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/RelLookupTableConverter.h" 231 #include "llvm/Transforms/Utils/StripGCRelocates.h" 232 #include "llvm/Transforms/Utils/StripNonLineTableDebugInfo.h" 233 #include "llvm/Transforms/Utils/SymbolRewriter.h" 234 #include "llvm/Transforms/Utils/UnifyFunctionExitNodes.h" 235 #include "llvm/Transforms/Utils/UnifyLoopExits.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 static cl::opt<InliningAdvisorMode> UseInlineAdvisor( 244 "enable-ml-inliner", cl::init(InliningAdvisorMode::Default), cl::Hidden, 245 cl::desc("Enable ML policy for inliner. Currently trained for -Oz only"), 246 cl::values(clEnumValN(InliningAdvisorMode::Default, "default", 247 "Heuristics-based inliner version."), 248 clEnumValN(InliningAdvisorMode::Development, "development", 249 "Use development mode (runtime-loadable model)."), 250 clEnumValN(InliningAdvisorMode::Release, "release", 251 "Use release mode (AOT-compiled model)."))); 252 253 static cl::opt<bool> EnableSyntheticCounts( 254 "enable-npm-synthetic-counts", cl::init(false), cl::Hidden, cl::ZeroOrMore, 255 cl::desc("Run synthetic function entry count generation " 256 "pass")); 257 258 static const Regex DefaultAliasRegex( 259 "^(default|thinlto-pre-link|thinlto|lto-pre-link|lto)<(O[0123sz])>$"); 260 261 /// Flag to enable inline deferral during PGO. 262 static cl::opt<bool> 263 EnablePGOInlineDeferral("enable-npm-pgo-inline-deferral", cl::init(true), 264 cl::Hidden, 265 cl::desc("Enable inline deferral during PGO")); 266 267 static cl::opt<bool> EnableMemProfiler("enable-mem-prof", cl::init(false), 268 cl::Hidden, cl::ZeroOrMore, 269 cl::desc("Enable memory profiler")); 270 271 static cl::opt<bool> PerformMandatoryInliningsFirst( 272 "mandatory-inlining-first", cl::init(true), cl::Hidden, cl::ZeroOrMore, 273 cl::desc("Perform mandatory inlinings module-wide, before performing " 274 "inlining.")); 275 276 static cl::opt<bool> EnableO3NonTrivialUnswitching( 277 "enable-npm-O3-nontrivial-unswitch", cl::init(true), cl::Hidden, 278 cl::ZeroOrMore, cl::desc("Enable non-trivial loop unswitching for -O3")); 279 280 static cl::opt<bool> DoNotRerunFunctionPasses( 281 "cgscc-npm-no-fp-rerun", cl::init(false), 282 cl::desc("Do not rerun function passes wrapped by the scc pass adapter, if " 283 "they were run already and the function hasn't changed.")); 284 285 PipelineTuningOptions::PipelineTuningOptions() { 286 LoopInterleaving = true; 287 LoopVectorization = true; 288 SLPVectorization = false; 289 LoopUnrolling = true; 290 ForgetAllSCEVInLoopUnroll = ForgetSCEVInLoopUnroll; 291 Coroutines = false; 292 LicmMssaOptCap = SetLicmMssaOptCap; 293 LicmMssaNoAccForPromotionCap = SetLicmMssaNoAccForPromotionCap; 294 CallGraphProfile = true; 295 MergeFunctions = false; 296 } 297 298 namespace llvm { 299 extern cl::opt<unsigned> MaxDevirtIterations; 300 extern cl::opt<bool> EnableConstraintElimination; 301 extern cl::opt<bool> EnableGVNHoist; 302 extern cl::opt<bool> EnableGVNSink; 303 extern cl::opt<bool> EnableHotColdSplit; 304 extern cl::opt<bool> EnableIROutliner; 305 extern cl::opt<bool> EnableOrderFileInstrumentation; 306 extern cl::opt<bool> EnableCHR; 307 extern cl::opt<bool> EnableLoopInterchange; 308 extern cl::opt<bool> EnableUnrollAndJam; 309 extern cl::opt<bool> EnableLoopFlatten; 310 extern cl::opt<bool> RunNewGVN; 311 extern cl::opt<bool> RunPartialInlining; 312 extern cl::opt<bool> ExtraVectorizerPasses; 313 314 extern cl::opt<bool> FlattenedProfileUsed; 315 316 extern cl::opt<AttributorRunOption> AttributorRun; 317 extern cl::opt<bool> EnableKnowledgeRetention; 318 319 extern cl::opt<bool> EnableMatrix; 320 321 extern cl::opt<bool> DisablePreInliner; 322 extern cl::opt<int> PreInlineThreshold; 323 } // namespace llvm 324 325 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O0 = { 326 /*SpeedLevel*/ 0, 327 /*SizeLevel*/ 0}; 328 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O1 = { 329 /*SpeedLevel*/ 1, 330 /*SizeLevel*/ 0}; 331 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O2 = { 332 /*SpeedLevel*/ 2, 333 /*SizeLevel*/ 0}; 334 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O3 = { 335 /*SpeedLevel*/ 3, 336 /*SizeLevel*/ 0}; 337 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::Os = { 338 /*SpeedLevel*/ 2, 339 /*SizeLevel*/ 1}; 340 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::Oz = { 341 /*SpeedLevel*/ 2, 342 /*SizeLevel*/ 2}; 343 344 namespace { 345 346 // The following passes/analyses have custom names, otherwise their name will 347 // include `(anonymous namespace)`. These are special since they are only for 348 // testing purposes and don't live in a header file. 349 350 /// No-op module pass which does nothing. 351 struct NoOpModulePass : PassInfoMixin<NoOpModulePass> { 352 PreservedAnalyses run(Module &M, ModuleAnalysisManager &) { 353 return PreservedAnalyses::all(); 354 } 355 356 static StringRef name() { return "NoOpModulePass"; } 357 }; 358 359 /// No-op module analysis. 360 class NoOpModuleAnalysis : public AnalysisInfoMixin<NoOpModuleAnalysis> { 361 friend AnalysisInfoMixin<NoOpModuleAnalysis>; 362 static AnalysisKey Key; 363 364 public: 365 struct Result {}; 366 Result run(Module &, ModuleAnalysisManager &) { return Result(); } 367 static StringRef name() { return "NoOpModuleAnalysis"; } 368 }; 369 370 /// No-op CGSCC pass which does nothing. 371 struct NoOpCGSCCPass : PassInfoMixin<NoOpCGSCCPass> { 372 PreservedAnalyses run(LazyCallGraph::SCC &C, CGSCCAnalysisManager &, 373 LazyCallGraph &, CGSCCUpdateResult &UR) { 374 return PreservedAnalyses::all(); 375 } 376 static StringRef name() { return "NoOpCGSCCPass"; } 377 }; 378 379 /// No-op CGSCC analysis. 380 class NoOpCGSCCAnalysis : public AnalysisInfoMixin<NoOpCGSCCAnalysis> { 381 friend AnalysisInfoMixin<NoOpCGSCCAnalysis>; 382 static AnalysisKey Key; 383 384 public: 385 struct Result {}; 386 Result run(LazyCallGraph::SCC &, CGSCCAnalysisManager &, LazyCallGraph &G) { 387 return Result(); 388 } 389 static StringRef name() { return "NoOpCGSCCAnalysis"; } 390 }; 391 392 /// No-op function pass which does nothing. 393 struct NoOpFunctionPass : PassInfoMixin<NoOpFunctionPass> { 394 PreservedAnalyses run(Function &F, FunctionAnalysisManager &) { 395 return PreservedAnalyses::all(); 396 } 397 static StringRef name() { return "NoOpFunctionPass"; } 398 }; 399 400 /// No-op function analysis. 401 class NoOpFunctionAnalysis : public AnalysisInfoMixin<NoOpFunctionAnalysis> { 402 friend AnalysisInfoMixin<NoOpFunctionAnalysis>; 403 static AnalysisKey Key; 404 405 public: 406 struct Result {}; 407 Result run(Function &, FunctionAnalysisManager &) { return Result(); } 408 static StringRef name() { return "NoOpFunctionAnalysis"; } 409 }; 410 411 /// No-op loop pass which does nothing. 412 struct NoOpLoopPass : PassInfoMixin<NoOpLoopPass> { 413 PreservedAnalyses run(Loop &L, LoopAnalysisManager &, 414 LoopStandardAnalysisResults &, LPMUpdater &) { 415 return PreservedAnalyses::all(); 416 } 417 static StringRef name() { return "NoOpLoopPass"; } 418 }; 419 420 /// No-op loop analysis. 421 class NoOpLoopAnalysis : public AnalysisInfoMixin<NoOpLoopAnalysis> { 422 friend AnalysisInfoMixin<NoOpLoopAnalysis>; 423 static AnalysisKey Key; 424 425 public: 426 struct Result {}; 427 Result run(Loop &, LoopAnalysisManager &, LoopStandardAnalysisResults &) { 428 return Result(); 429 } 430 static StringRef name() { return "NoOpLoopAnalysis"; } 431 }; 432 433 AnalysisKey NoOpModuleAnalysis::Key; 434 AnalysisKey NoOpCGSCCAnalysis::Key; 435 AnalysisKey NoOpFunctionAnalysis::Key; 436 AnalysisKey NoOpLoopAnalysis::Key; 437 438 /// Whether or not we should populate a PassInstrumentationCallbacks's class to 439 /// pass name map. 440 /// 441 /// This is for optimization purposes so we don't populate it if we never use 442 /// it. This should be updated if new pass instrumentation wants to use the map. 443 /// We currently only use this for --print-before/after. 444 bool shouldPopulateClassToPassNames() { 445 return !printBeforePasses().empty() || !printAfterPasses().empty(); 446 } 447 448 } // namespace 449 450 PassBuilder::PassBuilder(TargetMachine *TM, PipelineTuningOptions PTO, 451 Optional<PGOOptions> PGOOpt, 452 PassInstrumentationCallbacks *PIC) 453 : TM(TM), PTO(PTO), PGOOpt(PGOOpt), PIC(PIC) { 454 if (TM) 455 TM->registerPassBuilderCallbacks(*this); 456 if (PIC && shouldPopulateClassToPassNames()) { 457 #define MODULE_PASS(NAME, CREATE_PASS) \ 458 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 459 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 460 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 461 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 462 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 463 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 464 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 465 #define LOOP_PASS(NAME, CREATE_PASS) \ 466 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 467 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 468 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 469 #define CGSCC_PASS(NAME, CREATE_PASS) \ 470 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 471 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 472 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 473 #include "PassRegistry.def" 474 } 475 } 476 477 void PassBuilder::invokePeepholeEPCallbacks( 478 FunctionPassManager &FPM, PassBuilder::OptimizationLevel Level) { 479 for (auto &C : PeepholeEPCallbacks) 480 C(FPM, Level); 481 } 482 483 void PassBuilder::registerModuleAnalyses(ModuleAnalysisManager &MAM) { 484 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 485 MAM.registerPass([&] { return CREATE_PASS; }); 486 #include "PassRegistry.def" 487 488 for (auto &C : ModuleAnalysisRegistrationCallbacks) 489 C(MAM); 490 } 491 492 void PassBuilder::registerCGSCCAnalyses(CGSCCAnalysisManager &CGAM) { 493 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 494 CGAM.registerPass([&] { return CREATE_PASS; }); 495 #include "PassRegistry.def" 496 497 for (auto &C : CGSCCAnalysisRegistrationCallbacks) 498 C(CGAM); 499 } 500 501 void PassBuilder::registerFunctionAnalyses(FunctionAnalysisManager &FAM) { 502 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 503 FAM.registerPass([&] { return CREATE_PASS; }); 504 #include "PassRegistry.def" 505 506 for (auto &C : FunctionAnalysisRegistrationCallbacks) 507 C(FAM); 508 } 509 510 void PassBuilder::registerLoopAnalyses(LoopAnalysisManager &LAM) { 511 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 512 LAM.registerPass([&] { return CREATE_PASS; }); 513 #include "PassRegistry.def" 514 515 for (auto &C : LoopAnalysisRegistrationCallbacks) 516 C(LAM); 517 } 518 519 // Helper to add AnnotationRemarksPass. 520 static void addAnnotationRemarksPass(ModulePassManager &MPM) { 521 FunctionPassManager FPM; 522 FPM.addPass(AnnotationRemarksPass()); 523 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 524 } 525 526 // Helper to check if the current compilation phase is preparing for LTO 527 static bool isLTOPreLink(ThinOrFullLTOPhase Phase) { 528 return Phase == ThinOrFullLTOPhase::ThinLTOPreLink || 529 Phase == ThinOrFullLTOPhase::FullLTOPreLink; 530 } 531 532 // TODO: Investigate the cost/benefit of tail call elimination on debugging. 533 FunctionPassManager 534 PassBuilder::buildO1FunctionSimplificationPipeline(OptimizationLevel Level, 535 ThinOrFullLTOPhase Phase) { 536 537 FunctionPassManager FPM; 538 539 // Form SSA out of local memory accesses after breaking apart aggregates into 540 // scalars. 541 FPM.addPass(SROA()); 542 543 // Catch trivial redundancies 544 FPM.addPass(EarlyCSEPass(true /* Enable mem-ssa. */)); 545 546 // Hoisting of scalars and load expressions. 547 FPM.addPass(SimplifyCFGPass()); 548 FPM.addPass(InstCombinePass()); 549 550 FPM.addPass(LibCallsShrinkWrapPass()); 551 552 invokePeepholeEPCallbacks(FPM, Level); 553 554 FPM.addPass(SimplifyCFGPass()); 555 556 // Form canonically associated expression trees, and simplify the trees using 557 // basic mathematical properties. For example, this will form (nearly) 558 // minimal multiplication trees. 559 FPM.addPass(ReassociatePass()); 560 561 // Add the primary loop simplification pipeline. 562 // FIXME: Currently this is split into two loop pass pipelines because we run 563 // some function passes in between them. These can and should be removed 564 // and/or replaced by scheduling the loop pass equivalents in the correct 565 // positions. But those equivalent passes aren't powerful enough yet. 566 // Specifically, `SimplifyCFGPass` and `InstCombinePass` are currently still 567 // used. We have `LoopSimplifyCFGPass` which isn't yet powerful enough yet to 568 // fully replace `SimplifyCFGPass`, and the closest to the other we have is 569 // `LoopInstSimplify`. 570 LoopPassManager LPM1, LPM2; 571 572 // Simplify the loop body. We do this initially to clean up after other loop 573 // passes run, either when iterating on a loop or on inner loops with 574 // implications on the outer loop. 575 LPM1.addPass(LoopInstSimplifyPass()); 576 LPM1.addPass(LoopSimplifyCFGPass()); 577 578 // Try to remove as much code from the loop header as possible, 579 // to reduce amount of IR that will have to be duplicated. 580 // TODO: Investigate promotion cap for O1. 581 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 582 583 LPM1.addPass(LoopRotatePass(/* Disable header duplication */ true, 584 isLTOPreLink(Phase))); 585 // TODO: Investigate promotion cap for O1. 586 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 587 LPM1.addPass(SimpleLoopUnswitchPass()); 588 589 LPM2.addPass(LoopIdiomRecognizePass()); 590 LPM2.addPass(IndVarSimplifyPass()); 591 592 for (auto &C : LateLoopOptimizationsEPCallbacks) 593 C(LPM2, Level); 594 595 LPM2.addPass(LoopDeletionPass()); 596 597 if (EnableLoopInterchange) 598 LPM2.addPass(LoopInterchangePass()); 599 600 // Do not enable unrolling in PreLinkThinLTO phase during sample PGO 601 // because it changes IR to makes profile annotation in back compile 602 // inaccurate. The normal unroller doesn't pay attention to forced full unroll 603 // attributes so we need to make sure and allow the full unroll pass to pay 604 // attention to it. 605 if (Phase != ThinOrFullLTOPhase::ThinLTOPreLink || !PGOOpt || 606 PGOOpt->Action != PGOOptions::SampleUse) 607 LPM2.addPass(LoopFullUnrollPass(Level.getSpeedupLevel(), 608 /* OnlyWhenForced= */ !PTO.LoopUnrolling, 609 PTO.ForgetAllSCEVInLoopUnroll)); 610 611 for (auto &C : LoopOptimizerEndEPCallbacks) 612 C(LPM2, Level); 613 614 // We provide the opt remark emitter pass for LICM to use. We only need to do 615 // this once as it is immutable. 