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/AliasAnalysisEvaluator.h" 20 #include "llvm/Analysis/AliasSetTracker.h" 21 #include "llvm/Analysis/AssumptionCache.h" 22 #include "llvm/Analysis/BasicAliasAnalysis.h" 23 #include "llvm/Analysis/BlockFrequencyInfo.h" 24 #include "llvm/Analysis/BranchProbabilityInfo.h" 25 #include "llvm/Analysis/CFGPrinter.h" 26 #include "llvm/Analysis/CFLAndersAliasAnalysis.h" 27 #include "llvm/Analysis/CFLSteensAliasAnalysis.h" 28 #include "llvm/Analysis/CGSCCPassManager.h" 29 #include "llvm/Analysis/CallGraph.h" 30 #include "llvm/Analysis/DDG.h" 31 #include "llvm/Analysis/DDGPrinter.h" 32 #include "llvm/Analysis/Delinearization.h" 33 #include "llvm/Analysis/DemandedBits.h" 34 #include "llvm/Analysis/DependenceAnalysis.h" 35 #include "llvm/Analysis/DivergenceAnalysis.h" 36 #include "llvm/Analysis/DominanceFrontier.h" 37 #include "llvm/Analysis/FunctionPropertiesAnalysis.h" 38 #include "llvm/Analysis/GlobalsModRef.h" 39 #include "llvm/Analysis/IRSimilarityIdentifier.h" 40 #include "llvm/Analysis/IVUsers.h" 41 #include "llvm/Analysis/InlineAdvisor.h" 42 #include "llvm/Analysis/InlineSizeEstimatorAnalysis.h" 43 #include "llvm/Analysis/InstCount.h" 44 #include "llvm/Analysis/LazyCallGraph.h" 45 #include "llvm/Analysis/LazyValueInfo.h" 46 #include "llvm/Analysis/Lint.h" 47 #include "llvm/Analysis/LoopAccessAnalysis.h" 48 #include "llvm/Analysis/LoopCacheAnalysis.h" 49 #include "llvm/Analysis/LoopInfo.h" 50 #include "llvm/Analysis/LoopNestAnalysis.h" 51 #include "llvm/Analysis/MemDerefPrinter.h" 52 #include "llvm/Analysis/MemoryDependenceAnalysis.h" 53 #include "llvm/Analysis/MemorySSA.h" 54 #include "llvm/Analysis/ModuleDebugInfoPrinter.h" 55 #include "llvm/Analysis/ModuleSummaryAnalysis.h" 56 #include "llvm/Analysis/MustExecute.h" 57 #include "llvm/Analysis/ObjCARCAliasAnalysis.h" 58 #include "llvm/Analysis/OptimizationRemarkEmitter.h" 59 #include "llvm/Analysis/PhiValues.h" 60 #include "llvm/Analysis/PostDominators.h" 61 #include "llvm/Analysis/ProfileSummaryInfo.h" 62 #include "llvm/Analysis/RegionInfo.h" 63 #include "llvm/Analysis/ScalarEvolution.h" 64 #include "llvm/Analysis/ScalarEvolutionAliasAnalysis.h" 65 #include "llvm/Analysis/ScopedNoAliasAA.h" 66 #include "llvm/Analysis/StackLifetime.h" 67 #include "llvm/Analysis/StackSafetyAnalysis.h" 68 #include "llvm/Analysis/TargetLibraryInfo.h" 69 #include "llvm/Analysis/TargetTransformInfo.h" 70 #include "llvm/Analysis/TypeBasedAliasAnalysis.h" 71 #include "llvm/IR/Dominators.h" 72 #include "llvm/IR/IRPrintingPasses.h" 73 #include "llvm/IR/PassManager.h" 74 #include "llvm/IR/PrintPasses.h" 75 #include "llvm/IR/SafepointIRVerifier.h" 76 #include "llvm/IR/Verifier.h" 77 #include "llvm/Support/CommandLine.h" 78 #include "llvm/Support/Debug.h" 79 #include "llvm/Support/ErrorHandling.h" 80 #include "llvm/Support/FormatVariadic.h" 81 #include "llvm/Support/Regex.h" 82 #include "llvm/Target/TargetMachine.h" 83 #include "llvm/Transforms/AggressiveInstCombine/AggressiveInstCombine.h" 84 #include "llvm/Transforms/Coroutines/CoroCleanup.h" 85 #include "llvm/Transforms/Coroutines/CoroEarly.h" 86 #include "llvm/Transforms/Coroutines/CoroElide.h" 87 #include "llvm/Transforms/Coroutines/CoroSplit.h" 88 #include "llvm/Transforms/IPO/AlwaysInliner.h" 89 #include "llvm/Transforms/IPO/Annotation2Metadata.h" 90 #include "llvm/Transforms/IPO/ArgumentPromotion.h" 91 #include "llvm/Transforms/IPO/Attributor.h" 92 #include "llvm/Transforms/IPO/BlockExtractor.h" 93 #include "llvm/Transforms/IPO/CalledValuePropagation.h" 94 #include "llvm/Transforms/IPO/ConstantMerge.h" 95 #include "llvm/Transforms/IPO/CrossDSOCFI.h" 96 #include "llvm/Transforms/IPO/DeadArgumentElimination.h" 97 #include "llvm/Transforms/IPO/ElimAvailExtern.h" 98 #include "llvm/Transforms/IPO/ForceFunctionAttrs.h" 99 #include "llvm/Transforms/IPO/FunctionAttrs.h" 100 #include "llvm/Transforms/IPO/FunctionImport.h" 101 #include "llvm/Transforms/IPO/GlobalDCE.h" 102 #include "llvm/Transforms/IPO/GlobalOpt.h" 103 #include "llvm/Transforms/IPO/GlobalSplit.h" 104 #include "llvm/Transforms/IPO/HotColdSplitting.h" 105 #include "llvm/Transforms/IPO/IROutliner.h" 106 #include "llvm/Transforms/IPO/InferFunctionAttrs.h" 107 #include "llvm/Transforms/IPO/Inliner.h" 108 #include "llvm/Transforms/IPO/Internalize.h" 109 #include "llvm/Transforms/IPO/LoopExtractor.h" 110 #include "llvm/Transforms/IPO/LowerTypeTests.h" 111 #include "llvm/Transforms/IPO/MergeFunctions.h" 112 #include "llvm/Transforms/IPO/OpenMPOpt.h" 113 #include "llvm/Transforms/IPO/PartialInlining.h" 114 #include "llvm/Transforms/IPO/SCCP.h" 115 #include "llvm/Transforms/IPO/SampleProfile.h" 116 #include "llvm/Transforms/IPO/SampleProfileProbe.h" 117 #include "llvm/Transforms/IPO/StripDeadPrototypes.h" 118 #include "llvm/Transforms/IPO/StripSymbols.h" 119 #include "llvm/Transforms/IPO/SyntheticCountsPropagation.h" 120 #include "llvm/Transforms/IPO/WholeProgramDevirt.h" 121 #include "llvm/Transforms/InstCombine/InstCombine.h" 122 #include "llvm/Transforms/Instrumentation.h" 123 #include "llvm/Transforms/Instrumentation/AddressSanitizer.h" 124 #include "llvm/Transforms/Instrumentation/BoundsChecking.h" 125 #include "llvm/Transforms/Instrumentation/CGProfile.h" 126 #include "llvm/Transforms/Instrumentation/ControlHeightReduction.h" 127 #include "llvm/Transforms/Instrumentation/DataFlowSanitizer.h" 128 #include "llvm/Transforms/Instrumentation/GCOVProfiler.h" 129 #include "llvm/Transforms/Instrumentation/HWAddressSanitizer.h" 130 #include "llvm/Transforms/Instrumentation/InstrOrderFile.h" 131 #include "llvm/Transforms/Instrumentation/InstrProfiling.h" 132 #include "llvm/Transforms/Instrumentation/MemProfiler.h" 133 #include "llvm/Transforms/Instrumentation/MemorySanitizer.h" 134 #include "llvm/Transforms/Instrumentation/PGOInstrumentation.h" 135 #include "llvm/Transforms/Instrumentation/PoisonChecking.h" 136 #include "llvm/Transforms/Instrumentation/SanitizerCoverage.h" 137 #include "llvm/Transforms/Instrumentation/ThreadSanitizer.h" 138 #include "llvm/Transforms/ObjCARC.h" 139 #include "llvm/Transforms/Scalar/ADCE.h" 140 #include "llvm/Transforms/Scalar/AlignmentFromAssumptions.h" 141 #include "llvm/Transforms/Scalar/AnnotationRemarks.h" 142 #include "llvm/Transforms/Scalar/BDCE.h" 143 #include "llvm/Transforms/Scalar/CallSiteSplitting.h" 144 #include "llvm/Transforms/Scalar/ConstantHoisting.h" 145 #include "llvm/Transforms/Scalar/ConstraintElimination.h" 146 #include "llvm/Transforms/Scalar/CorrelatedValuePropagation.h" 147 #include "llvm/Transforms/Scalar/DCE.h" 148 #include "llvm/Transforms/Scalar/DeadStoreElimination.h" 149 #include "llvm/Transforms/Scalar/DivRemPairs.h" 150 #include "llvm/Transforms/Scalar/EarlyCSE.h" 151 #include "llvm/Transforms/Scalar/Float2Int.h" 152 #include "llvm/Transforms/Scalar/GVN.h" 153 #include "llvm/Transforms/Scalar/GuardWidening.h" 154 #include "llvm/Transforms/Scalar/IVUsersPrinter.h" 155 #include "llvm/Transforms/Scalar/IndVarSimplify.h" 156 #include "llvm/Transforms/Scalar/InductiveRangeCheckElimination.h" 157 #include "llvm/Transforms/Scalar/InferAddressSpaces.h" 158 #include "llvm/Transforms/Scalar/InstSimplifyPass.h" 159 #include "llvm/Transforms/Scalar/JumpThreading.h" 160 #include "llvm/Transforms/Scalar/LICM.h" 161 #include "llvm/Transforms/Scalar/LoopAccessAnalysisPrinter.h" 162 #include "llvm/Transforms/Scalar/LoopDataPrefetch.h" 163 #include "llvm/Transforms/Scalar/LoopDeletion.h" 164 #include "llvm/Transforms/Scalar/LoopDistribute.h" 165 #include "llvm/Transforms/Scalar/LoopFlatten.h" 166 #include "llvm/Transforms/Scalar/LoopFuse.h" 167 #include "llvm/Transforms/Scalar/LoopIdiomRecognize.h" 168 #include "llvm/Transforms/Scalar/LoopInstSimplify.h" 169 #include "llvm/Transforms/Scalar/LoopInterchange.h" 170 #include "llvm/Transforms/Scalar/LoopLoadElimination.h" 171 #include "llvm/Transforms/Scalar/LoopPassManager.h" 172 #include "llvm/Transforms/Scalar/LoopPredication.h" 173 #include "llvm/Transforms/Scalar/LoopReroll.h" 174 #include "llvm/Transforms/Scalar/LoopRotation.h" 175 #include "llvm/Transforms/Scalar/LoopSimplifyCFG.h" 176 #include "llvm/Transforms/Scalar/LoopSink.h" 177 #include "llvm/Transforms/Scalar/LoopStrengthReduce.h" 178 #include "llvm/Transforms/Scalar/LoopUnrollAndJamPass.h" 179 #include "llvm/Transforms/Scalar/LoopUnrollPass.h" 180 #include "llvm/Transforms/Scalar/LoopVersioningLICM.h" 181 #include "llvm/Transforms/Scalar/LowerAtomic.h" 182 #include "llvm/Transforms/Scalar/LowerConstantIntrinsics.h" 183 #include "llvm/Transforms/Scalar/LowerExpectIntrinsic.h" 184 #include "llvm/Transforms/Scalar/LowerGuardIntrinsic.h" 185 #include "llvm/Transforms/Scalar/LowerMatrixIntrinsics.h" 186 #include "llvm/Transforms/Scalar/LowerWidenableCondition.h" 187 #include "llvm/Transforms/Scalar/MakeGuardsExplicit.h" 188 #include "llvm/Transforms/Scalar/MemCpyOptimizer.h" 189 #include "llvm/Transforms/Scalar/MergeICmps.h" 190 #include "llvm/Transforms/Scalar/MergedLoadStoreMotion.h" 191 #include "llvm/Transforms/Scalar/NaryReassociate.h" 192 #include "llvm/Transforms/Scalar/NewGVN.h" 193 #include "llvm/Transforms/Scalar/PartiallyInlineLibCalls.h" 194 #include "llvm/Transforms/Scalar/Reassociate.h" 195 #include "llvm/Transforms/Scalar/Reg2Mem.h" 196 #include "llvm/Transforms/Scalar/RewriteStatepointsForGC.h" 197 #include "llvm/Transforms/Scalar/SCCP.h" 198 #include "llvm/Transforms/Scalar/SROA.h" 199 #include "llvm/Transforms/Scalar/ScalarizeMaskedMemIntrin.h" 200 #include "llvm/Transforms/Scalar/Scalarizer.h" 201 #include "llvm/Transforms/Scalar/SeparateConstOffsetFromGEP.h" 202 #include "llvm/Transforms/Scalar/SimpleLoopUnswitch.h" 203 #include "llvm/Transforms/Scalar/SimplifyCFG.h" 204 #include "llvm/Transforms/Scalar/Sink.h" 205 #include "llvm/Transforms/Scalar/SpeculateAroundPHIs.h" 206 #include "llvm/Transforms/Scalar/SpeculativeExecution.h" 207 #include "llvm/Transforms/Scalar/StraightLineStrengthReduce.h" 208 #include "llvm/Transforms/Scalar/StructurizeCFG.h" 209 #include "llvm/Transforms/Scalar/TailRecursionElimination.h" 210 #include "llvm/Transforms/Scalar/WarnMissedTransforms.h" 211 #include "llvm/Transforms/Utils/AddDiscriminators.h" 212 #include "llvm/Transforms/Utils/AssumeBundleBuilder.h" 213 #include "llvm/Transforms/Utils/BreakCriticalEdges.h" 214 #include "llvm/Transforms/Utils/CanonicalizeAliases.h" 215 #include "llvm/Transforms/Utils/CanonicalizeFreezeInLoops.h" 216 #include "llvm/Transforms/Utils/EntryExitInstrumenter.h" 217 #include "llvm/Transforms/Utils/FixIrreducible.h" 218 #include "llvm/Transforms/Utils/HelloWorld.h" 219 #include "llvm/Transforms/Utils/InjectTLIMappings.h" 220 #include "llvm/Transforms/Utils/InstructionNamer.h" 221 #include "llvm/Transforms/Utils/LCSSA.h" 222 #include "llvm/Transforms/Utils/LibCallsShrinkWrap.h" 223 #include "llvm/Transforms/Utils/LoopSimplify.h" 224 #include "llvm/Transforms/Utils/LoopVersioning.h" 225 #include "llvm/Transforms/Utils/LowerInvoke.h" 226 #include "llvm/Transforms/Utils/LowerSwitch.h" 227 #include "llvm/Transforms/Utils/Mem2Reg.h" 228 #include "llvm/Transforms/Utils/MetaRenamer.h" 229 #include "llvm/Transforms/Utils/NameAnonGlobals.h" 230 #include "llvm/Transforms/Utils/StripGCRelocates.h" 231 #include "llvm/Transforms/Utils/StripNonLineTableDebugInfo.h" 232 #include "llvm/Transforms/Utils/SymbolRewriter.h" 233 #include "llvm/Transforms/Utils/UnifyFunctionExitNodes.h" 234 #include "llvm/Transforms/Utils/UnifyLoopExits.h" 235 #include "llvm/Transforms/Utils/UniqueInternalLinkageNames.h" 236 #include "llvm/Transforms/Vectorize/LoadStoreVectorizer.h" 237 #include "llvm/Transforms/Vectorize/LoopVectorize.h" 238 #include "llvm/Transforms/Vectorize/SLPVectorizer.h" 239 #include "llvm/Transforms/Vectorize/VectorCombine.h" 240 241 using namespace llvm; 242 243 extern cl::opt<unsigned> MaxDevirtIterations; 244 245 static cl::opt<InliningAdvisorMode> UseInlineAdvisor( 246 "enable-ml-inliner", cl::init(InliningAdvisorMode::Default), cl::Hidden, 247 cl::desc("Enable ML policy for inliner. Currently trained for -Oz only"), 248 cl::values(clEnumValN(InliningAdvisorMode::Default, "default", 249 "Heuristics-based inliner version."), 250 clEnumValN(InliningAdvisorMode::Development, "development", 251 "Use development mode (runtime-loadable model)."), 252 clEnumValN(InliningAdvisorMode::Release, "release", 253 "Use release mode (AOT-compiled model)."))); 254 255 static cl::opt<bool> EnableSyntheticCounts( 256 "enable-npm-synthetic-counts", cl::init(false), cl::Hidden, cl::ZeroOrMore, 257 cl::desc("Run synthetic function entry count generation " 258 "pass")); 259 260 static const Regex DefaultAliasRegex( 261 "^(default|thinlto-pre-link|thinlto|lto-pre-link|lto)<(O[0123sz])>$"); 262 263 /// Flag to enable inline deferral during PGO. 264 static cl::opt<bool> 265 EnablePGOInlineDeferral("enable-npm-pgo-inline-deferral", cl::init(true), 266 cl::Hidden, 267 cl::desc("Enable inline deferral during PGO")); 268 269 static cl::opt<bool> EnableMemProfiler("enable-mem-prof", cl::init(false), 270 cl::Hidden, cl::ZeroOrMore, 271 cl::desc("Enable memory profiler")); 272 273 static cl::opt<bool> PerformMandatoryInliningsFirst( 274 "mandatory-inlining-first", cl::init(true), cl::Hidden, cl::ZeroOrMore, 275 cl::desc("Perform mandatory inlinings module-wide, before performing " 276 "inlining.")); 277 278 static cl::opt<bool> EnableO3NonTrivialUnswitching( 279 "enable-npm-O3-nontrivial-unswitch", cl::init(true), cl::Hidden, 280 cl::ZeroOrMore, cl::desc("Enable non-trivial loop unswitching for -O3")); 281 282 PipelineTuningOptions::PipelineTuningOptions() { 283 LoopInterleaving = true; 284 LoopVectorization = true; 285 SLPVectorization = false; 286 LoopUnrolling = true; 287 ForgetAllSCEVInLoopUnroll = ForgetSCEVInLoopUnroll; 288 Coroutines = false; 289 LicmMssaOptCap = SetLicmMssaOptCap; 290 LicmMssaNoAccForPromotionCap = SetLicmMssaNoAccForPromotionCap; 291 CallGraphProfile = true; 292 MergeFunctions = false; 293 UniqueLinkageNames = false; 294 } 295 extern cl::opt<bool> ExtraVectorizerPasses; 296 297 extern cl::opt<bool> EnableConstraintElimination; 298 extern cl::opt<bool> EnableGVNHoist; 299 extern cl::opt<bool> EnableGVNSink; 300 extern cl::opt<bool> EnableHotColdSplit; 301 extern cl::opt<bool> EnableIROutliner; 302 extern cl::opt<bool> EnableOrderFileInstrumentation; 303 extern cl::opt<bool> EnableCHR; 304 extern cl::opt<bool> EnableLoopInterchange; 305 extern cl::opt<bool> EnableUnrollAndJam; 306 extern cl::opt<bool> EnableLoopFlatten; 307 extern cl::opt<bool> RunNewGVN; 308 extern cl::opt<bool> RunPartialInlining; 309 310 extern cl::opt<bool> FlattenedProfileUsed; 311 312 extern cl::opt<AttributorRunOption> AttributorRun; 313 extern cl::opt<bool> EnableKnowledgeRetention; 314 315 extern cl::opt<bool> EnableMatrix; 316 317 extern cl::opt<bool> DisablePreInliner; 318 extern cl::opt<int> PreInlineThreshold; 319 320 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O0 = { 321 /*SpeedLevel*/ 0, 322 /*SizeLevel*/ 0}; 323 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O1 = { 324 /*SpeedLevel*/ 1, 325 /*SizeLevel*/ 0}; 326 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O2 = { 327 /*SpeedLevel*/ 2, 328 /*SizeLevel*/ 0}; 329 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O3 = { 330 /*SpeedLevel*/ 3, 331 /*SizeLevel*/ 0}; 332 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::Os = { 333 /*SpeedLevel*/ 2, 334 /*SizeLevel*/ 1}; 335 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::Oz = { 336 /*SpeedLevel*/ 2, 337 /*SizeLevel*/ 2}; 338 339 namespace { 340 341 // The following passes/analyses have custom names, otherwise their name will 342 // include `(anonymous namespace)`. These are special since they are only for 343 // testing purposes and don't live in a header file. 344 345 /// No-op module pass which does nothing. 