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/LoopBoundSplit.h" 163 #include "llvm/Transforms/Scalar/LoopDataPrefetch.h" 164 #include "llvm/Transforms/Scalar/LoopDeletion.h" 165 #include "llvm/Transforms/Scalar/LoopDistribute.h" 166 #include "llvm/Transforms/Scalar/LoopFlatten.h" 167 #include "llvm/Transforms/Scalar/LoopFuse.h" 168 #include "llvm/Transforms/Scalar/LoopIdiomRecognize.h" 169 #include "llvm/Transforms/Scalar/LoopInstSimplify.h" 170 #include "llvm/Transforms/Scalar/LoopInterchange.h" 171 #include "llvm/Transforms/Scalar/LoopLoadElimination.h" 172 #include "llvm/Transforms/Scalar/LoopPassManager.h" 173 #include "llvm/Transforms/Scalar/LoopPredication.h" 174 #include "llvm/Transforms/Scalar/LoopReroll.h" 175 #include "llvm/Transforms/Scalar/LoopRotation.h" 176 #include "llvm/Transforms/Scalar/LoopSimplifyCFG.h" 177 #include "llvm/Transforms/Scalar/LoopSink.h" 178 #include "llvm/Transforms/Scalar/LoopStrengthReduce.h" 179 #include "llvm/Transforms/Scalar/LoopUnrollAndJamPass.h" 180 #include "llvm/Transforms/Scalar/LoopUnrollPass.h" 181 #include "llvm/Transforms/Scalar/LoopVersioningLICM.h" 182 #include "llvm/Transforms/Scalar/LowerAtomic.h" 183 #include "llvm/Transforms/Scalar/LowerConstantIntrinsics.h" 184 #include "llvm/Transforms/Scalar/LowerExpectIntrinsic.h" 185 #include "llvm/Transforms/Scalar/LowerGuardIntrinsic.h" 186 #include "llvm/Transforms/Scalar/LowerMatrixIntrinsics.h" 187 #include "llvm/Transforms/Scalar/LowerWidenableCondition.h" 188 #include "llvm/Transforms/Scalar/MakeGuardsExplicit.h" 189 #include "llvm/Transforms/Scalar/MemCpyOptimizer.h" 190 #include "llvm/Transforms/Scalar/MergeICmps.h" 191 #include "llvm/Transforms/Scalar/MergedLoadStoreMotion.h" 192 #include "llvm/Transforms/Scalar/NaryReassociate.h" 193 #include "llvm/Transforms/Scalar/NewGVN.h" 194 #include "llvm/Transforms/Scalar/PartiallyInlineLibCalls.h" 195 #include "llvm/Transforms/Scalar/Reassociate.h" 196 #include "llvm/Transforms/Scalar/Reg2Mem.h" 197 #include "llvm/Transforms/Scalar/RewriteStatepointsForGC.h" 198 #include "llvm/Transforms/Scalar/SCCP.h" 199 #include "llvm/Transforms/Scalar/SROA.h" 200 #include "llvm/Transforms/Scalar/ScalarizeMaskedMemIntrin.h" 201 #include "llvm/Transforms/Scalar/Scalarizer.h" 202 #include "llvm/Transforms/Scalar/SeparateConstOffsetFromGEP.h" 203 #include "llvm/Transforms/Scalar/SimpleLoopUnswitch.h" 204 #include "llvm/Transforms/Scalar/SimplifyCFG.h" 205 #include "llvm/Transforms/Scalar/Sink.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 PipelineTuningOptions::PipelineTuningOptions() { 281 LoopInterleaving = true; 282 LoopVectorization = true; 283 SLPVectorization = false; 284 LoopUnrolling = true; 285 ForgetAllSCEVInLoopUnroll = ForgetSCEVInLoopUnroll; 286 Coroutines = false; 287 LicmMssaOptCap = SetLicmMssaOptCap; 288 LicmMssaNoAccForPromotionCap = SetLicmMssaNoAccForPromotionCap; 289 CallGraphProfile = true; 290 MergeFunctions = false; 291 } 292 293 namespace llvm { 294 extern cl::opt<unsigned> MaxDevirtIterations; 295 extern cl::opt<bool> EnableConstraintElimination; 296 extern cl::opt<bool> EnableFunctionSpecialization; 297 extern cl::opt<bool> EnableGVNHoist; 298 extern cl::opt<bool> EnableGVNSink; 299 extern cl::opt<bool> EnableHotColdSplit; 300 extern cl::opt<bool> EnableIROutliner; 301 extern cl::opt<bool> EnableOrderFileInstrumentation; 302 extern cl::opt<bool> EnableCHR; 303 extern cl::opt<bool> EnableLoopInterchange; 304 extern cl::opt<bool> EnableUnrollAndJam; 305 extern cl::opt<bool> EnableLoopFlatten; 306 extern cl::opt<bool> RunNewGVN; 307 extern cl::opt<bool> RunPartialInlining; 308 extern cl::opt<bool> ExtraVectorizerPasses; 309 310 extern cl::opt<bool> FlattenedProfileUsed; 311 312 extern cl::opt<AttributorRunOption> AttributorRun; 313 extern cl::opt<bool> EnableKnowledgeRetention; 314 315 extern cl::opt<bool> EnableMatrix; 316 317 extern cl::opt<bool> DisablePreInliner; 318 extern cl::opt<int> PreInlineThreshold; 319 } // namespace llvm 320 321 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O0 = { 322 /*SpeedLevel*/ 0, 323 /*SizeLevel*/ 0}; 324 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O1 = { 325 /*SpeedLevel*/ 1, 326 /*SizeLevel*/ 0}; 327 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O2 = { 328 /*SpeedLevel*/ 2, 329 /*SizeLevel*/ 0}; 330 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O3 = { 331 /*SpeedLevel*/ 3, 332 /*SizeLevel*/ 0}; 333 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::Os = { 334 /*SpeedLevel*/ 2, 335 /*SizeLevel*/ 1}; 336 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::Oz = { 337 /*SpeedLevel*/ 2, 338 /*SizeLevel*/ 2}; 339 340 namespace { 341 342 // The following passes/analyses have custom names, otherwise their name will 343 // include `(anonymous namespace)`. These are special since they are only for 344 // testing purposes and don't live in a header file. 345 346 /// No-op module pass which does nothing. 347 struct NoOpModulePass : PassInfoMixin<NoOpModulePass> { 348 PreservedAnalyses run(Module &M, ModuleAnalysisManager &) { 349 return PreservedAnalyses::all(); 350 } 351 352 static StringRef name() { return "NoOpModulePass"; } 353 }; 354 355 /// No-op module analysis. 356 class NoOpModuleAnalysis : public AnalysisInfoMixin<NoOpModuleAnalysis> { 357 friend AnalysisInfoMixin<NoOpModuleAnalysis>; 358 static AnalysisKey Key; 359 360 public: 361 struct Result {}; 362 Result run(Module &, ModuleAnalysisManager &) { return Result(); } 363 static StringRef name() { return "NoOpModuleAnalysis"; } 364 }; 365 366 /// No-op CGSCC pass which does nothing. 