616 FPM.addPass( 617 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 618 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM1), 619 EnableMSSALoopDependency, 620 /*UseBlockFrequencyInfo=*/true)); 621 FPM.addPass(SimplifyCFGPass()); 622 FPM.addPass(InstCombinePass()); 623 if (EnableLoopFlatten) 624 FPM.addPass(LoopFlattenPass()); 625 // The loop passes in LPM2 (LoopFullUnrollPass) do not preserve MemorySSA. 626 // *All* loop passes must preserve it, in order to be able to use it. 627 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM2), 628 /*UseMemorySSA=*/false, 629 /*UseBlockFrequencyInfo=*/false)); 630 631 // Delete small array after loop unroll. 632 FPM.addPass(SROA()); 633 634 // Specially optimize memory movement as it doesn't look like dataflow in SSA. 635 FPM.addPass(MemCpyOptPass()); 636 637 // Sparse conditional constant propagation. 638 // FIXME: It isn't clear why we do this *after* loop passes rather than 639 // before... 640 FPM.addPass(SCCPPass()); 641 642 // Delete dead bit computations (instcombine runs after to fold away the dead 643 // computations, and then ADCE will run later to exploit any new DCE 644 // opportunities that creates). 645 FPM.addPass(BDCEPass()); 646 647 // Run instcombine after redundancy and dead bit elimination to exploit 648 // opportunities opened up by them. 649 FPM.addPass(InstCombinePass()); 650 invokePeepholeEPCallbacks(FPM, Level); 651 652 if (PTO.Coroutines) 653 FPM.addPass(CoroElidePass()); 654 655 for (auto &C : ScalarOptimizerLateEPCallbacks) 656 C(FPM, Level); 657 658 // Finally, do an expensive DCE pass to catch all the dead code exposed by 659 // the simplifications and basic cleanup after all the simplifications. 660 // TODO: Investigate if this is too expensive. 661 FPM.addPass(ADCEPass()); 662 FPM.addPass(SimplifyCFGPass()); 663 FPM.addPass(InstCombinePass()); 664 invokePeepholeEPCallbacks(FPM, Level); 665 666 return FPM; 667 } 668 669 FunctionPassManager 670 PassBuilder::buildFunctionSimplificationPipeline(OptimizationLevel Level, 671 ThinOrFullLTOPhase Phase) { 672 assert(Level != OptimizationLevel::O0 && "Must request optimizations!"); 673 674 // The O1 pipeline has a separate pipeline creation function to simplify 675 // construction readability. 676 if (Level.getSpeedupLevel() == 1) 677 return buildO1FunctionSimplificationPipeline(Level, Phase); 678 679 FunctionPassManager FPM; 680 681 // Form SSA out of local memory accesses after breaking apart aggregates into 682 // scalars. 683 FPM.addPass(SROA()); 684 685 // Catch trivial redundancies 686 FPM.addPass(EarlyCSEPass(true /* Enable mem-ssa. */)); 687 if (EnableKnowledgeRetention) 688 FPM.addPass(AssumeSimplifyPass()); 689 690 // Hoisting of scalars and load expressions. 691 if (EnableGVNHoist) 692 FPM.addPass(GVNHoistPass()); 693 694 // Global value numbering based sinking. 695 if (EnableGVNSink) { 696 FPM.addPass(GVNSinkPass()); 697 FPM.addPass(SimplifyCFGPass()); 698 } 699 700 if (EnableConstraintElimination) 701 FPM.addPass(ConstraintEliminationPass()); 702 703 // Speculative execution if the target has divergent branches; otherwise nop. 704 FPM.addPass(SpeculativeExecutionPass(/* OnlyIfDivergentTarget =*/true)); 705 706 // Optimize based on known information about branches, and cleanup afterward. 707 FPM.addPass(JumpThreadingPass()); 708 FPM.addPass(CorrelatedValuePropagationPass()); 709 710 FPM.addPass(SimplifyCFGPass()); 711 if (Level == OptimizationLevel::O3) 712 FPM.addPass(AggressiveInstCombinePass()); 713 FPM.addPass(InstCombinePass()); 714 715 if (!Level.isOptimizingForSize()) 716 FPM.addPass(LibCallsShrinkWrapPass()); 717 718 invokePeepholeEPCallbacks(FPM, Level); 719 720 // For PGO use pipeline, try to optimize memory intrinsics such as memcpy 721 // using the size value profile. Don't perform this when optimizing for size. 722 if (PGOOpt && PGOOpt->Action == PGOOptions::IRUse && 723 !Level.isOptimizingForSize()) 724 FPM.addPass(PGOMemOPSizeOpt()); 725 726 FPM.addPass(TailCallElimPass()); 727 FPM.addPass(SimplifyCFGPass()); 728 729 // Form canonically associated expression trees, and simplify the trees using 730 // basic mathematical properties. For example, this will form (nearly) 731 // minimal multiplication trees. 732 FPM.addPass(ReassociatePass()); 733 734 // Add the primary loop simplification pipeline. 735 // FIXME: Currently this is split into two loop pass pipelines because we run 736 // some function passes in between them. These can and should be removed 737 // and/or replaced by scheduling the loop pass equivalents in the correct 738 // positions. But those equivalent passes aren't powerful enough yet. 739 // Specifically, `SimplifyCFGPass` and `InstCombinePass` are currently still 740 // used. We have `LoopSimplifyCFGPass` which isn't yet powerful enough yet to 741 // fully replace `SimplifyCFGPass`, and the closest to the other we have is 742 // `LoopInstSimplify`. 743 LoopPassManager LPM1, LPM2; 744 745 // Simplify the loop body. We do this initially to clean up after other loop 746 // passes run, either when iterating on a loop or on inner loops with 747 // implications on the outer loop. 748 LPM1.addPass(LoopInstSimplifyPass()); 749 LPM1.addPass(LoopSimplifyCFGPass()); 750 751 // Try to remove as much code from the loop header as possible, 752 // to reduce amount of IR that will have to be duplicated. 753 // TODO: Investigate promotion cap for O1. 754 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 755 756 // Disable header duplication in loop rotation at -Oz. 757 LPM1.addPass( 758 LoopRotatePass(Level != OptimizationLevel::Oz, isLTOPreLink(Phase))); 759 // TODO: Investigate promotion cap for O1. 760 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 761 LPM1.addPass( 762 SimpleLoopUnswitchPass(/* NonTrivial */ Level == OptimizationLevel::O3 && 763 EnableO3NonTrivialUnswitching)); 764 LPM2.addPass(LoopIdiomRecognizePass()); 765 LPM2.addPass(IndVarSimplifyPass()); 766 767 for (auto &C : LateLoopOptimizationsEPCallbacks) 768 C(LPM2, Level); 769 770 LPM2.addPass(LoopDeletionPass()); 771 772 if (EnableLoopInterchange) 773 LPM2.addPass(LoopInterchangePass()); 774 775 // Do not enable unrolling in PreLinkThinLTO phase during sample PGO 776 // because it changes IR to makes profile annotation in back compile 777 // inaccurate. The normal unroller doesn't pay attention to forced full unroll 778 // attributes so we need to make sure and allow the full unroll pass to pay 779 // attention to it. 780 if (Phase != ThinOrFullLTOPhase::ThinLTOPreLink || !PGOOpt || 781 PGOOpt->Action != PGOOptions::SampleUse) 782 LPM2.addPass(LoopFullUnrollPass(Level.getSpeedupLevel(), 783 /* OnlyWhenForced= */ !PTO.LoopUnrolling, 784 PTO.ForgetAllSCEVInLoopUnroll)); 785 786 for (auto &C : LoopOptimizerEndEPCallbacks) 787 C(LPM2, Level); 788 789 // We provide the opt remark emitter pass for LICM to use. We only need to do 790 // this once as it is immutable. 791 FPM.addPass( 792 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 793 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM1), 794 EnableMSSALoopDependency, 795 /*UseBlockFrequencyInfo=*/true)); 796 FPM.addPass(SimplifyCFGPass()); 797 FPM.addPass(InstCombinePass()); 798 if (EnableLoopFlatten) 799 FPM.addPass(LoopFlattenPass()); 800 // The loop passes in LPM2 (LoopIdiomRecognizePass, IndVarSimplifyPass, 801 // LoopDeletionPass and LoopFullUnrollPass) do not preserve MemorySSA. 802 // *All* loop passes must preserve it, in order to be able to use it. 803 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM2), 804 /*UseMemorySSA=*/false, 805 /*UseBlockFrequencyInfo=*/false)); 806 807 // Delete small array after loop unroll. 808 FPM.addPass(SROA()); 809 810 // Eliminate redundancies. 811 FPM.addPass(MergedLoadStoreMotionPass()); 812 if (RunNewGVN) 813 FPM.addPass(NewGVNPass()); 814 else 815 FPM.addPass(GVN()); 816 817 // Sparse conditional constant propagation. 818 // FIXME: It isn't clear why we do this *after* loop passes rather than 819 // before... 820 FPM.addPass(SCCPPass()); 821 822 // Delete dead bit computations (instcombine runs after to fold away the dead 823 // computations, and then ADCE will run later to exploit any new DCE 824 // opportunities that creates). 825 FPM.addPass(BDCEPass()); 826 827 // Run instcombine after redundancy and dead bit elimination to exploit 828 // opportunities opened up by them. 829 FPM.addPass(InstCombinePass()); 830 invokePeepholeEPCallbacks(FPM, Level); 831 832 // Re-consider control flow based optimizations after redundancy elimination, 833 // redo DCE, etc. 834 FPM.addPass(JumpThreadingPass()); 835 FPM.addPass(CorrelatedValuePropagationPass()); 836 837 // Finally, do an expensive DCE pass to catch all the dead code exposed by 838 // the simplifications and basic cleanup after all the simplifications. 839 // TODO: Investigate if this is too expensive. 840 FPM.addPass(ADCEPass()); 841 842 // Specially optimize memory movement as it doesn't look like dataflow in SSA. 843 FPM.addPass(MemCpyOptPass()); 844 845 FPM.addPass(DSEPass()); 846 FPM.addPass(createFunctionToLoopPassAdaptor( 847 LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap), 848 EnableMSSALoopDependency, /*UseBlockFrequencyInfo=*/true)); 849 850 if (PTO.Coroutines) 851 FPM.addPass(CoroElidePass()); 852 853 for (auto &C : ScalarOptimizerLateEPCallbacks) 854 C(FPM, Level); 855 856 FPM.addPass(SimplifyCFGPass( 857 SimplifyCFGOptions().hoistCommonInsts(true).sinkCommonInsts(true))); 858 FPM.addPass(InstCombinePass()); 859 invokePeepholeEPCallbacks(FPM, Level); 860 861 if (EnableCHR && Level == OptimizationLevel::O3 && PGOOpt && 862 (PGOOpt->Action == PGOOptions::IRUse || 863 PGOOpt->Action == PGOOptions::SampleUse)) 864 FPM.addPass(ControlHeightReductionPass()); 865 866 return FPM; 867 } 868 869 void PassBuilder::addRequiredLTOPreLinkPasses(ModulePassManager &MPM) { 870 MPM.addPass(CanonicalizeAliasesPass()); 871 MPM.addPass(NameAnonGlobalPass()); 872 } 873 874 void PassBuilder::addPGOInstrPasses(ModulePassManager &MPM, 875 PassBuilder::OptimizationLevel Level, 876 bool RunProfileGen, bool IsCS, 877 std::string ProfileFile, 878 std::string ProfileRemappingFile) { 879 assert(Level != OptimizationLevel::O0 && "Not expecting O0 here!"); 880 if (!IsCS && !DisablePreInliner) { 881 InlineParams IP; 882 883 IP.DefaultThreshold = PreInlineThreshold; 884 885 // FIXME: The hint threshold has the same value used by the regular inliner 886 // when not optimzing for size. This should probably be lowered after 887 // performance testing. 888 // FIXME: this comment is cargo culted from the old pass manager, revisit). 889 IP.HintThreshold = Level.isOptimizingForSize() ? PreInlineThreshold : 325; 890 ModuleInlinerWrapperPass MIWP(IP); 891 CGSCCPassManager &CGPipeline = MIWP.getPM(); 892 893 FunctionPassManager FPM; 894 FPM.addPass(SROA()); 895 FPM.addPass(EarlyCSEPass()); // Catch trivial redundancies. 896 FPM.addPass(SimplifyCFGPass()); // Merge & remove basic blocks. 897 FPM.addPass(InstCombinePass()); // Combine silly sequences. 898 invokePeepholeEPCallbacks(FPM, Level); 899 900 CGPipeline.addPass(createCGSCCToFunctionPassAdaptor(std::move(FPM))); 901 902 MPM.addPass(std::move(MIWP)); 903 904 // Delete anything that is now dead to make sure that we don't instrument 905 // dead code. Instrumentation can end up keeping dead code around and 906 // dramatically increase code size. 907 MPM.addPass(GlobalDCEPass()); 908 } 909 910 if (!RunProfileGen) { 911 assert(!ProfileFile.empty() && "Profile use expecting a profile file!"); 912 MPM.addPass(PGOInstrumentationUse(ProfileFile, ProfileRemappingFile, IsCS)); 913 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 914 // RequireAnalysisPass for PSI before subsequent non-module passes. 915 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 916 return; 917 } 918 919 // Perform PGO instrumentation. 920 MPM.addPass(PGOInstrumentationGen(IsCS)); 921 922 FunctionPassManager FPM; 923 // Disable header duplication in loop rotation at -Oz. 924 FPM.addPass(createFunctionToLoopPassAdaptor( 925 LoopRotatePass(Level != OptimizationLevel::Oz), EnableMSSALoopDependency, 926 /*UseBlockFrequencyInfo=*/false)); 927 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 928 929 // Add the profile lowering pass. 930 InstrProfOptions Options; 931 if (!ProfileFile.empty()) 932 Options.InstrProfileOutput = ProfileFile; 933 // Do counter promotion at Level greater than O0. 934 Options.DoCounterPromotion = true; 935 Options.UseBFIInPromotion = IsCS; 936 MPM.addPass(InstrProfiling(Options, IsCS)); 937 } 938 939 void PassBuilder::addPGOInstrPassesForO0(ModulePassManager &MPM, 940 bool RunProfileGen, bool IsCS, 941 std::string ProfileFile, 942 std::string ProfileRemappingFile) { 943 if (!RunProfileGen) { 944 assert(!ProfileFile.empty() && "Profile use expecting a profile file!"); 945 MPM.addPass(PGOInstrumentationUse(ProfileFile, ProfileRemappingFile, IsCS)); 946 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 947 // RequireAnalysisPass for PSI before subsequent non-module passes. 948 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 949 return; 950 } 951 952 // Perform PGO instrumentation. 953 MPM.addPass(PGOInstrumentationGen(IsCS)); 954 // Add the profile lowering pass. 955 InstrProfOptions Options; 956 if (!ProfileFile.empty()) 957 Options.InstrProfileOutput = ProfileFile; 958 // Do not do counter promotion at O0. 959 Options.DoCounterPromotion = false; 960 Options.UseBFIInPromotion = IsCS; 961 MPM.addPass(InstrProfiling(Options, IsCS)); 962 } 963 964 static InlineParams 965 getInlineParamsFromOptLevel(PassBuilder::OptimizationLevel Level) { 966 return getInlineParams(Level.getSpeedupLevel(), Level.getSizeLevel()); 967 } 968 969 ModuleInlinerWrapperPass 970 PassBuilder::buildInlinerPipeline(OptimizationLevel Level, 971 ThinOrFullLTOPhase Phase) { 972 InlineParams IP = getInlineParamsFromOptLevel(Level); 973 if (Phase == ThinOrFullLTOPhase::ThinLTOPreLink && PGOOpt && 974 PGOOpt->Action == PGOOptions::SampleUse) 975 IP.HotCallSiteThreshold = 0; 976 977 if (PGOOpt) 978 IP.EnableDeferral = EnablePGOInlineDeferral; 979 980 ModuleInlinerWrapperPass MIWP(IP, PerformMandatoryInliningsFirst, 981 UseInlineAdvisor, MaxDevirtIterations); 982 983 // Require the GlobalsAA analysis for the module so we can query it within 984 // the CGSCC pipeline. 