346 struct NoOpModulePass : PassInfoMixin<NoOpModulePass> { 347 PreservedAnalyses run(Module &M, ModuleAnalysisManager &) { 348 return PreservedAnalyses::all(); 349 } 350 351 static StringRef name() { return "NoOpModulePass"; } 352 }; 353 354 /// No-op module analysis. 355 class NoOpModuleAnalysis : public AnalysisInfoMixin<NoOpModuleAnalysis> { 356 friend AnalysisInfoMixin<NoOpModuleAnalysis>; 357 static AnalysisKey Key; 358 359 public: 360 struct Result {}; 361 Result run(Module &, ModuleAnalysisManager &) { return Result(); } 362 static StringRef name() { return "NoOpModuleAnalysis"; } 363 }; 364 365 /// No-op CGSCC pass which does nothing. 366 struct NoOpCGSCCPass : PassInfoMixin<NoOpCGSCCPass> { 367 PreservedAnalyses run(LazyCallGraph::SCC &C, CGSCCAnalysisManager &, 368 LazyCallGraph &, CGSCCUpdateResult &UR) { 369 return PreservedAnalyses::all(); 370 } 371 static StringRef name() { return "NoOpCGSCCPass"; } 372 }; 373 374 /// No-op CGSCC analysis. 375 class NoOpCGSCCAnalysis : public AnalysisInfoMixin<NoOpCGSCCAnalysis> { 376 friend AnalysisInfoMixin<NoOpCGSCCAnalysis>; 377 static AnalysisKey Key; 378 379 public: 380 struct Result {}; 381 Result run(LazyCallGraph::SCC &, CGSCCAnalysisManager &, LazyCallGraph &G) { 382 return Result(); 383 } 384 static StringRef name() { return "NoOpCGSCCAnalysis"; } 385 }; 386 387 /// No-op function pass which does nothing. 388 struct NoOpFunctionPass : PassInfoMixin<NoOpFunctionPass> { 389 PreservedAnalyses run(Function &F, FunctionAnalysisManager &) { 390 return PreservedAnalyses::all(); 391 } 392 static StringRef name() { return "NoOpFunctionPass"; } 393 }; 394 395 /// No-op function analysis. 396 class NoOpFunctionAnalysis : public AnalysisInfoMixin<NoOpFunctionAnalysis> { 397 friend AnalysisInfoMixin<NoOpFunctionAnalysis>; 398 static AnalysisKey Key; 399 400 public: 401 struct Result {}; 402 Result run(Function &, FunctionAnalysisManager &) { return Result(); } 403 static StringRef name() { return "NoOpFunctionAnalysis"; } 404 }; 405 406 /// No-op loop pass which does nothing. 407 struct NoOpLoopPass : PassInfoMixin<NoOpLoopPass> { 408 PreservedAnalyses run(Loop &L, LoopAnalysisManager &, 409 LoopStandardAnalysisResults &, LPMUpdater &) { 410 return PreservedAnalyses::all(); 411 } 412 static StringRef name() { return "NoOpLoopPass"; } 413 }; 414 415 /// No-op loop analysis. 416 class NoOpLoopAnalysis : public AnalysisInfoMixin<NoOpLoopAnalysis> { 417 friend AnalysisInfoMixin<NoOpLoopAnalysis>; 418 static AnalysisKey Key; 419 420 public: 421 struct Result {}; 422 Result run(Loop &, LoopAnalysisManager &, LoopStandardAnalysisResults &) { 423 return Result(); 424 } 425 static StringRef name() { return "NoOpLoopAnalysis"; } 426 }; 427 428 AnalysisKey NoOpModuleAnalysis::Key; 429 AnalysisKey NoOpCGSCCAnalysis::Key; 430 AnalysisKey NoOpFunctionAnalysis::Key; 431 AnalysisKey NoOpLoopAnalysis::Key; 432 433 /// Whether or not we should populate a PassInstrumentationCallbacks's class to 434 /// pass name map. 435 /// 436 /// This is for optimization purposes so we don't populate it if we never use 437 /// it. This should be updated if new pass instrumentation wants to use the map. 438 /// We currently only use this for --print-before/after. 439 bool shouldPopulateClassToPassNames() { 440 return !printBeforePasses().empty() || !printAfterPasses().empty(); 441 } 442 443 } // namespace 444 445 PassBuilder::PassBuilder(bool DebugLogging, TargetMachine *TM, 446 PipelineTuningOptions PTO, Optional<PGOOptions> PGOOpt, 447 PassInstrumentationCallbacks *PIC) 448 : DebugLogging(DebugLogging), TM(TM), PTO(PTO), PGOOpt(PGOOpt), PIC(PIC) { 449 if (TM) 450 TM->registerPassBuilderCallbacks(*this, DebugLogging); 451 if (PIC && shouldPopulateClassToPassNames()) { 452 #define MODULE_PASS(NAME, CREATE_PASS) \ 453 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 454 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 455 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 456 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 457 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 458 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 459 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 460 #define LOOP_PASS(NAME, CREATE_PASS) \ 461 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 462 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 463 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 464 #define CGSCC_PASS(NAME, CREATE_PASS) \ 465 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 466 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 467 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 468 #include "PassRegistry.def" 469 } 470 } 471 472 void PassBuilder::invokePeepholeEPCallbacks( 473 FunctionPassManager &FPM, PassBuilder::OptimizationLevel Level) { 474 for (auto &C : PeepholeEPCallbacks) 475 C(FPM, Level); 476 } 477 478 void PassBuilder::registerModuleAnalyses(ModuleAnalysisManager &MAM) { 479 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 480 MAM.registerPass([&] { return CREATE_PASS; }); 481 #include "PassRegistry.def" 482 483 for (auto &C : ModuleAnalysisRegistrationCallbacks) 484 C(MAM); 485 } 486 487 void PassBuilder::registerCGSCCAnalyses(CGSCCAnalysisManager &CGAM) { 488 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 489 CGAM.registerPass([&] { return CREATE_PASS; }); 490 #include "PassRegistry.def" 491 492 for (auto &C : CGSCCAnalysisRegistrationCallbacks) 493 C(CGAM); 494 } 495 496 void PassBuilder::registerFunctionAnalyses(FunctionAnalysisManager &FAM) { 497 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 498 FAM.registerPass([&] { return CREATE_PASS; }); 499 #include "PassRegistry.def" 500 501 for (auto &C : FunctionAnalysisRegistrationCallbacks) 502 C(FAM); 503 } 504 505 void PassBuilder::registerLoopAnalyses(LoopAnalysisManager &LAM) { 506 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 507 LAM.registerPass([&] { return CREATE_PASS; }); 508 #include "PassRegistry.def" 509 510 for (auto &C : LoopAnalysisRegistrationCallbacks) 511 C(LAM); 512 } 513 514 // Helper to add AnnotationRemarksPass. 515 static void addAnnotationRemarksPass(ModulePassManager &MPM) { 516 FunctionPassManager FPM; 517 FPM.addPass(AnnotationRemarksPass()); 518 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 519 } 520 521 // Helper to check if the current compilation phase is preparing for LTO 522 static bool isLTOPreLink(ThinOrFullLTOPhase Phase) { 523 return Phase == ThinOrFullLTOPhase::ThinLTOPreLink || 524 Phase == ThinOrFullLTOPhase::ThinLTOPreLink; 525 } 526 527 // TODO: Investigate the cost/benefit of tail call elimination on debugging. 528 FunctionPassManager 529 PassBuilder::buildO1FunctionSimplificationPipeline(OptimizationLevel Level, 530 ThinOrFullLTOPhase Phase) { 531 532 FunctionPassManager FPM(DebugLogging); 533 534 // Form SSA out of local memory accesses after breaking apart aggregates into 535 // scalars. 536 FPM.addPass(SROA()); 537 538 // Catch trivial redundancies 539 FPM.addPass(EarlyCSEPass(true /* Enable mem-ssa. */)); 540 541 // Hoisting of scalars and load expressions. 542 FPM.addPass(SimplifyCFGPass()); 543 FPM.addPass(InstCombinePass()); 544 545 FPM.addPass(LibCallsShrinkWrapPass()); 546 547 invokePeepholeEPCallbacks(FPM, Level); 548 549 FPM.addPass(SimplifyCFGPass()); 550 551 // Form canonically associated expression trees, and simplify the trees using 552 // basic mathematical properties. For example, this will form (nearly) 553 // minimal multiplication trees. 554 FPM.addPass(ReassociatePass()); 555 556 // Add the primary loop simplification pipeline. 557 // FIXME: Currently this is split into two loop pass pipelines because we run 558 // some function passes in between them. These can and should be removed 559 // and/or replaced by scheduling the loop pass equivalents in the correct 560 // positions. But those equivalent passes aren't powerful enough yet. 561 // Specifically, `SimplifyCFGPass` and `InstCombinePass` are currently still 562 // used. We have `LoopSimplifyCFGPass` which isn't yet powerful enough yet to 563 // fully replace `SimplifyCFGPass`, and the closest to the other we have is 564 // `LoopInstSimplify`. 565 LoopPassManager LPM1(DebugLogging), LPM2(DebugLogging); 566 567 // Simplify the loop body. We do this initially to clean up after other loop 568 // passes run, either when iterating on a loop or on inner loops with 569 // implications on the outer loop. 570 LPM1.addPass(LoopInstSimplifyPass()); 571 LPM1.addPass(LoopSimplifyCFGPass()); 572 573 LPM1.addPass(LoopRotatePass(/* Disable header duplication */ true, 574 isLTOPreLink(Phase))); 575 // TODO: Investigate promotion cap for O1. 576 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 577 LPM1.addPass(SimpleLoopUnswitchPass()); 578 579 LPM2.addPass(LoopIdiomRecognizePass()); 580 LPM2.addPass(IndVarSimplifyPass()); 581 582 for (auto &C : LateLoopOptimizationsEPCallbacks) 583 C(LPM2, Level); 584 585 LPM2.addPass(LoopDeletionPass()); 586 587 if (EnableLoopInterchange) 588 LPM2.addPass(LoopInterchangePass()); 589 590 // Do not enable unrolling in PreLinkThinLTO phase during sample PGO 591 // because it changes IR to makes profile annotation in back compile 592 // inaccurate. The normal unroller doesn't pay attention to forced full unroll 593 // attributes so we need to make sure and allow the full unroll pass to pay 594 // attention to it. 595 if (Phase != ThinOrFullLTOPhase::ThinLTOPreLink || !PGOOpt || 596 PGOOpt->Action != PGOOptions::SampleUse) 597 LPM2.addPass(LoopFullUnrollPass(Level.getSpeedupLevel(), 598 /* OnlyWhenForced= */ !PTO.LoopUnrolling, 599 PTO.ForgetAllSCEVInLoopUnroll)); 600 601 for (auto &C : LoopOptimizerEndEPCallbacks) 602 C(LPM2, Level); 603 604 // We provide the opt remark emitter pass for LICM to use. We only need to do 605 // this once as it is immutable. 606 FPM.addPass( 607 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 608 FPM.addPass(createFunctionToLoopPassAdaptor( 609 std::move(LPM1), EnableMSSALoopDependency, /*UseBlockFrequencyInfo=*/true, 610 DebugLogging)); 611 FPM.addPass(SimplifyCFGPass()); 612 FPM.addPass(InstCombinePass()); 613 if (EnableLoopFlatten) 614 FPM.addPass(LoopFlattenPass()); 615 // The loop passes in LPM2 (LoopFullUnrollPass) do not preserve MemorySSA. 616 // *All* loop passes must preserve it, in order to be able to use it. 617 FPM.addPass(createFunctionToLoopPassAdaptor( 618 std::move(LPM2), /*UseMemorySSA=*/false, /*UseBlockFrequencyInfo=*/false, 619 DebugLogging)); 620 621 // Delete small array after loop unroll. 622 FPM.addPass(SROA()); 623 624 // Specially optimize memory movement as it doesn't look like dataflow in SSA. 625 FPM.addPass(MemCpyOptPass()); 626 627 // Sparse conditional constant propagation. 628 // FIXME: It isn't clear why we do this *after* loop passes rather than 629 // before... 630 FPM.addPass(SCCPPass()); 631 632 // Delete dead bit computations (instcombine runs after to fold away the dead 633 // computations, and then ADCE will run later to exploit any new DCE 634 // opportunities that creates). 635 FPM.addPass(BDCEPass()); 636 637 // Run instcombine after redundancy and dead bit elimination to exploit 638 // opportunities opened up by them. 639 FPM.addPass(InstCombinePass()); 640 invokePeepholeEPCallbacks(FPM, Level); 641 642 if (PTO.Coroutines) 643 FPM.addPass(CoroElidePass()); 644 645 for (auto &C : ScalarOptimizerLateEPCallbacks) 646 C(FPM, Level); 647 648 // Finally, do an expensive DCE pass to catch all the dead code exposed by 649 // the simplifications and basic cleanup after all the simplifications. 650 // TODO: Investigate if this is too expensive. 651 FPM.addPass(ADCEPass()); 652 FPM.addPass(SimplifyCFGPass()); 653 FPM.addPass(InstCombinePass()); 654 invokePeepholeEPCallbacks(FPM, Level); 655 656 return FPM; 657 } 658 659 FunctionPassManager 660 PassBuilder::buildFunctionSimplificationPipeline(OptimizationLevel Level, 661 ThinOrFullLTOPhase Phase) { 662 assert(Level != OptimizationLevel::O0 && "Must request optimizations!"); 663 664 // The O1 pipeline has a separate pipeline creation function to simplify 665 // construction readability. 666 if (Level.getSpeedupLevel() == 1) 667 return buildO1FunctionSimplificationPipeline(Level, Phase); 668 669 FunctionPassManager FPM(DebugLogging); 670 671 // Form SSA out of local memory accesses after breaking apart aggregates into 672 // scalars. 673 FPM.addPass(SROA()); 674 675 // Catch trivial redundancies 676 FPM.addPass(EarlyCSEPass(true /* Enable mem-ssa. */)); 677 if (EnableKnowledgeRetention) 678 FPM.addPass(AssumeSimplifyPass()); 679 680 // Hoisting of scalars and load expressions. 681 if (EnableGVNHoist) 682 FPM.addPass(GVNHoistPass()); 683 684 // Global value numbering based sinking. 685 if (EnableGVNSink) { 686 FPM.addPass(GVNSinkPass()); 687 FPM.addPass(SimplifyCFGPass()); 688 } 689 690 if (EnableConstraintElimination) 691 FPM.addPass(ConstraintEliminationPass()); 692 693 // Speculative execution if the target has divergent branches; otherwise nop. 694 FPM.addPass(SpeculativeExecutionPass(/* OnlyIfDivergentTarget =*/true)); 695 696 // Optimize based on known information about branches, and cleanup afterward. 697 FPM.addPass(JumpThreadingPass()); 698 FPM.addPass(CorrelatedValuePropagationPass()); 699 700 FPM.addPass(SimplifyCFGPass()); 701 if (Level == OptimizationLevel::O3) 702 FPM.addPass(AggressiveInstCombinePass()); 703 FPM.addPass(InstCombinePass()); 704 705 if (!Level.isOptimizingForSize()) 706 FPM.addPass(LibCallsShrinkWrapPass()); 707 708 invokePeepholeEPCallbacks(FPM, Level); 709 710 // For PGO use pipeline, try to optimize memory intrinsics such as memcpy 711 // using the size value profile. Don't perform this when optimizing for size. 712 if (PGOOpt && PGOOpt->Action == PGOOptions::IRUse && 713 !Level.isOptimizingForSize()) 714 FPM.addPass(PGOMemOPSizeOpt()); 715 716 FPM.addPass(TailCallElimPass()); 717 FPM.addPass(SimplifyCFGPass()); 718 719 // Form canonically associated expression trees, and simplify the trees using 720 // basic mathematical properties. For example, this will form (nearly) 721 // minimal multiplication trees. 722 FPM.addPass(ReassociatePass()); 723 724 // Add the primary loop simplification pipeline. 725 // FIXME: Currently this is split into two loop pass pipelines because we run 726 // some function passes in between them. These can and should be removed 727 // and/or replaced by scheduling the loop pass equivalents in the correct 728 // positions. But those equivalent passes aren't powerful enough yet. 729 // Specifically, `SimplifyCFGPass` and `InstCombinePass` are currently still 730 // used. We have `LoopSimplifyCFGPass` which isn't yet powerful enough yet to 731 // fully replace `SimplifyCFGPass`, and the closest to the other we have is 732 // `LoopInstSimplify`. 733 LoopPassManager LPM1(DebugLogging), LPM2(DebugLogging); 734 735 // Simplify the loop body. We do this initially to clean up after other loop 736 // passes run, either when iterating on a loop or on inner loops with 737 // implications on the outer loop. 738 LPM1.addPass(LoopInstSimplifyPass()); 739 LPM1.addPass(LoopSimplifyCFGPass()); 740 741 // Disable header duplication in loop rotation at -Oz. 742 LPM1.addPass( 743 LoopRotatePass(Level != OptimizationLevel::Oz, isLTOPreLink(Phase))); 744 // TODO: Investigate promotion cap for O1. 