params.split(';'); 2051 2052 bool Enable = !ParamName.consume_front("no-"); 2053 if (ParamName == "pre") { 2054 Result.setPRE(Enable); 2055 } else if (ParamName == "load-pre") { 2056 Result.setLoadPRE(Enable); 2057 } else if (ParamName == "split-backedge-load-pre") { 2058 Result.setLoadPRESplitBackedge(Enable); 2059 } else if (ParamName == "memdep") { 2060 Result.setMemDep(Enable); 2061 } else { 2062 return make_error<StringError>( 2063 formatv("invalid GVN pass parameter '{0}' ", ParamName).str(), 2064 inconvertibleErrorCode()); 2065 } 2066 } 2067 return Result; 2068 } 2069 2070 Expected<StackLifetime::LivenessType> 2071 parseStackLifetimeOptions(StringRef Params) { 2072 StackLifetime::LivenessType Result = StackLifetime::LivenessType::May; 2073 while (!Params.empty()) { 2074 StringRef ParamName; 2075 std::tie(ParamName, Params) = Params.split(';'); 2076 2077 if (ParamName == "may") { 2078 Result = StackLifetime::LivenessType::May; 2079 } else if (ParamName == "must") { 2080 Result = StackLifetime::LivenessType::Must; 2081 } else { 2082 return make_error<StringError>( 2083 formatv("invalid StackLifetime parameter '{0}' ", ParamName).str(), 2084 inconvertibleErrorCode()); 2085 } 2086 } 2087 return Result; 2088 } 2089 2090 } // namespace 2091 2092 /// Tests whether a pass name starts with a valid prefix for a default pipeline 2093 /// alias. 2094 static bool startsWithDefaultPipelineAliasPrefix(StringRef Name) { 2095 return Name.startswith("default") || Name.startswith("thinlto") || 2096 Name.startswith("lto"); 2097 } 2098 2099 /// Tests whether registered callbacks will accept a given pass name. 2100 /// 2101 /// When parsing a pipeline text, the type of the outermost pipeline may be 2102 /// omitted, in which case the type is automatically determined from the first 2103 /// pass name in the text. This may be a name that is handled through one of the 2104 /// callbacks. We check this through the oridinary parsing callbacks by setting 2105 /// up a dummy PassManager in order to not force the client to also handle this 2106 /// type of query. 2107 template <typename PassManagerT, typename CallbacksT> 2108 static bool callbacksAcceptPassName(StringRef Name, CallbacksT &Callbacks) { 2109 if (!Callbacks.empty()) { 2110 PassManagerT DummyPM; 2111 for (auto &CB : Callbacks) 2112 if (CB(Name, DummyPM, {})) 2113 return true; 2114 } 2115 return false; 2116 } 2117 2118 template <typename CallbacksT> 2119 static bool isModulePassName(StringRef Name, CallbacksT &Callbacks) { 2120 // Manually handle aliases for pre-configured pipeline fragments. 2121 if (startsWithDefaultPipelineAliasPrefix(Name)) 2122 return DefaultAliasRegex.match(Name); 2123 2124 // Explicitly handle pass manager names. 2125 if (Name == "module") 2126 return true; 2127 if (Name == "cgscc") 2128 return true; 2129 if (Name == "function") 2130 return true; 2131 2132 // Explicitly handle custom-parsed pass names. 2133 if (parseRepeatPassName(Name)) 2134 return true; 2135 2136 #define MODULE_PASS(NAME, CREATE_PASS) \ 2137 if (Name == NAME) \ 2138 return true; 2139 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 2140 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2141 return true; 2142 #include "PassRegistry.def" 2143 2144 return callbacksAcceptPassName<ModulePassManager>(Name, Callbacks); 2145 } 2146 2147 template <typename CallbacksT> 2148 static bool isCGSCCPassName(StringRef Name, CallbacksT &Callbacks) { 2149 // Explicitly handle pass manager names. 2150 if (Name == "cgscc") 2151 return true; 2152 if (Name == "function") 2153 return true; 2154 2155 // Explicitly handle custom-parsed pass names. 2156 if (parseRepeatPassName(Name)) 2157 return true; 2158 if (parseDevirtPassName(Name)) 2159 return true; 2160 2161 #define CGSCC_PASS(NAME, CREATE_PASS) \ 2162 if (Name == NAME) \ 2163 return true; 2164 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 2165 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2166 return true; 2167 #include "PassRegistry.def" 2168 2169 return callbacksAcceptPassName<CGSCCPassManager>(Name, Callbacks); 2170 } 2171 2172 template <typename CallbacksT> 2173 static bool isFunctionPassName(StringRef Name, CallbacksT &Callbacks) { 2174 // Explicitly handle pass manager names. 