367 struct NoOpCGSCCPass : PassInfoMixin<NoOpCGSCCPass> { 368 PreservedAnalyses run(LazyCallGraph::SCC &C, CGSCCAnalysisManager &, 369 LazyCallGraph &, CGSCCUpdateResult &UR) { 370 return PreservedAnalyses::all(); 371 } 372 static StringRef name() { return "NoOpCGSCCPass"; } 373 }; 374 375 /// No-op CGSCC analysis. 376 class NoOpCGSCCAnalysis : public AnalysisInfoMixin<NoOpCGSCCAnalysis> { 377 friend AnalysisInfoMixin<NoOpCGSCCAnalysis>; 378 static AnalysisKey Key; 379 380 public: 381 struct Result {}; 382 Result run(LazyCallGraph::SCC &, CGSCCAnalysisManager &, LazyCallGraph &G) { 383 return Result(); 384 } 385 static StringRef name() { return "NoOpCGSCCAnalysis"; } 386 }; 387 388 /// No-op function pass which does nothing. 389 struct NoOpFunctionPass : PassInfoMixin<NoOpFunctionPass> { 390 PreservedAnalyses run(Function &F, FunctionAnalysisManager &) { 391 return PreservedAnalyses::all(); 392 } 393 static StringRef name() { return "NoOpFunctionPass"; } 394 }; 395 396 /// No-op function analysis. 397 class NoOpFunctionAnalysis : public AnalysisInfoMixin<NoOpFunctionAnalysis> { 398 friend AnalysisInfoMixin<NoOpFunctionAnalysis>; 399 static AnalysisKey Key; 400 401 public: 402 struct Result {}; 403 Result run(Function &, FunctionAnalysisManager &) { return Result(); } 404 static StringRef name() { return "NoOpFunctionAnalysis"; } 405 }; 406 407 /// No-op loop pass which does nothing. 408 struct NoOpLoopPass : PassInfoMixin<NoOpLoopPass> { 409 PreservedAnalyses run(Loop &L, LoopAnalysisManager &, 410 LoopStandardAnalysisResults &, LPMUpdater &) { 411 return PreservedAnalyses::all(); 412 } 413 static StringRef name() { return "NoOpLoopPass"; } 414 }; 415 416 /// No-op loop analysis. 417 class NoOpLoopAnalysis : public AnalysisInfoMixin<NoOpLoopAnalysis> { 418 friend AnalysisInfoMixin<NoOpLoopAnalysis>; 419 static AnalysisKey Key; 420 421 public: 422 struct Result {}; 423 Result run(Loop &, LoopAnalysisManager &, LoopStandardAnalysisResults &) { 424 return Result(); 425 } 426 static StringRef name() { return "NoOpLoopAnalysis"; } 427 }; 428 429 AnalysisKey NoOpModuleAnalysis::Key; 430 AnalysisKey NoOpCGSCCAnalysis::Key; 431 AnalysisKey NoOpFunctionAnalysis::Key; 432 AnalysisKey NoOpLoopAnalysis::Key; 433 434 /// Whether or not we should populate a PassInstrumentationCallbacks's class to 435 /// pass name map. 436 /// 437 /// This is for optimization purposes so we don't populate it if we never use 438 /// it. This should be updated if new pass instrumentation wants to use the map. 439 /// We currently only use this for --print-before/after. 440 bool shouldPopulateClassToPassNames() { 441 return !printBeforePasses().empty() || !printAfterPasses().empty(); 442 } 443 444 } // namespace 445 446 PassBuilder::PassBuilder(TargetMachine *TM, PipelineTuningOptions PTO, 447 Optional<PGOOptions> PGOOpt, 448 PassInstrumentationCallbacks *PIC) 449 : TM(TM), PTO(PTO), PGOOpt(PGOOpt), PIC(PIC) { 450 if (TM) 451 TM->registerPassBuilderCallbacks(*this); 452 if (PIC && shouldPopulateClassToPassNames()) { 453 #define MODULE_PASS(NAME, CREATE_PASS) \ 454 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 455 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 456 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 457 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 458 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 459 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 460 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 461 #define LOOP_PASS(NAME, CREATE_PASS) \ 462 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 463 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 464 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 465 #define CGSCC_PASS(NAME, CREATE_PASS) \ 466 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 467 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 468 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 469 #include "PassRegistry.def" 470 } 471 } 472 473 void PassBuilder::invokePeepholeEPCallbacks( 474 FunctionPassManager &FPM, PassBuilder::OptimizationLevel Level) { 475 for (auto &C : PeepholeEPCallbacks) 476 C(FPM, Level); 477 } 478 479 void PassBuilder::registerModuleAnalyses(ModuleAnalysisManager &MAM) { 480 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 481 MAM.registerPass([&] { return CREATE_PASS; }); 482 #include "PassRegistry.def" 483 484 for (auto &C : ModuleAnalysisRegistrationCallbacks) 485 C(MAM); 486 } 487 488 void PassBuilder::registerCGSCCAnalyses(CGSCCAnalysisManager &CGAM) { 489 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 490 CGAM.registerPass([&] { return CREATE_PASS; }); 491 #include "PassRegistry.def" 492 493 for (auto &C : CGSCCAnalysisRegistrationCallbacks) 494 C(CGAM); 495 } 496 497 void PassBuilder::registerFunctionAnalyses(FunctionAnalysisManager &FAM) { 498 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 499 FAM.registerPass([&] { return CREATE_PASS; }); 500 #include "PassRegistry.def" 501 502 for (auto &C : FunctionAnalysisRegistrationCallbacks) 503 C(FAM); 504 } 505 506 void PassBuilder::registerLoopAnalyses(LoopAnalysisManager &LAM) { 507 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 508 LAM.registerPass([&] { return CREATE_PASS; }); 509 #include "PassRegistry.def" 510 511 for (auto &C : LoopAnalysisRegistrationCallbacks) 512 C(LAM); 513 } 514 515 // Helper to add AnnotationRemarksPass. 