745 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 746 LPM1.addPass( 747 SimpleLoopUnswitchPass(/* NonTrivial */ Level == OptimizationLevel::O3 && 748 EnableO3NonTrivialUnswitching)); 749 LPM2.addPass(LoopIdiomRecognizePass()); 750 LPM2.addPass(IndVarSimplifyPass()); 751 752 for (auto &C : LateLoopOptimizationsEPCallbacks) 753 C(LPM2, Level); 754 755 LPM2.addPass(LoopDeletionPass()); 756 757 if (EnableLoopInterchange) 758 LPM2.addPass(LoopInterchangePass()); 759 760 // Do not enable unrolling in PreLinkThinLTO phase during sample PGO 761 // because it changes IR to makes profile annotation in back compile 762 // inaccurate. The normal unroller doesn't pay attention to forced full unroll 763 // attributes so we need to make sure and allow the full unroll pass to pay 764 // attention to it. 765 if (Phase != ThinOrFullLTOPhase::ThinLTOPreLink || !PGOOpt || 766 PGOOpt->Action != PGOOptions::SampleUse) 767 LPM2.addPass(LoopFullUnrollPass(Level.getSpeedupLevel(), 768 /* OnlyWhenForced= */ !PTO.LoopUnrolling, 769 PTO.ForgetAllSCEVInLoopUnroll)); 770 771 for (auto &C : LoopOptimizerEndEPCallbacks) 772 C(LPM2, Level); 773 774 // We provide the opt remark emitter pass for LICM to use. We only need to do 775 // this once as it is immutable. 776 FPM.addPass( 777 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 778 FPM.addPass(createFunctionToLoopPassAdaptor( 779 std::move(LPM1), EnableMSSALoopDependency, /*UseBlockFrequencyInfo=*/true, 780 DebugLogging)); 781 FPM.addPass(SimplifyCFGPass()); 782 FPM.addPass(InstCombinePass()); 783 if (EnableLoopFlatten) 784 FPM.addPass(LoopFlattenPass()); 785 // The loop passes in LPM2 (LoopIdiomRecognizePass, IndVarSimplifyPass, 786 // LoopDeletionPass and LoopFullUnrollPass) do not preserve MemorySSA. 787 // *All* loop passes must preserve it, in order to be able to use it. 788 FPM.addPass(createFunctionToLoopPassAdaptor( 789 std::move(LPM2), /*UseMemorySSA=*/false, /*UseBlockFrequencyInfo=*/false, 790 DebugLogging)); 791 792 // Delete small array after loop unroll. 793 FPM.addPass(SROA()); 794 795 // Eliminate redundancies. 796 FPM.addPass(MergedLoadStoreMotionPass()); 797 if (RunNewGVN) 798 FPM.addPass(NewGVNPass()); 799 else 800 FPM.addPass(GVN()); 801 802 // Sparse conditional constant propagation. 803 // FIXME: It isn't clear why we do this *after* loop passes rather than 804 // before... 805 FPM.addPass(SCCPPass()); 806 807 // Delete dead bit computations (instcombine runs after to fold away the dead 808 // computations, and then ADCE will run later to exploit any new DCE 809 // opportunities that creates). 810 FPM.addPass(BDCEPass()); 811 812 // Run instcombine after redundancy and dead bit elimination to exploit 813 // opportunities opened up by them. 814 FPM.addPass(InstCombinePass()); 815 invokePeepholeEPCallbacks(FPM, Level); 816 817 // Re-consider control flow based optimizations after redundancy elimination, 818 // redo DCE, etc. 819 FPM.addPass(JumpThreadingPass()); 820 FPM.addPass(CorrelatedValuePropagationPass()); 821 822 // Finally, do an expensive DCE pass to catch all the dead code exposed by 823 // the simplifications and basic cleanup after all the simplifications. 824 // TODO: Investigate if this is too expensive. 825 FPM.addPass(ADCEPass()); 826 827 // Specially optimize memory movement as it doesn't look like dataflow in SSA. 828 FPM.addPass(MemCpyOptPass()); 829 830 FPM.addPass(DSEPass()); 831 FPM.addPass(createFunctionToLoopPassAdaptor( 832 LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap), 833 EnableMSSALoopDependency, /*UseBlockFrequencyInfo=*/true, DebugLogging)); 834 835 if (PTO.Coroutines) 836 FPM.addPass(CoroElidePass()); 837 838 for (auto &C : ScalarOptimizerLateEPCallbacks) 839 C(FPM, Level); 840 841 FPM.addPass(SimplifyCFGPass()); 842 FPM.addPass(InstCombinePass()); 843 invokePeepholeEPCallbacks(FPM, Level); 844 845 if (EnableCHR && Level == OptimizationLevel::O3 && PGOOpt && 846 (PGOOpt->Action == PGOOptions::IRUse || 847 PGOOpt->Action == PGOOptions::SampleUse)) 848 FPM.addPass(ControlHeightReductionPass()); 849 850 return FPM; 851 } 852 853 void PassBuilder::addRequiredLTOPreLinkPasses(ModulePassManager &MPM) { 854 MPM.addPass(CanonicalizeAliasesPass()); 855 MPM.addPass(NameAnonGlobalPass()); 856 } 857 858 void PassBuilder::addPGOInstrPasses(ModulePassManager &MPM, 859 PassBuilder::OptimizationLevel Level, 860 bool RunProfileGen, bool IsCS, 861 std::string ProfileFile, 862 std::string ProfileRemappingFile) { 863 assert(Level != OptimizationLevel::O0 && "Not expecting O0 here!"); 864 if (!IsCS && !DisablePreInliner) { 865 InlineParams IP; 866 867 IP.DefaultThreshold = PreInlineThreshold; 868 869 // FIXME: The hint threshold has the same value used by the regular inliner 870 // when not optimzing for size. This should probably be lowered after 871 // performance testing. 872 // FIXME: this comment is cargo culted from the old pass manager, revisit). 873 IP.HintThreshold = Level.isOptimizingForSize() ? PreInlineThreshold : 325; 874 ModuleInlinerWrapperPass MIWP(IP, DebugLogging); 875 CGSCCPassManager &CGPipeline = MIWP.getPM(); 876 877 FunctionPassManager FPM; 878 FPM.addPass(SROA()); 879 FPM.addPass(EarlyCSEPass()); // Catch trivial redundancies. 880 FPM.addPass(SimplifyCFGPass()); // Merge & remove basic blocks. 881 FPM.addPass(InstCombinePass()); // Combine silly sequences. 882 invokePeepholeEPCallbacks(FPM, Level); 883 884 CGPipeline.addPass(createCGSCCToFunctionPassAdaptor(std::move(FPM))); 885 886 MPM.addPass(std::move(MIWP)); 887 888 // Delete anything that is now dead to make sure that we don't instrument 889 // dead code. Instrumentation can end up keeping dead code around and 890 // dramatically increase code size. 891 MPM.addPass(GlobalDCEPass()); 892 } 893 894 if (!RunProfileGen) { 895 assert(!ProfileFile.empty() && "Profile use expecting a profile file!"); 896 MPM.addPass(PGOInstrumentationUse(ProfileFile, ProfileRemappingFile, IsCS)); 897 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 898 // RequireAnalysisPass for PSI before subsequent non-module passes. 899 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 900 return; 901 } 902 903 // Perform PGO instrumentation. 904 MPM.addPass(PGOInstrumentationGen(IsCS)); 905 906 FunctionPassManager FPM; 907 // Disable header duplication in loop rotation at -Oz. 908 FPM.addPass(createFunctionToLoopPassAdaptor( 909 LoopRotatePass(Level != OptimizationLevel::Oz), EnableMSSALoopDependency, 910 /*UseBlockFrequencyInfo=*/false, DebugLogging)); 911 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 912 913 // Add the profile lowering pass. 914 InstrProfOptions Options; 915 if (!ProfileFile.empty()) 916 Options.InstrProfileOutput = ProfileFile; 917 // Do counter promotion at Level greater than O0. 918 Options.DoCounterPromotion = true; 919 Options.UseBFIInPromotion = IsCS; 920 MPM.addPass(InstrProfiling(Options, IsCS)); 921 } 922 923 void PassBuilder::addPGOInstrPassesForO0(ModulePassManager &MPM, 924 bool RunProfileGen, bool IsCS, 925 std::string ProfileFile, 926 std::string ProfileRemappingFile) { 927 if (!RunProfileGen) { 928 assert(!ProfileFile.empty() && "Profile use expecting a profile file!"); 929 MPM.addPass(PGOInstrumentationUse(ProfileFile, ProfileRemappingFile, IsCS)); 930 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 931 // RequireAnalysisPass for PSI before subsequent non-module passes. 932 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 933 return; 934 } 935 936 // Perform PGO instrumentation. 937 MPM.addPass(PGOInstrumentationGen(IsCS)); 938 // Add the profile lowering pass. 939 InstrProfOptions Options; 940 if (!ProfileFile.empty()) 941 Options.InstrProfileOutput = ProfileFile; 942 // Do not do counter promotion at O0. 943 Options.DoCounterPromotion = false; 944 Options.UseBFIInPromotion = IsCS; 945 MPM.addPass(InstrProfiling(Options, IsCS)); 946 } 947 948 static InlineParams 949 getInlineParamsFromOptLevel(PassBuilder::OptimizationLevel Level) { 950 return getInlineParams(Level.getSpeedupLevel(), Level.getSizeLevel()); 951 } 952 953 ModuleInlinerWrapperPass 954 PassBuilder::buildInlinerPipeline(OptimizationLevel Level, 955 ThinOrFullLTOPhase Phase) { 956 InlineParams IP = getInlineParamsFromOptLevel(Level); 957 if (Phase == ThinOrFullLTOPhase::ThinLTOPreLink && PGOOpt && 958 PGOOpt->Action == PGOOptions::SampleUse) 959 IP.HotCallSiteThreshold = 0; 960 961 if (PGOOpt) 962 IP.EnableDeferral = EnablePGOInlineDeferral; 963 964 ModuleInlinerWrapperPass MIWP(IP, DebugLogging, 965 PerformMandatoryInliningsFirst, 966 UseInlineAdvisor, MaxDevirtIterations); 967 968 // Require the GlobalsAA analysis for the module so we can query it within 969 // the CGSCC pipeline. 970 MIWP.addRequiredModuleAnalysis<GlobalsAA>(); 971 972 // Require the ProfileSummaryAnalysis for the module so we can query it within 973 // the inliner pass. 974 MIWP.addRequiredModuleAnalysis<ProfileSummaryAnalysis>(); 975 976 // Now begin the main postorder CGSCC pipeline. 977 // FIXME: The current CGSCC pipeline has its origins in the legacy pass 978 // manager and trying to emulate its precise behavior. Much of this doesn't 979 // make a lot of sense and we should revisit the core CGSCC structure. 980 CGSCCPassManager &MainCGPipeline = MIWP.getPM(); 981 982 // Note: historically, the PruneEH pass was run first to deduce nounwind and 983 // generally clean up exception handling overhead. It isn't clear this is 984 // valuable as the inliner doesn't currently care whether it is inlining an 985 // invoke or a call. 986 987 if (AttributorRun & AttributorRunOption::CGSCC) 988 MainCGPipeline.addPass(AttributorCGSCCPass()); 989 990 if (PTO.Coroutines) 991 MainCGPipeline.addPass(CoroSplitPass(Level != OptimizationLevel::O0)); 992 993 // Now deduce any function attributes based in the current code. 994 MainCGPipeline.addPass(PostOrderFunctionAttrsPass()); 995 996 // When at O3 add argument promotion to the pass pipeline. 997 // FIXME: It isn't at all clear why this should be limited to O3. 998 if (Level == OptimizationLevel::O3) 999 MainCGPipeline.addPass(ArgumentPromotionPass()); 1000 1001 // Try to perform OpenMP specific optimizations. This is a (quick!) no-op if 1002 // there are no OpenMP runtime calls present in the module. 1003 if (Level == OptimizationLevel::O2 || Level == OptimizationLevel::O3) 1004 MainCGPipeline.addPass(OpenMPOptPass()); 1005 1006 for (auto &C : CGSCCOptimizerLateEPCallbacks) 1007 C(MainCGPipeline, Level); 1008 1009 // Lastly, add the core function simplification pipeline nested inside the 1010 // CGSCC walk. 1011 MainCGPipeline.addPass(createCGSCCToFunctionPassAdaptor( 1012 buildFunctionSimplificationPipeline(Level, Phase))); 1013 1014 return MIWP; 1015 } 1016 1017 ModulePassManager 1018 PassBuilder::buildModuleSimplificationPipeline(OptimizationLevel Level, 1019 ThinOrFullLTOPhase Phase) { 1020 ModulePassManager MPM(DebugLogging); 1021 1022 // Add UniqueInternalLinkageNames Pass which renames internal linkage 1023 // symbols with unique names. 1024 if (PTO.UniqueLinkageNames) 1025 MPM.addPass(UniqueInternalLinkageNamesPass()); 1026 1027 // Place pseudo probe instrumentation as the first pass of the pipeline to 1028 // minimize the impact of optimization changes. 1029 if (PGOOpt && PGOOpt->PseudoProbeForProfiling && 1030 Phase != ThinOrFullLTOPhase::ThinLTOPostLink) 1031 MPM.addPass(SampleProfileProbePass(TM)); 1032 1033 bool HasSampleProfile = PGOOpt && (PGOOpt->Action == PGOOptions::SampleUse); 1034 1035 // In ThinLTO mode, when flattened profile is used, all the available 1036 // profile information will be annotated in PreLink phase so there is 1037 // no need to load the profile again in PostLink. 1038 bool LoadSampleProfile = 1039 HasSampleProfile && 1040 !(FlattenedProfileUsed && Phase == ThinOrFullLTOPhase::ThinLTOPostLink); 1041 1042 // During the ThinLTO backend phase we perform early indirect call promotion 1043 // here, before globalopt. Otherwise imported available_externally functions 1044 // look unreferenced and are removed. If we are going to load the sample 1045 // profile then defer until later. 1046 // TODO: See if we can move later and consolidate with the location where 1047 // we perform ICP when we are loading a sample profile. 1048 // TODO: We pass HasSampleProfile (whether there was a sample profile file 1049 // passed to the compile) to the SamplePGO flag of ICP. This is used to 1050 // determine whether the new direct calls are annotated with prof metadata. 1051 // Ideally this should be determined from whether the IR is annotated with 1052 // sample profile, and not whether the a sample profile was provided on the 1053 // command line. E.g. for flattened profiles where we will not be reloading 1054 // the sample profile in the ThinLTO backend, we ideally shouldn't have to 1055 // provide the sample profile file. 1056 if (Phase == ThinOrFullLTOPhase::ThinLTOPostLink && !LoadSampleProfile) 1057 MPM.addPass(PGOIndirectCallPromotion(true /* InLTO */, HasSampleProfile)); 1058 1059 // Do basic inference of function attributes from known properties of system 1060 // libraries and other oracles. 1061 MPM.addPass(InferFunctionAttrsPass()); 1062 1063 // Create an early function pass manager to cleanup the output of the 1064 // frontend. 1065 FunctionPassManager EarlyFPM(DebugLogging); 1066 EarlyFPM.addPass(SimplifyCFGPass()); 1067 EarlyFPM.addPass(SROA()); 1068 EarlyFPM.addPass(EarlyCSEPass()); 1069 EarlyFPM.addPass(LowerExpectIntrinsicPass()); 1070 if (PTO.Coroutines) 1071 EarlyFPM.addPass(CoroEarlyPass()); 1072 if (Level == OptimizationLevel::O3) 1073 EarlyFPM.addPass(CallSiteSplittingPass()); 1074 1075 // In SamplePGO ThinLTO backend, we need instcombine before profile annotation 1076 // to convert bitcast to direct calls so that they can be inlined during the 1077 // profile annotation prepration step. 1078 // More details about SamplePGO design can be found in: 1079 // https://research.google.com/pubs/pub45290.html 1080 // FIXME: revisit how SampleProfileLoad/Inliner/ICP is structured. 1081 if (LoadSampleProfile) 1082 EarlyFPM.addPass(InstCombinePass()); 1083 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(EarlyFPM))); 1084 1085 if (LoadSampleProfile) { 1086 // Annotate sample profile right after early FPM to ensure freshness of 1087 // the debug info. 1088 MPM.addPass(SampleProfileLoaderPass(PGOOpt->ProfileFile, 1089 PGOOpt->ProfileRemappingFile, Phase)); 1090 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 1091 // RequireAnalysisPass for PSI before subsequent non-module passes. 1092 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 1093 // Do not invoke ICP in the LTOPrelink phase as it makes it hard 1094 // for the profile annotation to be accurate in the LTO backend. 1095 if (Phase != ThinOrFullLTOPhase::ThinLTOPreLink && 1096 Phase != ThinOrFullLTOPhase::FullLTOPreLink) 1097 // We perform early indirect call promotion here, before globalopt. 1098 // This is important for the ThinLTO backend phase because otherwise 1099 // imported available_externally functions look unreferenced and are 1100 // removed. 1101 MPM.addPass( 1102 PGOIndirectCallPromotion(true /* IsInLTO */, true /* SamplePGO */)); 1103 } 1104 1105 if (AttributorRun & AttributorRunOption::MODULE) 1106 MPM.addPass(AttributorPass()); 1107 1108 // Lower type metadata and the type.test intrinsic in the ThinLTO 1109 // post link pipeline after ICP. This is to enable usage of the type 1110 // tests in ICP sequences. 1111 if (Phase == ThinOrFullLTOPhase::ThinLTOPostLink) 1112 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1113 1114 for (auto &C : PipelineEarlySimplificationEPCallbacks) 1115 C(MPM, Level); 1116 1117 // Interprocedural constant propagation now that basic cleanup has occurred 1118 // and prior to optimizing globals. 1119 // FIXME: This position in the pipeline hasn't been carefully considered in 1120 // years, it should be re-analyzed. 1121 MPM.addPass(IPSCCPPass()); 1122 1123 // Attach metadata to indirect call sites indicating the set of functions 1124 // they may target at run-time. This should follow IPSCCP. 1125 MPM.addPass(CalledValuePropagationPass()); 1126 1127 // Optimize globals to try and fold them into constants. 1128 MPM.addPass(GlobalOptPass()); 1129 1130 // Promote any localized globals to SSA registers. 