2175 if (Name == "function") 2176 return true; 2177 if (Name == "loop" || Name == "loop-mssa") 2178 return true; 2179 2180 // Explicitly handle custom-parsed pass names. 2181 if (parseRepeatPassName(Name)) 2182 return true; 2183 2184 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2185 if (Name == NAME) \ 2186 return true; 2187 #define FUNCTION_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2188 if (checkParametrizedPassName(Name, NAME)) \ 2189 return true; 2190 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 2191 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2192 return true; 2193 #include "PassRegistry.def" 2194 2195 return callbacksAcceptPassName<FunctionPassManager>(Name, Callbacks); 2196 } 2197 2198 template <typename CallbacksT> 2199 static bool isLoopPassName(StringRef Name, CallbacksT &Callbacks) { 2200 // Explicitly handle pass manager names. 2201 if (Name == "loop" || Name == "loop-mssa") 2202 return true; 2203 2204 // Explicitly handle custom-parsed pass names. 2205 if (parseRepeatPassName(Name)) 2206 return true; 2207 2208 #define LOOP_PASS(NAME, CREATE_PASS) \ 2209 if (Name == NAME) \ 2210 return true; 2211 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2212 if (checkParametrizedPassName(Name, NAME)) \ 2213 return true; 2214 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 2215 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2216 return true; 2217 #include "PassRegistry.def" 2218 2219 return callbacksAcceptPassName<LoopPassManager>(Name, Callbacks); 2220 } 2221 2222 Optional<std::vector<PassBuilder::PipelineElement>> 2223 PassBuilder::parsePipelineText(StringRef Text) { 2224 std::vector<PipelineElement> ResultPipeline; 2225 2226 SmallVector<std::vector<PipelineElement> *, 4> PipelineStack = { 2227 &ResultPipeline}; 2228 for (;;) { 2229 std::vector<PipelineElement> &Pipeline = *PipelineStack.back(); 2230 size_t Pos = Text.find_first_of(",()"); 2231 Pipeline.push_back({Text.substr(0, Pos), {}}); 2232 2233 // If we have a single terminating name, we're done. 2234 if (Pos == Text.npos) 2235 break; 2236 2237 char Sep = Text[Pos]; 2238 Text = Text.substr(Pos + 1); 2239 if (Sep == ',') 2240 // Just a name ending in a comma, continue. 2241 continue; 2242 2243 if (Sep == '(') { 2244 // Push the inner pipeline onto the stack to continue processing. 2245 PipelineStack.push_back(&Pipeline.back().InnerPipeline); 2246 continue; 2247 } 2248 2249 assert(Sep == ')' && "Bogus separator!"); 2250 // When handling the close parenthesis, we greedily consume them to avoid 2251 // empty strings in the pipeline. 2252 do { 2253 // If we try to pop the outer pipeline we have unbalanced parentheses. 2254 if (PipelineStack.size() == 1) 2255 return None; 2256 2257 PipelineStack.pop_back(); 2258 } while (Text.consume_front(")")); 2259 2260 // Check if we've finished parsing. 2261 if (Text.empty()) 2262 break; 2263 2264 // Otherwise, the end of an inner pipeline always has to be followed by 2265 // a comma, and then we can continue. 2266 if (!Text.consume_front(",")) 2267 return None; 2268 } 2269 2270 if (PipelineStack.size() > 1) 2271 // Unbalanced paretheses. 2272 return None; 2273 2274 assert(PipelineStack.back() == &ResultPipeline && 2275 "Wrong pipeline at the bottom of the stack!"); 2276 return {std::move(ResultPipeline)}; 2277 } 2278 2279 Error PassBuilder::parseModulePass(ModulePassManager &MPM, 2280 const PipelineElement &E) { 2281 auto &Name = E.Name; 2282 auto &InnerPipeline = E.InnerPipeline; 2283 2284 // First handle complex passes like the pass managers which carry pipelines. 