516 static void addAnnotationRemarksPass(ModulePassManager &MPM) { 517 FunctionPassManager FPM; 518 FPM.addPass(AnnotationRemarksPass()); 519 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 520 } 521 522 // Helper to check if the current compilation phase is preparing for LTO 523 static bool isLTOPreLink(ThinOrFullLTOPhase Phase) { 524 return Phase == ThinOrFullLTOPhase::ThinLTOPreLink || 525 Phase == ThinOrFullLTOPhase::FullLTOPreLink; 526 } 527 528 // TODO: Investigate the cost/benefit of tail call elimination on debugging. 529 FunctionPassManager 530 PassBuilder::buildO1FunctionSimplificationPipeline(OptimizationLevel Level, 531 ThinOrFullLTOPhase Phase) { 532 533 FunctionPassManager FPM; 534 535 // Form SSA out of local memory accesses after breaking apart aggregates into 536 // scalars. 537 FPM.addPass(SROA()); 538 539 // Catch trivial redundancies 540 FPM.addPass(EarlyCSEPass(true /* Enable mem-ssa. */)); 541 542 // Hoisting of scalars and load expressions. 543 FPM.addPass(SimplifyCFGPass()); 544 FPM.addPass(InstCombinePass()); 545 546 FPM.addPass(LibCallsShrinkWrapPass()); 547 548 invokePeepholeEPCallbacks(FPM, Level); 549 550 FPM.addPass(SimplifyCFGPass()); 551 552 // Form canonically associated expression trees, and simplify the trees using 553 // basic mathematical properties. For example, this will form (nearly) 554 // minimal multiplication trees. 555 FPM.addPass(ReassociatePass()); 556 557 // Add the primary loop simplification pipeline. 558 // FIXME: Currently this is split into two loop pass pipelines because we run 559 // some function passes in between them. These can and should be removed 560 // and/or replaced by scheduling the loop pass equivalents in the correct 561 // positions. But those equivalent passes aren't powerful enough yet. 562 // Specifically, `SimplifyCFGPass` and `InstCombinePass` are currently still 563 // used. We have `LoopSimplifyCFGPass` which isn't yet powerful enough yet to 564 // fully replace `SimplifyCFGPass`, and the closest to the other we have is 565 // `LoopInstSimplify`. 566 LoopPassManager LPM1, LPM2; 567 568 // Simplify the loop body. We do this initially to clean up after other loop 569 // passes run, either when iterating on a loop or on inner loops with 570 // implications on the outer loop. 571 LPM1.addPass(LoopInstSimplifyPass()); 572 LPM1.addPass(LoopSimplifyCFGPass()); 573 574 // Try to remove as much code from the loop header as possible, 575 // to reduce amount of IR that will have to be duplicated. 576 // TODO: Investigate promotion cap for O1. 577 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 578 579 LPM1.addPass(LoopRotatePass(/* Disable header duplication */ true, 580 isLTOPreLink(Phase))); 581 // TODO: Investigate promotion cap for O1. 582 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 583 LPM1.addPass(SimpleLoopUnswitchPass()); 584 585 LPM2.addPass(LoopIdiomRecognizePass()); 586 LPM2.addPass(IndVarSimplifyPass()); 587 588 for (auto &C : LateLoopOptimizationsEPCallbacks) 589 C(LPM2, Level); 590 591 LPM2.addPass(LoopDeletionPass()); 592 593 if (EnableLoopInterchange) 594 LPM2.addPass(LoopInterchangePass()); 595 596 // Do not enable unrolling in PreLinkThinLTO phase during sample PGO 597 // because it changes IR to makes profile annotation in back compile 598 // inaccurate. The normal unroller doesn't pay attention to forced full unroll 599 // attributes so we need to make sure and allow the full unroll pass to pay 600 // attention to it. 601 if (Phase != ThinOrFullLTOPhase::ThinLTOPreLink || !PGOOpt || 602 PGOOpt->Action != PGOOptions::SampleUse) 603 LPM2.addPass(LoopFullUnrollPass(Level.getSpeedupLevel(), 604 /* OnlyWhenForced= */ !PTO.LoopUnrolling, 605 PTO.ForgetAllSCEVInLoopUnroll)); 606 607 for (auto &C : LoopOptimizerEndEPCallbacks) 608 C(LPM2, Level); 609 610 // We provide the opt remark emitter pass for LICM to use. We only need to do 611 // this once as it is immutable. 612 FPM.addPass( 613 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 614 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM1), 615 EnableMSSALoopDependency, 616 /*UseBlockFrequencyInfo=*/true)); 617 FPM.addPass(SimplifyCFGPass()); 618 FPM.addPass(InstCombinePass()); 619 if (EnableLoopFlatten) 620 FPM.addPass(createFunctionToLoopPassAdaptor(LoopFlattenPass())); 621 // The loop passes in LPM2 (LoopFullUnrollPass) do not preserve MemorySSA. 622 // *All* loop passes must preserve it, in order to be able to use it. 623 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM2), 624 /*UseMemorySSA=*/false, 625 /*UseBlockFrequencyInfo=*/false)); 626 627 // Delete small array after loop unroll. 628 FPM.addPass(SROA()); 629 630 // Specially optimize memory movement as it doesn't look like dataflow in SSA. 631 FPM.addPass(MemCpyOptPass()); 632 633 // Sparse conditional constant propagation. 634 // FIXME: It isn't clear why we do this *after* loop passes rather than 635 // before... 636 FPM.addPass(SCCPPass()); 637 638 // Delete dead bit computations (instcombine runs after to fold away the dead 639 // computations, and then ADCE will run later to exploit any new DCE 640 // opportunities that creates). 641 FPM.addPass(BDCEPass()); 642 643 // Run instcombine after redundancy and dead bit elimination to exploit 644 // opportunities opened up by them. 645 FPM.addPass(InstCombinePass()); 646 invokePeepholeEPCallbacks(FPM, Level); 647 648 if (PTO.Coroutines) 649 FPM.addPass(CoroElidePass()); 650 651 for (auto &C : ScalarOptimizerLateEPCallbacks) 652 C(FPM, Level); 653 654 // Finally, do an expensive DCE pass to catch all the dead code exposed by 655 // the simplifications and basic cleanup after all the simplifications. 