985 MIWP.addModulePass(RequireAnalysisPass<GlobalsAA, Module>()); 986 // Invalidate AAManager so it can be recreated and pick up the newly available 987 // GlobalsAA. 988 MIWP.addModulePass( 989 createModuleToFunctionPassAdaptor(InvalidateAnalysisPass<AAManager>())); 990 991 // Require the ProfileSummaryAnalysis for the module so we can query it within 992 // the inliner pass. 993 MIWP.addModulePass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 994 995 // Now begin the main postorder CGSCC pipeline. 996 // FIXME: The current CGSCC pipeline has its origins in the legacy pass 997 // manager and trying to emulate its precise behavior. Much of this doesn't 998 // make a lot of sense and we should revisit the core CGSCC structure. 999 CGSCCPassManager &MainCGPipeline = MIWP.getPM(); 1000 1001 // Note: historically, the PruneEH pass was run first to deduce nounwind and 1002 // generally clean up exception handling overhead. It isn't clear this is 1003 // valuable as the inliner doesn't currently care whether it is inlining an 1004 // invoke or a call. 1005 1006 if (AttributorRun & AttributorRunOption::CGSCC) 1007 MainCGPipeline.addPass(AttributorCGSCCPass()); 1008 1009 if (PTO.Coroutines) 1010 MainCGPipeline.addPass(CoroSplitPass(Level != OptimizationLevel::O0)); 1011 1012 // Now deduce any function attributes based in the current code. 1013 MainCGPipeline.addPass(PostOrderFunctionAttrsPass()); 1014 1015 // When at O3 add argument promotion to the pass pipeline. 1016 // FIXME: It isn't at all clear why this should be limited to O3. 1017 if (Level == OptimizationLevel::O3) 1018 MainCGPipeline.addPass(ArgumentPromotionPass()); 1019 1020 // Try to perform OpenMP specific optimizations. This is a (quick!) no-op if 1021 // there are no OpenMP runtime calls present in the module. 1022 if (Level == OptimizationLevel::O2 || Level == OptimizationLevel::O3) 1023 MainCGPipeline.addPass(OpenMPOptCGSCCPass()); 1024 1025 for (auto &C : CGSCCOptimizerLateEPCallbacks) 1026 C(MainCGPipeline, Level); 1027 1028 // Lastly, add the core function simplification pipeline nested inside the 1029 // CGSCC walk. 1030 auto FSP = buildFunctionSimplificationPipeline(Level, Phase); 1031 if (DoNotRerunFunctionPasses) 1032 FSP.addPass(RequireAnalysisPass<FunctionStatusAnalysis, Function>()); 1033 MainCGPipeline.addPass(createCGSCCToFunctionPassAdaptor(std::move(FSP))); 1034 1035 return MIWP; 1036 } 1037 1038 ModulePassManager 1039 PassBuilder::buildModuleSimplificationPipeline(OptimizationLevel Level, 1040 ThinOrFullLTOPhase Phase) { 1041 ModulePassManager MPM; 1042 1043 // Place pseudo probe instrumentation as the first pass of the pipeline to 1044 // minimize the impact of optimization changes. 1045 if (PGOOpt && PGOOpt->PseudoProbeForProfiling && 1046 Phase != ThinOrFullLTOPhase::ThinLTOPostLink) 1047 MPM.addPass(SampleProfileProbePass(TM)); 1048 1049 bool HasSampleProfile = PGOOpt && (PGOOpt->Action == PGOOptions::SampleUse); 1050 1051 // In ThinLTO mode, when flattened profile is used, all the available 1052 // profile information will be annotated in PreLink phase so there is 1053 // no need to load the profile again in PostLink. 1054 bool LoadSampleProfile = 1055 HasSampleProfile && 1056 !(FlattenedProfileUsed && Phase == ThinOrFullLTOPhase::ThinLTOPostLink); 1057 1058 // During the ThinLTO backend phase we perform early indirect call promotion 1059 // here, before globalopt. Otherwise imported available_externally functions 1060 // look unreferenced and are removed. If we are going to load the sample 1061 // profile then defer until later. 1062 // TODO: See if we can move later and consolidate with the location where 1063 // we perform ICP when we are loading a sample profile. 1064 // TODO: We pass HasSampleProfile (whether there was a sample profile file 1065 // passed to the compile) to the SamplePGO flag of ICP. This is used to 1066 // determine whether the new direct calls are annotated with prof metadata. 1067 // Ideally this should be determined from whether the IR is annotated with 1068 // sample profile, and not whether the a sample profile was provided on the 1069 // command line. E.g. for flattened profiles where we will not be reloading 1070 // the sample profile in the ThinLTO backend, we ideally shouldn't have to 1071 // provide the sample profile file. 1072 if (Phase == ThinOrFullLTOPhase::ThinLTOPostLink && !LoadSampleProfile) 1073 MPM.addPass(PGOIndirectCallPromotion(true /* InLTO */, HasSampleProfile)); 1074 1075 // Do basic inference of function attributes from known properties of system 1076 // libraries and other oracles. 1077 MPM.addPass(InferFunctionAttrsPass()); 1078 1079 // Create an early function pass manager to cleanup the output of the 1080 // frontend. 1081 FunctionPassManager EarlyFPM; 1082 // Lower llvm.expect to metadata before attempting transforms. 1083 // Compare/branch metadata may alter the behavior of passes like SimplifyCFG. 1084 EarlyFPM.addPass(LowerExpectIntrinsicPass()); 1085 EarlyFPM.addPass(SimplifyCFGPass()); 1086 EarlyFPM.addPass(SROA()); 1087 EarlyFPM.addPass(EarlyCSEPass()); 1088 if (PTO.Coroutines) 1089 EarlyFPM.addPass(CoroEarlyPass()); 1090 if (Level == OptimizationLevel::O3) 1091 EarlyFPM.addPass(CallSiteSplittingPass()); 1092 1093 // In SamplePGO ThinLTO backend, we need instcombine before profile annotation 1094 // to convert bitcast to direct calls so that they can be inlined during the 1095 // profile annotation prepration step. 1096 // More details about SamplePGO design can be found in: 1097 // https://research.google.com/pubs/pub45290.html 1098 // FIXME: revisit how SampleProfileLoad/Inliner/ICP is structured. 1099 if (LoadSampleProfile) 1100 EarlyFPM.addPass(InstCombinePass()); 1101 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(EarlyFPM))); 1102 1103 if (LoadSampleProfile) { 1104 // Annotate sample profile right after early FPM to ensure freshness of 1105 // the debug info. 1106 MPM.addPass(SampleProfileLoaderPass(PGOOpt->ProfileFile, 1107 PGOOpt->ProfileRemappingFile, Phase)); 1108 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 1109 // RequireAnalysisPass for PSI before subsequent non-module passes. 1110 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 1111 // Do not invoke ICP in the LTOPrelink phase as it makes it hard 1112 // for the profile annotation to be accurate in the LTO backend. 1113 if (Phase != ThinOrFullLTOPhase::ThinLTOPreLink && 1114 Phase != ThinOrFullLTOPhase::FullLTOPreLink) 1115 // We perform early indirect call promotion here, before globalopt. 1116 // This is important for the ThinLTO backend phase because otherwise 1117 // imported available_externally functions look unreferenced and are 1118 // removed. 1119 MPM.addPass( 1120 PGOIndirectCallPromotion(true /* IsInLTO */, true /* SamplePGO */)); 1121 } 1122 1123 // Try to perform OpenMP specific optimizations on the module. This is a 1124 // (quick!) no-op if there are no OpenMP runtime calls present in the module. 1125 if (Level == OptimizationLevel::O2 || Level == OptimizationLevel::O3) 1126 MPM.addPass(OpenMPOptPass()); 1127 1128 if (AttributorRun & AttributorRunOption::MODULE) 1129 MPM.addPass(AttributorPass()); 1130 1131 // Lower type metadata and the type.test intrinsic in the ThinLTO 1132 // post link pipeline after ICP. This is to enable usage of the type 1133 // tests in ICP sequences. 1134 if (Phase == ThinOrFullLTOPhase::ThinLTOPostLink) 1135 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1136 1137 for (auto &C : PipelineEarlySimplificationEPCallbacks) 1138 C(MPM, Level); 1139 1140 // Interprocedural constant propagation now that basic cleanup has occurred 1141 // and prior to optimizing globals. 1142 // FIXME: This position in the pipeline hasn't been carefully considered in 1143 // years, it should be re-analyzed. 1144 MPM.addPass(IPSCCPPass()); 1145 1146 // Attach metadata to indirect call sites indicating the set of functions 1147 // they may target at run-time. This should follow IPSCCP. 1148 MPM.addPass(CalledValuePropagationPass()); 1149 1150 // Optimize globals to try and fold them into constants. 1151 MPM.addPass(GlobalOptPass()); 1152 1153 // Promote any localized globals to SSA registers. 1154 // FIXME: Should this instead by a run of SROA? 1155 // FIXME: We should probably run instcombine and simplify-cfg afterward to 1156 // delete control flows that are dead once globals have been folded to 1157 // constants. 1158 MPM.addPass(createModuleToFunctionPassAdaptor(PromotePass())); 1159 1160 // Remove any dead arguments exposed by cleanups and constant folding 1161 // globals. 1162 MPM.addPass(DeadArgumentEliminationPass()); 1163 1164 // Create a small function pass pipeline to cleanup after all the global 1165 // optimizations. 1166 FunctionPassManager GlobalCleanupPM; 1167 GlobalCleanupPM.addPass(InstCombinePass()); 1168 invokePeepholeEPCallbacks(GlobalCleanupPM, Level); 1169 1170 GlobalCleanupPM.addPass(SimplifyCFGPass()); 1171 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(GlobalCleanupPM))); 1172 1173 // Add all the requested passes for instrumentation PGO, if requested. 1174 if (PGOOpt && Phase != ThinOrFullLTOPhase::ThinLTOPostLink && 1175 (PGOOpt->Action == PGOOptions::IRInstr || 1176 PGOOpt->Action == PGOOptions::IRUse)) { 1177 addPGOInstrPasses(MPM, Level, 1178 /* RunProfileGen */ PGOOpt->Action == PGOOptions::IRInstr, 1179 /* IsCS */ false, PGOOpt->ProfileFile, 1180 PGOOpt->ProfileRemappingFile); 1181 MPM.addPass(PGOIndirectCallPromotion(false, false)); 1182 } 1183 if (PGOOpt && Phase != ThinOrFullLTOPhase::ThinLTOPostLink && 1184 PGOOpt->CSAction == PGOOptions::CSIRInstr) 1185 MPM.addPass(PGOInstrumentationGenCreateVar(PGOOpt->CSProfileGenFile)); 1186 1187 // Synthesize function entry counts for non-PGO compilation. 1188 if (EnableSyntheticCounts && !PGOOpt) 1189 MPM.addPass(SyntheticCountsPropagation()); 1190 1191 MPM.addPass(buildInlinerPipeline(Level, Phase)); 1192 if (DoNotRerunFunctionPasses) 1193 MPM.addPass(createModuleToFunctionPassAdaptor( 1194 InvalidateAnalysisPass<FunctionStatusAnalysis>())); 1195 1196 if (EnableMemProfiler && Phase != ThinOrFullLTOPhase::ThinLTOPreLink) { 1197 MPM.addPass(createModuleToFunctionPassAdaptor(MemProfilerPass())); 1198 MPM.addPass(ModuleMemProfilerPass()); 1199 } 1200 1201 return MPM; 1202 } 1203 1204 /// TODO: Should LTO cause any differences to this set of passes? 1205 void PassBuilder::addVectorPasses(OptimizationLevel Level, 1206 FunctionPassManager &FPM, bool IsLTO) { 1207 FPM.addPass(LoopVectorizePass( 1208 LoopVectorizeOptions(!PTO.LoopInterleaving, !PTO.LoopVectorization))); 1209 1210 if (IsLTO) { 1211 // The vectorizer may have significantly shortened a loop body; unroll 1212 // again. Unroll small loops to hide loop backedge latency and saturate any 1213 // parallel execution resources of an out-of-order processor. We also then 1214 // need to clean up redundancies and loop invariant code. 1215 // FIXME: It would be really good to use a loop-integrated instruction 1216 // combiner for cleanup here so that the unrolling and LICM can be pipelined 1217 // across the loop nests. 1218 // We do UnrollAndJam in a separate LPM to ensure it happens before unroll 1219 if (EnableUnrollAndJam && PTO.LoopUnrolling) 1220 FPM.addPass(LoopUnrollAndJamPass(Level.getSpeedupLevel())); 1221 FPM.addPass(LoopUnrollPass(LoopUnrollOptions( 1222 Level.getSpeedupLevel(), /*OnlyWhenForced=*/!PTO.LoopUnrolling, 1223 PTO.ForgetAllSCEVInLoopUnroll))); 1224 FPM.addPass(WarnMissedTransformationsPass()); 1225 } 1226 1227 if (!IsLTO) { 1228 // Eliminate loads by forwarding stores from the previous iteration to loads 1229 // of the current iteration. 1230 FPM.addPass(LoopLoadEliminationPass()); 1231 } 1232 // Cleanup after the loop optimization passes. 1233 FPM.addPass(InstCombinePass()); 1234 1235 if (Level.getSpeedupLevel() > 1 && ExtraVectorizerPasses) { 1236 // At higher optimization levels, try to clean up any runtime overlap and 1237 // alignment checks inserted by the vectorizer. We want to track correlated 1238 // runtime checks for two inner loops in the same outer loop, fold any 1239 // common computations, hoist loop-invariant aspects out of any outer loop, 1240 // and unswitch the runtime checks if possible. Once hoisted, we may have 1241 // dead (or speculatable) control flows or more combining opportunities. 1242 FPM.addPass(EarlyCSEPass()); 1243 FPM.addPass(CorrelatedValuePropagationPass()); 1244 FPM.addPass(InstCombinePass()); 1245 LoopPassManager LPM; 1246 LPM.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 1247 LPM.addPass(SimpleLoopUnswitchPass(/* NonTrivial */ Level == 1248 OptimizationLevel::O3)); 1249 FPM.addPass( 1250 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 1251 FPM.addPass(createFunctionToLoopPassAdaptor( 1252 std::move(LPM), EnableMSSALoopDependency, 1253 /*UseBlockFrequencyInfo=*/true)); 1254 FPM.addPass(SimplifyCFGPass()); 1255 FPM.addPass(InstCombinePass()); 1256 } 1257 1258 if (IsLTO) { 1259 FPM.addPass(SimplifyCFGPass(SimplifyCFGOptions().hoistCommonInsts(true))); 1260 } else { 1261 // Now that we've formed fast to execute loop structures, we do further 1262 // optimizations. These are run afterward as they might block doing complex 1263 // analyses and transforms such as what are needed for loop vectorization. 1264 1265 // Cleanup after loop vectorization, etc. Simplification passes like CVP and 1266 // GVN, loop transforms, and others have already run, so it's now better to 1267 // convert to more optimized IR using more aggressive simplify CFG options. 1268 // The extra sinking transform can create larger basic blocks, so do this 1269 // before SLP vectorization. 1270 FPM.addPass(SimplifyCFGPass(SimplifyCFGOptions() 1271 .forwardSwitchCondToPhi(true) 1272 .convertSwitchToLookupTable(true) 1273 .needCanonicalLoops(false) 1274 .hoistCommonInsts(true) 1275 .sinkCommonInsts(true))); 1276 } 1277 if (IsLTO) { 1278 FPM.addPass(SCCPPass()); 1279 FPM.addPass(InstCombinePass()); 1280 FPM.addPass(BDCEPass()); 1281 } 1282 1283 // Optimize parallel scalar instruction chains into SIMD instructions. 1284 if (PTO.SLPVectorization) { 1285 FPM.addPass(SLPVectorizerPass()); 1286 if (Level.getSpeedupLevel() > 1 && ExtraVectorizerPasses) { 1287 FPM.addPass(EarlyCSEPass()); 1288 } 1289 } 1290 // Enhance/cleanup vector code. 1291 FPM.addPass(VectorCombinePass()); 1292 1293 if (!IsLTO) { 1294 FPM.addPass(InstCombinePass()); 1295 // Unroll small loops to hide loop backedge latency and saturate any 1296 // parallel execution resources of an out-of-order processor. We also then 1297 // need to clean up redundancies and loop invariant code. 1298 // FIXME: It would be really good to use a loop-integrated instruction 1299 // combiner for cleanup here so that the unrolling and LICM can be pipelined 1300 // across the loop nests. 