2285 if (!InnerPipeline.empty()) { 2286 if (Name == "module") { 2287 ModulePassManager NestedMPM(DebugLogging); 2288 if (auto Err = parseModulePassPipeline(NestedMPM, InnerPipeline)) 2289 return Err; 2290 MPM.addPass(std::move(NestedMPM)); 2291 return Error::success(); 2292 } 2293 if (Name == "cgscc") { 2294 CGSCCPassManager CGPM(DebugLogging); 2295 if (auto Err = parseCGSCCPassPipeline(CGPM, InnerPipeline)) 2296 return Err; 2297 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPM))); 2298 return Error::success(); 2299 } 2300 if (Name == "function") { 2301 FunctionPassManager FPM(DebugLogging); 2302 if (auto Err = parseFunctionPassPipeline(FPM, InnerPipeline)) 2303 return Err; 2304 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 2305 return Error::success(); 2306 } 2307 if (auto Count = parseRepeatPassName(Name)) { 2308 ModulePassManager NestedMPM(DebugLogging); 2309 if (auto Err = parseModulePassPipeline(NestedMPM, InnerPipeline)) 2310 return Err; 2311 MPM.addPass(createRepeatedPass(*Count, std::move(NestedMPM))); 2312 return Error::success(); 2313 } 2314 2315 for (auto &C : ModulePipelineParsingCallbacks) 2316 if (C(Name, MPM, InnerPipeline)) 2317 return Error::success(); 2318 2319 // Normal passes can't have pipelines. 2320 return make_error<StringError>( 2321 formatv("invalid use of '{0}' pass as module pipeline", Name).str(), 2322 inconvertibleErrorCode()); 2323 ; 2324 } 2325 2326 // Manually handle aliases for pre-configured pipeline fragments. 2327 if (startsWithDefaultPipelineAliasPrefix(Name)) { 2328 SmallVector<StringRef, 3> Matches; 2329 if (!DefaultAliasRegex.match(Name, &Matches)) 2330 return make_error<StringError>( 2331 formatv("unknown default pipeline alias '{0}'", Name).str(), 2332 inconvertibleErrorCode()); 2333 2334 assert(Matches.size() == 3 && "Must capture two matched strings!"); 2335 2336 OptimizationLevel L = StringSwitch<OptimizationLevel>(Matches[2]) 2337 .Case("O0", OptimizationLevel::O0) 2338 .Case("O1", OptimizationLevel::O1) 2339 .Case("O2", OptimizationLevel::O2) 2340 .Case("O3", OptimizationLevel::O3) 2341 .Case("Os", OptimizationLevel::Os) 2342 .Case("Oz", OptimizationLevel::Oz); 2343 if (L == OptimizationLevel::O0 && Matches[1] != "thinlto" && 2344 Matches[1] != "lto") { 2345 MPM.addPass(buildO0DefaultPipeline(L, Matches[1] == "thinlto-pre-link" || 2346 Matches[1] == "lto-pre-link")); 2347 return Error::success(); 2348 } 2349 2350 // This is consistent with old pass manager invoked via opt, but 2351 // inconsistent with clang. Clang doesn't enable loop vectorization 2352 // but does enable slp vectorization at Oz. 2353 PTO.LoopVectorization = 2354 L.getSpeedupLevel() > 1 && L != OptimizationLevel::Oz; 2355 PTO.SLPVectorization = 2356 L.getSpeedupLevel() > 1 && L != OptimizationLevel::Oz; 2357 2358 if (Matches[1] == "default") { 2359 MPM.addPass(buildPerModuleDefaultPipeline(L)); 2360 } else if (Matches[1] == "thinlto-pre-link") { 2361 MPM.addPass(buildThinLTOPreLinkDefaultPipeline(L)); 2362 } else if (Matches[1] == "thinlto") { 2363 MPM.addPass(buildThinLTODefaultPipeline(L, nullptr)); 2364 } else if (Matches[1] == "lto-pre-link") { 2365 MPM.addPass(buildLTOPreLinkDefaultPipeline(L)); 2366 } else { 2367 assert(Matches[1] == "lto" && "Not one of the matched options!"); 2368 MPM.addPass(buildLTODefaultPipeline(L, nullptr)); 2369 } 2370 return Error::success(); 2371 } 2372 2373 // Finally expand the basic registered passes from the .inc file. 2374 #define MODULE_PASS(NAME, CREATE_PASS) \ 2375 if (Name == NAME) { \ 2376 MPM.addPass(CREATE_PASS); \ 2377 return Error::success(); \ 2378 } 2379 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 2380 if (Name == "require<" NAME ">") { \ 2381 MPM.addPass( \ 2382 RequireAnalysisPass< \ 2383 std::remove_reference<decltype(CREATE_PASS)>::type, Module>()); \ 2384 return Error::success(); \ 2385 } \ 2386 if (Name == "invalidate<" NAME ">") { \ 2387 MPM.addPass(InvalidateAnalysisPass< \ 2388 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2389 return Error::success(); \ 2390 } 2391 #define CGSCC_PASS(NAME, CREATE_PASS) \ 2392 if (Name == NAME) { \ 2393 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(CREATE_PASS)); \ 2394 return Error::success(); \ 2395 } 2396 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2397 if (Name == NAME) { \ 2398 MPM.addPass(createModuleToFunctionPassAdaptor(CREATE_PASS)); \ 2399 return Error::success(); \ 