656 // TODO: Investigate if this is too expensive. 657 FPM.addPass(ADCEPass()); 658 FPM.addPass(SimplifyCFGPass()); 659 FPM.addPass(InstCombinePass()); 660 invokePeepholeEPCallbacks(FPM, Level); 661 662 return FPM; 663 } 664 665 FunctionPassManager 666 PassBuilder::buildFunctionSimplificationPipeline(OptimizationLevel Level, 667 ThinOrFullLTOPhase Phase) { 668 assert(Level != OptimizationLevel::O0 && "Must request optimizations!"); 669 670 // The O1 pipeline has a separate pipeline creation function to simplify 671 // construction readability. 672 if (Level.getSpeedupLevel() == 1) 673 return buildO1FunctionSimplificationPipeline(Level, Phase); 674 675 FunctionPassManager FPM; 676 677 // Form SSA out of local memory accesses after breaking apart aggregates into 678 // scalars. 679 FPM.addPass(SROA()); 680 681 // Catch trivial redundancies 682 FPM.addPass(EarlyCSEPass(true /* Enable mem-ssa. */)); 683 if (EnableKnowledgeRetention) 684 FPM.addPass(AssumeSimplifyPass()); 685 686 // Hoisting of scalars and load expressions. 687 if (EnableGVNHoist) 688 FPM.addPass(GVNHoistPass()); 689 690 // Global value numbering based sinking. 691 if (EnableGVNSink) { 692 FPM.addPass(GVNSinkPass()); 693 FPM.addPass(SimplifyCFGPass()); 694 } 695 696 if (EnableConstraintElimination) 697 FPM.addPass(ConstraintEliminationPass()); 698 699 // Speculative execution if the target has divergent branches; otherwise nop. 700 FPM.addPass(SpeculativeExecutionPass(/* OnlyIfDivergentTarget =*/true)); 701 702 // Optimize based on known information about branches, and cleanup afterward. 703 FPM.addPass(JumpThreadingPass()); 704 FPM.addPass(CorrelatedValuePropagationPass()); 705 706 FPM.addPass(SimplifyCFGPass()); 707 if (Level == OptimizationLevel::O3) 708 FPM.addPass(AggressiveInstCombinePass()); 709 FPM.addPass(InstCombinePass()); 710 711 if (!Level.isOptimizingForSize()) 712 FPM.addPass(LibCallsShrinkWrapPass()); 713 714 invokePeepholeEPCallbacks(FPM, Level); 715 716 // For PGO use pipeline, try to optimize memory intrinsics such as memcpy 717 // using the size value profile. Don't perform this when optimizing for size. 718 if (PGOOpt && PGOOpt->Action == PGOOptions::IRUse && 719 !Level.isOptimizingForSize()) 720 FPM.addPass(PGOMemOPSizeOpt()); 721 722 FPM.addPass(TailCallElimPass()); 723 FPM.addPass(SimplifyCFGPass()); 724 725 // Form canonically associated expression trees, and simplify the trees using 726 // basic mathematical properties. For example, this will form (nearly) 727 // minimal multiplication trees. 728 FPM.addPass(ReassociatePass()); 729 730 // Add the primary loop simplification pipeline. 731 // FIXME: Currently this is split into two loop pass pipelines because we run 732 // some function passes in between them. These can and should be removed 733 // and/or replaced by scheduling the loop pass equivalents in the correct 734 // positions. But those equivalent passes aren't powerful enough yet. 735 // Specifically, `SimplifyCFGPass` and `InstCombinePass` are currently still 736 // used. We have `LoopSimplifyCFGPass` which isn't yet powerful enough yet to 737 // fully replace `SimplifyCFGPass`, and the closest to the other we have is 738 // `LoopInstSimplify`. 739 LoopPassManager LPM1, LPM2; 740 741 // Simplify the loop body. We do this initially to clean up after other loop 742 // passes run, either when iterating on a loop or on inner loops with 743 // implications on the outer loop. 744 LPM1.addPass(LoopInstSimplifyPass()); 745 LPM1.addPass(LoopSimplifyCFGPass()); 746 747 // Try to remove as much code from the loop header as possible, 748 // to reduce amount of IR that will have to be duplicated. 749 // TODO: Investigate promotion cap for O1. 750 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 751 752 // Disable header duplication in loop rotation at -Oz. 753 LPM1.addPass( 754 LoopRotatePass(Level != OptimizationLevel::Oz, isLTOPreLink(Phase))); 755 // TODO: Investigate promotion cap for O1. 756 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 757 LPM1.addPass( 758 SimpleLoopUnswitchPass(/* NonTrivial */ Level == OptimizationLevel::O3 && 759 EnableO3NonTrivialUnswitching)); 760 LPM2.addPass(LoopIdiomRecognizePass()); 761 LPM2.addPass(IndVarSimplifyPass()); 762 763 for (auto &C : LateLoopOptimizationsEPCallbacks) 764 C(LPM2, Level); 765 766 LPM2.addPass(LoopDeletionPass()); 767 768 if (EnableLoopInterchange) 769 LPM2.addPass(LoopInterchangePass()); 770 771 // Do not enable unrolling in PreLinkThinLTO phase during sample PGO 772 // because it changes IR to makes profile annotation in back compile 773 // inaccurate. The normal unroller doesn't pay attention to forced full unroll 774 // attributes so we need to make sure and allow the full unroll pass to pay 775 // attention to it. 776 if (Phase != ThinOrFullLTOPhase::ThinLTOPreLink || !PGOOpt || 777 PGOOpt->Action != PGOOptions::SampleUse) 778 LPM2.addPass(LoopFullUnrollPass(Level.getSpeedupLevel(), 779 /* OnlyWhenForced= */ !PTO.LoopUnrolling, 780 PTO.ForgetAllSCEVInLoopUnroll)); 781 782 for (auto &C : LoopOptimizerEndEPCallbacks) 783 C(LPM2, Level); 784 785 // We provide the opt remark emitter pass for LICM to use. We only need to do 786 // this once as it is immutable. 787 FPM.addPass( 788 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 789 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM1), 790 EnableMSSALoopDependency, 791 /*UseBlockFrequencyInfo=*/true)); 792 FPM.addPass(SimplifyCFGPass()); 793 FPM.addPass(InstCombinePass()); 794 if (EnableLoopFlatten) 795 FPM.addPass(createFunctionToLoopPassAdaptor(LoopFlattenPass())); 796 // The loop passes in LPM2 (LoopIdiomRecognizePass, IndVarSimplifyPass, 797 // LoopDeletionPass and LoopFullUnrollPass) do not preserve MemorySSA. 