1301 // We do UnrollAndJam in a separate LPM to ensure it happens before unroll 1302 if (EnableUnrollAndJam && PTO.LoopUnrolling) { 1303 FPM.addPass(LoopUnrollAndJamPass(Level.getSpeedupLevel())); 1304 } 1305 FPM.addPass(LoopUnrollPass(LoopUnrollOptions( 1306 Level.getSpeedupLevel(), /*OnlyWhenForced=*/!PTO.LoopUnrolling, 1307 PTO.ForgetAllSCEVInLoopUnroll))); 1308 FPM.addPass(WarnMissedTransformationsPass()); 1309 FPM.addPass(InstCombinePass()); 1310 FPM.addPass( 1311 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 1312 FPM.addPass(createFunctionToLoopPassAdaptor( 1313 LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap), 1314 EnableMSSALoopDependency, /*UseBlockFrequencyInfo=*/true)); 1315 } 1316 1317 // Now that we've vectorized and unrolled loops, we may have more refined 1318 // alignment information, try to re-derive it here. 1319 FPM.addPass(AlignmentFromAssumptionsPass()); 1320 1321 if (IsLTO) 1322 FPM.addPass(InstCombinePass()); 1323 } 1324 1325 ModulePassManager 1326 PassBuilder::buildModuleOptimizationPipeline(OptimizationLevel Level, 1327 bool LTOPreLink) { 1328 ModulePassManager MPM; 1329 1330 // Optimize globals now that the module is fully simplified. 1331 MPM.addPass(GlobalOptPass()); 1332 MPM.addPass(GlobalDCEPass()); 1333 1334 // Run partial inlining pass to partially inline functions that have 1335 // large bodies. 1336 if (RunPartialInlining) 1337 MPM.addPass(PartialInlinerPass()); 1338 1339 // Remove avail extern fns and globals definitions since we aren't compiling 1340 // an object file for later LTO. For LTO we want to preserve these so they 1341 // are eligible for inlining at link-time. Note if they are unreferenced they 1342 // will be removed by GlobalDCE later, so this only impacts referenced 1343 // available externally globals. Eventually they will be suppressed during 1344 // codegen, but eliminating here enables more opportunity for GlobalDCE as it 1345 // may make globals referenced by available external functions dead and saves 1346 // running remaining passes on the eliminated functions. These should be 1347 // preserved during prelinking for link-time inlining decisions. 1348 if (!LTOPreLink) 1349 MPM.addPass(EliminateAvailableExternallyPass()); 1350 1351 if (EnableOrderFileInstrumentation) 1352 MPM.addPass(InstrOrderFilePass()); 1353 1354 // Do RPO function attribute inference across the module to forward-propagate 1355 // attributes where applicable. 1356 // FIXME: Is this really an optimization rather than a canonicalization? 1357 MPM.addPass(ReversePostOrderFunctionAttrsPass()); 1358 1359 // Do a post inline PGO instrumentation and use pass. This is a context 1360 // sensitive PGO pass. We don't want to do this in LTOPreLink phrase as 1361 // cross-module inline has not been done yet. The context sensitive 1362 // instrumentation is after all the inlines are done. 1363 if (!LTOPreLink && PGOOpt) { 1364 if (PGOOpt->CSAction == PGOOptions::CSIRInstr) 1365 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ true, 1366 /* IsCS */ true, PGOOpt->CSProfileGenFile, 1367 PGOOpt->ProfileRemappingFile); 1368 else if (PGOOpt->CSAction == PGOOptions::CSIRUse) 1369 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ false, 1370 /* IsCS */ true, PGOOpt->ProfileFile, 1371 PGOOpt->ProfileRemappingFile); 1372 } 1373 1374 // Re-require GloblasAA here prior to function passes. This is particularly 1375 // useful as the above will have inlined, DCE'ed, and function-attr 1376 // propagated everything. We should at this point have a reasonably minimal 1377 // and richly annotated call graph. By computing aliasing and mod/ref 1378 // information for all local globals here, the late loop passes and notably 1379 // the vectorizer will be able to use them to help recognize vectorizable 1380 // memory operations. 1381 MPM.addPass(RequireAnalysisPass<GlobalsAA, Module>()); 1382 1383 FunctionPassManager OptimizePM; 1384 OptimizePM.addPass(Float2IntPass()); 1385 OptimizePM.addPass(LowerConstantIntrinsicsPass()); 1386 1387 if (EnableMatrix) { 1388 OptimizePM.addPass(LowerMatrixIntrinsicsPass()); 1389 OptimizePM.addPass(EarlyCSEPass()); 1390 } 1391 1392 // FIXME: We need to run some loop optimizations to re-rotate loops after 1393 // simplify-cfg and others undo their rotation. 1394 1395 // Optimize the loop execution. These passes operate on entire loop nests 1396 // rather than on each loop in an inside-out manner, and so they are actually 1397 // function passes. 1398 1399 for (auto &C : VectorizerStartEPCallbacks) 1400 C(OptimizePM, Level); 1401 1402 // First rotate loops that may have been un-rotated by prior passes. 1403 // Disable header duplication at -Oz. 1404 OptimizePM.addPass(createFunctionToLoopPassAdaptor( 1405 LoopRotatePass(Level != OptimizationLevel::Oz, LTOPreLink), 1406 EnableMSSALoopDependency, 1407 /*UseBlockFrequencyInfo=*/false)); 1408 1409 // Distribute loops to allow partial vectorization. I.e. isolate dependences 1410 // into separate loop that would otherwise inhibit vectorization. This is 1411 // currently only performed for loops marked with the metadata 1412 // llvm.loop.distribute=true or when -enable-loop-distribute is specified. 1413 OptimizePM.addPass(LoopDistributePass()); 1414 1415 // Populates the VFABI attribute with the scalar-to-vector mappings 1416 // from the TargetLibraryInfo. 1417 OptimizePM.addPass(InjectTLIMappings()); 1418 1419 addVectorPasses(Level, OptimizePM, /* IsLTO */ false); 1420 1421 // Split out cold code. Splitting is done late to avoid hiding context from 1422 // other optimizations and inadvertently regressing performance. The tradeoff 1423 // is that this has a higher code size cost than splitting early. 1424 if (EnableHotColdSplit && !LTOPreLink) 1425 MPM.addPass(HotColdSplittingPass()); 1426 1427 // Search the code for similar regions of code. If enough similar regions can 1428 // be found where extracting the regions into their own function will decrease 1429 // the size of the program, we extract the regions, a deduplicate the 1430 // structurally similar regions. 1431 if (EnableIROutliner) 1432 MPM.addPass(IROutlinerPass()); 1433 1434 // Merge functions if requested. 1435 if (PTO.MergeFunctions) 1436 MPM.addPass(MergeFunctionsPass()); 1437 1438 // LoopSink pass sinks instructions hoisted by LICM, which serves as a 1439 // canonicalization pass that enables other optimizations. As a result, 1440 // LoopSink pass needs to be a very late IR pass to avoid undoing LICM 1441 // result too early. 1442 OptimizePM.addPass(LoopSinkPass()); 1443 1444 // And finally clean up LCSSA form before generating code. 1445 OptimizePM.addPass(InstSimplifyPass()); 1446 1447 // This hoists/decomposes div/rem ops. It should run after other sink/hoist 1448 // passes to avoid re-sinking, but before SimplifyCFG because it can allow 1449 // flattening of blocks. 1450 OptimizePM.addPass(DivRemPairsPass()); 1451 1452 // LoopSink (and other loop passes since the last simplifyCFG) might have 1453 // resulted in single-entry-single-exit or empty blocks. Clean up the CFG. 1454 OptimizePM.addPass(SimplifyCFGPass()); 1455 1456 // Optimize PHIs by speculating around them when profitable. Note that this 1457 // pass needs to be run after any PRE or similar pass as it is essentially 1458 // inserting redundancies into the program. This even includes SimplifyCFG. 1459 OptimizePM.addPass(SpeculateAroundPHIsPass()); 1460 1461 if (PTO.Coroutines) 1462 OptimizePM.addPass(CoroCleanupPass()); 1463 1464 // Add the core optimizing pipeline. 1465 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(OptimizePM))); 1466 1467 for (auto &C : OptimizerLastEPCallbacks) 1468 C(MPM, Level); 1469 1470 if (PTO.CallGraphProfile) 1471 MPM.addPass(CGProfilePass()); 1472 1473 // Now we need to do some global optimization transforms. 1474 // FIXME: It would seem like these should come first in the optimization 1475 // pipeline and maybe be the bottom of the canonicalization pipeline? Weird 1476 // ordering here. 1477 MPM.addPass(GlobalDCEPass()); 1478 MPM.addPass(ConstantMergePass()); 1479 1480 // TODO: Relative look table converter pass caused an issue when full lto is 1481 // enabled. See https://reviews.llvm.org/D94355 for more details. 1482 // Until the issue fixed, disable this pass during pre-linking phase. 1483 if (!LTOPreLink) 1484 MPM.addPass(RelLookupTableConverterPass()); 1485 1486 return MPM; 1487 } 1488 1489 ModulePassManager 1490 PassBuilder::buildPerModuleDefaultPipeline(OptimizationLevel Level, 1491 bool LTOPreLink) { 1492 assert(Level != OptimizationLevel::O0 && 1493 "Must request optimizations for the default pipeline!"); 1494 1495 ModulePassManager MPM; 1496 1497 // Convert @llvm.global.annotations to !annotation metadata. 1498 MPM.addPass(Annotation2MetadataPass()); 1499 1500 // Force any function attributes we want the rest of the pipeline to observe. 1501 MPM.addPass(ForceFunctionAttrsPass()); 1502 1503 // Apply module pipeline start EP callback. 1504 for (auto &C : PipelineStartEPCallbacks) 1505 C(MPM, Level); 1506 1507 if (PGOOpt && PGOOpt->DebugInfoForProfiling) 1508 MPM.addPass(createModuleToFunctionPassAdaptor(AddDiscriminatorsPass())); 1509 1510 // Add the core simplification pipeline. 1511 MPM.addPass(buildModuleSimplificationPipeline( 1512 Level, LTOPreLink ? ThinOrFullLTOPhase::FullLTOPreLink 1513 : ThinOrFullLTOPhase::None)); 1514 1515 // Now add the optimization pipeline. 1516 MPM.addPass(buildModuleOptimizationPipeline(Level, LTOPreLink)); 1517 1518 if (PGOOpt && PGOOpt->PseudoProbeForProfiling) 1519 MPM.addPass(PseudoProbeUpdatePass()); 1520 1521 // Emit annotation remarks. 1522 addAnnotationRemarksPass(MPM); 1523 1524 if (LTOPreLink) 1525 addRequiredLTOPreLinkPasses(MPM); 1526 1527 return MPM; 1528 } 1529 1530 ModulePassManager 1531 PassBuilder::buildThinLTOPreLinkDefaultPipeline(OptimizationLevel Level) { 1532 assert(Level != OptimizationLevel::O0 && 1533 "Must request optimizations for the default pipeline!"); 1534 1535 ModulePassManager MPM; 1536 1537 // Convert @llvm.global.annotations to !annotation metadata. 1538 MPM.addPass(Annotation2MetadataPass()); 1539 1540 // Force any function attributes we want the rest of the pipeline to observe. 1541 MPM.addPass(ForceFunctionAttrsPass()); 1542 1543 if (PGOOpt && PGOOpt->DebugInfoForProfiling) 1544 MPM.addPass(createModuleToFunctionPassAdaptor(AddDiscriminatorsPass())); 1545 1546 // Apply module pipeline start EP callback. 1547 for (auto &C : PipelineStartEPCallbacks) 1548 C(MPM, Level); 1549 1550 // If we are planning to perform ThinLTO later, we don't bloat the code with 1551 // unrolling/vectorization/... now. Just simplify the module as much as we 1552 // can. 1553 MPM.addPass(buildModuleSimplificationPipeline( 1554 Level, ThinOrFullLTOPhase::ThinLTOPreLink)); 1555 1556 // Run partial inlining pass to partially inline functions that have 1557 // large bodies. 1558 // FIXME: It isn't clear whether this is really the right place to run this 1559 // in ThinLTO. Because there is another canonicalization and simplification 1560 // phase that will run after the thin link, running this here ends up with 1561 // less information than will be available later and it may grow functions in 1562 // ways that aren't beneficial. 1563 if (RunPartialInlining) 1564 MPM.addPass(PartialInlinerPass()); 1565 1566 // Reduce the size of the IR as much as possible. 1567 MPM.addPass(GlobalOptPass()); 1568 1569 // Module simplification splits coroutines, but does not fully clean up 1570 // coroutine intrinsics. To ensure ThinLTO optimization passes don't trip up 1571 // on these, we schedule the cleanup here. 1572 if (PTO.Coroutines) 1573 MPM.addPass(createModuleToFunctionPassAdaptor(CoroCleanupPass())); 1574 1575 if (PGOOpt && PGOOpt->PseudoProbeForProfiling) 1576 MPM.addPass(PseudoProbeUpdatePass()); 1577 1578 // Handle OptimizerLastEPCallbacks added by clang on PreLink. Actual 1579 // optimization is going to be done in PostLink stage, but clang can't 1580 // add callbacks there in case of in-process ThinLTO called by linker. 1581 for (auto &C : OptimizerLastEPCallbacks) 1582 C(MPM, Level); 1583 1584 // Emit annotation remarks. 1585 addAnnotationRemarksPass(MPM); 1586 1587 addRequiredLTOPreLinkPasses(MPM); 1588 1589 return MPM; 1590 } 1591 1592 ModulePassManager PassBuilder::buildThinLTODefaultPipeline( 1593 OptimizationLevel Level, const ModuleSummaryIndex *ImportSummary) { 1594 ModulePassManager MPM; 1595 1596 // Convert @llvm.global.annotations to !annotation metadata. 1597 MPM.addPass(Annotation2MetadataPass()); 1598 1599 if (ImportSummary) { 1600 // These passes import type identifier resolutions for whole-program 1601 // devirtualization and CFI. They must run early because other passes may 1602 // disturb the specific instruction patterns that these passes look for, 1603 // creating dependencies on resolutions that may not appear in the summary. 1604 // 1605 // For example, GVN may transform the pattern assume(type.test) appearing in 1606 // two basic blocks into assume(phi(type.test, type.test)), which would 1607 // transform a dependency on a WPD resolution into a dependency on a type 1608 // identifier resolution for CFI. 1609 // 1610 // Also, WPD has access to more precise information than ICP and can 1611 // devirtualize more effectively, so it should operate on the IR first. 1612 // 1613 // The WPD and LowerTypeTest passes need to run at -O0 to lower type 1614 // metadata and intrinsics. 1615 MPM.addPass(WholeProgramDevirtPass(nullptr, ImportSummary)); 1616 MPM.addPass(LowerTypeTestsPass(nullptr, ImportSummary)); 1617 } 1618 1619 if (Level == OptimizationLevel::O0) { 1620 // Run a second time to clean up any type tests left behind by WPD for use 1621 // in ICP. 1622 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1623 // Drop available_externally and unreferenced globals. This is necessary 1624 // with ThinLTO in order to avoid leaving undefined references to dead 1625 // globals in the object file. 1626 MPM.addPass(EliminateAvailableExternallyPass()); 1627 MPM.addPass(GlobalDCEPass()); 1628 return MPM; 1629 } 1630 1631 // Force any function attributes we want the rest of the pipeline to observe. 1632 MPM.addPass(ForceFunctionAttrsPass()); 1633 1634 // Add the core simplification pipeline. 