2400 } 2401 #define FUNCTION_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2402 if (checkParametrizedPassName(Name, NAME)) { \ 2403 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2404 if (!Params) \ 2405 return Params.takeError(); \ 2406 MPM.addPass(createModuleToFunctionPassAdaptor(CREATE_PASS(Params.get()))); \ 2407 return Error::success(); \ 2408 } 2409 #define LOOP_PASS(NAME, CREATE_PASS) \ 2410 if (Name == NAME) { \ 2411 MPM.addPass( \ 2412 createModuleToFunctionPassAdaptor(createFunctionToLoopPassAdaptor( \ 2413 CREATE_PASS, false, false, DebugLogging))); \ 2414 return Error::success(); \ 2415 } 2416 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2417 if (checkParametrizedPassName(Name, NAME)) { \ 2418 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2419 if (!Params) \ 2420 return Params.takeError(); \ 2421 MPM.addPass( \ 2422 createModuleToFunctionPassAdaptor(createFunctionToLoopPassAdaptor( \ 2423 CREATE_PASS(Params.get()), false, false, DebugLogging))); \ 2424 return Error::success(); \ 2425 } 2426 #include "PassRegistry.def" 2427 2428 for (auto &C : ModulePipelineParsingCallbacks) 2429 if (C(Name, MPM, InnerPipeline)) 2430 return Error::success(); 2431 return make_error<StringError>( 2432 formatv("unknown module pass '{0}'", Name).str(), 2433 inconvertibleErrorCode()); 2434 } 2435 2436 Error PassBuilder::parseCGSCCPass(CGSCCPassManager &CGPM, 2437 const PipelineElement &E) { 2438 auto &Name = E.Name; 2439 auto &InnerPipeline = E.InnerPipeline; 2440 2441 // First handle complex passes like the pass managers which carry pipelines. 2442 if (!InnerPipeline.empty()) { 2443 if (Name == "cgscc") { 2444 CGSCCPassManager NestedCGPM(DebugLogging); 2445 if (auto Err = parseCGSCCPassPipeline(NestedCGPM, InnerPipeline)) 2446 return Err; 2447 // Add the nested pass manager with the appropriate adaptor. 2448 CGPM.addPass(std::move(NestedCGPM)); 2449 return Error::success(); 2450 } 2451 if (Name == "function") { 2452 FunctionPassManager FPM(DebugLogging); 2453 if (auto Err = parseFunctionPassPipeline(FPM, InnerPipeline)) 2454 return Err; 2455 // Add the nested pass manager with the appropriate adaptor. 2456 CGPM.addPass(createCGSCCToFunctionPassAdaptor(std::move(FPM))); 2457 return Error::success(); 2458 } 2459 if (auto Count = parseRepeatPassName(Name)) { 2460 CGSCCPassManager NestedCGPM(DebugLogging); 2461 if (auto Err = parseCGSCCPassPipeline(NestedCGPM, InnerPipeline)) 2462 return Err; 2463 CGPM.addPass(createRepeatedPass(*Count, std::move(NestedCGPM))); 2464 return Error::success(); 2465 } 2466 if (auto MaxRepetitions = parseDevirtPassName(Name)) { 2467 CGSCCPassManager NestedCGPM(DebugLogging); 2468 if (auto Err = parseCGSCCPassPipeline(NestedCGPM, InnerPipeline)) 2469 return Err; 2470 CGPM.addPass( 2471 createDevirtSCCRepeatedPass(std::move(NestedCGPM), *MaxRepetitions)); 2472 return Error::success(); 2473 } 2474 2475 for (auto &C : CGSCCPipelineParsingCallbacks) 2476 if (C(Name, CGPM, InnerPipeline)) 2477 return Error::success(); 2478 2479 // Normal passes can't have pipelines. 2480 return make_error<StringError>( 2481 formatv("invalid use of '{0}' pass as cgscc pipeline", Name).str(), 2482 inconvertibleErrorCode()); 2483 } 2484 2485 // Now expand the basic registered passes from the .inc file. 2486 #define CGSCC_PASS(NAME, CREATE_PASS) \ 2487 if (Name == NAME) { \ 2488 CGPM.addPass(CREATE_PASS); \ 2489 return Error::success(); \ 2490 } 2491 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 2492 if (Name == "require<" NAME ">") { \ 2493 CGPM.addPass(RequireAnalysisPass< \ 2494 std::remove_reference<decltype(CREATE_PASS)>::type, \ 2495 LazyCallGraph::SCC, CGSCCAnalysisManager, LazyCallGraph &, \ 2496 CGSCCUpdateResult &>()); \ 2497 return Error::success(); \ 2498 } \ 2499 if (Name == "invalidate<" NAME ">") { \ 2500 CGPM.addPass(InvalidateAnalysisPass< \ 2501 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2502 return Error::success(); \ 2503 } 2504 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2505 if (Name == NAME) { \ 2506 