798 // *All* loop passes must preserve it, in order to be able to use it. 799 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM2), 800 /*UseMemorySSA=*/false, 801 /*UseBlockFrequencyInfo=*/false)); 802 803 // Delete small array after loop unroll. 804 FPM.addPass(SROA()); 805 806 // Eliminate redundancies. 807 FPM.addPass(MergedLoadStoreMotionPass()); 808 if (RunNewGVN) 809 FPM.addPass(NewGVNPass()); 810 else 811 FPM.addPass(GVN()); 812 813 // Sparse conditional constant propagation. 814 // FIXME: It isn't clear why we do this *after* loop passes rather than 815 // before... 816 FPM.addPass(SCCPPass()); 817 818 // Delete dead bit computations (instcombine runs after to fold away the dead 819 // computations, and then ADCE will run later to exploit any new DCE 820 // opportunities that creates). 821 FPM.addPass(BDCEPass()); 822 823 // Run instcombine after redundancy and dead bit elimination to exploit 824 // opportunities opened up by them. 825 FPM.addPass(InstCombinePass()); 826 invokePeepholeEPCallbacks(FPM, Level); 827 828 // Re-consider control flow based optimizations after redundancy elimination, 829 // redo DCE, etc. 830 FPM.addPass(JumpThreadingPass()); 831 FPM.addPass(CorrelatedValuePropagationPass()); 832 833 // Finally, do an expensive DCE pass to catch all the dead code exposed by 834 // the simplifications and basic cleanup after all the simplifications. 835 // TODO: Investigate if this is too expensive. 836 FPM.addPass(ADCEPass()); 837 838 // Specially optimize memory movement as it doesn't look like dataflow in SSA. 839 FPM.addPass(MemCpyOptPass()); 840 841 FPM.addPass(DSEPass()); 842 FPM.addPass(createFunctionToLoopPassAdaptor( 843 LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap), 844 EnableMSSALoopDependency, /*UseBlockFrequencyInfo=*/true)); 845 846 if (PTO.Coroutines) 847 FPM.addPass(CoroElidePass()); 848 849 for (auto &C : ScalarOptimizerLateEPCallbacks) 850 C(FPM, Level); 851 852 FPM.addPass(SimplifyCFGPass( 853 SimplifyCFGOptions().hoistCommonInsts(true).sinkCommonInsts(true))); 854 FPM.addPass(InstCombinePass()); 855 invokePeepholeEPCallbacks(FPM, Level); 856 857 if (EnableCHR && Level == OptimizationLevel::O3 && PGOOpt && 858 (PGOOpt->Action == PGOOptions::IRUse || 859 PGOOpt->Action == PGOOptions::SampleUse)) 860 FPM.addPass(ControlHeightReductionPass()); 861 862 return FPM; 863 } 864 865 void PassBuilder::addRequiredLTOPreLinkPasses(ModulePassManager &MPM) { 866 MPM.addPass(CanonicalizeAliasesPass()); 867 MPM.addPass(NameAnonGlobalPass()); 868 } 869 870 void PassBuilder::addPGOInstrPasses(ModulePassManager &MPM, 871 PassBuilder::OptimizationLevel Level, 872 bool RunProfileGen, bool IsCS, 873 std::string ProfileFile, 874 std::string ProfileRemappingFile) { 875 assert(Level != OptimizationLevel::O0 && "Not expecting O0 here!"); 876 if (!IsCS && !DisablePreInliner) { 877 InlineParams IP; 878 879 IP.DefaultThreshold = PreInlineThreshold; 880 881 // FIXME: The hint threshold has the same value used by the regular inliner 882 // when not optimzing for size. This should probably be lowered after 883 // performance testing. 884 // FIXME: this comment is cargo culted from the old pass manager, revisit). 885 IP.HintThreshold = Level.isOptimizingForSize() ? PreInlineThreshold : 325; 886 ModuleInlinerWrapperPass MIWP(IP); 887 CGSCCPassManager &CGPipeline = MIWP.getPM(); 888 889 FunctionPassManager FPM; 890 FPM.addPass(SROA()); 891 FPM.addPass(EarlyCSEPass()); // Catch trivial redundancies. 892 FPM.addPass(SimplifyCFGPass()); // Merge & remove basic blocks. 893 FPM.addPass(InstCombinePass()); // Combine silly sequences. 894 invokePeepholeEPCallbacks(FPM, Level); 895 896 CGPipeline.addPass(createCGSCCToFunctionPassAdaptor(std::move(FPM))); 897 898 MPM.addPass(std::move(MIWP)); 899 900 // Delete anything that is now dead to make sure that we don't instrument 901 // dead code. Instrumentation can end up keeping dead code around and 902 // dramatically increase code size. 903 MPM.addPass(GlobalDCEPass()); 904 } 905 906 if (!RunProfileGen) { 907 assert(!ProfileFile.empty() && "Profile use expecting a profile file!"); 908 MPM.addPass(PGOInstrumentationUse(ProfileFile, ProfileRemappingFile, IsCS)); 909 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 910 // RequireAnalysisPass for PSI before subsequent non-module passes. 911 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 912 return; 913 } 914 915 // Perform PGO instrumentation. 916 MPM.addPass(PGOInstrumentationGen(IsCS)); 917 918 FunctionPassManager FPM; 919 // Disable header duplication in loop rotation at -Oz. 920 FPM.addPass(createFunctionToLoopPassAdaptor( 921 LoopRotatePass(Level != OptimizationLevel::Oz), EnableMSSALoopDependency, 922 /*UseBlockFrequencyInfo=*/false)); 923 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 924 925 // Add the profile lowering pass. 926 InstrProfOptions Options; 927 if (!ProfileFile.empty()) 928 Options.InstrProfileOutput = ProfileFile; 929 // Do counter promotion at Level greater than O0. 