1131 // FIXME: Should this instead by a run of SROA? 1132 // FIXME: We should probably run instcombine and simplify-cfg afterward to 1133 // delete control flows that are dead once globals have been folded to 1134 // constants. 1135 MPM.addPass(createModuleToFunctionPassAdaptor(PromotePass())); 1136 1137 // Remove any dead arguments exposed by cleanups and constant folding 1138 // globals. 1139 MPM.addPass(DeadArgumentEliminationPass()); 1140 1141 // Create a small function pass pipeline to cleanup after all the global 1142 // optimizations. 1143 FunctionPassManager GlobalCleanupPM(DebugLogging); 1144 GlobalCleanupPM.addPass(InstCombinePass()); 1145 invokePeepholeEPCallbacks(GlobalCleanupPM, Level); 1146 1147 GlobalCleanupPM.addPass(SimplifyCFGPass()); 1148 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(GlobalCleanupPM))); 1149 1150 // Add all the requested passes for instrumentation PGO, if requested. 1151 if (PGOOpt && Phase != ThinOrFullLTOPhase::ThinLTOPostLink && 1152 (PGOOpt->Action == PGOOptions::IRInstr || 1153 PGOOpt->Action == PGOOptions::IRUse)) { 1154 addPGOInstrPasses(MPM, Level, 1155 /* RunProfileGen */ PGOOpt->Action == PGOOptions::IRInstr, 1156 /* IsCS */ false, PGOOpt->ProfileFile, 1157 PGOOpt->ProfileRemappingFile); 1158 MPM.addPass(PGOIndirectCallPromotion(false, false)); 1159 } 1160 if (PGOOpt && Phase != ThinOrFullLTOPhase::ThinLTOPostLink && 1161 PGOOpt->CSAction == PGOOptions::CSIRInstr) 1162 MPM.addPass(PGOInstrumentationGenCreateVar(PGOOpt->CSProfileGenFile)); 1163 1164 // Synthesize function entry counts for non-PGO compilation. 1165 if (EnableSyntheticCounts && !PGOOpt) 1166 MPM.addPass(SyntheticCountsPropagation()); 1167 1168 MPM.addPass(buildInlinerPipeline(Level, Phase)); 1169 1170 if (EnableMemProfiler && Phase != ThinOrFullLTOPhase::ThinLTOPreLink) { 1171 MPM.addPass(createModuleToFunctionPassAdaptor(MemProfilerPass())); 1172 MPM.addPass(ModuleMemProfilerPass()); 1173 } 1174 1175 return MPM; 1176 } 1177 1178 ModulePassManager 1179 PassBuilder::buildModuleOptimizationPipeline(OptimizationLevel Level, 1180 bool LTOPreLink) { 1181 ModulePassManager MPM(DebugLogging); 1182 1183 // Optimize globals now that the module is fully simplified. 1184 MPM.addPass(GlobalOptPass()); 1185 MPM.addPass(GlobalDCEPass()); 1186 1187 // Run partial inlining pass to partially inline functions that have 1188 // large bodies. 1189 if (RunPartialInlining) 1190 MPM.addPass(PartialInlinerPass()); 1191 1192 // Remove avail extern fns and globals definitions since we aren't compiling 1193 // an object file for later LTO. For LTO we want to preserve these so they 1194 // are eligible for inlining at link-time. Note if they are unreferenced they 1195 // will be removed by GlobalDCE later, so this only impacts referenced 1196 // available externally globals. Eventually they will be suppressed during 1197 // codegen, but eliminating here enables more opportunity for GlobalDCE as it 1198 // may make globals referenced by available external functions dead and saves 1199 // running remaining passes on the eliminated functions. These should be 1200 // preserved during prelinking for link-time inlining decisions. 1201 if (!LTOPreLink) 1202 MPM.addPass(EliminateAvailableExternallyPass()); 1203 1204 if (EnableOrderFileInstrumentation) 1205 MPM.addPass(InstrOrderFilePass()); 1206 1207 // Do RPO function attribute inference across the module to forward-propagate 1208 // attributes where applicable. 1209 // FIXME: Is this really an optimization rather than a canonicalization? 1210 MPM.addPass(ReversePostOrderFunctionAttrsPass()); 1211 1212 // Do a post inline PGO instrumentation and use pass. This is a context 1213 // sensitive PGO pass. We don't want to do this in LTOPreLink phrase as 1214 // cross-module inline has not been done yet. The context sensitive 1215 // instrumentation is after all the inlines are done. 1216 if (!LTOPreLink && PGOOpt) { 1217 if (PGOOpt->CSAction == PGOOptions::CSIRInstr) 1218 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ true, 1219 /* IsCS */ true, PGOOpt->CSProfileGenFile, 1220 PGOOpt->ProfileRemappingFile); 1221 else if (PGOOpt->CSAction == PGOOptions::CSIRUse) 1222 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ false, 1223 /* IsCS */ true, PGOOpt->ProfileFile, 1224 PGOOpt->ProfileRemappingFile); 1225 } 1226 1227 // Re-require GloblasAA here prior to function passes. This is particularly 1228 // useful as the above will have inlined, DCE'ed, and function-attr 1229 // propagated everything. We should at this point have a reasonably minimal 1230 // and richly annotated call graph. By computing aliasing and mod/ref 1231 // information for all local globals here, the late loop passes and notably 1232 // the vectorizer will be able to use them to help recognize vectorizable 1233 // memory operations. 1234 MPM.addPass(RequireAnalysisPass<GlobalsAA, Module>()); 1235 1236 FunctionPassManager OptimizePM(DebugLogging); 1237 OptimizePM.addPass(Float2IntPass()); 1238 OptimizePM.addPass(LowerConstantIntrinsicsPass()); 1239 1240 if (EnableMatrix) { 1241 OptimizePM.addPass(LowerMatrixIntrinsicsPass()); 1242 OptimizePM.addPass(EarlyCSEPass()); 1243 } 1244 1245 // FIXME: We need to run some loop optimizations to re-rotate loops after 1246 // simplify-cfg and others undo their rotation. 1247 1248 // Optimize the loop execution. These passes operate on entire loop nests 1249 // rather than on each loop in an inside-out manner, and so they are actually 1250 // function passes. 1251 1252 for (auto &C : VectorizerStartEPCallbacks) 1253 C(OptimizePM, Level); 1254 1255 // First rotate loops that may have been un-rotated by prior passes. 1256 // Disable header duplication at -Oz. 1257 OptimizePM.addPass(createFunctionToLoopPassAdaptor( 1258 LoopRotatePass(Level != OptimizationLevel::Oz, LTOPreLink), 1259 EnableMSSALoopDependency, 1260 /*UseBlockFrequencyInfo=*/false, DebugLogging)); 1261 1262 // Distribute loops to allow partial vectorization. I.e. isolate dependences 1263 // into separate loop that would otherwise inhibit vectorization. This is 1264 // currently only performed for loops marked with the metadata 1265 // llvm.loop.distribute=true or when -enable-loop-distribute is specified. 1266 OptimizePM.addPass(LoopDistributePass()); 1267 1268 // Populates the VFABI attribute with the scalar-to-vector mappings 1269 // from the TargetLibraryInfo. 1270 OptimizePM.addPass(InjectTLIMappings()); 1271 1272 // Now run the core loop vectorizer. 1273 OptimizePM.addPass(LoopVectorizePass( 1274 LoopVectorizeOptions(!PTO.LoopInterleaving, !PTO.LoopVectorization))); 1275 1276 // Eliminate loads by forwarding stores from the previous iteration to loads 1277 // of the current iteration. 1278 OptimizePM.addPass(LoopLoadEliminationPass()); 1279 1280 // Cleanup after the loop optimization passes. 1281 OptimizePM.addPass(InstCombinePass()); 1282 1283 if (Level.getSpeedupLevel() > 1 && ExtraVectorizerPasses) { 1284 // At higher optimization levels, try to clean up any runtime overlap and 1285 // alignment checks inserted by the vectorizer. We want to track correlated 1286 // runtime checks for two inner loops in the same outer loop, fold any 1287 // common computations, hoist loop-invariant aspects out of any outer loop, 1288 // and unswitch the runtime checks if possible. Once hoisted, we may have 1289 // dead (or speculatable) control flows or more combining opportunities. 1290 OptimizePM.addPass(EarlyCSEPass()); 1291 OptimizePM.addPass(CorrelatedValuePropagationPass()); 1292 OptimizePM.addPass(InstCombinePass()); 1293 LoopPassManager LPM(DebugLogging); 1294 LPM.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 1295 LPM.addPass( 1296 SimpleLoopUnswitchPass(/* NonTrivial */ Level == OptimizationLevel::O3)); 1297 OptimizePM.addPass(RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 1298 OptimizePM.addPass(createFunctionToLoopPassAdaptor( 1299 std::move(LPM), EnableMSSALoopDependency, /*UseBlockFrequencyInfo=*/true, 1300 DebugLogging)); 1301 OptimizePM.addPass(SimplifyCFGPass()); 1302 OptimizePM.addPass(InstCombinePass()); 1303 } 1304 1305 // Now that we've formed fast to execute loop structures, we do further 1306 // optimizations. These are run afterward as they might block doing complex 1307 // analyses and transforms such as what are needed for loop vectorization. 1308 1309 // Cleanup after loop vectorization, etc. Simplification passes like CVP and 1310 // GVN, loop transforms, and others have already run, so it's now better to 1311 // convert to more optimized IR using more aggressive simplify CFG options. 1312 // The extra sinking transform can create larger basic blocks, so do this 1313 // before SLP vectorization. 1314 // FIXME: study whether hoisting and/or sinking of common instructions should 1315 // be delayed until after SLP vectorizer. 1316 OptimizePM.addPass(SimplifyCFGPass(SimplifyCFGOptions() 1317 .forwardSwitchCondToPhi(true) 1318 .convertSwitchToLookupTable(true) 1319 .needCanonicalLoops(false) 1320 .hoistCommonInsts(true) 1321 .sinkCommonInsts(true))); 1322 1323 // Optimize parallel scalar instruction chains into SIMD instructions. 1324 if (PTO.SLPVectorization) { 1325 OptimizePM.addPass(SLPVectorizerPass()); 1326 if (Level.getSpeedupLevel() > 1 && ExtraVectorizerPasses) { 1327 OptimizePM.addPass(EarlyCSEPass()); 1328 } 1329 } 1330 1331 // Enhance/cleanup vector code. 1332 OptimizePM.addPass(VectorCombinePass()); 1333 OptimizePM.addPass(InstCombinePass()); 1334 1335 // Unroll small loops to hide loop backedge latency and saturate any parallel 1336 // execution resources of an out-of-order processor. We also then need to 1337 // clean up redundancies and loop invariant code. 1338 // FIXME: It would be really good to use a loop-integrated instruction 1339 // combiner for cleanup here so that the unrolling and LICM can be pipelined 1340 // across the loop nests. 1341 // We do UnrollAndJam in a separate LPM to ensure it happens before unroll 1342 if (EnableUnrollAndJam && PTO.LoopUnrolling) { 1343 OptimizePM.addPass(LoopUnrollAndJamPass(Level.getSpeedupLevel())); 1344 } 1345 OptimizePM.addPass(LoopUnrollPass(LoopUnrollOptions( 1346 Level.getSpeedupLevel(), /*OnlyWhenForced=*/!PTO.LoopUnrolling, 1347 PTO.ForgetAllSCEVInLoopUnroll))); 1348 OptimizePM.addPass(WarnMissedTransformationsPass()); 1349 OptimizePM.addPass(InstCombinePass()); 1350 OptimizePM.addPass(RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 1351 OptimizePM.addPass(createFunctionToLoopPassAdaptor( 1352 LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap), 1353 EnableMSSALoopDependency, /*UseBlockFrequencyInfo=*/true, DebugLogging)); 1354 1355 // Now that we've vectorized and unrolled loops, we may have more refined 1356 // alignment information, try to re-derive it here. 1357 OptimizePM.addPass(AlignmentFromAssumptionsPass()); 1358 1359 // Split out cold code. Splitting is done late to avoid hiding context from 1360 // other optimizations and inadvertently regressing performance. The tradeoff 1361 // is that this has a higher code size cost than splitting early. 1362 if (EnableHotColdSplit && !LTOPreLink) 1363 MPM.addPass(HotColdSplittingPass()); 1364 1365 // Search the code for similar regions of code. If enough similar regions can 1366 // be found where extracting the regions into their own function will decrease 1367 // the size of the program, we extract the regions, a deduplicate the 1368 // structurally similar regions. 1369 if (EnableIROutliner) 1370 MPM.addPass(IROutlinerPass()); 1371 1372 // Merge functions if requested. 1373 if (PTO.MergeFunctions) 1374 MPM.addPass(MergeFunctionsPass()); 1375 1376 // LoopSink pass sinks instructions hoisted by LICM, which serves as a 1377 // canonicalization pass that enables other optimizations. As a result, 1378 // LoopSink pass needs to be a very late IR pass to avoid undoing LICM 1379 // result too early. 1380 OptimizePM.addPass(LoopSinkPass()); 1381 1382 // And finally clean up LCSSA form before generating code. 1383 OptimizePM.addPass(InstSimplifyPass()); 1384 1385 // This hoists/decomposes div/rem ops. It should run after other sink/hoist 1386 // passes to avoid re-sinking, but before SimplifyCFG because it can allow 1387 // flattening of blocks. 1388 OptimizePM.addPass(DivRemPairsPass()); 1389 1390 // LoopSink (and other loop passes since the last simplifyCFG) might have 1391 // resulted in single-entry-single-exit or empty blocks. Clean up the CFG. 1392 OptimizePM.addPass(SimplifyCFGPass()); 1393 1394 // Optimize PHIs by speculating around them when profitable. Note that this 1395 // pass needs to be run after any PRE or similar pass as it is essentially 1396 // inserting redundancies into the program. This even includes SimplifyCFG. 1397 OptimizePM.addPass(SpeculateAroundPHIsPass()); 1398 1399 if (PTO.Coroutines) 1400 OptimizePM.addPass(CoroCleanupPass()); 1401 1402 // Add the core optimizing pipeline. 1403 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(OptimizePM))); 1404 1405 for (auto &C : OptimizerLastEPCallbacks) 1406 C(MPM, Level); 1407 1408 if (PTO.CallGraphProfile) 1409 MPM.addPass(CGProfilePass()); 1410 1411 // Now we need to do some global optimization transforms. 1412 // FIXME: It would seem like these should come first in the optimization 1413 // pipeline and maybe be the bottom of the canonicalization pipeline? Weird 1414 // ordering here. 1415 MPM.addPass(GlobalDCEPass()); 1416 MPM.addPass(ConstantMergePass()); 1417 1418 return MPM; 1419 } 1420 1421 ModulePassManager 1422 PassBuilder::buildPerModuleDefaultPipeline(OptimizationLevel Level, 1423 bool LTOPreLink) { 1424 assert(Level != OptimizationLevel::O0 && 1425 "Must request optimizations for the default pipeline!"); 1426 1427 ModulePassManager MPM(DebugLogging); 1428 1429 // Convert @llvm.global.annotations to !annotation metadata. 1430 MPM.addPass(Annotation2MetadataPass()); 1431 1432 // Force any function attributes we want the rest of the pipeline to observe. 1433 MPM.addPass(ForceFunctionAttrsPass()); 1434 1435 // Apply module pipeline start EP callback. 1436 for (auto &C : PipelineStartEPCallbacks) 1437 C(MPM, Level); 1438 1439 if (PGOOpt && PGOOpt->DebugInfoForProfiling) 1440 MPM.addPass(createModuleToFunctionPassAdaptor(AddDiscriminatorsPass())); 1441 1442 // Add the core simplification pipeline. 1443 MPM.addPass(buildModuleSimplificationPipeline( 1444 Level, LTOPreLink ? ThinOrFullLTOPhase::FullLTOPreLink 1445 : ThinOrFullLTOPhase::None)); 1446 1447 // Now add the optimization pipeline. 1448 MPM.addPass(buildModuleOptimizationPipeline(Level, LTOPreLink)); 1449 1450 if (PGOOpt && PGOOpt->PseudoProbeForProfiling) 1451 MPM.addPass(PseudoProbeUpdatePass()); 1452 1453 // Emit annotation remarks. 1454 addAnnotationRemarksPass(MPM); 1455 1456 if (LTOPreLink) 1457 addRequiredLTOPreLinkPasses(MPM); 1458 1459 return MPM; 1460 } 1461 1462 ModulePassManager 1463 PassBuilder::buildThinLTOPreLinkDefaultPipeline(OptimizationLevel Level) { 1464 assert(Level != OptimizationLevel::O0 && 1465 "Must request optimizations for the default pipeline!"); 1466 1467 ModulePassManager MPM(DebugLogging); 1468 1469 // Convert @llvm.global.annotations to !annotation metadata. 1470 MPM.addPass(Annotation2MetadataPass()); 1471 1472 // Force any function attributes we want the rest of the pipeline to observe. 1473 MPM.addPass(ForceFunctionAttrsPass()); 1474 1475 if (PGOOpt && PGOOpt->DebugInfoForProfiling) 1476 MPM.addPass(createModuleToFunctionPassAdaptor(AddDiscriminatorsPass())); 1477 1478 // Apply module pipeline start EP callback. 1479 for (auto &C : PipelineStartEPCallbacks) 1480 C(MPM, Level); 1481 1482 // If we are planning to perform ThinLTO later, we don't bloat the code with 1483 // unrolling/vectorization/... now. Just simplify the module as much as we 1484 // can. 1485 MPM.addPass(buildModuleSimplificationPipeline( 1486 Level, ThinOrFullLTOPhase::ThinLTOPreLink)); 1487 1488 // Run partial inlining pass to partially inline functions that have 1489 // large bodies. 