CGPM.addPass(createCGSCCToFunctionPassAdaptor(CREATE_PASS)); \ 2507 return Error::success(); \ 2508 } 2509 #define FUNCTION_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2510 if (checkParametrizedPassName(Name, NAME)) { \ 2511 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2512 if (!Params) \ 2513 return Params.takeError(); \ 2514 CGPM.addPass(createCGSCCToFunctionPassAdaptor(CREATE_PASS(Params.get()))); \ 2515 return Error::success(); \ 2516 } 2517 #define LOOP_PASS(NAME, CREATE_PASS) \ 2518 if (Name == NAME) { \ 2519 CGPM.addPass( \ 2520 createCGSCCToFunctionPassAdaptor(createFunctionToLoopPassAdaptor( \ 2521 CREATE_PASS, false, false, DebugLogging))); \ 2522 return Error::success(); \ 2523 } 2524 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2525 if (checkParametrizedPassName(Name, NAME)) { \ 2526 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2527 if (!Params) \ 2528 return Params.takeError(); \ 2529 CGPM.addPass( \ 2530 createCGSCCToFunctionPassAdaptor(createFunctionToLoopPassAdaptor( \ 2531 CREATE_PASS(Params.get()), false, false, DebugLogging))); \ 2532 return Error::success(); \ 2533 } 2534 #include "PassRegistry.def" 2535 2536 for (auto &C : CGSCCPipelineParsingCallbacks) 2537 if (C(Name, CGPM, InnerPipeline)) 2538 return Error::success(); 2539 return make_error<StringError>( 2540 formatv("unknown cgscc pass '{0}'", Name).str(), 2541 inconvertibleErrorCode()); 2542 } 2543 2544 Error PassBuilder::parseFunctionPass(FunctionPassManager &FPM, 2545 const PipelineElement &E) { 2546 auto &Name = E.Name; 2547 auto &InnerPipeline = E.InnerPipeline; 2548 2549 // First handle complex passes like the pass managers which carry pipelines. 2550 if (!InnerPipeline.empty()) { 2551 if (Name == "function") { 2552 FunctionPassManager NestedFPM(DebugLogging); 2553 if (auto Err = parseFunctionPassPipeline(NestedFPM, InnerPipeline)) 2554 return Err; 2555 // Add the nested pass manager with the appropriate adaptor. 2556 FPM.addPass(std::move(NestedFPM)); 2557 return Error::success(); 2558 } 2559 if (Name == "loop" || Name == "loop-mssa") { 2560 LoopPassManager LPM(DebugLogging); 2561 if (auto Err = parseLoopPassPipeline(LPM, InnerPipeline)) 2562 return Err; 2563 // Add the nested pass manager with the appropriate adaptor. 2564 bool UseMemorySSA = (Name == "loop-mssa"); 2565 bool UseBFI = 2566 std::any_of(InnerPipeline.begin(), InnerPipeline.end(), 2567 [](auto Pipeline) { return Pipeline.Name == "licm"; }); 2568 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM), UseMemorySSA, 2569 UseBFI, DebugLogging)); 2570 return Error::success(); 2571 } 2572 if (auto Count = parseRepeatPassName(Name)) { 2573 FunctionPassManager NestedFPM(DebugLogging); 2574 if (auto Err = parseFunctionPassPipeline(NestedFPM, InnerPipeline)) 2575 return Err; 2576 FPM.addPass(createRepeatedPass(*Count, std::move(NestedFPM))); 2577 return Error::success(); 2578 } 2579 2580 for (auto &C : FunctionPipelineParsingCallbacks) 2581 if (C(Name, FPM, InnerPipeline)) 2582 return Error::success(); 2583 2584 // Normal passes can't have pipelines. 2585 return make_error<StringError>( 2586 formatv("invalid use of '{0}' pass as function pipeline", Name).str(), 2587 inconvertibleErrorCode()); 2588 } 2589 2590 // Now expand the basic registered passes from the .inc file. 2591 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2592 if (Name == NAME) { \ 2593 FPM.addPass(CREATE_PASS); \ 2594 return Error::success(); \ 2595 } 2596 #define FUNCTION_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2597 if (checkParametrizedPassName(Name, NAME)) { \ 2598 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2599 if (!Params) \ 2600 return Params.takeError(); \ 2601 FPM.addPass(CREATE_PASS(Params.get())); \ 2602 return Error::success(); \ 2603 } 2604 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 2605 if (Name == "require<" NAME ">") { \ 2606 FPM.addPass( \ 2607 RequireAnalysisPass< \ 2608 std::remove_reference<decltype(CREATE_PASS)>::type, Function>()); \ 2609 return Error::success(); \ 2610 } \ 2611 if (Name == "invalidate<" NAME ">") { \ 2612 FPM.addPass(InvalidateAnalysisPass< \ 2613 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2614 return Error::success(); \ 2615 } 2616 // FIXME: UseMemorySSA is set to false. Maybe we could do things like: 2617 // bool UseMemorySSA = !