930 Options.DoCounterPromotion = true; 931 Options.UseBFIInPromotion = IsCS; 932 MPM.addPass(InstrProfiling(Options, IsCS)); 933 } 934 935 void PassBuilder::addPGOInstrPassesForO0(ModulePassManager &MPM, 936 bool RunProfileGen, bool IsCS, 937 std::string ProfileFile, 938 std::string ProfileRemappingFile) { 939 if (!RunProfileGen) { 940 assert(!ProfileFile.empty() && "Profile use expecting a profile file!"); 941 MPM.addPass(PGOInstrumentationUse(ProfileFile, ProfileRemappingFile, IsCS)); 942 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 943 // RequireAnalysisPass for PSI before subsequent non-module passes. 944 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 945 return; 946 } 947 948 // Perform PGO instrumentation. 949 MPM.addPass(PGOInstrumentationGen(IsCS)); 950 // Add the profile lowering pass. 951 InstrProfOptions Options; 952 if (!ProfileFile.empty()) 953 Options.InstrProfileOutput = ProfileFile; 954 // Do not do counter promotion at O0. 955 Options.DoCounterPromotion = false; 956 Options.UseBFIInPromotion = IsCS; 957 MPM.addPass(InstrProfiling(Options, IsCS)); 958 } 959 960 static InlineParams 961 getInlineParamsFromOptLevel(PassBuilder::OptimizationLevel Level) { 962 return getInlineParams(Level.getSpeedupLevel(), Level.getSizeLevel()); 963 } 964 965 ModuleInlinerWrapperPass 966 PassBuilder::buildInlinerPipeline(OptimizationLevel Level, 967 ThinOrFullLTOPhase Phase) { 968 InlineParams IP = getInlineParamsFromOptLevel(Level); 969 if (Phase == ThinOrFullLTOPhase::ThinLTOPreLink && PGOOpt && 970 PGOOpt->Action == PGOOptions::SampleUse) 971 IP.HotCallSiteThreshold = 0; 972 973 if (PGOOpt) 974 IP.EnableDeferral = EnablePGOInlineDeferral; 975 976 ModuleInlinerWrapperPass MIWP(IP, PerformMandatoryInliningsFirst, 977 UseInlineAdvisor, MaxDevirtIterations); 978 979 // Require the GlobalsAA analysis for the module so we can query it within 980 // the CGSCC pipeline. 981 MIWP.addModulePass(RequireAnalysisPass<GlobalsAA, Module>()); 982 // Invalidate AAManager so it can be recreated and pick up the newly available 983 // GlobalsAA. 984 MIWP.addModulePass( 985 createModuleToFunctionPassAdaptor(InvalidateAnalysisPass<AAManager>())); 986 987 // Require the ProfileSummaryAnalysis for the module so we can query it within 988 // the inliner pass. 989 MIWP.addModulePass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 990 991 // Now begin the main postorder CGSCC pipeline. 992 // FIXME: The current CGSCC pipeline has its origins in the legacy pass 993 // manager and trying to emulate its precise behavior. Much of this doesn't 994 // make a lot of sense and we should revisit the core CGSCC structure. 995 CGSCCPassManager &MainCGPipeline = MIWP.getPM(); 996 997 // Note: historically, the PruneEH pass was run first to deduce nounwind and 998 // generally clean up exception handling overhead. It isn't clear this is 999 // valuable as the inliner doesn't currently care whether it is inlining an 1000 // invoke or a call. 1001 1002 if (AttributorRun & AttributorRunOption::CGSCC) 1003 MainCGPipeline.addPass(AttributorCGSCCPass()); 1004 1005 if (PTO.Coroutines) 1006 MainCGPipeline.addPass(CoroSplitPass(Level != OptimizationLevel::O0)); 1007 1008 // Now deduce any function attributes based in the current code. 1009 MainCGPipeline.addPass(PostOrderFunctionAttrsPass()); 1010 1011 // When at O3 add argument promotion to the pass pipeline. 1012 // FIXME: It isn't at all clear why this should be limited to O3. 1013 if (Level == OptimizationLevel::O3) 1014 MainCGPipeline.addPass(ArgumentPromotionPass()); 1015 1016 // Try to perform OpenMP specific optimizations. This is a (quick!) no-op if 1017 // there are no OpenMP runtime calls present in the module. 1018 if (Level == OptimizationLevel::O2 || Level == OptimizationLevel::O3) 1019 MainCGPipeline.addPass(OpenMPOptCGSCCPass()); 1020 1021 for (auto &C : CGSCCOptimizerLateEPCallbacks) 1022 C(MainCGPipeline, Level); 1023 1024 // Lastly, add the core function simplification pipeline nested inside the 1025 // CGSCC walk. 1026 MainCGPipeline.addPass(createCGSCCToFunctionPassAdaptor( 1027 buildFunctionSimplificationPipeline(Level, Phase))); 1028 1029 return MIWP; 1030 } 1031 1032 ModulePassManager 1033 PassBuilder::buildModuleSimplificationPipeline(OptimizationLevel Level, 1034 ThinOrFullLTOPhase Phase) { 1035 ModulePassManager MPM; 1036 1037 // Place pseudo probe instrumentation as the first pass of the pipeline to 1038 // minimize the impact of optimization changes. 1039 if (PGOOpt && PGOOpt->PseudoProbeForProfiling && 1040 Phase != ThinOrFullLTOPhase::ThinLTOPostLink) 1041 MPM.addPass(SampleProfileProbePass(TM)); 1042 1043 bool HasSampleProfile = PGOOpt && (PGOOpt->Action == PGOOptions::SampleUse); 1044 1045 // In ThinLTO mode, when flattened profile is used, all the available 1046 // profile information will be annotated in PreLink phase so there is 1047 // no need to load the profile again in PostLink. 1048 bool LoadSampleProfile = 1049 HasSampleProfile && 1050 !(FlattenedProfileUsed && Phase == ThinOrFullLTOPhase::ThinLTOPostLink); 1051 1052 // During the ThinLTO backend phase we perform early indirect call promotion 1053 // here, before globalopt. Otherwise imported available_externally functions 1054 // look unreferenced and are removed. If we are going to load the sample 1055 // profile then defer until later. 1056 // TODO: See if we can move later and consolidate with the location where 1057 // we perform ICP when we are loading a sample profile. 1058 // TODO: We pass HasSampleProfile (whether there was a sample profile file 1059 // passed to the compile) to the SamplePGO flag of ICP. This is used to 1060 // determine whether the new direct calls are annotated with prof metadata. 