1635 MPM.addPass(buildModuleSimplificationPipeline( 1636 Level, ThinOrFullLTOPhase::ThinLTOPostLink)); 1637 1638 // Now add the optimization pipeline. 1639 MPM.addPass(buildModuleOptimizationPipeline(Level)); 1640 1641 // Emit annotation remarks. 1642 addAnnotationRemarksPass(MPM); 1643 1644 return MPM; 1645 } 1646 1647 ModulePassManager 1648 PassBuilder::buildLTOPreLinkDefaultPipeline(OptimizationLevel Level) { 1649 assert(Level != OptimizationLevel::O0 && 1650 "Must request optimizations for the default pipeline!"); 1651 // FIXME: We should use a customized pre-link pipeline! 1652 return buildPerModuleDefaultPipeline(Level, 1653 /* LTOPreLink */ true); 1654 } 1655 1656 ModulePassManager 1657 PassBuilder::buildLTODefaultPipeline(OptimizationLevel Level, 1658 ModuleSummaryIndex *ExportSummary) { 1659 ModulePassManager MPM; 1660 1661 // Convert @llvm.global.annotations to !annotation metadata. 1662 MPM.addPass(Annotation2MetadataPass()); 1663 1664 if (Level == OptimizationLevel::O0) { 1665 // The WPD and LowerTypeTest passes need to run at -O0 to lower type 1666 // metadata and intrinsics. 1667 MPM.addPass(WholeProgramDevirtPass(ExportSummary, nullptr)); 1668 MPM.addPass(LowerTypeTestsPass(ExportSummary, nullptr)); 1669 // Run a second time to clean up any type tests left behind by WPD for use 1670 // in ICP. 1671 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1672 1673 // Emit annotation remarks. 1674 addAnnotationRemarksPass(MPM); 1675 1676 return MPM; 1677 } 1678 1679 if (PGOOpt && PGOOpt->Action == PGOOptions::SampleUse) { 1680 // Load sample profile before running the LTO optimization pipeline. 1681 MPM.addPass(SampleProfileLoaderPass(PGOOpt->ProfileFile, 1682 PGOOpt->ProfileRemappingFile, 1683 ThinOrFullLTOPhase::FullLTOPostLink)); 1684 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 1685 // RequireAnalysisPass for PSI before subsequent non-module passes. 1686 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 1687 } 1688 1689 // Remove unused virtual tables to improve the quality of code generated by 1690 // whole-program devirtualization and bitset lowering. 1691 MPM.addPass(GlobalDCEPass()); 1692 1693 // Force any function attributes we want the rest of the pipeline to observe. 1694 MPM.addPass(ForceFunctionAttrsPass()); 1695 1696 // Do basic inference of function attributes from known properties of system 1697 // libraries and other oracles. 1698 MPM.addPass(InferFunctionAttrsPass()); 1699 1700 if (Level.getSpeedupLevel() > 1) { 1701 FunctionPassManager EarlyFPM; 1702 EarlyFPM.addPass(CallSiteSplittingPass()); 1703 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(EarlyFPM))); 1704 1705 // Indirect call promotion. This should promote all the targets that are 1706 // left by the earlier promotion pass that promotes intra-module targets. 1707 // This two-step promotion is to save the compile time. For LTO, it should 1708 // produce the same result as if we only do promotion here. 1709 MPM.addPass(PGOIndirectCallPromotion( 1710 true /* InLTO */, PGOOpt && PGOOpt->Action == PGOOptions::SampleUse)); 1711 // Propagate constants at call sites into the functions they call. This 1712 // opens opportunities for globalopt (and inlining) by substituting function 1713 // pointers passed as arguments to direct uses of functions. 1714 MPM.addPass(IPSCCPPass()); 1715 1716 // Attach metadata to indirect call sites indicating the set of functions 1717 // they may target at run-time. This should follow IPSCCP. 1718 MPM.addPass(CalledValuePropagationPass()); 1719 } 1720 1721 // Now deduce any function attributes based in the current code. 1722 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor( 1723 PostOrderFunctionAttrsPass())); 1724 1725 // Do RPO function attribute inference across the module to forward-propagate 1726 // attributes where applicable. 1727 // FIXME: Is this really an optimization rather than a canonicalization? 1728 MPM.addPass(ReversePostOrderFunctionAttrsPass()); 1729 1730 // Use in-range annotations on GEP indices to split globals where beneficial. 1731 MPM.addPass(GlobalSplitPass()); 1732 1733 // Run whole program optimization of virtual call when the list of callees 1734 // is fixed. 1735 MPM.addPass(WholeProgramDevirtPass(ExportSummary, nullptr)); 1736 1737 // Stop here at -O1. 1738 if (Level == OptimizationLevel::O1) { 1739 // The LowerTypeTestsPass needs to run to lower type metadata and the 1740 // type.test intrinsics. The pass does nothing if CFI is disabled. 1741 MPM.addPass(LowerTypeTestsPass(ExportSummary, nullptr)); 1742 // Run a second time to clean up any type tests left behind by WPD for use 1743 // in ICP (which is performed earlier than this in the regular LTO 1744 // pipeline). 1745 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1746 1747 // Emit annotation remarks. 1748 addAnnotationRemarksPass(MPM); 1749 1750 return MPM; 1751 } 1752 1753 // Optimize globals to try and fold them into constants. 1754 MPM.addPass(GlobalOptPass()); 1755 1756 // Promote any localized globals to SSA registers. 1757 MPM.addPass(createModuleToFunctionPassAdaptor(PromotePass())); 1758 1759 // Linking modules together can lead to duplicate global constant, only 1760 // keep one copy of each constant. 1761 MPM.addPass(ConstantMergePass()); 1762 1763 // Remove unused arguments from functions. 1764 MPM.addPass(DeadArgumentEliminationPass()); 1765 1766 // Reduce the code after globalopt and ipsccp. Both can open up significant 1767 // simplification opportunities, and both can propagate functions through 1768 // function pointers. When this happens, we often have to resolve varargs 1769 // calls, etc, so let instcombine do this. 1770 FunctionPassManager PeepholeFPM; 1771 if (Level == OptimizationLevel::O3) 1772 PeepholeFPM.addPass(AggressiveInstCombinePass()); 1773 PeepholeFPM.addPass(InstCombinePass()); 1774 invokePeepholeEPCallbacks(PeepholeFPM, Level); 1775 1776 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(PeepholeFPM))); 1777 1778 // Note: historically, the PruneEH pass was run first to deduce nounwind and 1779 // generally clean up exception handling overhead. It isn't clear this is 1780 // valuable as the inliner doesn't currently care whether it is inlining an 1781 // invoke or a call. 1782 // Run the inliner now. 1783 MPM.addPass(ModuleInlinerWrapperPass(getInlineParamsFromOptLevel(Level))); 1784 1785 // Optimize globals again after we ran the inliner. 1786 MPM.addPass(GlobalOptPass()); 1787 1788 // Garbage collect dead functions. 1789 // FIXME: Add ArgumentPromotion pass after once it's ported. 1790 MPM.addPass(GlobalDCEPass()); 1791 1792 FunctionPassManager FPM; 1793 // The IPO Passes may leave cruft around. Clean up after them. 1794 FPM.addPass(InstCombinePass()); 1795 invokePeepholeEPCallbacks(FPM, Level); 1796 1797 FPM.addPass(JumpThreadingPass(/*InsertFreezeWhenUnfoldingSelect*/ true)); 1798 1799 // Do a post inline PGO instrumentation and use pass. This is a context 1800 // sensitive PGO pass. 1801 if (PGOOpt) { 1802 if (PGOOpt->CSAction == PGOOptions::CSIRInstr) 1803 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ true, 1804 /* IsCS */ true, PGOOpt->CSProfileGenFile, 1805 PGOOpt->ProfileRemappingFile); 1806 else if (PGOOpt->CSAction == PGOOptions::CSIRUse) 1807 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ false, 1808 /* IsCS */ true, PGOOpt->ProfileFile, 1809 PGOOpt->ProfileRemappingFile); 1810 } 1811 1812 // Break up allocas 1813 FPM.addPass(SROA()); 1814 1815 // LTO provides additional opportunities for tailcall elimination due to 1816 // link-time inlining, and visibility of nocapture attribute. 1817 FPM.addPass(TailCallElimPass()); 1818 1819 // Run a few AA driver optimizations here and now to cleanup the code. 1820 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 1821 1822 MPM.addPass( 1823 createModuleToPostOrderCGSCCPassAdaptor(PostOrderFunctionAttrsPass())); 1824 1825 // Require the GlobalsAA analysis for the module so we can query it within 1826 // MainFPM. 1827 MPM.addPass(RequireAnalysisPass<GlobalsAA, Module>()); 1828 // Invalidate AAManager so it can be recreated and pick up the newly available 1829 // GlobalsAA. 1830 MPM.addPass( 1831 createModuleToFunctionPassAdaptor(InvalidateAnalysisPass<AAManager>())); 1832 1833 FunctionPassManager MainFPM; 1834 MainFPM.addPass(createFunctionToLoopPassAdaptor( 1835 LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap), 1836 EnableMSSALoopDependency, /*UseBlockFrequencyInfo=*/true)); 1837 1838 if (RunNewGVN) 1839 MainFPM.addPass(NewGVNPass()); 1840 else 1841 MainFPM.addPass(GVN()); 1842 1843 // Remove dead memcpy()'s. 1844 MainFPM.addPass(MemCpyOptPass()); 1845 1846 // Nuke dead stores. 1847 MainFPM.addPass(DSEPass()); 1848 MainFPM.addPass(MergedLoadStoreMotionPass()); 1849 1850 // More loops are countable; try to optimize them. 1851 if (EnableLoopFlatten && Level.getSpeedupLevel() > 1) 1852 MainFPM.addPass(LoopFlattenPass()); 1853 1854 if (EnableConstraintElimination) 1855 MainFPM.addPass(ConstraintEliminationPass()); 1856 1857 LoopPassManager LPM; 1858 LPM.addPass(IndVarSimplifyPass()); 1859 LPM.addPass(LoopDeletionPass()); 1860 // FIXME: Add loop interchange. 1861 1862 // Unroll small loops and perform peeling. 1863 LPM.addPass(LoopFullUnrollPass(Level.getSpeedupLevel(), 1864 /* OnlyWhenForced= */ !PTO.LoopUnrolling, 1865 PTO.ForgetAllSCEVInLoopUnroll)); 1866 // The loop passes in LPM (LoopFullUnrollPass) do not preserve MemorySSA. 1867 // *All* loop passes must preserve it, in order to be able to use it. 1868 MainFPM.addPass(createFunctionToLoopPassAdaptor( 1869 std::move(LPM), /*UseMemorySSA=*/false, /*UseBlockFrequencyInfo=*/true)); 1870 1871 MainFPM.addPass(LoopDistributePass()); 1872 1873 addVectorPasses(Level, MainFPM, /* IsLTO */ true); 1874 1875 invokePeepholeEPCallbacks(MainFPM, Level); 1876 MainFPM.addPass(JumpThreadingPass(/*InsertFreezeWhenUnfoldingSelect*/ true)); 1877 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(MainFPM))); 1878 1879 // Create a function that performs CFI checks for cross-DSO calls with 1880 // targets in the current module. 1881 MPM.addPass(CrossDSOCFIPass()); 1882 1883 // Lower type metadata and the type.test intrinsic. This pass supports 1884 // clang's control flow integrity mechanisms (-fsanitize=cfi*) and needs 1885 // to be run at link time if CFI is enabled. This pass does nothing if 1886 // CFI is disabled. 1887 MPM.addPass(LowerTypeTestsPass(ExportSummary, nullptr)); 1888 // Run a second time to clean up any type tests left behind by WPD for use 1889 // in ICP (which is performed earlier than this in the regular LTO pipeline). 1890 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1891 1892 // Enable splitting late in the FullLTO post-link pipeline. This is done in 1893 // the same stage in the old pass manager (\ref addLateLTOOptimizationPasses). 1894 if (EnableHotColdSplit) 1895 MPM.addPass(HotColdSplittingPass()); 1896 1897 // Add late LTO optimization passes. 1898 // Delete basic blocks, which optimization passes may have killed. 1899 MPM.addPass(createModuleToFunctionPassAdaptor( 1900 SimplifyCFGPass(SimplifyCFGOptions().hoistCommonInsts(true)))); 1901 1902 // Drop bodies of available eternally objects to improve GlobalDCE. 1903 MPM.addPass(EliminateAvailableExternallyPass()); 1904 1905 // Now that we have optimized the program, discard unreachable functions. 1906 MPM.addPass(GlobalDCEPass()); 1907 1908 if (PTO.MergeFunctions) 1909 MPM.addPass(MergeFunctionsPass()); 1910 1911 // Emit annotation remarks. 1912 addAnnotationRemarksPass(MPM); 1913 1914 return MPM; 1915 } 1916 1917 ModulePassManager PassBuilder::buildO0DefaultPipeline(OptimizationLevel Level, 1918 bool LTOPreLink) { 1919 assert(Level == OptimizationLevel::O0 && 1920 "buildO0DefaultPipeline should only be used with O0"); 1921 1922 ModulePassManager MPM; 1923 1924 if (PGOOpt && (PGOOpt->Action == PGOOptions::IRInstr || 1925 PGOOpt->Action == PGOOptions::IRUse)) 1926 addPGOInstrPassesForO0( 1927 MPM, 1928 /* RunProfileGen */ (PGOOpt->Action == PGOOptions::IRInstr), 1929 /* IsCS */ false, PGOOpt->ProfileFile, PGOOpt->ProfileRemappingFile); 1930 1931 for (auto &C : PipelineStartEPCallbacks) 1932 C(MPM, Level); 1933 for (auto &C : PipelineEarlySimplificationEPCallbacks) 1934 C(MPM, Level); 1935 1936 // Build a minimal pipeline based on the semantics required by LLVM, 1937 // which is just that always inlining occurs. Further, disable generating 1938 // lifetime intrinsics to avoid enabling further optimizations during 1939 // code generation. 1940 // However, we need to insert lifetime intrinsics to avoid invalid access 1941 // caused by multithreaded coroutines. 1942 MPM.addPass(AlwaysInlinerPass( 1943 /*InsertLifetimeIntrinsics=*/PTO.Coroutines)); 1944 1945 if (PTO.MergeFunctions) 1946 MPM.addPass(MergeFunctionsPass()); 1947 1948 if (EnableMatrix) 1949 MPM.addPass( 1950 createModuleToFunctionPassAdaptor(LowerMatrixIntrinsicsPass(true))); 1951 1952 if (!CGSCCOptimizerLateEPCallbacks.empty()) { 1953 CGSCCPassManager CGPM; 1954 for (auto &C : CGSCCOptimizerLateEPCallbacks) 1955 C(CGPM, Level); 1956 if (!CGPM.isEmpty()) 1957 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPM))); 1958 } 1959 if (!LateLoopOptimizationsEPCallbacks.empty()) { 1960 LoopPassManager LPM; 1961 for (auto &C : LateLoopOptimizationsEPCallbacks) 1962 C(LPM, Level); 1963 if (!LPM.isEmpty()) { 1964 MPM.addPass(createModuleToFunctionPassAdaptor( 1965 createFunctionToLoopPassAdaptor(std::move(LPM)))); 1966 } 1967 } 1968 if (!LoopOptimizerEndEPCallbacks.empty()) { 1969 LoopPassManager LPM; 1970 for (auto &C : LoopOptimizerEndEPCallbacks) 1971 C(LPM, Level); 1972 if (!LPM.isEmpty()) { 1973 MPM.addPass(createModuleToFunctionPassAdaptor( 1974 createFunctionToLoopPassAdaptor(std::move(LPM)))); 1975 } 1976 } 1977 if (!ScalarOptimizerLateEPCallbacks.empty()) { 1978 FunctionPassManager FPM; 1979 for (auto &C : ScalarOptimizerLateEPCallbacks) 1980 C(FPM, Level); 1981 if (!FPM.isEmpty()) 1982 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 1983 } 1984 if (!VectorizerStartEPCallbacks.empty()) { 1985 FunctionPassManager FPM; 1986 for (auto &C : VectorizerStartEPCallbacks) 1987 C(FPM, Level); 1988 if (!FPM.isEmpty()) 1989 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 1990 } 1991 1992 if (PTO.Coroutines) { 1993 MPM.addPass(createModuleToFunctionPassAdaptor(CoroEarlyPass())); 1994 1995 CGSCCPassManager CGPM; 1996 CGPM.addPass(CoroSplitPass()); 1997 CGPM.addPass(createCGSCCToFunctionPassAdaptor(CoroElidePass())); 1998 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPM))); 1999 2000 MPM.addPass(createModuleToFunctionPassAdaptor(CoroCleanupPass())); 2001 } 2002 2003 for (auto &C : OptimizerLastEPCallbacks) 2004 C(MPM, Level); 2005 2006 if (LTOPreLink) 2007 addRequiredLTOPreLinkPasses(MPM); 2008 2009 return MPM; 2010 } 2011 2012 AAManager PassBuilder::buildDefaultAAPipeline() { 2013 AAManager AA; 2014 2015 // The order in which these are registered determines their priority when 2016 // being queried. 