("canon-freeze" || "loop-predication" || 2618 // "guard-widening"); 2619 // The risk is that it may become obsolete if we're not careful. 2620 #define LOOP_PASS(NAME, CREATE_PASS) \ 2621 if (Name == NAME) { \ 2622 FPM.addPass(createFunctionToLoopPassAdaptor(CREATE_PASS, false, false, \ 2623 DebugLogging)); \ 2624 return Error::success(); \ 2625 } 2626 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2627 if (checkParametrizedPassName(Name, NAME)) { \ 2628 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2629 if (!Params) \ 2630 return Params.takeError(); \ 2631 FPM.addPass(createFunctionToLoopPassAdaptor(CREATE_PASS(Params.get()), \ 2632 false, false, DebugLogging)); \ 2633 return Error::success(); \ 2634 } 2635 #include "PassRegistry.def" 2636 2637 for (auto &C : FunctionPipelineParsingCallbacks) 2638 if (C(Name, FPM, InnerPipeline)) 2639 return Error::success(); 2640 return make_error<StringError>( 2641 formatv("unknown function pass '{0}'", Name).str(), 2642 inconvertibleErrorCode()); 2643 } 2644 2645 Error PassBuilder::parseLoopPass(LoopPassManager &LPM, 2646 const PipelineElement &E) { 2647 StringRef Name = E.Name; 2648 auto &InnerPipeline = E.InnerPipeline; 2649 2650 // First handle complex passes like the pass managers which carry pipelines. 2651 if (!InnerPipeline.empty()) { 2652 if (Name == "loop") { 2653 LoopPassManager NestedLPM(DebugLogging); 2654 if (auto Err = parseLoopPassPipeline(NestedLPM, InnerPipeline)) 2655 return Err; 2656 // Add the nested pass manager with the appropriate adaptor. 2657 LPM.addPass(std::move(NestedLPM)); 2658 return Error::success(); 2659 } 2660 if (auto Count = parseRepeatPassName(Name)) { 2661 LoopPassManager NestedLPM(DebugLogging); 2662 if (auto Err = parseLoopPassPipeline(NestedLPM, InnerPipeline)) 2663 return Err; 2664 LPM.addPass(createRepeatedPass(*Count, std::move(NestedLPM))); 2665 return Error::success(); 2666 } 2667 2668 for (auto &C : LoopPipelineParsingCallbacks) 2669 if (C(Name, LPM, InnerPipeline)) 2670 return Error::success(); 2671 2672 // Normal passes can't have pipelines. 2673 return make_error<StringError>( 2674 formatv("invalid use of '{0}' pass as loop pipeline", Name).str(), 2675 inconvertibleErrorCode()); 2676 } 2677 2678 // Now expand the basic registered passes from the .inc file. 2679 #define LOOP_PASS(NAME, CREATE_PASS) \ 2680 if (Name == NAME) { \ 2681 LPM.addPass(CREATE_PASS); \ 2682 return Error::success(); \ 2683 } 2684 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2685 if (checkParametrizedPassName(Name, NAME)) { \ 2686 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2687 if (!Params) \ 2688 return Params.takeError(); \ 2689 LPM.addPass(CREATE_PASS(Params.get())); \ 2690 return Error::success(); \ 2691 } 2692 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 2693 if (Name == "require<" NAME ">") { \ 2694 LPM.addPass(RequireAnalysisPass< \ 2695 std::remove_reference<decltype(CREATE_PASS)>::type, Loop, \ 2696 LoopAnalysisManager, LoopStandardAnalysisResults &, \ 2697 LPMUpdater &>()); \ 2698 return Error::success(); \ 2699 } \ 2700 if (Name == "invalidate<" NAME ">") { \ 2701 LPM.addPass(InvalidateAnalysisPass< \ 2702 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2703 return Error::success(); \ 2704 } 2705 #include "PassRegistry.def" 2706 2707 for (auto &C : LoopPipelineParsingCallbacks) 2708 if (C(Name, LPM, InnerPipeline)) 2709 return Error::success(); 2710 return make_error<StringError>(formatv("unknown loop pass '{0}'", Name).str(), 2711 inconvertibleErrorCode()); 2712 } 2713 2714 bool PassBuilder::parseAAPassName(AAManager &AA, StringRef Name) { 2715 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 2716 if (Name == NAME) { \ 2717 AA.registerModuleAnalysis< \ 2718 std::remove_reference<decltype(CREATE_PASS)>::type>(); \ 2719 return true; \ 2720 } 2721 