1061 // Ideally this should be determined from whether the IR is annotated with 1062 // sample profile, and not whether the a sample profile was provided on the 1063 // command line. E.g. for flattened profiles where we will not be reloading 1064 // the sample profile in the ThinLTO backend, we ideally shouldn't have to 1065 // provide the sample profile file. 1066 if (Phase == ThinOrFullLTOPhase::ThinLTOPostLink && !LoadSampleProfile) 1067 MPM.addPass(PGOIndirectCallPromotion(true /* InLTO */, HasSampleProfile)); 1068 1069 // Do basic inference of function attributes from known properties of system 1070 // libraries and other oracles. 1071 MPM.addPass(InferFunctionAttrsPass()); 1072 1073 // Create an early function pass manager to cleanup the output of the 1074 // frontend. 1075 FunctionPassManager EarlyFPM; 1076 // Lower llvm.expect to metadata before attempting transforms. 1077 // Compare/branch metadata may alter the behavior of passes like SimplifyCFG. 1078 EarlyFPM.addPass(LowerExpectIntrinsicPass()); 1079 EarlyFPM.addPass(SimplifyCFGPass()); 1080 EarlyFPM.addPass(SROA()); 1081 EarlyFPM.addPass(EarlyCSEPass()); 1082 if (PTO.Coroutines) 1083 EarlyFPM.addPass(CoroEarlyPass()); 1084 if (Level == OptimizationLevel::O3) 1085 EarlyFPM.addPass(CallSiteSplittingPass()); 1086 1087 // In SamplePGO ThinLTO backend, we need instcombine before profile annotation 1088 // to convert bitcast to direct calls so that they can be inlined during the 1089 // profile annotation prepration step. 1090 // More details about SamplePGO design can be found in: 1091 // https://research.google.com/pubs/pub45290.html 1092 // FIXME: revisit how SampleProfileLoad/Inliner/ICP is structured. 1093 if (LoadSampleProfile) 1094 EarlyFPM.addPass(InstCombinePass()); 1095 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(EarlyFPM))); 1096 1097 if (LoadSampleProfile) { 1098 // Annotate sample profile right after early FPM to ensure freshness of 1099 // the debug info. 1100 MPM.addPass(SampleProfileLoaderPass(PGOOpt->ProfileFile, 1101 PGOOpt->ProfileRemappingFile, Phase)); 1102 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 1103 // RequireAnalysisPass for PSI before subsequent non-module passes. 1104 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 1105 // Do not invoke ICP in the LTOPrelink phase as it makes it hard 1106 // for the profile annotation to be accurate in the LTO backend. 1107 if (Phase != ThinOrFullLTOPhase::ThinLTOPreLink && 1108 Phase != ThinOrFullLTOPhase::FullLTOPreLink) 1109 // We perform early indirect call promotion here, before globalopt. 1110 // This is important for the ThinLTO backend phase because otherwise 1111 // imported available_externally functions look unreferenced and are 1112 // removed. 1113 MPM.addPass( 1114 PGOIndirectCallPromotion(true /* IsInLTO */, true /* SamplePGO */)); 1115 } 1116 1117 // Try to perform OpenMP specific optimizations on the module. This is a 1118 // (quick!) no-op if there are no OpenMP runtime calls present in the module. 1119 if (Level == OptimizationLevel::O2 || Level == OptimizationLevel::O3) 1120 MPM.addPass(OpenMPOptPass()); 1121 1122 if (AttributorRun & AttributorRunOption::MODULE) 1123 MPM.addPass(AttributorPass()); 1124 1125 // Lower type metadata and the type.test intrinsic in the ThinLTO 1126 // post link pipeline after ICP. This is to enable usage of the type 1127 // tests in ICP sequences. 1128 if (Phase == ThinOrFullLTOPhase::ThinLTOPostLink) 1129 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1130 1131 for (auto &C : PipelineEarlySimplificationEPCallbacks) 1132 C(MPM, Level); 1133 1134 // Specialize functions with IPSCCP. 1135 if (EnableFunctionSpecialization) 1136 MPM.addPass(FunctionSpecializationPass()); 1137 1138 // Interprocedural constant propagation now that basic cleanup has occurred 1139 // and prior to optimizing globals. 1140 // FIXME: This position in the pipeline hasn't been carefully considered in 1141 // years, it should be re-analyzed. 1142 MPM.addPass(IPSCCPPass()); 1143 1144 // Attach metadata to indirect call sites indicating the set of functions 1145 // they may target at run-time. This should follow IPSCCP. 1146 MPM.addPass(CalledValuePropagationPass()); 1147 1148 // Optimize globals to try and fold them into constants. 1149 MPM.addPass(GlobalOptPass()); 1150 1151 // Promote any localized globals to SSA registers. 1152 // FIXME: Should this instead by a run of SROA? 1153 // FIXME: We should probably run instcombine and simplify-cfg afterward to 1154 // delete control flows that are dead once globals have been folded to 1155 // constants. 1156 MPM.addPass(createModuleToFunctionPassAdaptor(PromotePass())); 1157 1158 // Remove any dead arguments exposed by cleanups and constant folding 1159 // globals. 1160 MPM.addPass(DeadArgumentEliminationPass()); 1161 1162 // Create a small function pass pipeline to cleanup after all the global 1163 // optimizations. 1164 FunctionPassManager GlobalCleanupPM; 1165 GlobalCleanupPM.addPass(InstCombinePass()); 1166 invokePeepholeEPCallbacks(GlobalCleanupPM, Level); 1167 1168 GlobalCleanupPM.addPass(SimplifyCFGPass()); 1169 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(GlobalCleanupPM))); 1170 1171 // Add all the requested passes for instrumentation PGO, if requested. 1172 if (PGOOpt && Phase != ThinOrFullLTOPhase::ThinLTOPostLink && 1173 (PGOOpt->Action == PGOOptions::IRInstr || 1174 PGOOpt->Action == PGOOptions::IRUse)) { 1175 addPGOInstrPasses(MPM, Level, 1176 /* RunProfileGen */ PGOOpt->Action == PGOOptions::IRInstr, 1177 /* IsCS */ false, PGOOpt->ProfileFile, 1178 PGOOpt->ProfileRemappingFile); 1179 MPM.addPass(PGOIndirectCallPromotion(false, false)); 1180 } 1181 if (PGOOpt && Phase != ThinOrFullLTOPhase::ThinLTOPostLink && 1182 PGOOpt->CSAction == PGOOptions::CSIRInstr) 1183 MPM.addPass(PGOInstrumentationGenCreateVar(PGOOpt->CSProfileGenFile)); 1184 1185 // Synthesize function entry counts for non-PGO compilation. 