1490 // FIXME: It isn't clear whether this is really the right place to run this 1491 // in ThinLTO. Because there is another canonicalization and simplification 1492 // phase that will run after the thin link, running this here ends up with 1493 // less information than will be available later and it may grow functions in 1494 // ways that aren't beneficial. 1495 if (RunPartialInlining) 1496 MPM.addPass(PartialInlinerPass()); 1497 1498 // Reduce the size of the IR as much as possible. 1499 MPM.addPass(GlobalOptPass()); 1500 1501 // Module simplification splits coroutines, but does not fully clean up 1502 // coroutine intrinsics. To ensure ThinLTO optimization passes don't trip up 1503 // on these, we schedule the cleanup here. 1504 if (PTO.Coroutines) 1505 MPM.addPass(createModuleToFunctionPassAdaptor(CoroCleanupPass())); 1506 1507 if (PGOOpt && PGOOpt->PseudoProbeForProfiling) 1508 MPM.addPass(PseudoProbeUpdatePass()); 1509 1510 // Handle OptimizerLastEPCallbacks added by clang on PreLink. Actual 1511 // optimization is going to be done in PostLink stage, but clang can't 1512 // add callbacks there in case of in-process ThinLTO called by linker. 1513 for (auto &C : OptimizerLastEPCallbacks) 1514 C(MPM, Level); 1515 1516 // Emit annotation remarks. 1517 addAnnotationRemarksPass(MPM); 1518 1519 addRequiredLTOPreLinkPasses(MPM); 1520 1521 return MPM; 1522 } 1523 1524 ModulePassManager PassBuilder::buildThinLTODefaultPipeline( 1525 OptimizationLevel Level, const ModuleSummaryIndex *ImportSummary) { 1526 ModulePassManager MPM(DebugLogging); 1527 1528 // Convert @llvm.global.annotations to !annotation metadata. 1529 MPM.addPass(Annotation2MetadataPass()); 1530 1531 if (ImportSummary) { 1532 // These passes import type identifier resolutions for whole-program 1533 // devirtualization and CFI. They must run early because other passes may 1534 // disturb the specific instruction patterns that these passes look for, 1535 // creating dependencies on resolutions that may not appear in the summary. 1536 // 1537 // For example, GVN may transform the pattern assume(type.test) appearing in 1538 // two basic blocks into assume(phi(type.test, type.test)), which would 1539 // transform a dependency on a WPD resolution into a dependency on a type 1540 // identifier resolution for CFI. 1541 // 1542 // Also, WPD has access to more precise information than ICP and can 1543 // devirtualize more effectively, so it should operate on the IR first. 1544 // 1545 // The WPD and LowerTypeTest passes need to run at -O0 to lower type 1546 // metadata and intrinsics. 1547 MPM.addPass(WholeProgramDevirtPass(nullptr, ImportSummary)); 1548 MPM.addPass(LowerTypeTestsPass(nullptr, ImportSummary)); 1549 } 1550 1551 if (Level == OptimizationLevel::O0) { 1552 // Run a second time to clean up any type tests left behind by WPD for use 1553 // in ICP. 1554 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1555 // Drop available_externally and unreferenced globals. This is necessary 1556 // with ThinLTO in order to avoid leaving undefined references to dead 1557 // globals in the object file. 1558 MPM.addPass(EliminateAvailableExternallyPass()); 1559 MPM.addPass(GlobalDCEPass()); 1560 return MPM; 1561 } 1562 1563 // Force any function attributes we want the rest of the pipeline to observe. 1564 MPM.addPass(ForceFunctionAttrsPass()); 1565 1566 // Add the core simplification pipeline. 1567 MPM.addPass(buildModuleSimplificationPipeline( 1568 Level, ThinOrFullLTOPhase::ThinLTOPostLink)); 1569 1570 // Now add the optimization pipeline. 1571 MPM.addPass(buildModuleOptimizationPipeline(Level)); 1572 1573 // Emit annotation remarks. 1574 addAnnotationRemarksPass(MPM); 1575 1576 return MPM; 1577 } 1578 1579 ModulePassManager 1580 PassBuilder::buildLTOPreLinkDefaultPipeline(OptimizationLevel Level) { 1581 assert(Level != OptimizationLevel::O0 && 1582 "Must request optimizations for the default pipeline!"); 1583 // FIXME: We should use a customized pre-link pipeline! 1584 return buildPerModuleDefaultPipeline(Level, 1585 /* LTOPreLink */ true); 1586 } 1587 1588 ModulePassManager 1589 PassBuilder::buildLTODefaultPipeline(OptimizationLevel Level, 1590 ModuleSummaryIndex *ExportSummary) { 1591 ModulePassManager MPM(DebugLogging); 1592 1593 // Convert @llvm.global.annotations to !annotation metadata. 1594 MPM.addPass(Annotation2MetadataPass()); 1595 1596 if (Level == OptimizationLevel::O0) { 1597 // The WPD and LowerTypeTest passes need to run at -O0 to lower type 1598 // metadata and intrinsics. 1599 MPM.addPass(WholeProgramDevirtPass(ExportSummary, nullptr)); 1600 MPM.addPass(LowerTypeTestsPass(ExportSummary, nullptr)); 1601 // Run a second time to clean up any type tests left behind by WPD for use 1602 // in ICP. 1603 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1604 1605 // Emit annotation remarks. 1606 addAnnotationRemarksPass(MPM); 1607 1608 return MPM; 1609 } 1610 1611 if (PGOOpt && PGOOpt->Action == PGOOptions::SampleUse) { 1612 // Load sample profile before running the LTO optimization pipeline. 1613 MPM.addPass(SampleProfileLoaderPass(PGOOpt->ProfileFile, 1614 PGOOpt->ProfileRemappingFile, 1615 ThinOrFullLTOPhase::FullLTOPostLink)); 1616 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 1617 // RequireAnalysisPass for PSI before subsequent non-module passes. 1618 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 1619 } 1620 1621 // Remove unused virtual tables to improve the quality of code generated by 1622 // whole-program devirtualization and bitset lowering. 1623 MPM.addPass(GlobalDCEPass()); 1624 1625 // Force any function attributes we want the rest of the pipeline to observe. 1626 MPM.addPass(ForceFunctionAttrsPass()); 1627 1628 // Do basic inference of function attributes from known properties of system 1629 // libraries and other oracles. 1630 MPM.addPass(InferFunctionAttrsPass()); 1631 1632 if (Level.getSpeedupLevel() > 1) { 1633 FunctionPassManager EarlyFPM(DebugLogging); 1634 EarlyFPM.addPass(CallSiteSplittingPass()); 1635 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(EarlyFPM))); 1636 1637 // Indirect call promotion. This should promote all the targets that are 1638 // left by the earlier promotion pass that promotes intra-module targets. 1639 // This two-step promotion is to save the compile time. For LTO, it should 1640 // produce the same result as if we only do promotion here. 1641 MPM.addPass(PGOIndirectCallPromotion( 1642 true /* InLTO */, PGOOpt && PGOOpt->Action == PGOOptions::SampleUse)); 1643 // Propagate constants at call sites into the functions they call. This 1644 // opens opportunities for globalopt (and inlining) by substituting function 1645 // pointers passed as arguments to direct uses of functions. 1646 MPM.addPass(IPSCCPPass()); 1647 1648 // Attach metadata to indirect call sites indicating the set of functions 1649 // they may target at run-time. This should follow IPSCCP. 1650 MPM.addPass(CalledValuePropagationPass()); 1651 } 1652 1653 // Now deduce any function attributes based in the current code. 1654 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor( 1655 PostOrderFunctionAttrsPass())); 1656 1657 // Do RPO function attribute inference across the module to forward-propagate 1658 // attributes where applicable. 1659 // FIXME: Is this really an optimization rather than a canonicalization? 1660 MPM.addPass(ReversePostOrderFunctionAttrsPass()); 1661 1662 // Use in-range annotations on GEP indices to split globals where beneficial. 1663 MPM.addPass(GlobalSplitPass()); 1664 1665 // Run whole program optimization of virtual call when the list of callees 1666 // is fixed. 1667 MPM.addPass(WholeProgramDevirtPass(ExportSummary, nullptr)); 1668 1669 // Stop here at -O1. 1670 if (Level == OptimizationLevel::O1) { 1671 // The LowerTypeTestsPass needs to run to lower type metadata and the 1672 // type.test intrinsics. The pass does nothing if CFI is disabled. 1673 MPM.addPass(LowerTypeTestsPass(ExportSummary, nullptr)); 1674 // Run a second time to clean up any type tests left behind by WPD for use 1675 // in ICP (which is performed earlier than this in the regular LTO 1676 // pipeline). 1677 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1678 1679 // Emit annotation remarks. 1680 addAnnotationRemarksPass(MPM); 1681 1682 return MPM; 1683 } 1684 1685 // Optimize globals to try and fold them into constants. 1686 MPM.addPass(GlobalOptPass()); 1687 1688 // Promote any localized globals to SSA registers. 1689 MPM.addPass(createModuleToFunctionPassAdaptor(PromotePass())); 1690 1691 // Linking modules together can lead to duplicate global constant, only 1692 // keep one copy of each constant. 1693 MPM.addPass(ConstantMergePass()); 1694 1695 // Remove unused arguments from functions. 1696 MPM.addPass(DeadArgumentEliminationPass()); 1697 1698 // Reduce the code after globalopt and ipsccp. Both can open up significant 1699 // simplification opportunities, and both can propagate functions through 1700 // function pointers. When this happens, we often have to resolve varargs 1701 // calls, etc, so let instcombine do this. 1702 FunctionPassManager PeepholeFPM(DebugLogging); 1703 if (Level == OptimizationLevel::O3) 1704 PeepholeFPM.addPass(AggressiveInstCombinePass()); 1705 PeepholeFPM.addPass(InstCombinePass()); 1706 invokePeepholeEPCallbacks(PeepholeFPM, Level); 1707 1708 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(PeepholeFPM))); 1709 1710 // Note: historically, the PruneEH pass was run first to deduce nounwind and 1711 // generally clean up exception handling overhead. It isn't clear this is 1712 // valuable as the inliner doesn't currently care whether it is inlining an 1713 // invoke or a call. 1714 // Run the inliner now. 1715 MPM.addPass(ModuleInlinerWrapperPass(getInlineParamsFromOptLevel(Level), 1716 DebugLogging)); 1717 1718 // Optimize globals again after we ran the inliner. 1719 MPM.addPass(GlobalOptPass()); 1720 1721 // Garbage collect dead functions. 1722 // FIXME: Add ArgumentPromotion pass after once it's ported. 1723 MPM.addPass(GlobalDCEPass()); 1724 1725 FunctionPassManager FPM(DebugLogging); 1726 // The IPO Passes may leave cruft around. Clean up after them. 1727 FPM.addPass(InstCombinePass()); 1728 invokePeepholeEPCallbacks(FPM, Level); 1729 1730 FPM.addPass(JumpThreadingPass(/*InsertFreezeWhenUnfoldingSelect*/ true)); 1731 1732 // Do a post inline PGO instrumentation and use pass. This is a context 1733 // sensitive PGO pass. 1734 if (PGOOpt) { 1735 if (PGOOpt->CSAction == PGOOptions::CSIRInstr) 1736 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ true, 1737 /* IsCS */ true, PGOOpt->CSProfileGenFile, 1738 PGOOpt->ProfileRemappingFile); 1739 else if (PGOOpt->CSAction == PGOOptions::CSIRUse) 1740 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ false, 1741 /* IsCS */ true, PGOOpt->ProfileFile, 1742 PGOOpt->ProfileRemappingFile); 1743 } 1744 1745 // Break up allocas 1746 FPM.addPass(SROA()); 1747 1748 // LTO provides additional opportunities for tailcall elimination due to 1749 // link-time inlining, and visibility of nocapture attribute. 1750 FPM.addPass(TailCallElimPass()); 1751 1752 // Run a few AA driver optimizations here and now to cleanup the code. 1753 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 1754 1755 MPM.addPass( 1756 createModuleToPostOrderCGSCCPassAdaptor(PostOrderFunctionAttrsPass())); 1757 // FIXME: here we run IP alias analysis in the legacy PM. 1758 1759 FunctionPassManager MainFPM; 1760 1761 MainFPM.addPass(createFunctionToLoopPassAdaptor( 1762 LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap))); 1763 1764 if (RunNewGVN) 1765 MainFPM.addPass(NewGVNPass()); 1766 else 1767 MainFPM.addPass(GVN()); 1768 1769 // Remove dead memcpy()'s. 1770 MainFPM.addPass(MemCpyOptPass()); 1771 1772 // Nuke dead stores. 1773 MainFPM.addPass(DSEPass()); 1774 MainFPM.addPass(MergedLoadStoreMotionPass()); 1775 1776 // More loops are countable; try to optimize them. 1777 if (EnableLoopFlatten && Level.getSpeedupLevel() > 1) 1778 MainFPM.addPass(LoopFlattenPass()); 1779 1780 if (EnableConstraintElimination) 1781 MainFPM.addPass(ConstraintEliminationPass()); 1782 1783 LoopPassManager LPM(DebugLogging); 1784 LPM.addPass(IndVarSimplifyPass()); 1785 LPM.addPass(LoopDeletionPass()); 1786 // FIXME: Add loop interchange. 1787 1788 // Unroll small loops and perform peeling. 1789 LPM.addPass(LoopFullUnrollPass(Level.getSpeedupLevel(), 1790 /* OnlyWhenForced= */ !PTO.LoopUnrolling, 1791 PTO.ForgetAllSCEVInLoopUnroll)); 1792 // The loop passes in LPM (LoopFullUnrollPass) do not preserve MemorySSA. 1793 // *All* loop passes must preserve it, in order to be able to use it. 1794 MainFPM.addPass(createFunctionToLoopPassAdaptor( 1795 std::move(LPM), /*UseMemorySSA=*/false, /*UseBlockFrequencyInfo=*/true, 1796 DebugLogging)); 1797 1798 MainFPM.addPass(LoopDistributePass()); 1799 MainFPM.addPass(LoopVectorizePass( 1800 LoopVectorizeOptions(!PTO.LoopInterleaving, !PTO.LoopVectorization))); 1801 // The vectorizer may have significantly shortened a loop body; unroll again. 1802 MainFPM.addPass(LoopUnrollPass(LoopUnrollOptions( 1803 Level.getSpeedupLevel(), /*OnlyWhenForced=*/!PTO.LoopUnrolling, 1804 PTO.ForgetAllSCEVInLoopUnroll))); 1805 1806 MainFPM.addPass(WarnMissedTransformationsPass()); 1807 1808 MainFPM.addPass(InstCombinePass()); 1809 MainFPM.addPass(SimplifyCFGPass(SimplifyCFGOptions().hoistCommonInsts(true))); 1810 MainFPM.addPass(SCCPPass()); 1811 MainFPM.addPass(InstCombinePass()); 1812 MainFPM.addPass(BDCEPass()); 1813 1814 // More scalar chains could be vectorized due to more alias information 1815 if (PTO.SLPVectorization) { 1816 MainFPM.addPass(SLPVectorizerPass()); 1817 if (Level.getSpeedupLevel() > 1 && ExtraVectorizerPasses) { 1818 MainFPM.addPass(EarlyCSEPass()); 1819 } 1820 } 1821 1822 MainFPM.addPass(VectorCombinePass()); // Clean up partial vectorization. 1823 1824 // After vectorization, assume intrinsics may tell us more about pointer 1825 // alignments. 1826 MainFPM.addPass(AlignmentFromAssumptionsPass()); 1827 1828 // FIXME: Conditionally run LoadCombine here, after it's ported 1829 // (in case we still have this pass, given its questionable usefulness). 1830 1831 MainFPM.addPass(InstCombinePass()); 1832 invokePeepholeEPCallbacks(MainFPM, Level); 1833 MainFPM.addPass(JumpThreadingPass(/*InsertFreezeWhenUnfoldingSelect*/ true)); 1834 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(MainFPM))); 1835 1836 // Create a function that performs CFI checks for cross-DSO calls with 1837 // targets in the current module. 1838 MPM.addPass(CrossDSOCFIPass()); 1839 1840 // Lower type metadata and the type.test intrinsic. This pass supports 1841 // clang's control flow integrity mechanisms (-fsanitize=cfi*) and needs 1842 // to be run at link time if CFI is enabled. This pass does nothing if 1843 // CFI is disabled. 1844 MPM.addPass(LowerTypeTestsPass(ExportSummary, nullptr)); 1845 // Run a second time to clean up any type tests left behind by WPD for use 1846 // in ICP (which is performed earlier than this in the regular LTO pipeline). 1847 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1848 1849 // Enable splitting late in the FullLTO post-link pipeline. This is done in 1850 // the same stage in the old pass manager (\ref addLateLTOOptimizationPasses). 1851 if (EnableHotColdSplit) 1852 MPM.addPass(HotColdSplittingPass()); 1853 1854 // Add late LTO optimization passes. 1855 // Delete basic blocks, which optimization passes may have killed. 1856 MPM.addPass(createModuleToFunctionPassAdaptor( 1857 SimplifyCFGPass(SimplifyCFGOptions().hoistCommonInsts(true)))); 1858 1859 // Drop bodies of available eternally objects to improve GlobalDCE. 