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 2722 if (Name == NAME) { \ 2723 AA.registerFunctionAnalysis< \ 2724 std::remove_reference<decltype(CREATE_PASS)>::type>(); \ 2725 return true; \ 2726 } 2727 #include "PassRegistry.def" 2728 2729 for (auto &C : AAParsingCallbacks) 2730 if (C(Name, AA)) 2731 return true; 2732 return false; 2733 } 2734 2735 Error PassBuilder::parseLoopPassPipeline(LoopPassManager &LPM, 2736 ArrayRef<PipelineElement> Pipeline) { 2737 for (const auto &Element : Pipeline) { 2738 if (auto Err = parseLoopPass(LPM, Element)) 2739 return Err; 2740 } 2741 return Error::success(); 2742 } 2743 2744 Error PassBuilder::parseFunctionPassPipeline( 2745 FunctionPassManager &FPM, ArrayRef<PipelineElement> Pipeline) { 2746 for (const auto &Element : Pipeline) { 2747 if (auto Err = parseFunctionPass(FPM, Element)) 2748 return Err; 2749 } 2750 return Error::success(); 2751 } 2752 2753 Error PassBuilder::parseCGSCCPassPipeline(CGSCCPassManager &CGPM, 2754 ArrayRef<PipelineElement> Pipeline) { 2755 for (const auto &Element : Pipeline) { 2756 if (auto Err = parseCGSCCPass(CGPM, Element)) 2757 return Err; 2758 } 2759 return Error::success(); 2760 } 2761 2762 void PassBuilder::crossRegisterProxies(LoopAnalysisManager &LAM, 2763 FunctionAnalysisManager &FAM, 2764 CGSCCAnalysisManager &CGAM, 2765 ModuleAnalysisManager &MAM) { 2766 MAM.registerPass([&] { return FunctionAnalysisManagerModuleProxy(FAM); }); 2767 MAM.registerPass([&] { return CGSCCAnalysisManagerModuleProxy(CGAM); }); 2768 CGAM.registerPass([&] { return ModuleAnalysisManagerCGSCCProxy(MAM); }); 2769 FAM.registerPass([&] { return CGSCCAnalysisManagerFunctionProxy(CGAM); }); 2770 FAM.registerPass([&] { return ModuleAnalysisManagerFunctionProxy(MAM); }); 2771 FAM.registerPass([&] { return LoopAnalysisManagerFunctionProxy(LAM); }); 2772 LAM.registerPass([&] { return FunctionAnalysisManagerLoopProxy(FAM); }); 2773 } 2774 2775 Error PassBuilder::parseModulePassPipeline(ModulePassManager &MPM, 2776 ArrayRef<PipelineElement> Pipeline) { 2777 for (const auto &Element : Pipeline) { 2778 if (auto Err = parseModulePass(MPM, Element)) 2779 return Err; 2780 } 2781 return Error::success(); 2782 } 2783 2784 // Primary pass pipeline description parsing routine for a \c ModulePassManager 2785 // FIXME: Should this routine accept a TargetMachine or require the caller to 2786 // pre-populate the analysis managers with target-specific stuff? 2787 Error PassBuilder::parsePassPipeline(ModulePassManager &MPM, 2788 StringRef PipelineText) { 2789 auto Pipeline = parsePipelineText(PipelineText); 2790 if (!Pipeline || Pipeline->empty()) 2791 return make_error<StringError>( 2792 formatv("invalid pipeline '{0}'", PipelineText).str(), 2793 inconvertibleErrorCode()); 2794 2795 // If the first name isn't at the module layer, wrap the pipeline up 2796 // automatically. 2797 StringRef FirstName = Pipeline->front().Name; 2798 2799 if (!isModulePassName(FirstName, ModulePipelineParsingCallbacks)) { 2800 if (isCGSCCPassName(FirstName, CGSCCPipelineParsingCallbacks)) { 2801 Pipeline = {{"cgscc", std::move(*Pipeline)}}; 2802 } else if (isFunctionPassName(FirstName, 2803 FunctionPipelineParsingCallbacks)) { 2804 Pipeline = {{"function", std::move(*Pipeline)}}; 2805 } else if (isLoopPassName(FirstName, LoopPipelineParsingCallbacks)) { 2806 Pipeline = {{"function", {{"loop", std::move(*Pipeline)}}}}; 2807 } else { 2808 for (auto &C : TopLevelPipelineParsingCallbacks) 2809 if (C(MPM, *Pipeline, DebugLogging)) 2810 return Error::success(); 2811 2812 // Unknown pass or pipeline name! 2813 auto &InnerPipeline = Pipeline->front().InnerPipeline; 2814 return make_error<StringError>( 2815 formatv("unknown {0} name '{1}'", 2816 (InnerPipeline.empty() ? "pass" : "pipeline"), FirstName) 2817 .str(), 2818 inconvertibleErrorCode()); 2819 } 2820 } 2821 2822 if (auto Err = parseModulePassPipeline(MPM, *Pipeline)) 2823 return Err; 2824 return Error::success(); 2825 } 2826 2827 // Primary pass pipeline description parsing routine for a \c CGSCCPassManager 2828 Error PassBuilder::parsePassPipeline(CGSCCPassManager &CGPM, 2829 StringRef PipelineText) { 2830 auto Pipeline = parsePipelineText(PipelineText); 2831 if (!Pipeline || Pipeline->empty()) 2832 return make_error<StringError>( 2833 formatv("invalid pipeline '{0}'", PipelineText).str(), 2834 inconvertibleErrorCode()); 2835 2836 StringRef FirstName = Pipeline->front().Name; 2837 if (!isCGSCCPassName(FirstName, CGSCCPipelineParsingCallbacks)) 2838 return make_error<StringError>( 2839 formatv("unknown cgscc pass '{0}' in pipeline '{1}'", FirstName, 2840 PipelineText) 2841 .str(), 2842 inconvertibleErrorCode()); 2843 2844 if (auto Err = parseCGSCCPassPipeline(CGPM, *Pipeline)) 2845 return Err; 2846 return Error::success(); 2847 } 2848 2849 // Primary pass pipeline description parsing routine for a \c 2850 // FunctionPassManager 2851 Error PassBuilder::parsePassPipeline(FunctionPassManager &FPM, 2852 StringRef PipelineText) { 2853 auto Pipeline = parsePipelineText(PipelineText); 2854 if (!Pipeline || Pipeline->empty()) 2855 return make_error<StringError>( 2856 formatv("invalid pipeline '{0}'", PipelineText).str(), 2857 inconvertibleErrorCode()); 2858 2859 StringRef FirstName = Pipeline->front().Name; 2860 if (!isFunctionPassName(FirstName, FunctionPipelineParsingCallbacks)) 2861 return make_error<StringError>( 2862 formatv("unknown function pass '{0}' in pipeline '{1}'", FirstName, 2863 PipelineText) 2864 .str(), 2865 inconvertibleErrorCode()); 2866 2867 if (auto Err = parseFunctionPassPipeline(FPM, *Pipeline)) 2868 return Err; 2869 return Error::success(); 2870 } 2871 2872 // Primary pass pipeline description parsing routine for a \c LoopPassManager 2873 Error PassBuilder::parsePassPipeline(LoopPassManager &CGPM, 2874 StringRef PipelineText) { 2875 auto Pipeline = parsePipelineText(PipelineText); 2876 if (!Pipeline || Pipeline->empty()) 2877 return make_error<StringError>( 2878 formatv("invalid pipeline '{0}'", PipelineText).str(), 2879 inconvertibleErrorCode()); 2880 2881 if (auto Err = parseLoopPassPipeline(CGPM, *Pipeline)) 2882 return Err; 2883 2884 return Error::success(); 2885 } 2886 2887 Error PassBuilder::parseAAPipeline(AAManager &AA, StringRef PipelineText) { 2888 // If the pipeline just consists of the word 'default' just replace the AA 2889 // manager with our default one. 2890 if (PipelineText == "default") { 2891 AA = buildDefaultAAPipeline(); 2892 return Error::success(); 2893 } 2894 2895 while (!PipelineText.empty()) { 2896 StringRef Name; 2897 std::tie(Name, PipelineText) = PipelineText.split(','); 2898 if (!parseAAPassName(AA, Name)) 2899 return make_error<StringError>( 2900 formatv("unknown alias analysis name '{0}'", Name).str(), 2901 inconvertibleErrorCode()); 2902 } 2903 2904 return Error::success(); 2905 } 2906 2907 bool PassBuilder::isAAPassName(StringRef PassName) { 2908 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 2909 if (PassName == NAME) \ 2910 return true; 2911 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 2912 if (PassName == NAME) \ 2913 return true; 2914 #include "PassRegistry.def" 2915 return false; 2916 } 2917 2918 bool PassBuilder::isAnalysisPassName(StringRef PassName) { 2919 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 2920 if (PassName == NAME) \ 2921 return true; 2922 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 2923 if (PassName == NAME) \ 2924 return true; 2925 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 2926 if (PassName == NAME) \ 2927 return true; 2928 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 2929 if (PassName == NAME) \ 2930 return true; 2931 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 2932 if (PassName == NAME) \ 2933 return true; 2934 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 2935 if (PassName == NAME) \ 2936 return true; 2937 #include "PassRegistry.def" 2938 return false; 2939 } 2940 2941 void PassBuilder::registerParseTopLevelPipelineCallback( 2942 const std::function<bool(ModulePassManager &, ArrayRef<PipelineElement>, 2943 bool DebugLogging)> &C) { 2944 TopLevelPipelineParsingCallbacks.push_back(C); 2945 } 2946