2017 2018 // First we register the basic alias analysis that provides the majority of 2019 // per-function local AA logic. This is a stateless, on-demand local set of 2020 // AA techniques. 2021 AA.registerFunctionAnalysis<BasicAA>(); 2022 2023 // Next we query fast, specialized alias analyses that wrap IR-embedded 2024 // information about aliasing. 2025 AA.registerFunctionAnalysis<ScopedNoAliasAA>(); 2026 AA.registerFunctionAnalysis<TypeBasedAA>(); 2027 2028 // Add support for querying global aliasing information when available. 2029 // Because the `AAManager` is a function analysis and `GlobalsAA` is a module 2030 // analysis, all that the `AAManager` can do is query for any *cached* 2031 // results from `GlobalsAA` through a readonly proxy. 2032 AA.registerModuleAnalysis<GlobalsAA>(); 2033 2034 // Add target-specific alias analyses. 2035 if (TM) 2036 TM->registerDefaultAliasAnalyses(AA); 2037 2038 return AA; 2039 } 2040 2041 static Optional<int> parseRepeatPassName(StringRef Name) { 2042 if (!Name.consume_front("repeat<") || !Name.consume_back(">")) 2043 return None; 2044 int Count; 2045 if (Name.getAsInteger(0, Count) || Count <= 0) 2046 return None; 2047 return Count; 2048 } 2049 2050 static Optional<int> parseDevirtPassName(StringRef Name) { 2051 if (!Name.consume_front("devirt<") || !Name.consume_back(">")) 2052 return None; 2053 int Count; 2054 if (Name.getAsInteger(0, Count) || Count < 0) 2055 return None; 2056 return Count; 2057 } 2058 2059 static bool checkParametrizedPassName(StringRef Name, StringRef PassName) { 2060 if (!Name.consume_front(PassName)) 2061 return false; 2062 // normal pass name w/o parameters == default parameters 2063 if (Name.empty()) 2064 return true; 2065 return Name.startswith("<") && Name.endswith(">"); 2066 } 2067 2068 namespace { 2069 2070 /// This performs customized parsing of pass name with parameters. 2071 /// 2072 /// We do not need parametrization of passes in textual pipeline very often, 2073 /// yet on a rare occasion ability to specify parameters right there can be 2074 /// useful. 2075 /// 2076 /// \p Name - parameterized specification of a pass from a textual pipeline 2077 /// is a string in a form of : 2078 /// PassName '<' parameter-list '>' 2079 /// 2080 /// Parameter list is being parsed by the parser callable argument, \p Parser, 2081 /// It takes a string-ref of parameters and returns either StringError or a 2082 /// parameter list in a form of a custom parameters type, all wrapped into 2083 /// Expected<> template class. 2084 /// 2085 template <typename ParametersParseCallableT> 2086 auto parsePassParameters(ParametersParseCallableT &&Parser, StringRef Name, 2087 StringRef PassName) -> decltype(Parser(StringRef{})) { 2088 using ParametersT = typename decltype(Parser(StringRef{}))::value_type; 2089 2090 StringRef Params = Name; 2091 if (!Params.consume_front(PassName)) { 2092 assert(false && 2093 "unable to strip pass name from parametrized pass specification"); 2094 } 2095 if (Params.empty()) 2096 return ParametersT{}; 2097 if (!Params.consume_front("<") || !Params.consume_back(">")) { 2098 assert(false && "invalid format for parametrized pass name"); 2099 } 2100 2101 Expected<ParametersT> Result = Parser(Params); 2102 assert((Result || Result.template errorIsA<StringError>()) && 2103 "Pass parameter parser can only return StringErrors."); 2104 return Result; 2105 } 2106 2107 /// Parser of parameters for LoopUnroll pass. 2108 Expected<LoopUnrollOptions> parseLoopUnrollOptions(StringRef Params) { 2109 LoopUnrollOptions UnrollOpts; 2110 while (!Params.empty()) { 2111 StringRef ParamName; 2112 std::tie(ParamName, Params) = Params.split(';'); 2113 int OptLevel = StringSwitch<int>(ParamName) 2114 .Case("O0", 0) 2115 .Case("O1", 1) 2116 .Case("O2", 2) 2117 .Case("O3", 3) 2118 .Default(-1); 2119 if (OptLevel >= 0) { 2120 UnrollOpts.setOptLevel(OptLevel); 2121 continue; 2122 } 2123 if (ParamName.consume_front("full-unroll-max=")) { 2124 int Count; 2125 if (ParamName.getAsInteger(0, Count)) 2126 return make_error<StringError>( 2127 formatv("invalid LoopUnrollPass parameter '{0}' ", ParamName).str(), 2128 inconvertibleErrorCode()); 2129 UnrollOpts.setFullUnrollMaxCount(Count); 2130 continue; 2131 } 2132 2133 bool Enable = !ParamName.consume_front("no-"); 2134 if (ParamName == "partial") { 2135 UnrollOpts.setPartial(Enable); 2136 } else if (ParamName == "peeling") { 2137 UnrollOpts.setPeeling(Enable); 2138 } else if (ParamName == "profile-peeling") { 2139 UnrollOpts.setProfileBasedPeeling(Enable); 2140 } else if (ParamName == "runtime") { 2141 UnrollOpts.setRuntime(Enable); 2142 } else if (ParamName == "upperbound") { 2143 UnrollOpts.setUpperBound(Enable); 2144 } else { 2145 return make_error<StringError>( 2146 formatv("invalid LoopUnrollPass parameter '{0}' ", ParamName).str(), 2147 inconvertibleErrorCode()); 2148 } 2149 } 2150 return UnrollOpts; 2151 } 2152 2153 Expected<MemorySanitizerOptions> parseMSanPassOptions(StringRef Params) { 2154 MemorySanitizerOptions Result; 2155 while (!Params.empty()) { 2156 StringRef ParamName; 2157 std::tie(ParamName, Params) = Params.split(';'); 2158 2159 if (ParamName == "recover") { 2160 Result.Recover = true; 2161 } else if (ParamName == "kernel") { 2162 Result.Kernel = true; 2163 } else if (ParamName.consume_front("track-origins=")) { 2164 if (ParamName.getAsInteger(0, Result.TrackOrigins)) 2165 return make_error<StringError>( 2166 formatv("invalid argument to MemorySanitizer pass track-origins " 2167 "parameter: '{0}' ", 2168 ParamName) 2169 .str(), 2170 inconvertibleErrorCode()); 2171 } else { 2172 return make_error<StringError>( 2173 formatv("invalid MemorySanitizer pass parameter '{0}' ", ParamName) 2174 .str(), 2175 inconvertibleErrorCode()); 2176 } 2177 } 2178 return Result; 2179 } 2180 2181 /// Parser of parameters for SimplifyCFG pass. 2182 Expected<SimplifyCFGOptions> parseSimplifyCFGOptions(StringRef Params) { 2183 SimplifyCFGOptions Result; 2184 while (!Params.empty()) { 2185 StringRef ParamName; 2186 std::tie(ParamName, Params) = Params.split(';'); 2187 2188 bool Enable = !ParamName.consume_front("no-"); 2189 if (ParamName == "forward-switch-cond") { 2190 Result.forwardSwitchCondToPhi(Enable); 2191 } else if (ParamName == "switch-to-lookup") { 2192 Result.convertSwitchToLookupTable(Enable); 2193 } else if (ParamName == "keep-loops") { 2194 Result.needCanonicalLoops(Enable); 2195 } else if (ParamName == "hoist-common-insts") { 2196 Result.hoistCommonInsts(Enable); 2197 } else if (ParamName == "sink-common-insts") { 2198 Result.sinkCommonInsts(Enable); 2199 } else if (Enable && ParamName.consume_front("bonus-inst-threshold=")) { 2200 APInt BonusInstThreshold; 2201 if (ParamName.getAsInteger(0, BonusInstThreshold)) 2202 return make_error<StringError>( 2203 formatv("invalid argument to SimplifyCFG pass bonus-threshold " 2204 "parameter: '{0}' ", 2205 ParamName).str(), 2206 inconvertibleErrorCode()); 2207 Result.bonusInstThreshold(BonusInstThreshold.getSExtValue()); 2208 } else { 2209 return make_error<StringError>( 2210 formatv("invalid SimplifyCFG pass parameter '{0}' ", ParamName).str(), 2211 inconvertibleErrorCode()); 2212 } 2213 } 2214 return Result; 2215 } 2216 2217 /// Parser of parameters for LoopVectorize pass. 2218 Expected<LoopVectorizeOptions> parseLoopVectorizeOptions(StringRef Params) { 2219 LoopVectorizeOptions Opts; 2220 while (!Params.empty()) { 2221 StringRef ParamName; 2222 std::tie(ParamName, Params) = Params.split(';'); 2223 2224 bool Enable = !ParamName.consume_front("no-"); 2225 if (ParamName == "interleave-forced-only") { 2226 Opts.setInterleaveOnlyWhenForced(Enable); 2227 } else if (ParamName == "vectorize-forced-only") { 2228 Opts.setVectorizeOnlyWhenForced(Enable); 2229 } else { 2230 return make_error<StringError>( 2231 formatv("invalid LoopVectorize parameter '{0}' ", ParamName).str(), 2232 inconvertibleErrorCode()); 2233 } 2234 } 2235 return Opts; 2236 } 2237 2238 Expected<bool> parseLoopUnswitchOptions(StringRef Params) { 2239 bool Result = false; 2240 while (!Params.empty()) { 2241 StringRef ParamName; 2242 std::tie(ParamName, Params) = Params.split(';'); 2243 2244 bool Enable = !ParamName.consume_front("no-"); 2245 if (ParamName == "nontrivial") { 2246 Result = Enable; 2247 } else { 2248 return make_error<StringError>( 2249 formatv("invalid LoopUnswitch pass parameter '{0}' ", ParamName) 2250 .str(), 2251 inconvertibleErrorCode()); 2252 } 2253 } 2254 return Result; 2255 } 2256 2257 Expected<bool> parseMergedLoadStoreMotionOptions(StringRef Params) { 2258 bool Result = false; 2259 while (!Params.empty()) { 2260 StringRef ParamName; 2261 std::tie(ParamName, Params) = Params.split(';'); 2262 2263 bool Enable = !ParamName.consume_front("no-"); 2264 if (ParamName == "split-footer-bb") { 2265 Result = Enable; 2266 } else { 2267 return make_error<StringError>( 2268 formatv("invalid MergedLoadStoreMotion pass parameter '{0}' ", 2269 ParamName) 2270 .str(), 2271 inconvertibleErrorCode()); 2272 } 2273 } 2274 return Result; 2275 } 2276 2277 Expected<GVNOptions> parseGVNOptions(StringRef Params) { 2278 GVNOptions Result; 2279 while (!Params.empty()) { 2280 StringRef ParamName; 2281 std::tie(ParamName, Params) = Params.split(';'); 2282 2283 bool Enable = !ParamName.consume_front("no-"); 2284 if (ParamName == "pre") { 2285 Result.setPRE(Enable); 2286 } else if (ParamName == "load-pre") { 2287 Result.setLoadPRE(Enable); 2288 } else if (ParamName == "split-backedge-load-pre") { 2289 Result.setLoadPRESplitBackedge(Enable); 2290 } else if (ParamName == "memdep") { 2291 Result.setMemDep(Enable); 2292 } else { 2293 return make_error<StringError>( 2294 formatv("invalid GVN pass parameter '{0}' ", ParamName).str(), 2295 inconvertibleErrorCode()); 2296 } 2297 } 2298 return Result; 2299 } 2300 2301 Expected<StackLifetime::LivenessType> 2302 parseStackLifetimeOptions(StringRef Params) { 2303 StackLifetime::LivenessType Result = StackLifetime::LivenessType::May; 2304 while (!Params.empty()) { 2305 StringRef ParamName; 2306 std::tie(ParamName, Params) = Params.split(';'); 2307 2308 if (ParamName == "may") { 2309 Result = StackLifetime::LivenessType::May; 2310 } else if (ParamName == "must") { 2311 Result = StackLifetime::LivenessType::Must; 2312 } else { 2313 return make_error<StringError>( 2314 formatv("invalid StackLifetime parameter '{0}' ", ParamName).str(), 2315 inconvertibleErrorCode()); 2316 } 2317 } 2318 return Result; 2319 } 2320 2321 } // namespace 2322 2323 /// Tests whether a pass name starts with a valid prefix for a default pipeline 2324 /// alias. 2325 static bool startsWithDefaultPipelineAliasPrefix(StringRef Name) { 2326 return Name.startswith("default") || Name.startswith("thinlto") || 2327 Name.startswith("lto"); 2328 } 2329 2330 /// Tests whether registered callbacks will accept a given pass name. 2331 /// 2332 /// When parsing a pipeline text, the type of the outermost pipeline may be 2333 /// omitted, in which case the type is automatically determined from the first 2334 /// pass name in the text. This may be a name that is handled through one of the 2335 /// callbacks. We check this through the oridinary parsing callbacks by setting 2336 /// up a dummy PassManager in order to not force the client to also handle this 2337 /// type of query. 2338 template <typename PassManagerT, typename CallbacksT> 2339 static bool callbacksAcceptPassName(StringRef Name, CallbacksT &Callbacks) { 2340 if (!Callbacks.empty()) { 2341 PassManagerT DummyPM; 2342 for (auto &CB : Callbacks) 2343 if (CB(Name, DummyPM, {})) 2344 return true; 2345 } 2346 return false; 2347 } 2348 2349 template <typename CallbacksT> 2350 static bool isModulePassName(StringRef Name, CallbacksT &Callbacks) { 2351 // Manually handle aliases for pre-configured pipeline fragments. 2352 if (startsWithDefaultPipelineAliasPrefix(Name)) 2353 return DefaultAliasRegex.match(Name); 2354 2355 // Explicitly handle pass manager names. 2356 if (Name == "module") 2357 return true; 2358 if (Name == "cgscc") 2359 return true; 2360 if (Name == "function") 2361 return true; 2362 2363 // Explicitly handle custom-parsed pass names. 2364 if (parseRepeatPassName(Name)) 2365 return true; 2366 2367 #define MODULE_PASS(NAME, CREATE_PASS) \ 2368 if (Name == NAME) \ 2369 return true; 2370 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 2371 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2372 return true; 2373 #include "PassRegistry.def" 2374 2375 return callbacksAcceptPassName<ModulePassManager>(Name, Callbacks); 2376 } 2377 2378 template <typename CallbacksT> 2379 static bool isCGSCCPassName(StringRef Name, CallbacksT &Callbacks) { 2380 // Explicitly handle pass manager names. 2381 if (Name == "cgscc") 2382 return true; 2383 if (Name == "function") 2384 return true; 2385 2386 // Explicitly handle custom-parsed pass names. 2387 if (parseRepeatPassName(Name)) 2388 return true; 2389 if (parseDevirtPassName(Name)) 2390 return true; 2391 2392 #define CGSCC_PASS(NAME, CREATE_PASS) \ 2393 if (Name == NAME) \ 2394 return true; 2395 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 2396 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2397 return true; 2398 #include "PassRegistry.def" 2399 2400 return callbacksAcceptPassName<CGSCCPassManager>(Name, Callbacks); 2401 } 2402 2403 template <typename CallbacksT> 2404 static bool isFunctionPassName(StringRef Name, CallbacksT &Callbacks) { 2405 // Explicitly handle pass manager names. 2406 if (Name == "function") 2407 return true; 2408 if (Name == "loop" || Name == "loop-mssa") 2409 return true; 2410 2411 // Explicitly handle custom-parsed pass names. 2412 if (parseRepeatPassName(Name)) 2413 return true; 2414 2415 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2416 if (Name == NAME) \ 2417 return true; 2418 #define FUNCTION_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2419 if (checkParametrizedPassName(Name, NAME)) \ 2420 return true; 2421 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 2422 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2423 return true; 2424 #include "PassRegistry.def" 2425 2426 return callbacksAcceptPassName<FunctionPassManager>(Name, Callbacks); 2427 } 2428 2429 template <typename CallbacksT> 2430 static bool isLoopPassName(StringRef Name, CallbacksT &Callbacks) { 2431 // Explicitly handle pass manager names. 