1860 MPM.addPass(EliminateAvailableExternallyPass()); 1861 1862 // Now that we have optimized the program, discard unreachable functions. 1863 MPM.addPass(GlobalDCEPass()); 1864 1865 if (PTO.MergeFunctions) 1866 MPM.addPass(MergeFunctionsPass()); 1867 1868 // Emit annotation remarks. 1869 addAnnotationRemarksPass(MPM); 1870 1871 return MPM; 1872 } 1873 1874 ModulePassManager PassBuilder::buildO0DefaultPipeline(OptimizationLevel Level, 1875 bool LTOPreLink) { 1876 assert(Level == OptimizationLevel::O0 && 1877 "buildO0DefaultPipeline should only be used with O0"); 1878 1879 ModulePassManager MPM(DebugLogging); 1880 1881 // Add UniqueInternalLinkageNames Pass which renames internal linkage 1882 // symbols with unique names. 1883 if (PTO.UniqueLinkageNames) 1884 MPM.addPass(UniqueInternalLinkageNamesPass()); 1885 1886 if (PGOOpt && (PGOOpt->Action == PGOOptions::IRInstr || 1887 PGOOpt->Action == PGOOptions::IRUse)) 1888 addPGOInstrPassesForO0( 1889 MPM, 1890 /* RunProfileGen */ (PGOOpt->Action == PGOOptions::IRInstr), 1891 /* IsCS */ false, PGOOpt->ProfileFile, PGOOpt->ProfileRemappingFile); 1892 1893 for (auto &C : PipelineStartEPCallbacks) 1894 C(MPM, Level); 1895 for (auto &C : PipelineEarlySimplificationEPCallbacks) 1896 C(MPM, Level); 1897 1898 // Build a minimal pipeline based on the semantics required by LLVM, 1899 // which is just that always inlining occurs. Further, disable generating 1900 // lifetime intrinsics to avoid enabling further optimizations during 1901 // code generation. 1902 // However, we need to insert lifetime intrinsics to avoid invalid access 1903 // caused by multithreaded coroutines. 1904 MPM.addPass(AlwaysInlinerPass( 1905 /*InsertLifetimeIntrinsics=*/PTO.Coroutines)); 1906 1907 if (PTO.MergeFunctions) 1908 MPM.addPass(MergeFunctionsPass()); 1909 1910 if (EnableMatrix) 1911 MPM.addPass( 1912 createModuleToFunctionPassAdaptor(LowerMatrixIntrinsicsPass(true))); 1913 1914 if (!CGSCCOptimizerLateEPCallbacks.empty()) { 1915 CGSCCPassManager CGPM(DebugLogging); 1916 for (auto &C : CGSCCOptimizerLateEPCallbacks) 1917 C(CGPM, Level); 1918 if (!CGPM.isEmpty()) 1919 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPM))); 1920 } 1921 if (!LateLoopOptimizationsEPCallbacks.empty()) { 1922 LoopPassManager LPM(DebugLogging); 1923 for (auto &C : LateLoopOptimizationsEPCallbacks) 1924 C(LPM, Level); 1925 if (!LPM.isEmpty()) { 1926 MPM.addPass(createModuleToFunctionPassAdaptor( 1927 createFunctionToLoopPassAdaptor(std::move(LPM)))); 1928 } 1929 } 1930 if (!LoopOptimizerEndEPCallbacks.empty()) { 1931 LoopPassManager LPM(DebugLogging); 1932 for (auto &C : LoopOptimizerEndEPCallbacks) 1933 C(LPM, Level); 1934 if (!LPM.isEmpty()) { 1935 MPM.addPass(createModuleToFunctionPassAdaptor( 1936 createFunctionToLoopPassAdaptor(std::move(LPM)))); 1937 } 1938 } 1939 if (!ScalarOptimizerLateEPCallbacks.empty()) { 1940 FunctionPassManager FPM(DebugLogging); 1941 for (auto &C : ScalarOptimizerLateEPCallbacks) 1942 C(FPM, Level); 1943 if (!FPM.isEmpty()) 1944 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 1945 } 1946 if (!VectorizerStartEPCallbacks.empty()) { 1947 FunctionPassManager FPM(DebugLogging); 1948 for (auto &C : VectorizerStartEPCallbacks) 1949 C(FPM, Level); 1950 if (!FPM.isEmpty()) 1951 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 1952 } 1953 1954 if (PTO.Coroutines) { 1955 MPM.addPass(createModuleToFunctionPassAdaptor(CoroEarlyPass())); 1956 1957 CGSCCPassManager CGPM(DebugLogging); 1958 CGPM.addPass(CoroSplitPass()); 1959 CGPM.addPass(createCGSCCToFunctionPassAdaptor(CoroElidePass())); 1960 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPM))); 1961 1962 MPM.addPass(createModuleToFunctionPassAdaptor(CoroCleanupPass())); 1963 } 1964 1965 for (auto &C : OptimizerLastEPCallbacks) 1966 C(MPM, Level); 1967 1968 if (LTOPreLink) 1969 addRequiredLTOPreLinkPasses(MPM); 1970 1971 return MPM; 1972 } 1973 1974 AAManager PassBuilder::buildDefaultAAPipeline() { 1975 AAManager AA; 1976 1977 // The order in which these are registered determines their priority when 1978 // being queried. 1979 1980 // First we register the basic alias analysis that provides the majority of 1981 // per-function local AA logic. This is a stateless, on-demand local set of 1982 // AA techniques. 1983 AA.registerFunctionAnalysis<BasicAA>(); 1984 1985 // Next we query fast, specialized alias analyses that wrap IR-embedded 1986 // information about aliasing. 1987 AA.registerFunctionAnalysis<ScopedNoAliasAA>(); 1988 AA.registerFunctionAnalysis<TypeBasedAA>(); 1989 1990 // Add support for querying global aliasing information when available. 1991 // Because the `AAManager` is a function analysis and `GlobalsAA` is a module 1992 // analysis, all that the `AAManager` can do is query for any *cached* 1993 // results from `GlobalsAA` through a readonly proxy. 1994 AA.registerModuleAnalysis<GlobalsAA>(); 1995 1996 // Add target-specific alias analyses. 1997 if (TM) 1998 TM->registerDefaultAliasAnalyses(AA); 1999 2000 return AA; 2001 } 2002 2003 static Optional<int> parseRepeatPassName(StringRef Name) { 2004 if (!Name.consume_front("repeat<") || !Name.consume_back(">")) 2005 return None; 2006 int Count; 2007 if (Name.getAsInteger(0, Count) || Count <= 0) 2008 return None; 2009 return Count; 2010 } 2011 2012 static Optional<int> parseDevirtPassName(StringRef Name) { 2013 if (!Name.consume_front("devirt<") || !Name.consume_back(">")) 2014 return None; 2015 int Count; 2016 if (Name.getAsInteger(0, Count) || Count < 0) 2017 return None; 2018 return Count; 2019 } 2020 2021 static bool checkParametrizedPassName(StringRef Name, StringRef PassName) { 2022 if (!Name.consume_front(PassName)) 2023 return false; 2024 // normal pass name w/o parameters == default parameters 2025 if (Name.empty()) 2026 return true; 2027 return Name.startswith("<") && Name.endswith(">"); 2028 } 2029 2030 namespace { 2031 2032 /// This performs customized parsing of pass name with parameters. 2033 /// 2034 /// We do not need parametrization of passes in textual pipeline very often, 2035 /// yet on a rare occasion ability to specify parameters right there can be 2036 /// useful. 2037 /// 2038 /// \p Name - parameterized specification of a pass from a textual pipeline 2039 /// is a string in a form of : 2040 /// PassName '<' parameter-list '>' 2041 /// 2042 /// Parameter list is being parsed by the parser callable argument, \p Parser, 2043 /// It takes a string-ref of parameters and returns either StringError or a 2044 /// parameter list in a form of a custom parameters type, all wrapped into 2045 /// Expected<> template class. 2046 /// 2047 template <typename ParametersParseCallableT> 2048 auto parsePassParameters(ParametersParseCallableT &&Parser, StringRef Name, 2049 StringRef PassName) -> decltype(Parser(StringRef{})) { 2050 using ParametersT = typename decltype(Parser(StringRef{}))::value_type; 2051 2052 StringRef Params = Name; 2053 if (!Params.consume_front(PassName)) { 2054 assert(false && 2055 "unable to strip pass name from parametrized pass specification"); 2056 } 2057 if (Params.empty()) 2058 return ParametersT{}; 2059 if (!Params.consume_front("<") || !Params.consume_back(">")) { 2060 assert(false && "invalid format for parametrized pass name"); 2061 } 2062 2063 Expected<ParametersT> Result = Parser(Params); 2064 assert((Result || Result.template errorIsA<StringError>()) && 2065 "Pass parameter parser can only return StringErrors."); 2066 return Result; 2067 } 2068 2069 /// Parser of parameters for LoopUnroll pass. 2070 Expected<LoopUnrollOptions> parseLoopUnrollOptions(StringRef Params) { 2071 LoopUnrollOptions UnrollOpts; 2072 while (!Params.empty()) { 2073 StringRef ParamName; 2074 std::tie(ParamName, Params) = Params.split(';'); 2075 int OptLevel = StringSwitch<int>(ParamName) 2076 .Case("O0", 0) 2077 .Case("O1", 1) 2078 .Case("O2", 2) 2079 .Case("O3", 3) 2080 .Default(-1); 2081 if (OptLevel >= 0) { 2082 UnrollOpts.setOptLevel(OptLevel); 2083 continue; 2084 } 2085 if (ParamName.consume_front("full-unroll-max=")) { 2086 int Count; 2087 if (ParamName.getAsInteger(0, Count)) 2088 return make_error<StringError>( 2089 formatv("invalid LoopUnrollPass parameter '{0}' ", ParamName).str(), 2090 inconvertibleErrorCode()); 2091 UnrollOpts.setFullUnrollMaxCount(Count); 2092 continue; 2093 } 2094 2095 bool Enable = !ParamName.consume_front("no-"); 2096 if (ParamName == "partial") { 2097 UnrollOpts.setPartial(Enable); 2098 } else if (ParamName == "peeling") { 2099 UnrollOpts.setPeeling(Enable); 2100 } else if (ParamName == "profile-peeling") { 2101 UnrollOpts.setProfileBasedPeeling(Enable); 2102 } else if (ParamName == "runtime") { 2103 UnrollOpts.setRuntime(Enable); 2104 } else if (ParamName == "upperbound") { 2105 UnrollOpts.setUpperBound(Enable); 2106 } else { 2107 return make_error<StringError>( 2108 formatv("invalid LoopUnrollPass parameter '{0}' ", ParamName).str(), 2109 inconvertibleErrorCode()); 2110 } 2111 } 2112 return UnrollOpts; 2113 } 2114 2115 Expected<MemorySanitizerOptions> parseMSanPassOptions(StringRef Params) { 2116 MemorySanitizerOptions Result; 2117 while (!Params.empty()) { 2118 StringRef ParamName; 2119 std::tie(ParamName, Params) = Params.split(';'); 2120 2121 if (ParamName == "recover") { 2122 Result.Recover = true; 2123 } else if (ParamName == "kernel") { 2124 Result.Kernel = true; 2125 } else if (ParamName.consume_front("track-origins=")) { 2126 if (ParamName.getAsInteger(0, Result.TrackOrigins)) 2127 return make_error<StringError>( 2128 formatv("invalid argument to MemorySanitizer pass track-origins " 2129 "parameter: '{0}' ", 2130 ParamName) 2131 .str(), 2132 inconvertibleErrorCode()); 2133 } else { 2134 return make_error<StringError>( 2135 formatv("invalid MemorySanitizer pass parameter '{0}' ", ParamName) 2136 .str(), 2137 inconvertibleErrorCode()); 2138 } 2139 } 2140 return Result; 2141 } 2142 2143 /// Parser of parameters for SimplifyCFG pass. 2144 Expected<SimplifyCFGOptions> parseSimplifyCFGOptions(StringRef Params) { 2145 SimplifyCFGOptions Result; 2146 while (!Params.empty()) { 2147 StringRef ParamName; 2148 std::tie(ParamName, Params) = Params.split(';'); 2149 2150 bool Enable = !ParamName.consume_front("no-"); 2151 if (ParamName == "forward-switch-cond") { 2152 Result.forwardSwitchCondToPhi(Enable); 2153 } else if (ParamName == "switch-to-lookup") { 2154 Result.convertSwitchToLookupTable(Enable); 2155 } else if (ParamName == "keep-loops") { 2156 Result.needCanonicalLoops(Enable); 2157 } else if (ParamName == "hoist-common-insts") { 2158 Result.hoistCommonInsts(Enable); 2159 } else if (ParamName == "sink-common-insts") { 2160 Result.sinkCommonInsts(Enable); 2161 } else if (Enable && ParamName.consume_front("bonus-inst-threshold=")) { 2162 APInt BonusInstThreshold; 2163 if (ParamName.getAsInteger(0, BonusInstThreshold)) 2164 return make_error<StringError>( 2165 formatv("invalid argument to SimplifyCFG pass bonus-threshold " 2166 "parameter: '{0}' ", 2167 ParamName).str(), 2168 inconvertibleErrorCode()); 2169 Result.bonusInstThreshold(BonusInstThreshold.getSExtValue()); 2170 } else { 2171 return make_error<StringError>( 2172 formatv("invalid SimplifyCFG pass parameter '{0}' ", ParamName).str(), 2173 inconvertibleErrorCode()); 2174 } 2175 } 2176 return Result; 2177 } 2178 2179 /// Parser of parameters for LoopVectorize pass. 2180 Expected<LoopVectorizeOptions> parseLoopVectorizeOptions(StringRef Params) { 2181 LoopVectorizeOptions Opts; 2182 while (!Params.empty()) { 2183 StringRef ParamName; 2184 std::tie(ParamName, Params) = Params.split(';'); 2185 2186 bool Enable = !ParamName.consume_front("no-"); 2187 if (ParamName == "interleave-forced-only") { 2188 Opts.setInterleaveOnlyWhenForced(Enable); 2189 } else if (ParamName == "vectorize-forced-only") { 2190 Opts.setVectorizeOnlyWhenForced(Enable); 2191 } else { 2192 return make_error<StringError>( 2193 formatv("invalid LoopVectorize parameter '{0}' ", ParamName).str(), 2194 inconvertibleErrorCode()); 2195 } 2196 } 2197 return Opts; 2198 } 2199 2200 Expected<bool> parseLoopUnswitchOptions(StringRef Params) { 2201 bool Result = false; 2202 while (!Params.empty()) { 2203 StringRef ParamName; 2204 std::tie(ParamName, Params) = Params.split(';'); 2205 2206 bool Enable = !ParamName.consume_front("no-"); 2207 if (ParamName == "nontrivial") { 2208 Result = Enable; 2209 } else { 2210 return make_error<StringError>( 2211 formatv("invalid LoopUnswitch pass parameter '{0}' ", ParamName) 2212 .str(), 2213 inconvertibleErrorCode()); 2214 } 2215 } 2216 return Result; 2217 } 2218 2219 Expected<bool> parseMergedLoadStoreMotionOptions(StringRef Params) { 2220 bool Result = false; 2221 while (!Params.empty()) { 2222 StringRef ParamName; 2223 std::tie(ParamName, Params) = Params.split(';'); 2224 2225 bool Enable = !ParamName.consume_front("no-"); 2226 if (ParamName == "split-footer-bb") { 2227 Result = Enable; 2228 } else { 2229 return make_error<StringError>( 2230 formatv("invalid MergedLoadStoreMotion pass parameter '{0}' ", 2231 ParamName) 2232 .str(), 2233 inconvertibleErrorCode()); 2234 } 2235 } 2236 return Result; 2237 } 2238 2239 Expected<GVNOptions> parseGVNOptions(StringRef Params) { 2240 GVNOptions Result; 2241 while (!Params.empty()) { 2242 StringRef ParamName; 2243 std::tie(ParamName, Params) = Params.split(';'); 2244 2245 bool Enable = !ParamName.consume_front("no-"); 2246 if (ParamName == "pre") { 2247 Result.setPRE(Enable); 2248 } else if (ParamName == "load-pre") { 2249 Result.setLoadPRE(Enable); 2250 } else if (ParamName == "split-backedge-load-pre") { 2251 Result.setLoadPRESplitBackedge(Enable); 2252 } else if (ParamName == "memdep") { 2253 Result.setMemDep(Enable); 2254 } else { 2255 return make_error<StringError>( 2256 formatv("invalid GVN pass parameter '{0}' ", ParamName).str(), 2257 inconvertibleErrorCode()); 2258 } 2259 } 2260 return Result; 2261 } 2262 2263 Expected<StackLifetime::LivenessType> 2264 parseStackLifetimeOptions(StringRef Params) { 2265 StackLifetime::LivenessType Result = StackLifetime::LivenessType::May; 2266 while (!Params.empty()) { 2267 StringRef ParamName; 2268 std::tie(ParamName, Params) = Params.split(';'); 2269 2270 if (ParamName == "may") { 2271 Result = StackLifetime::LivenessType::May; 2272 } else if (ParamName == "must") { 2273 Result = StackLifetime::LivenessType::Must; 2274 } else { 2275 return make_error<StringError>( 2276 formatv("invalid StackLifetime parameter '{0}' ", ParamName).str(), 2277 inconvertibleErrorCode()); 2278 } 2279 } 2280 return Result; 2281 } 2282 2283 } // namespace 2284 2285 /// Tests whether a pass name starts with a valid prefix for a default pipeline 2286 /// alias. 2287 static bool startsWithDefaultPipelineAliasPrefix(StringRef Name) { 2288 return Name.startswith("default") || Name.startswith("thinlto") || 2289 Name.startswith("lto"); 2290 } 2291 2292 /// Tests whether registered callbacks will accept a given pass name. 2293 /// 2294 /// When parsing a pipeline text, the type of the outermost pipeline may be 2295 /// omitted, in which case the type is automatically determined from the first 2296 /// pass name in the text. This may be a name that is handled through one of the 2297 /// callbacks. We check this through the oridinary parsing callbacks by setting 2298 /// up a dummy PassManager in order to not force the client to also handle this 2299 /// type of query. 2300 template <typename PassManagerT, typename CallbacksT> 2301 static bool callbacksAcceptPassName(StringRef Name, CallbacksT &Callbacks) { 2302 if (!Callbacks.empty()) { 2303 PassManagerT DummyPM; 2304 for (auto &CB : Callbacks) 2305 if (CB(Name, DummyPM, {})) 2306 return true; 2307 } 2308 return false; 2309 } 2310 2311 template <typename CallbacksT> 2312 static bool isModulePassName(StringRef Name, CallbacksT &Callbacks) { 2313 // Manually handle aliases for pre-configured pipeline fragments. 2314 if (startsWithDefaultPipelineAliasPrefix(Name)) 2315 return DefaultAliasRegex.match(Name); 2316 2317 // Explicitly handle pass manager names. 2318 if (Name == "module") 2319 return true; 2320 if (Name == "cgscc") 2321 return true; 2322 if (Name == "function") 2323 return true; 2324 2325 // Explicitly handle custom-parsed pass names. 2326 if (parseRepeatPassName(Name)) 2327 return true; 2328 2329 #define MODULE_PASS(NAME, CREATE_PASS) \ 2330 if (Name == NAME) \ 2331 return true; 2332 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 2333 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2334 return true; 2335 #include "PassRegistry.def" 2336 2337 return callbacksAcceptPassName<ModulePassManager>(Name, Callbacks); 2338 } 2339 2340 template <typename CallbacksT> 2341 static bool isCGSCCPassName(StringRef Name, CallbacksT &Callbacks) { 2342 // Explicitly handle pass manager names. 2343 if (Name == "cgscc") 2344 return true; 2345 if (Name == "function") 2346 return true; 2347 2348 // Explicitly handle custom-parsed pass names. 2349 if (parseRepeatPassName(Name)) 2350 return true; 2351 if (parseDevirtPassName(Name)) 2352 return true; 2353 2354 #define CGSCC_PASS(NAME, CREATE_PASS) \ 2355 if (Name == NAME) \ 2356 return true; 2357 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 2358 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2359 return true; 2360 #include "PassRegistry.def" 2361 2362 return callbacksAcceptPassName<CGSCCPassManager>(Name, Callbacks); 2363 } 2364 2365 template <typename CallbacksT> 2366 static bool isFunctionPassName(StringRef Name, CallbacksT &Callbacks) { 2367 // Explicitly handle pass manager names. 2368 if (Name == "function") 2369 return true; 2370 if (Name == "loop" || Name == "loop-mssa") 2371 return true; 2372 2373 // Explicitly handle custom-parsed pass names. 2374 if (parseRepeatPassName(Name)) 2375 return true; 2376 2377 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2378 if (Name == NAME) \ 2379 return true; 2380 #define FUNCTION_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2381 if (checkParametrizedPassName(Name, NAME)) \ 2382 return true; 2383 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 2384 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2385 return true; 2386 #include "PassRegistry.def" 2387 2388 return callbacksAcceptPassName<FunctionPassManager>(Name, Callbacks); 2389 } 2390 2391 template <typename CallbacksT> 2392 static bool isLoopPassName(StringRef Name, CallbacksT &Callbacks) { 2393 // Explicitly handle pass manager names. 2394 if (Name == "loop" || Name == "loop-mssa") 2395 return true; 2396 2397 // Explicitly handle custom-parsed pass names. 2398 if (parseRepeatPassName(Name)) 2399 return true; 2400 2401 #define LOOP_PASS(NAME, CREATE_PASS) \ 2402 if (Name == NAME) \ 2403 return true; 2404 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2405 if (checkParametrizedPassName(Name, NAME)) \ 2406 return true; 2407 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 2408 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2409 return true; 2410 #include "PassRegistry.def" 2411 2412 return callbacksAcceptPassName<LoopPassManager>(Name, Callbacks); 2413 } 2414 2415 Optional<std::vector<PassBuilder::PipelineElement>> 2416 PassBuilder::parsePipelineText(StringRef Text) { 2417 std::vector<PipelineElement> ResultPipeline; 2418 2419 SmallVector<std::vector<PipelineElement> *, 4> PipelineStack = { 2420 &ResultPipeline}; 2421 for (;;) { 2422 std::vector<PipelineElement> &Pipeline = *PipelineStack.back(); 2423 size_t Pos = Text.find_first_of(",()"); 2424 Pipeline.push_back({Text.substr(0, Pos), {}}); 2425 2426 // If we have a single terminating name, we're done. 2427 if (Pos == Text.npos) 2428 break; 2429 2430 char Sep = Text[Pos]; 2431 Text = Text.substr(Pos + 1); 2432 if (Sep == ',') 2433 // Just a name ending in a comma, continue. 2434 continue; 2435 2436 if (Sep == '(') { 2437 // Push the inner pipeline onto the stack to continue processing. 2438 PipelineStack.push_back(&Pipeline.back().InnerPipeline); 2439 continue; 2440 } 2441 2442 assert(Sep == ')' && "Bogus separator!"); 2443 // When handling the close parenthesis, we greedily consume them to avoid 2444 // empty strings in the pipeline. 2445 do { 2446 // If we try to pop the outer pipeline we have unbalanced parentheses. 2447 if (PipelineStack.size() == 1) 2448 return None; 2449 2450 PipelineStack.pop_back(); 2451 } while (Text.consume_front(")")); 2452 2453 // Check if we've finished parsing. 2454 if (Text.empty()) 2455 break; 2456 2457 // Otherwise, the end of an inner pipeline always has to be followed by 2458 // a comma, and then we can continue. 2459 if (!Text.consume_front(",")) 2460 return None; 2461 } 2462 2463 if (PipelineStack.size() > 1) 2464 // Unbalanced paretheses. 2465 return None; 2466 2467 assert(PipelineStack.back() == &ResultPipeline && 2468 "Wrong pipeline at the bottom of the stack!"); 2469 return {std::move(ResultPipeline)}; 2470 } 2471 2472 Error PassBuilder::parseModulePass(ModulePassManager &MPM, 2473 const PipelineElement &E) { 2474 auto &Name = E.Name; 2475 auto &InnerPipeline = E.InnerPipeline; 2476 2477 // First handle complex passes like the pass managers which carry pipelines. 2478 if (!InnerPipeline.empty()) { 2479 if (Name == "module") { 2480 ModulePassManager NestedMPM(DebugLogging); 2481 if (auto Err = parseModulePassPipeline(NestedMPM, InnerPipeline)) 2482 return Err; 2483 MPM.addPass(std::move(NestedMPM)); 2484 return Error::success(); 2485 } 2486 if (Name == "cgscc") { 2487 CGSCCPassManager CGPM(DebugLogging); 2488 if (auto Err = parseCGSCCPassPipeline(CGPM, InnerPipeline)) 2489 return Err; 2490 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPM))); 2491 return Error::success(); 2492 } 2493 if (Name == "function") { 2494 FunctionPassManager FPM(DebugLogging); 2495 if (auto Err = parseFunctionPassPipeline(FPM, InnerPipeline)) 2496 return Err; 2497 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 2498 return Error::success(); 2499 } 2500 if (auto Count = parseRepeatPassName(Name)) { 2501 ModulePassManager NestedMPM(DebugLogging); 2502 if (auto Err = parseModulePassPipeline(NestedMPM, InnerPipeline)) 2503 return Err; 2504 MPM.addPass(createRepeatedPass(*Count, std::move(NestedMPM))); 2505 return Error::success(); 2506 } 2507 2508 for (auto &C : ModulePipelineParsingCallbacks) 2509 if (C(Name, MPM, InnerPipeline)) 2510 return Error::success(); 2511 2512 // Normal passes can't have pipelines. 2513 return make_error<StringError>( 2514 formatv("invalid use of '{0}' pass as module pipeline", Name).str(), 2515 inconvertibleErrorCode()); 2516 ; 2517 } 2518 2519 // Manually handle aliases for pre-configured pipeline fragments. 2520 if (startsWithDefaultPipelineAliasPrefix(Name)) { 2521 SmallVector<StringRef, 3> Matches; 2522 if (!DefaultAliasRegex.match(Name, &Matches)) 2523 return make_error<StringError>( 2524 formatv("unknown default pipeline alias '{0}'", Name).str(), 2525 inconvertibleErrorCode()); 2526 2527 assert(Matches.size() == 3 && "Must capture two matched strings!"); 2528 2529 OptimizationLevel L = StringSwitch<OptimizationLevel>(Matches[2]) 2530 .Case("O0", OptimizationLevel::O0) 2531 .Case("O1", OptimizationLevel::O1) 2532 .Case("O2", OptimizationLevel::O2) 2533 .Case("O3", OptimizationLevel::O3) 2534 .Case("Os", OptimizationLevel::Os) 2535 .Case("Oz", OptimizationLevel::Oz); 2536 if (L == OptimizationLevel::O0 && Matches[1] != "thinlto" && 2537 Matches[1] != "lto") { 2538 MPM.addPass(buildO0DefaultPipeline(L, Matches[1] == "thinlto-pre-link" || 2539 Matches[1] == "lto-pre-link")); 2540 return Error::success(); 2541 } 2542 2543 // This is consistent with old pass manager invoked via opt, but 2544 // inconsistent with clang. Clang doesn't enable loop vectorization 2545 // but does enable slp vectorization at Oz. 2546 PTO.LoopVectorization = 2547 L.getSpeedupLevel() > 1 && L != OptimizationLevel::Oz; 2548 PTO.SLPVectorization = 2549 L.getSpeedupLevel() > 1 && L != OptimizationLevel::Oz; 2550 2551 if (Matches[1] == "default") { 2552 MPM.addPass(buildPerModuleDefaultPipeline(L)); 2553 } else if (Matches[1] == "thinlto-pre-link") { 2554 MPM.addPass(buildThinLTOPreLinkDefaultPipeline(L)); 2555 } else if (Matches[1] == "thinlto") { 2556 MPM.addPass(buildThinLTODefaultPipeline(L, nullptr)); 2557 } else if (Matches[1] == "lto-pre-link") { 2558 MPM.addPass(buildLTOPreLinkDefaultPipeline(L)); 2559 } else { 2560 assert(Matches[1] == "lto" && "Not one of the matched options!"); 2561 MPM.addPass(buildLTODefaultPipeline(L, nullptr)); 2562 } 2563 return Error::success(); 2564 } 2565 2566 // Finally expand the basic registered passes from the .inc file. 2567 #define MODULE_PASS(NAME, CREATE_PASS) \ 2568 if (Name == NAME) { \ 2569 MPM.addPass(CREATE_PASS); \ 2570 return Error::success(); \ 2571 } 2572 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 2573 if (Name == "require<" NAME ">") { \ 2574 MPM.addPass( \ 2575 RequireAnalysisPass< \ 2576 std::remove_reference<decltype(CREATE_PASS)>::type, Module>()); \ 2577 return Error::success(); \ 2578 } \ 2579 if (Name == "invalidate<" NAME ">") { \ 2580 MPM.addPass(InvalidateAnalysisPass< \ 2581 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2582 return Error::success(); \ 2583 } 2584 #define CGSCC_PASS(NAME, CREATE_PASS) \ 2585 if (Name == NAME) { \ 2586 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(CREATE_PASS)); \ 2587 return Error::success(); \ 2588 } 2589 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2590 if (Name == NAME) { \ 2591 MPM.addPass(createModuleToFunctionPassAdaptor(CREATE_PASS)); \ 2592 return Error::success(); \ 2593 } 2594 #define FUNCTION_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2595 if (checkParametrizedPassName(Name, NAME)) { \ 2596 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2597 if (!Params) \ 2598 return Params.takeError(); \ 2599 MPM.addPass(createModuleToFunctionPassAdaptor(CREATE_PASS(Params.get()))); \ 2600 return Error::success(); \ 2601 } 2602 #define LOOP_PASS(NAME, CREATE_PASS) \ 2603 if (Name == NAME) { \ 2604 MPM.addPass( \ 2605 createModuleToFunctionPassAdaptor(createFunctionToLoopPassAdaptor( \ 2606 CREATE_PASS, false, false, DebugLogging))); \ 2607 return Error::success(); \ 2608 } 2609 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2610 if (checkParametrizedPassName(Name, NAME)) { \ 2611 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2612 if (!Params) \ 2613 return Params.takeError(); \ 2614 MPM.addPass( \ 2615 createModuleToFunctionPassAdaptor(createFunctionToLoopPassAdaptor( \ 2616 CREATE_PASS(Params.get()), false, false, DebugLogging))); \ 2617 return Error::success(); \ 2618 } 2619 #include "PassRegistry.def" 2620 2621 for (auto &C : ModulePipelineParsingCallbacks) 2622 if (C(Name, MPM, InnerPipeline)) 2623 return Error::success(); 2624 return make_error<StringError>( 2625 formatv("unknown module pass '{0}'", Name).str(), 2626 inconvertibleErrorCode()); 2627 } 2628 2629 Error PassBuilder::parseCGSCCPass(CGSCCPassManager &CGPM, 2630 const PipelineElement &E) { 2631 auto &Name = E.Name; 2632 auto &InnerPipeline = E.InnerPipeline; 2633 2634 // First handle complex passes like the pass managers which carry pipelines. 2635 if (!InnerPipeline.empty()) { 2636 if (Name == "cgscc") { 2637 CGSCCPassManager NestedCGPM(DebugLogging); 2638 if (auto Err = parseCGSCCPassPipeline(NestedCGPM, InnerPipeline)) 2639 return Err; 2640 // Add the nested pass manager with the appropriate adaptor. 2641 CGPM.addPass(std::move(NestedCGPM)); 2642 return Error::success(); 2643 } 2644 if (Name == "function") { 2645 FunctionPassManager FPM(DebugLogging); 2646 if (auto Err = parseFunctionPassPipeline(FPM, InnerPipeline)) 2647 return Err; 2648 // Add the nested pass manager with the appropriate adaptor. 2649 CGPM.addPass(createCGSCCToFunctionPassAdaptor(std::move(FPM))); 2650 return Error::success(); 2651 } 2652 if (auto Count = parseRepeatPassName(Name)) { 2653 CGSCCPassManager NestedCGPM(DebugLogging); 2654 if (auto Err = parseCGSCCPassPipeline(NestedCGPM, InnerPipeline)) 2655 return Err; 2656 CGPM.addPass(createRepeatedPass(*Count, std::move(NestedCGPM))); 2657 return Error::success(); 2658 } 2659 if (auto MaxRepetitions = parseDevirtPassName(Name)) { 2660 CGSCCPassManager NestedCGPM(DebugLogging); 2661 if (auto Err = parseCGSCCPassPipeline(NestedCGPM, InnerPipeline)) 2662 return Err; 2663 CGPM.addPass( 2664 createDevirtSCCRepeatedPass(std::move(NestedCGPM), *MaxRepetitions)); 2665 return Error::success(); 2666 } 2667 2668 for (auto &C : CGSCCPipelineParsingCallbacks) 2669 if (C(Name, CGPM, 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 cgscc pipeline", Name).str(), 2675 inconvertibleErrorCode()); 2676 } 2677 2678 // Now expand the basic registered passes from the .inc file. 2679 #define CGSCC_PASS(NAME, CREATE_PASS) \ 2680 if (Name == NAME) { \ 2681 CGPM.addPass(CREATE_PASS); \ 2682 return Error::success(); \ 2683 } 2684 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 2685 if (Name == "require<" NAME ">") { \ 2686 CGPM.addPass(RequireAnalysisPass< \ 2687 std::remove_reference<decltype(CREATE_PASS)>::type, \ 2688 LazyCallGraph::SCC, CGSCCAnalysisManager, LazyCallGraph &, \ 2689 CGSCCUpdateResult &>()); \ 2690 return Error::success(); \ 2691 } \ 2692 if (Name == "invalidate<" NAME ">") { \ 2693 CGPM.addPass(InvalidateAnalysisPass< \ 2694 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2695 return Error::success(); \ 2696 } 2697 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2698 if (Name == NAME) { \ 2699 CGPM.addPass(createCGSCCToFunctionPassAdaptor(CREATE_PASS)); \ 2700 return Error::success(); \ 2701 } 2702 #define FUNCTION_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2703 if (checkParametrizedPassName(Name, NAME)) { \ 2704 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2705 if (!Params) \ 2706 return Params.takeError(); \ 2707 CGPM.addPass(createCGSCCToFunctionPassAdaptor(CREATE_PASS(Params.get()))); \ 2708 return Error::success(); \ 2709 } 2710 #define LOOP_PASS(NAME, CREATE_PASS) \ 2711 if (Name == NAME) { \ 2712 CGPM.addPass( \ 2713 createCGSCCToFunctionPassAdaptor(createFunctionToLoopPassAdaptor( \ 2714 CREATE_PASS, false, false, DebugLogging))); \ 2715 return Error::success(); \ 2716 } 2717 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2718 if (checkParametrizedPassName(Name, NAME)) { \ 2719 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2720 if (!Params) \ 2721 return Params.takeError(); \ 2722 CGPM.addPass( \ 2723 createCGSCCToFunctionPassAdaptor(createFunctionToLoopPassAdaptor( \ 2724 CREATE_PASS(Params.get()), false, false, DebugLogging))); \ 2725 return Error::success(); \ 2726 } 2727 #include "PassRegistry.def" 2728 2729 for (auto &C : CGSCCPipelineParsingCallbacks) 2730 if (C(Name, CGPM, InnerPipeline)) 2731 return Error::success(); 2732 return make_error<StringError>( 2733 formatv("unknown cgscc pass '{0}'", Name).str(), 2734 inconvertibleErrorCode()); 2735 } 2736 2737 Error PassBuilder::parseFunctionPass(FunctionPassManager &FPM, 2738 const PipelineElement &E) { 2739 auto &Name = E.Name; 2740 auto &InnerPipeline = E.InnerPipeline; 2741 2742 // First handle complex passes like the pass managers which carry pipelines. 2743 if (!InnerPipeline.empty()) { 2744 if (Name == "function") { 2745 FunctionPassManager NestedFPM(DebugLogging); 2746 if (auto Err = parseFunctionPassPipeline(NestedFPM, InnerPipeline)) 2747 return Err; 2748 // Add the nested pass manager with the appropriate adaptor. 