2432 if (Name == "loop" || Name == "loop-mssa") 2433 return true; 2434 2435 // Explicitly handle custom-parsed pass names. 2436 if (parseRepeatPassName(Name)) 2437 return true; 2438 2439 #define LOOP_PASS(NAME, CREATE_PASS) \ 2440 if (Name == NAME) \ 2441 return true; 2442 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2443 if (checkParametrizedPassName(Name, NAME)) \ 2444 return true; 2445 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 2446 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2447 return true; 2448 #include "PassRegistry.def" 2449 2450 return callbacksAcceptPassName<LoopPassManager>(Name, Callbacks); 2451 } 2452 2453 Optional<std::vector<PassBuilder::PipelineElement>> 2454 PassBuilder::parsePipelineText(StringRef Text) { 2455 std::vector<PipelineElement> ResultPipeline; 2456 2457 SmallVector<std::vector<PipelineElement> *, 4> PipelineStack = { 2458 &ResultPipeline}; 2459 for (;;) { 2460 std::vector<PipelineElement> &Pipeline = *PipelineStack.back(); 2461 size_t Pos = Text.find_first_of(",()"); 2462 Pipeline.push_back({Text.substr(0, Pos), {}}); 2463 2464 // If we have a single terminating name, we're done. 2465 if (Pos == Text.npos) 2466 break; 2467 2468 char Sep = Text[Pos]; 2469 Text = Text.substr(Pos + 1); 2470 if (Sep == ',') 2471 // Just a name ending in a comma, continue. 2472 continue; 2473 2474 if (Sep == '(') { 2475 // Push the inner pipeline onto the stack to continue processing. 2476 PipelineStack.push_back(&Pipeline.back().InnerPipeline); 2477 continue; 2478 } 2479 2480 assert(Sep == ')' && "Bogus separator!"); 2481 // When handling the close parenthesis, we greedily consume them to avoid 2482 // empty strings in the pipeline. 2483 do { 2484 // If we try to pop the outer pipeline we have unbalanced parentheses. 2485 if (PipelineStack.size() == 1) 2486 return None; 2487 2488 PipelineStack.pop_back(); 2489 } while (Text.consume_front(")")); 2490 2491 // Check if we've finished parsing. 2492 if (Text.empty()) 2493 break; 2494 2495 // Otherwise, the end of an inner pipeline always has to be followed by 2496 // a comma, and then we can continue. 2497 if (!Text.consume_front(",")) 2498 return None; 2499 } 2500 2501 if (PipelineStack.size() > 1) 2502 // Unbalanced paretheses. 2503 return None; 2504 2505 assert(PipelineStack.back() == &ResultPipeline && 2506 "Wrong pipeline at the bottom of the stack!"); 2507 return {std::move(ResultPipeline)}; 2508 } 2509 2510 Error PassBuilder::parseModulePass(ModulePassManager &MPM, 2511 const PipelineElement &E) { 2512 auto &Name = E.Name; 2513 auto &InnerPipeline = E.InnerPipeline; 2514 2515 // First handle complex passes like the pass managers which carry pipelines. 2516 if (!InnerPipeline.empty()) { 2517 if (Name == "module") { 2518 ModulePassManager NestedMPM; 2519 if (auto Err = parseModulePassPipeline(NestedMPM, InnerPipeline)) 2520 return Err; 2521 MPM.addPass(std::move(NestedMPM)); 2522 return Error::success(); 2523 } 2524 if (Name == "cgscc") { 2525 CGSCCPassManager CGPM; 2526 if (auto Err = parseCGSCCPassPipeline(CGPM, InnerPipeline)) 2527 return Err; 2528 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPM))); 2529 return Error::success(); 2530 } 2531 if (Name == "function") { 2532 FunctionPassManager FPM; 2533 if (auto Err = parseFunctionPassPipeline(FPM, InnerPipeline)) 2534 return Err; 2535 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 2536 return Error::success(); 2537 } 2538 if (auto Count = parseRepeatPassName(Name)) { 2539 ModulePassManager NestedMPM; 2540 if (auto Err = parseModulePassPipeline(NestedMPM, InnerPipeline)) 2541 return Err; 2542 MPM.addPass(createRepeatedPass(*Count, std::move(NestedMPM))); 2543 return Error::success(); 2544 } 2545 2546 for (auto &C : ModulePipelineParsingCallbacks) 2547 if (C(Name, MPM, InnerPipeline)) 2548 return Error::success(); 2549 2550 // Normal passes can't have pipelines. 2551 return make_error<StringError>( 2552 formatv("invalid use of '{0}' pass as module pipeline", Name).str(), 2553 inconvertibleErrorCode()); 2554 ; 2555 } 2556 2557 // Manually handle aliases for pre-configured pipeline fragments. 2558 if (startsWithDefaultPipelineAliasPrefix(Name)) { 2559 SmallVector<StringRef, 3> Matches; 2560 if (!DefaultAliasRegex.match(Name, &Matches)) 2561 return make_error<StringError>( 2562 formatv("unknown default pipeline alias '{0}'", Name).str(), 2563 inconvertibleErrorCode()); 2564 2565 assert(Matches.size() == 3 && "Must capture two matched strings!"); 2566 2567 OptimizationLevel L = StringSwitch<OptimizationLevel>(Matches[2]) 2568 .Case("O0", OptimizationLevel::O0) 2569 .Case("O1", OptimizationLevel::O1) 2570 .Case("O2", OptimizationLevel::O2) 2571 .Case("O3", OptimizationLevel::O3) 2572 .Case("Os", OptimizationLevel::Os) 2573 .Case("Oz", OptimizationLevel::Oz); 2574 if (L == OptimizationLevel::O0 && Matches[1] != "thinlto" && 2575 Matches[1] != "lto") { 2576 MPM.addPass(buildO0DefaultPipeline(L, Matches[1] == "thinlto-pre-link" || 2577 Matches[1] == "lto-pre-link")); 2578 return Error::success(); 2579 } 2580 2581 // This is consistent with old pass manager invoked via opt, but 2582 // inconsistent with clang. Clang doesn't enable loop vectorization 2583 // but does enable slp vectorization at Oz. 2584 PTO.LoopVectorization = 2585 L.getSpeedupLevel() > 1 && L != OptimizationLevel::Oz; 2586 PTO.SLPVectorization = 2587 L.getSpeedupLevel() > 1 && L != OptimizationLevel::Oz; 2588 2589 if (Matches[1] == "default") { 2590 MPM.addPass(buildPerModuleDefaultPipeline(L)); 2591 } else if (Matches[1] == "thinlto-pre-link") { 2592 MPM.addPass(buildThinLTOPreLinkDefaultPipeline(L)); 2593 } else if (Matches[1] == "thinlto") { 2594 MPM.addPass(buildThinLTODefaultPipeline(L, nullptr)); 2595 } else if (Matches[1] == "lto-pre-link") { 2596 MPM.addPass(buildLTOPreLinkDefaultPipeline(L)); 2597 } else { 2598 assert(Matches[1] == "lto" && "Not one of the matched options!"); 2599 MPM.addPass(buildLTODefaultPipeline(L, nullptr)); 2600 } 2601 return Error::success(); 2602 } 2603 2604 // Finally expand the basic registered passes from the .inc file. 2605 #define MODULE_PASS(NAME, CREATE_PASS) \ 2606 if (Name == NAME) { \ 2607 MPM.addPass(CREATE_PASS); \ 2608 return Error::success(); \ 2609 } 2610 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 2611 if (Name == "require<" NAME ">") { \ 2612 MPM.addPass( \ 2613 RequireAnalysisPass< \ 2614 std::remove_reference<decltype(CREATE_PASS)>::type, Module>()); \ 2615 return Error::success(); \ 2616 } \ 2617 if (Name == "invalidate<" NAME ">") { \ 2618 MPM.addPass(InvalidateAnalysisPass< \ 2619 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2620 return Error::success(); \ 2621 } 2622 #define CGSCC_PASS(NAME, CREATE_PASS) \ 2623 if (Name == NAME) { \ 2624 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(CREATE_PASS)); \ 2625 return Error::success(); \ 2626 } 2627 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2628 if (Name == NAME) { \ 2629 MPM.addPass(createModuleToFunctionPassAdaptor(CREATE_PASS)); \ 2630 return Error::success(); \ 2631 } 2632 #define FUNCTION_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2633 if (checkParametrizedPassName(Name, NAME)) { \ 2634 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2635 if (!Params) \ 2636 return Params.takeError(); \ 2637 MPM.addPass(createModuleToFunctionPassAdaptor(CREATE_PASS(Params.get()))); \ 2638 return Error::success(); \ 2639 } 2640 #define LOOP_PASS(NAME, CREATE_PASS) \ 2641 if (Name == NAME) { \ 2642 MPM.addPass(createModuleToFunctionPassAdaptor( \ 2643 createFunctionToLoopPassAdaptor(CREATE_PASS, false, false))); \ 2644 return Error::success(); \ 2645 } 2646 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2647 if (checkParametrizedPassName(Name, NAME)) { \ 2648 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2649 if (!Params) \ 2650 return Params.takeError(); \ 2651 MPM.addPass( \ 2652 createModuleToFunctionPassAdaptor(createFunctionToLoopPassAdaptor( \ 2653 CREATE_PASS(Params.get()), false, false))); \ 2654 return Error::success(); \ 2655 } 2656 #include "PassRegistry.def" 2657 2658 for (auto &C : ModulePipelineParsingCallbacks) 2659 if (C(Name, MPM, InnerPipeline)) 2660 return Error::success(); 2661 return make_error<StringError>( 2662 formatv("unknown module pass '{0}'", Name).str(), 2663 inconvertibleErrorCode()); 2664 } 2665 2666 Error PassBuilder::parseCGSCCPass(CGSCCPassManager &CGPM, 2667 const PipelineElement &E) { 2668 auto &Name = E.Name; 2669 auto &InnerPipeline = E.InnerPipeline; 2670 2671 // First handle complex passes like the pass managers which carry pipelines. 2672 if (!InnerPipeline.empty()) { 2673 if (Name == "cgscc") { 2674 CGSCCPassManager NestedCGPM; 2675 if (auto Err = parseCGSCCPassPipeline(NestedCGPM, InnerPipeline)) 2676 return Err; 2677 // Add the nested pass manager with the appropriate adaptor. 2678 CGPM.addPass(std::move(NestedCGPM)); 2679 return Error::success(); 2680 } 2681 if (Name == "function") { 2682 FunctionPassManager FPM; 2683 if (auto Err = parseFunctionPassPipeline(FPM, InnerPipeline)) 2684 return Err; 2685 // Add the nested pass manager with the appropriate adaptor. 2686 CGPM.addPass(createCGSCCToFunctionPassAdaptor(std::move(FPM))); 2687 return Error::success(); 2688 } 2689 if (auto Count = parseRepeatPassName(Name)) { 2690 CGSCCPassManager NestedCGPM; 2691 if (auto Err = parseCGSCCPassPipeline(NestedCGPM, InnerPipeline)) 2692 return Err; 2693 CGPM.addPass(createRepeatedPass(*Count, std::move(NestedCGPM))); 2694 return Error::success(); 2695 } 2696 if (auto MaxRepetitions = parseDevirtPassName(Name)) { 2697 CGSCCPassManager NestedCGPM; 2698 if (auto Err = parseCGSCCPassPipeline(NestedCGPM, InnerPipeline)) 2699 return Err; 2700 CGPM.addPass( 2701 createDevirtSCCRepeatedPass(std::move(NestedCGPM), *MaxRepetitions)); 2702 return Error::success(); 2703 } 2704 2705 for (auto &C : CGSCCPipelineParsingCallbacks) 2706 if (C(Name, CGPM, InnerPipeline)) 2707 return Error::success(); 2708 2709 // Normal passes can't have pipelines. 2710 return make_error<StringError>( 2711 formatv("invalid use of '{0}' pass as cgscc pipeline", Name).str(), 2712 inconvertibleErrorCode()); 2713 } 2714 2715 // Now expand the basic registered passes from the .inc file. 2716 #define CGSCC_PASS(NAME, CREATE_PASS) \ 2717 if (Name == NAME) { \ 2718 CGPM.addPass(CREATE_PASS); \ 2719 return Error::success(); \ 2720 } 2721 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 2722 if (Name == "require<" NAME ">") { \ 2723 CGPM.addPass(RequireAnalysisPass< \ 2724 std::remove_reference<decltype(CREATE_PASS)>::type, \ 2725 LazyCallGraph::SCC, CGSCCAnalysisManager, LazyCallGraph &, \ 2726 CGSCCUpdateResult &>()); \ 2727 return Error::success(); \ 2728 } \ 2729 if (Name == "invalidate<" NAME ">") { \ 2730 CGPM.addPass(InvalidateAnalysisPass< \ 2731 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2732 return Error::success(); \ 2733 } 2734 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2735 if (Name == NAME) { \ 2736 CGPM.addPass(createCGSCCToFunctionPassAdaptor(CREATE_PASS)); \ 2737 return Error::success(); \ 2738 } 2739 #define FUNCTION_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2740 if (checkParametrizedPassName(Name, NAME)) { \ 2741 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2742 if (!Params) \ 2743 return Params.takeError(); \ 2744 CGPM.addPass(createCGSCCToFunctionPassAdaptor(CREATE_PASS(Params.get()))); \ 2745 return Error::success(); \ 2746 } 2747 #define LOOP_PASS(NAME, CREATE_PASS) \ 2748 if (Name == NAME) { \ 2749 CGPM.addPass(createCGSCCToFunctionPassAdaptor( \ 2750 createFunctionToLoopPassAdaptor(CREATE_PASS, false, false))); \ 2751 return Error::success(); \ 2752 } 2753 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2754 if (checkParametrizedPassName(Name, NAME)) { \ 2755 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2756 if (!Params) \ 2757 return Params.takeError(); \ 2758 CGPM.addPass( \ 2759 createCGSCCToFunctionPassAdaptor(createFunctionToLoopPassAdaptor( \ 2760 CREATE_PASS(Params.get()), false, false))); \ 2761 return Error::success(); \ 2762 } 2763 #include "PassRegistry.def" 2764 2765 for (auto &C : CGSCCPipelineParsingCallbacks) 2766 if (C(Name, CGPM, InnerPipeline)) 2767 return Error::success(); 2768 return make_error<StringError>( 2769 formatv("unknown cgscc pass '{0}'", Name).str(), 2770 inconvertibleErrorCode()); 2771 } 2772 2773 Error PassBuilder::parseFunctionPass(FunctionPassManager &FPM, 2774 const PipelineElement &E) { 2775 auto &Name = E.Name; 2776 auto &InnerPipeline = E.InnerPipeline; 2777 2778 // First handle complex passes like the pass managers which carry pipelines. 2779 if (!InnerPipeline.empty()) { 2780 if (Name == "function") { 2781 FunctionPassManager NestedFPM; 2782 if (auto Err = parseFunctionPassPipeline(NestedFPM, InnerPipeline)) 2783 return Err; 2784 // Add the nested pass manager with the appropriate adaptor. 2785 FPM.addPass(std::move(NestedFPM)); 2786 return Error::success(); 2787 } 2788 if (Name == "loop" || Name == "loop-mssa") { 2789 LoopPassManager LPM; 2790 if (auto Err = parseLoopPassPipeline(LPM, InnerPipeline)) 2791 return Err; 2792 // Add the nested pass manager with the appropriate adaptor. 2793 bool UseMemorySSA = (Name == "loop-mssa"); 2794 bool UseBFI = llvm::any_of( 2795 InnerPipeline, [](auto Pipeline) { return Pipeline.Name == "licm"; }); 2796 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM), UseMemorySSA, 2797 UseBFI)); 2798 return Error::success(); 2799 } 2800 if (auto Count = parseRepeatPassName(Name)) { 2801 FunctionPassManager NestedFPM; 2802 if (auto Err = parseFunctionPassPipeline(NestedFPM, InnerPipeline)) 2803 return Err; 2804 FPM.addPass(createRepeatedPass(*Count, std::move(NestedFPM))); 2805 return Error::success(); 2806 } 2807 2808 for (auto &C : FunctionPipelineParsingCallbacks) 2809 if (C(Name, FPM, InnerPipeline)) 2810 return Error::success(); 2811 2812 // Normal passes can't have pipelines. 