2749 FPM.addPass(std::move(NestedFPM)); 2750 return Error::success(); 2751 } 2752 if (Name == "loop" || Name == "loop-mssa") { 2753 LoopPassManager LPM(DebugLogging); 2754 if (auto Err = parseLoopPassPipeline(LPM, InnerPipeline)) 2755 return Err; 2756 // Add the nested pass manager with the appropriate adaptor. 2757 bool UseMemorySSA = (Name == "loop-mssa"); 2758 bool UseBFI = llvm::any_of( 2759 InnerPipeline, [](auto Pipeline) { return Pipeline.Name == "licm"; }); 2760 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM), UseMemorySSA, 2761 UseBFI, DebugLogging)); 2762 return Error::success(); 2763 } 2764 if (auto Count = parseRepeatPassName(Name)) { 2765 FunctionPassManager NestedFPM(DebugLogging); 2766 if (auto Err = parseFunctionPassPipeline(NestedFPM, InnerPipeline)) 2767 return Err; 2768 FPM.addPass(createRepeatedPass(*Count, std::move(NestedFPM))); 2769 return Error::success(); 2770 } 2771 2772 for (auto &C : FunctionPipelineParsingCallbacks) 2773 if (C(Name, FPM, InnerPipeline)) 2774 return Error::success(); 2775 2776 // Normal passes can't have pipelines. 2777 return make_error<StringError>( 2778 formatv("invalid use of '{0}' pass as function pipeline", Name).str(), 2779 inconvertibleErrorCode()); 2780 } 2781 2782 // Now expand the basic registered passes from the .inc file. 2783 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2784 if (Name == NAME) { \ 2785 FPM.addPass(CREATE_PASS); \ 2786 return Error::success(); \ 2787 } 2788 #define FUNCTION_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2789 if (checkParametrizedPassName(Name, NAME)) { \ 2790 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2791 if (!Params) \ 2792 return Params.takeError(); \ 2793 FPM.addPass(CREATE_PASS(Params.get())); \ 2794 return Error::success(); \ 2795 } 2796 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 2797 if (Name == "require<" NAME ">") { \ 2798 FPM.addPass( \ 2799 RequireAnalysisPass< \ 2800 std::remove_reference<decltype(CREATE_PASS)>::type, Function>()); \ 2801 return Error::success(); \ 2802 } \ 2803 if (Name == "invalidate<" NAME ">") { \ 2804 FPM.addPass(InvalidateAnalysisPass< \ 2805 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2806 return Error::success(); \ 2807 } 2808 // FIXME: UseMemorySSA is set to false. Maybe we could do things like: 2809 // bool UseMemorySSA = !("canon-freeze" || "loop-predication" || 2810 // "guard-widening"); 2811 // The risk is that it may become obsolete if we're not careful. 2812 #define LOOP_PASS(NAME, CREATE_PASS) \ 2813 if (Name == NAME) { \ 2814 FPM.addPass(createFunctionToLoopPassAdaptor(CREATE_PASS, false, false, \ 2815 DebugLogging)); \ 2816 return Error::success(); \ 2817 } 2818 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2819 if (checkParametrizedPassName(Name, NAME)) { \ 2820 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2821 if (!Params) \ 2822 return Params.takeError(); \ 2823 FPM.addPass(createFunctionToLoopPassAdaptor(CREATE_PASS(Params.get()), \ 2824 false, false, DebugLogging)); \ 2825 return Error::success(); \ 2826 } 2827 #include "PassRegistry.def" 2828 2829 for (auto &C : FunctionPipelineParsingCallbacks) 2830 if (C(Name, FPM, InnerPipeline)) 2831 return Error::success(); 2832 return make_error<StringError>( 2833 formatv("unknown function pass '{0}'", Name).str(), 2834 inconvertibleErrorCode()); 2835 } 2836 2837 Error PassBuilder::parseLoopPass(LoopPassManager &LPM, 2838 const PipelineElement &E) { 2839 StringRef Name = E.Name; 2840 auto &InnerPipeline = E.InnerPipeline; 2841 2842 // First handle complex passes like the pass managers which carry pipelines. 2843 if (!InnerPipeline.empty()) { 2844 if (Name == "loop") { 2845 LoopPassManager NestedLPM(DebugLogging); 2846 if (auto Err = parseLoopPassPipeline(NestedLPM, InnerPipeline)) 2847 return Err; 2848 // Add the nested pass manager with the appropriate adaptor. 2849 LPM.addPass(std::move(NestedLPM)); 2850 return Error::success(); 2851 } 2852 if (auto Count = parseRepeatPassName(Name)) { 2853 LoopPassManager NestedLPM(DebugLogging); 2854 if (auto Err = parseLoopPassPipeline(NestedLPM, InnerPipeline)) 2855 return Err; 2856 LPM.addPass(createRepeatedPass(*Count, std::move(NestedLPM))); 2857 return Error::success(); 2858 } 2859 2860 for (auto &C : LoopPipelineParsingCallbacks) 2861 if (C(Name, LPM, InnerPipeline)) 2862 return Error::success(); 2863 2864 // Normal passes can't have pipelines. 2865 return make_error<StringError>( 2866 formatv("invalid use of '{0}' pass as loop pipeline", Name).str(), 2867 inconvertibleErrorCode()); 2868 } 2869 2870 // Now expand the basic registered passes from the .inc file. 2871 #define LOOP_PASS(NAME, CREATE_PASS) \ 2872 if (Name == NAME) { \ 2873 LPM.addPass(CREATE_PASS); \ 2874 return Error::success(); \ 2875 } 2876 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2877 if (checkParametrizedPassName(Name, NAME)) { \ 2878 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2879 if (!Params) \ 2880 return Params.takeError(); \ 2881 LPM.addPass(CREATE_PASS(Params.get())); \ 2882 return Error::success(); \ 2883 } 2884 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 2885 if (Name == "require<" NAME ">") { \ 2886 LPM.addPass(RequireAnalysisPass< \ 2887 std::remove_reference<decltype(CREATE_PASS)>::type, Loop, \ 2888 LoopAnalysisManager, LoopStandardAnalysisResults &, \ 2889 LPMUpdater &>()); \ 2890 return Error::success(); \ 2891 } \ 2892 if (Name == "invalidate<" NAME ">") { \ 2893 LPM.addPass(InvalidateAnalysisPass< \ 2894 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2895 return Error::success(); \ 2896 } 2897 #include "PassRegistry.def" 2898 2899 for (auto &C : LoopPipelineParsingCallbacks) 2900 if (C(Name, LPM, InnerPipeline)) 2901 return Error::success(); 2902 return make_error<StringError>(formatv("unknown loop pass '{0}'", Name).str(), 2903 inconvertibleErrorCode()); 2904 } 2905 2906 bool PassBuilder::parseAAPassName(AAManager &AA, StringRef Name) { 2907 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 2908 if (Name == NAME) { \ 2909 AA.registerModuleAnalysis< \ 2910 std::remove_reference<decltype(CREATE_PASS)>::type>(); \ 2911 return true; \ 2912 } 2913 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 2914 if (Name == NAME) { \ 2915 AA.registerFunctionAnalysis< \ 2916 std::remove_reference<decltype(CREATE_PASS)>::type>(); \ 2917 return true; \ 2918 } 2919 #include "PassRegistry.def" 2920 2921 for (auto &C : AAParsingCallbacks) 2922 if (C(Name, AA)) 2923 return true; 2924 return false; 2925 } 2926 2927 Error PassBuilder::parseLoopPassPipeline(LoopPassManager &LPM, 2928 ArrayRef<PipelineElement> Pipeline) { 2929 for (const auto &Element : Pipeline) { 2930 if (auto Err = parseLoopPass(LPM, Element)) 2931 return Err; 2932 } 2933 return Error::success(); 2934 } 2935 2936 Error PassBuilder::parseFunctionPassPipeline( 2937 FunctionPassManager &FPM, ArrayRef<PipelineElement> Pipeline) { 2938 for (const auto &Element : Pipeline) { 2939 if (auto Err = parseFunctionPass(FPM, Element)) 2940 return Err; 2941 } 2942 return Error::success(); 2943 } 2944 2945 Error PassBuilder::parseCGSCCPassPipeline(CGSCCPassManager &CGPM, 2946 ArrayRef<PipelineElement> Pipeline) { 2947 for (const auto &Element : Pipeline) { 2948 if (auto Err = parseCGSCCPass(CGPM, Element)) 2949 return Err; 2950 } 2951 return Error::success(); 2952 } 2953 2954 void PassBuilder::crossRegisterProxies(LoopAnalysisManager &LAM, 2955 FunctionAnalysisManager &FAM, 2956 CGSCCAnalysisManager &CGAM, 2957 ModuleAnalysisManager &MAM) { 2958 MAM.registerPass([&] { return FunctionAnalysisManagerModuleProxy(FAM); }); 2959 MAM.registerPass([&] { return CGSCCAnalysisManagerModuleProxy(CGAM); }); 2960 CGAM.registerPass([&] { return ModuleAnalysisManagerCGSCCProxy(MAM); }); 2961 FAM.registerPass([&] { return CGSCCAnalysisManagerFunctionProxy(CGAM); }); 2962 FAM.registerPass([&] { return ModuleAnalysisManagerFunctionProxy(MAM); }); 2963 FAM.registerPass([&] { return LoopAnalysisManagerFunctionProxy(LAM); }); 2964 LAM.registerPass([&] { return FunctionAnalysisManagerLoopProxy(FAM); }); 2965 } 2966 2967 Error PassBuilder::parseModulePassPipeline(ModulePassManager &MPM, 2968 ArrayRef<PipelineElement> Pipeline) { 2969 for (const auto &Element : Pipeline) { 2970 if (auto Err = parseModulePass(MPM, Element)) 2971 return Err; 2972 } 2973 return Error::success(); 2974 } 2975 2976 // Primary pass pipeline description parsing routine for a \c ModulePassManager 2977 // FIXME: Should this routine accept a TargetMachine or require the caller to 2978 // pre-populate the analysis managers with target-specific stuff? 2979 Error PassBuilder::parsePassPipeline(ModulePassManager &MPM, 2980 StringRef PipelineText) { 2981 auto Pipeline = parsePipelineText(PipelineText); 2982 if (!Pipeline || Pipeline->empty()) 2983 return make_error<StringError>( 2984 formatv("invalid pipeline '{0}'", PipelineText).str(), 2985 inconvertibleErrorCode()); 2986 2987 // If the first name isn't at the module layer, wrap the pipeline up 2988 // automatically. 2989 StringRef FirstName = Pipeline->front().Name; 2990 2991 if (!isModulePassName(FirstName, ModulePipelineParsingCallbacks)) { 2992 if (isCGSCCPassName(FirstName, CGSCCPipelineParsingCallbacks)) { 2993 Pipeline = {{"cgscc", std::move(*Pipeline)}}; 2994 } else if (isFunctionPassName(FirstName, 2995 FunctionPipelineParsingCallbacks)) { 2996 Pipeline = {{"function", std::move(*Pipeline)}}; 2997 } else if (isLoopPassName(FirstName, LoopPipelineParsingCallbacks)) { 2998 Pipeline = {{"function", {{"loop", std::move(*Pipeline)}}}}; 2999 } else { 3000 for (auto &C : TopLevelPipelineParsingCallbacks) 3001 if (C(MPM, *Pipeline, DebugLogging)) 3002 return Error::success(); 3003 3004 // Unknown pass or pipeline name! 3005 auto &InnerPipeline = Pipeline->front().InnerPipeline; 3006 return make_error<StringError>( 3007 formatv("unknown {0} name '{1}'", 3008 (InnerPipeline.empty() ? "pass" : "pipeline"), FirstName) 3009 .str(), 3010 inconvertibleErrorCode()); 3011 } 3012 } 3013 3014 if (auto Err = parseModulePassPipeline(MPM, *Pipeline)) 3015 return Err; 3016 return Error::success(); 3017 } 3018 3019 // Primary pass pipeline description parsing routine for a \c CGSCCPassManager 3020 Error PassBuilder::parsePassPipeline(CGSCCPassManager &CGPM, 3021 StringRef PipelineText) { 3022 auto Pipeline = parsePipelineText(PipelineText); 3023 if (!Pipeline || Pipeline->empty()) 3024 return make_error<StringError>( 3025 formatv("invalid pipeline '{0}'", PipelineText).str(), 3026 inconvertibleErrorCode()); 3027 3028 StringRef FirstName = Pipeline->front().Name; 3029 if (!isCGSCCPassName(FirstName, CGSCCPipelineParsingCallbacks)) 3030 return make_error<StringError>( 3031 formatv("unknown cgscc pass '{0}' in pipeline '{1}'", FirstName, 3032 PipelineText) 3033 .str(), 3034 inconvertibleErrorCode()); 3035 3036 if (auto Err = parseCGSCCPassPipeline(CGPM, *Pipeline)) 3037 return Err; 3038 return Error::success(); 3039 } 3040 3041 // Primary pass pipeline description parsing routine for a \c 3042 // FunctionPassManager 3043 Error PassBuilder::parsePassPipeline(FunctionPassManager &FPM, 3044 StringRef PipelineText) { 3045 auto Pipeline = parsePipelineText(PipelineText); 3046 if (!Pipeline || Pipeline->empty()) 3047 return make_error<StringError>( 3048 formatv("invalid pipeline '{0}'", PipelineText).str(), 3049 inconvertibleErrorCode()); 3050 3051 StringRef FirstName = Pipeline->front().Name; 3052 if (!isFunctionPassName(FirstName, FunctionPipelineParsingCallbacks)) 3053 return make_error<StringError>( 3054 formatv("unknown function pass '{0}' in pipeline '{1}'", FirstName, 3055 PipelineText) 3056 .str(), 3057 inconvertibleErrorCode()); 3058 3059 if (auto Err = parseFunctionPassPipeline(FPM, *Pipeline)) 3060 return Err; 3061 return Error::success(); 3062 } 3063 3064 // Primary pass pipeline description parsing routine for a \c LoopPassManager 3065 Error PassBuilder::parsePassPipeline(LoopPassManager &CGPM, 3066 StringRef PipelineText) { 3067 auto Pipeline = parsePipelineText(PipelineText); 3068 if (!Pipeline || Pipeline->empty()) 3069 return make_error<StringError>( 3070 formatv("invalid pipeline '{0}'", PipelineText).str(), 3071 inconvertibleErrorCode()); 3072 3073 if (auto Err = parseLoopPassPipeline(CGPM, *Pipeline)) 3074 return Err; 3075 3076 return Error::success(); 3077 } 3078 3079 Error PassBuilder::parseAAPipeline(AAManager &AA, StringRef PipelineText) { 3080 // If the pipeline just consists of the word 'default' just replace the AA 3081 // manager with our default one. 3082 if (PipelineText == "default") { 3083 AA = buildDefaultAAPipeline(); 3084 return Error::success(); 3085 } 3086 3087 while (!PipelineText.empty()) { 3088 StringRef Name; 3089 std::tie(Name, PipelineText) = PipelineText.split(','); 3090 if (!parseAAPassName(AA, Name)) 3091 return make_error<StringError>( 3092 formatv("unknown alias analysis name '{0}'", Name).str(), 3093 inconvertibleErrorCode()); 3094 } 3095 3096 return Error::success(); 3097 } 3098 3099 bool PassBuilder::isAAPassName(StringRef PassName) { 3100 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3101 if (PassName == NAME) \ 3102 return true; 3103 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3104 if (PassName == NAME) \ 3105 return true; 3106 #include "PassRegistry.def" 3107 return false; 3108 } 3109 3110 bool PassBuilder::isAnalysisPassName(StringRef PassName) { 3111 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 3112 if (PassName == NAME) \ 3113 return true; 3114 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 3115 if (PassName == NAME) \ 3116 return true; 3117 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 3118 if (PassName == NAME) \ 3119 return true; 3120 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 3121 if (PassName == NAME) \ 3122 return true; 3123 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3124 if (PassName == NAME) \ 3125 return true; 3126 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3127 if (PassName == NAME) \ 3128 return true; 3129 #include "PassRegistry.def" 3130 return false; 3131 } 3132 3133 static void printPassName(StringRef PassName, raw_ostream &OS) { 3134 OS << " " << PassName << "\n"; 3135 } 3136 3137 void PassBuilder::printPassNames(raw_ostream &OS) { 3138 // TODO: print pass descriptions when they are available 3139 3140 OS << "Module passes:\n"; 3141 #define MODULE_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3142 #include "PassRegistry.def" 3143 3144 OS << "Module analyses:\n"; 3145 #define MODULE_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3146 #include "PassRegistry.def" 3147 3148 OS << "Module alias analyses:\n"; 3149 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3150 #include "PassRegistry.def" 3151 3152 OS << "CGSCC passes:\n"; 3153 #define CGSCC_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3154 #include "PassRegistry.def" 3155 3156 OS << "CGSCC analyses:\n"; 3157 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3158 #include "PassRegistry.def" 3159 3160 OS << "Function passes:\n"; 3161 #define FUNCTION_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3162 #include "PassRegistry.def" 3163 3164 OS << "Function analyses:\n"; 3165 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3166 #include "PassRegistry.def" 3167 3168 OS << "Function alias analyses:\n"; 3169 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3170 #include "PassRegistry.def" 3171 3172 OS << "Loop passes:\n"; 3173 #define LOOP_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3174 #include "PassRegistry.def" 3175 3176 OS << "Loop analyses:\n"; 3177 #define LOOP_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3178 #include "PassRegistry.def" 3179 } 3180 3181 void PassBuilder::registerParseTopLevelPipelineCallback( 3182 const std::function<bool(ModulePassManager &, ArrayRef<PipelineElement>, 3183 bool DebugLogging)> &C) { 3184 TopLevelPipelineParsingCallbacks.push_back(C); 3185 } 3186