2813 return make_error<StringError>( 2814 formatv("invalid use of '{0}' pass as function pipeline", Name).str(), 2815 inconvertibleErrorCode()); 2816 } 2817 2818 // Now expand the basic registered passes from the .inc file. 2819 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2820 if (Name == NAME) { \ 2821 FPM.addPass(CREATE_PASS); \ 2822 return Error::success(); \ 2823 } 2824 #define FUNCTION_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2825 if (checkParametrizedPassName(Name, NAME)) { \ 2826 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2827 if (!Params) \ 2828 return Params.takeError(); \ 2829 FPM.addPass(CREATE_PASS(Params.get())); \ 2830 return Error::success(); \ 2831 } 2832 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 2833 if (Name == "require<" NAME ">") { \ 2834 FPM.addPass( \ 2835 RequireAnalysisPass< \ 2836 std::remove_reference<decltype(CREATE_PASS)>::type, Function>()); \ 2837 return Error::success(); \ 2838 } \ 2839 if (Name == "invalidate<" NAME ">") { \ 2840 FPM.addPass(InvalidateAnalysisPass< \ 2841 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2842 return Error::success(); \ 2843 } 2844 // FIXME: UseMemorySSA is set to false. Maybe we could do things like: 2845 // bool UseMemorySSA = !("canon-freeze" || "loop-predication" || 2846 // "guard-widening"); 2847 // The risk is that it may become obsolete if we're not careful. 2848 #define LOOP_PASS(NAME, CREATE_PASS) \ 2849 if (Name == NAME) { \ 2850 FPM.addPass(createFunctionToLoopPassAdaptor(CREATE_PASS, false, false)); \ 2851 return Error::success(); \ 2852 } 2853 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2854 if (checkParametrizedPassName(Name, NAME)) { \ 2855 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2856 if (!Params) \ 2857 return Params.takeError(); \ 2858 FPM.addPass(createFunctionToLoopPassAdaptor(CREATE_PASS(Params.get()), \ 2859 false, false)); \ 2860 return Error::success(); \ 2861 } 2862 #include "PassRegistry.def" 2863 2864 for (auto &C : FunctionPipelineParsingCallbacks) 2865 if (C(Name, FPM, InnerPipeline)) 2866 return Error::success(); 2867 return make_error<StringError>( 2868 formatv("unknown function pass '{0}'", Name).str(), 2869 inconvertibleErrorCode()); 2870 } 2871 2872 Error PassBuilder::parseLoopPass(LoopPassManager &LPM, 2873 const PipelineElement &E) { 2874 StringRef Name = E.Name; 2875 auto &InnerPipeline = E.InnerPipeline; 2876 2877 // First handle complex passes like the pass managers which carry pipelines. 2878 if (!InnerPipeline.empty()) { 2879 if (Name == "loop") { 2880 LoopPassManager NestedLPM; 2881 if (auto Err = parseLoopPassPipeline(NestedLPM, InnerPipeline)) 2882 return Err; 2883 // Add the nested pass manager with the appropriate adaptor. 2884 LPM.addPass(std::move(NestedLPM)); 2885 return Error::success(); 2886 } 2887 if (auto Count = parseRepeatPassName(Name)) { 2888 LoopPassManager NestedLPM; 2889 if (auto Err = parseLoopPassPipeline(NestedLPM, InnerPipeline)) 2890 return Err; 2891 LPM.addPass(createRepeatedPass(*Count, std::move(NestedLPM))); 2892 return Error::success(); 2893 } 2894 2895 for (auto &C : LoopPipelineParsingCallbacks) 2896 if (C(Name, LPM, InnerPipeline)) 2897 return Error::success(); 2898 2899 // Normal passes can't have pipelines. 2900 return make_error<StringError>( 2901 formatv("invalid use of '{0}' pass as loop pipeline", Name).str(), 2902 inconvertibleErrorCode()); 2903 } 2904 2905 // Now expand the basic registered passes from the .inc file. 2906 #define LOOP_PASS(NAME, CREATE_PASS) \ 2907 if (Name == NAME) { \ 2908 LPM.addPass(CREATE_PASS); \ 2909 return Error::success(); \ 2910 } 2911 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2912 if (checkParametrizedPassName(Name, NAME)) { \ 2913 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2914 if (!Params) \ 2915 return Params.takeError(); \ 2916 LPM.addPass(CREATE_PASS(Params.get())); \ 2917 return Error::success(); \ 2918 } 2919 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 2920 if (Name == "require<" NAME ">") { \ 2921 LPM.addPass(RequireAnalysisPass< \ 2922 std::remove_reference<decltype(CREATE_PASS)>::type, Loop, \ 2923 LoopAnalysisManager, LoopStandardAnalysisResults &, \ 2924 LPMUpdater &>()); \ 2925 return Error::success(); \ 2926 } \ 2927 if (Name == "invalidate<" NAME ">") { \ 2928 LPM.addPass(InvalidateAnalysisPass< \ 2929 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2930 return Error::success(); \ 2931 } 2932 #include "PassRegistry.def" 2933 2934 for (auto &C : LoopPipelineParsingCallbacks) 2935 if (C(Name, LPM, InnerPipeline)) 2936 return Error::success(); 2937 return make_error<StringError>(formatv("unknown loop pass '{0}'", Name).str(), 2938 inconvertibleErrorCode()); 2939 } 2940 2941 bool PassBuilder::parseAAPassName(AAManager &AA, StringRef Name) { 2942 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 2943 if (Name == NAME) { \ 2944 AA.registerModuleAnalysis< \ 2945 std::remove_reference<decltype(CREATE_PASS)>::type>(); \ 2946 return true; \ 2947 } 2948 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 2949 if (Name == NAME) { \ 2950 AA.registerFunctionAnalysis< \ 2951 std::remove_reference<decltype(CREATE_PASS)>::type>(); \ 2952 return true; \ 2953 } 2954 #include "PassRegistry.def" 2955 2956 for (auto &C : AAParsingCallbacks) 2957 if (C(Name, AA)) 2958 return true; 2959 return false; 2960 } 2961 2962 Error PassBuilder::parseLoopPassPipeline(LoopPassManager &LPM, 2963 ArrayRef<PipelineElement> Pipeline) { 2964 for (const auto &Element : Pipeline) { 2965 if (auto Err = parseLoopPass(LPM, Element)) 2966 return Err; 2967 } 2968 return Error::success(); 2969 } 2970 2971 Error PassBuilder::parseFunctionPassPipeline( 2972 FunctionPassManager &FPM, ArrayRef<PipelineElement> Pipeline) { 2973 for (const auto &Element : Pipeline) { 2974 if (auto Err = parseFunctionPass(FPM, Element)) 2975 return Err; 2976 } 2977 return Error::success(); 2978 } 2979 2980 Error PassBuilder::parseCGSCCPassPipeline(CGSCCPassManager &CGPM, 2981 ArrayRef<PipelineElement> Pipeline) { 2982 for (const auto &Element : Pipeline) { 2983 if (auto Err = parseCGSCCPass(CGPM, Element)) 2984 return Err; 2985 } 2986 return Error::success(); 2987 } 2988 2989 void PassBuilder::crossRegisterProxies(LoopAnalysisManager &LAM, 2990 FunctionAnalysisManager &FAM, 2991 CGSCCAnalysisManager &CGAM, 2992 ModuleAnalysisManager &MAM) { 2993 MAM.registerPass([&] { return FunctionAnalysisManagerModuleProxy(FAM); }); 2994 MAM.registerPass([&] { return CGSCCAnalysisManagerModuleProxy(CGAM); }); 2995 CGAM.registerPass([&] { return ModuleAnalysisManagerCGSCCProxy(MAM); }); 2996 FAM.registerPass([&] { return CGSCCAnalysisManagerFunctionProxy(CGAM); }); 2997 FAM.registerPass([&] { return ModuleAnalysisManagerFunctionProxy(MAM); }); 2998 FAM.registerPass([&] { return LoopAnalysisManagerFunctionProxy(LAM); }); 2999 LAM.registerPass([&] { return FunctionAnalysisManagerLoopProxy(FAM); }); 3000 } 3001 3002 Error PassBuilder::parseModulePassPipeline(ModulePassManager &MPM, 3003 ArrayRef<PipelineElement> Pipeline) { 3004 for (const auto &Element : Pipeline) { 3005 if (auto Err = parseModulePass(MPM, Element)) 3006 return Err; 3007 } 3008 return Error::success(); 3009 } 3010 3011 // Primary pass pipeline description parsing routine for a \c ModulePassManager 3012 // FIXME: Should this routine accept a TargetMachine or require the caller to 3013 // pre-populate the analysis managers with target-specific stuff? 3014 Error PassBuilder::parsePassPipeline(ModulePassManager &MPM, 3015 StringRef PipelineText) { 3016 auto Pipeline = parsePipelineText(PipelineText); 3017 if (!Pipeline || Pipeline->empty()) 3018 return make_error<StringError>( 3019 formatv("invalid pipeline '{0}'", PipelineText).str(), 3020 inconvertibleErrorCode()); 3021 3022 // If the first name isn't at the module layer, wrap the pipeline up 3023 // automatically. 3024 StringRef FirstName = Pipeline->front().Name; 3025 3026 if (!isModulePassName(FirstName, ModulePipelineParsingCallbacks)) { 3027 if (isCGSCCPassName(FirstName, CGSCCPipelineParsingCallbacks)) { 3028 Pipeline = {{"cgscc", std::move(*Pipeline)}}; 3029 } else if (isFunctionPassName(FirstName, 3030 FunctionPipelineParsingCallbacks)) { 3031 Pipeline = {{"function", std::move(*Pipeline)}}; 3032 } else if (isLoopPassName(FirstName, LoopPipelineParsingCallbacks)) { 3033 Pipeline = {{"function", {{"loop", std::move(*Pipeline)}}}}; 3034 } else { 3035 for (auto &C : TopLevelPipelineParsingCallbacks) 3036 if (C(MPM, *Pipeline)) 3037 return Error::success(); 3038 3039 // Unknown pass or pipeline name! 3040 auto &InnerPipeline = Pipeline->front().InnerPipeline; 3041 return make_error<StringError>( 3042 formatv("unknown {0} name '{1}'", 3043 (InnerPipeline.empty() ? "pass" : "pipeline"), FirstName) 3044 .str(), 3045 inconvertibleErrorCode()); 3046 } 3047 } 3048 3049 if (auto Err = parseModulePassPipeline(MPM, *Pipeline)) 3050 return Err; 3051 return Error::success(); 3052 } 3053 3054 // Primary pass pipeline description parsing routine for a \c CGSCCPassManager 3055 Error PassBuilder::parsePassPipeline(CGSCCPassManager &CGPM, 3056 StringRef PipelineText) { 3057 auto Pipeline = parsePipelineText(PipelineText); 3058 if (!Pipeline || Pipeline->empty()) 3059 return make_error<StringError>( 3060 formatv("invalid pipeline '{0}'", PipelineText).str(), 3061 inconvertibleErrorCode()); 3062 3063 StringRef FirstName = Pipeline->front().Name; 3064 if (!isCGSCCPassName(FirstName, CGSCCPipelineParsingCallbacks)) 3065 return make_error<StringError>( 3066 formatv("unknown cgscc pass '{0}' in pipeline '{1}'", FirstName, 3067 PipelineText) 3068 .str(), 3069 inconvertibleErrorCode()); 3070 3071 if (auto Err = parseCGSCCPassPipeline(CGPM, *Pipeline)) 3072 return Err; 3073 return Error::success(); 3074 } 3075 3076 // Primary pass pipeline description parsing routine for a \c 3077 // FunctionPassManager 3078 Error PassBuilder::parsePassPipeline(FunctionPassManager &FPM, 3079 StringRef PipelineText) { 3080 auto Pipeline = parsePipelineText(PipelineText); 3081 if (!Pipeline || Pipeline->empty()) 3082 return make_error<StringError>( 3083 formatv("invalid pipeline '{0}'", PipelineText).str(), 3084 inconvertibleErrorCode()); 3085 3086 StringRef FirstName = Pipeline->front().Name; 3087 if (!isFunctionPassName(FirstName, FunctionPipelineParsingCallbacks)) 3088 return make_error<StringError>( 3089 formatv("unknown function pass '{0}' in pipeline '{1}'", FirstName, 3090 PipelineText) 3091 .str(), 3092 inconvertibleErrorCode()); 3093 3094 if (auto Err = parseFunctionPassPipeline(FPM, *Pipeline)) 3095 return Err; 3096 return Error::success(); 3097 } 3098 3099 // Primary pass pipeline description parsing routine for a \c LoopPassManager 3100 Error PassBuilder::parsePassPipeline(LoopPassManager &CGPM, 3101 StringRef PipelineText) { 3102 auto Pipeline = parsePipelineText(PipelineText); 3103 if (!Pipeline || Pipeline->empty()) 3104 return make_error<StringError>( 3105 formatv("invalid pipeline '{0}'", PipelineText).str(), 3106 inconvertibleErrorCode()); 3107 3108 if (auto Err = parseLoopPassPipeline(CGPM, *Pipeline)) 3109 return Err; 3110 3111 return Error::success(); 3112 } 3113 3114 Error PassBuilder::parseAAPipeline(AAManager &AA, StringRef PipelineText) { 3115 // If the pipeline just consists of the word 'default' just replace the AA 3116 // manager with our default one. 3117 if (PipelineText == "default") { 3118 AA = buildDefaultAAPipeline(); 3119 return Error::success(); 3120 } 3121 3122 while (!PipelineText.empty()) { 3123 StringRef Name; 3124 std::tie(Name, PipelineText) = PipelineText.split(','); 3125 if (!parseAAPassName(AA, Name)) 3126 return make_error<StringError>( 3127 formatv("unknown alias analysis name '{0}'", Name).str(), 3128 inconvertibleErrorCode()); 3129 } 3130 3131 return Error::success(); 3132 } 3133 3134 bool PassBuilder::isAAPassName(StringRef PassName) { 3135 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3136 if (PassName == NAME) \ 3137 return true; 3138 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3139 if (PassName == NAME) \ 3140 return true; 3141 #include "PassRegistry.def" 3142 return false; 3143 } 3144 3145 bool PassBuilder::isAnalysisPassName(StringRef PassName) { 3146 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 3147 if (PassName == NAME) \ 3148 return true; 3149 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 3150 if (PassName == NAME) \ 3151 return true; 3152 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 3153 if (PassName == NAME) \ 3154 return true; 3155 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 3156 if (PassName == NAME) \ 3157 return true; 3158 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3159 if (PassName == NAME) \ 3160 return true; 3161 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3162 if (PassName == NAME) \ 3163 return true; 3164 #include "PassRegistry.def" 3165 return false; 3166 } 3167 3168 static void printPassName(StringRef PassName, raw_ostream &OS) { 3169 OS << " " << PassName << "\n"; 3170 } 3171 3172 void PassBuilder::printPassNames(raw_ostream &OS) { 3173 // TODO: print pass descriptions when they are available 3174 3175 OS << "Module passes:\n"; 3176 #define MODULE_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3177 #include "PassRegistry.def" 3178 3179 OS << "Module analyses:\n"; 3180 #define MODULE_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3181 #include "PassRegistry.def" 3182 3183 OS << "Module alias analyses:\n"; 3184 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3185 #include "PassRegistry.def" 3186 3187 OS << "CGSCC passes:\n"; 3188 #define CGSCC_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3189 #include "PassRegistry.def" 3190 3191 OS << "CGSCC analyses:\n"; 3192 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3193 #include "PassRegistry.def" 3194 3195 OS << "Function passes:\n"; 3196 #define FUNCTION_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3197 #include "PassRegistry.def" 3198 3199 OS << "Function analyses:\n"; 3200 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3201 #include "PassRegistry.def" 3202 3203 OS << "Function alias analyses:\n"; 3204 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3205 #include "PassRegistry.def" 3206 3207 OS << "Loop passes:\n"; 3208 #define LOOP_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3209 #include "PassRegistry.def" 3210 3211 OS << "Loop analyses:\n"; 3212 #define LOOP_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3213 #include "PassRegistry.def" 3214 } 3215 3216 void PassBuilder::registerParseTopLevelPipelineCallback( 3217 const std::function<bool(ModulePassManager &, ArrayRef<PipelineElement>)> 3218 &C) { 3219 TopLevelPipelineParsingCallbacks.push_back(C); 3220 } 3221