1186 if (EnableSyntheticCounts && !PGOOpt) 1187 MPM.addPass(SyntheticCountsPropagation()); 1188 1189 MPM.addPass(buildInlinerPipeline(Level, Phase)); 1190 1191 if (EnableMemProfiler && Phase != ThinOrFullLTOPhase::ThinLTOPreLink) { 1192 MPM.addPass(createModuleToFunctionPassAdaptor(MemProfilerPass())); 1193 MPM.addPass(ModuleMemProfilerPass()); 1194 } 1195 1196 return MPM; 1197 } 1198 1199 /// TODO: Should LTO cause any differences to this set of passes? 1200 void PassBuilder::addVectorPasses(OptimizationLevel Level, 1201 FunctionPassManager &FPM, bool IsLTO) { 1202 FPM.addPass(LoopVectorizePass( 1203 LoopVectorizeOptions(!PTO.LoopInterleaving, !PTO.LoopVectorization))); 1204 1205 if (IsLTO) { 1206 // The vectorizer may have significantly shortened a loop body; unroll 1207 // again. Unroll small loops to hide loop backedge latency and saturate any 1208 // parallel execution resources of an out-of-order processor. We also then 1209 // need to clean up redundancies and loop invariant code. 1210 // FIXME: It would be really good to use a loop-integrated instruction 1211 // combiner for cleanup here so that the unrolling and LICM can be pipelined 1212 // across the loop nests. 1213 // We do UnrollAndJam in a separate LPM to ensure it happens before unroll 1214 if (EnableUnrollAndJam && PTO.LoopUnrolling) 1215 FPM.addPass(createFunctionToLoopPassAdaptor( 1216 LoopUnrollAndJamPass(Level.getSpeedupLevel()))); 1217 FPM.addPass(LoopUnrollPass(LoopUnrollOptions( 1218 Level.getSpeedupLevel(), /*OnlyWhenForced=*/!PTO.LoopUnrolling, 1219 PTO.ForgetAllSCEVInLoopUnroll))); 1220 FPM.addPass(WarnMissedTransformationsPass()); 1221 } 1222 1223 if (!IsLTO) { 1224 // Eliminate loads by forwarding stores from the previous iteration to loads 1225 // of the current iteration. 1226 FPM.addPass(LoopLoadEliminationPass()); 1227 } 1228 // Cleanup after the loop optimization passes. 1229 FPM.addPass(InstCombinePass()); 1230 1231 if (Level.getSpeedupLevel() > 1 && ExtraVectorizerPasses) { 1232 // At higher optimization levels, try to clean up any runtime overlap and 1233 // alignment checks inserted by the vectorizer. We want to track correlated 1234 // runtime checks for two inner loops in the same outer loop, fold any 1235 // common computations, hoist loop-invariant aspects out of any outer loop, 1236 // and unswitch the runtime checks if possible. Once hoisted, we may have 1237 // dead (or speculatable) control flows or more combining opportunities. 1238 FPM.addPass(EarlyCSEPass()); 1239 FPM.addPass(CorrelatedValuePropagationPass()); 1240 FPM.addPass(InstCombinePass()); 1241 LoopPassManager LPM; 1242 LPM.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 1243 LPM.addPass(SimpleLoopUnswitchPass(/* NonTrivial */ Level == 1244 OptimizationLevel::O3)); 1245 FPM.addPass( 1246 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 1247 FPM.addPass(createFunctionToLoopPassAdaptor( 1248 std::move(LPM), EnableMSSALoopDependency, 1249 /*UseBlockFrequencyInfo=*/true)); 1250 FPM.addPass(SimplifyCFGPass()); 1251 FPM.addPass(InstCombinePass()); 1252 } 1253 1254 // Now that we've formed fast to execute loop structures, we do further 1255 // optimizations. These are run afterward as they might block doing complex 1256 // analyses and transforms such as what are needed for loop vectorization. 1257 1258 // Cleanup after loop vectorization, etc. Simplification passes like CVP and 1259 // GVN, loop transforms, and others have already run, so it's now better to 1260 // convert to more optimized IR using more aggressive simplify CFG options. 1261 // The extra sinking transform can create larger basic blocks, so do this 1262 // before SLP vectorization. 1263 FPM.addPass(SimplifyCFGPass(SimplifyCFGOptions() 1264 .forwardSwitchCondToPhi(true) 1265 .convertSwitchToLookupTable(true) 1266 .needCanonicalLoops(false) 1267 .hoistCommonInsts(true) 1268 .sinkCommonInsts(true))); 1269 1270 if (IsLTO) { 1271 FPM.addPass(SCCPPass()); 1272 FPM.addPass(InstCombinePass()); 1273 FPM.addPass(BDCEPass()); 1274 } 1275 1276 // Optimize parallel scalar instruction chains into SIMD instructions. 1277 if (PTO.SLPVectorization) { 1278 FPM.addPass(SLPVectorizerPass()); 1279 if (Level.getSpeedupLevel() > 1 && ExtraVectorizerPasses) { 1280 FPM.addPass(EarlyCSEPass()); 1281 } 1282 } 1283 // Enhance/cleanup vector code. 1284 FPM.addPass(VectorCombinePass()); 1285 1286 if (!IsLTO) { 1287 FPM.addPass(InstCombinePass()); 1288 // Unroll small loops to hide loop backedge latency and saturate any 1289 // parallel execution resources of an out-of-order processor. We also then 1290 // need to clean up redundancies and loop invariant code. 1291 // FIXME: It would be really good to use a loop-integrated instruction 1292 // combiner for cleanup here so that the unrolling and LICM can be pipelined 1293 // across the loop nests. 1294 // We do UnrollAndJam in a separate LPM to ensure it happens before unroll 1295 if (EnableUnrollAndJam && PTO.LoopUnrolling) { 1296 FPM.addPass(createFunctionToLoopPassAdaptor( 1297 LoopUnrollAndJamPass(Level.getSpeedupLevel()))); 1298 } 1299 FPM.addPass(LoopUnrollPass(LoopUnrollOptions( 1300 Level.getSpeedupLevel(), /*OnlyWhenForced=*/!PTO.LoopUnrolling, 1301 PTO.ForgetAllSCEVInLoopUnroll))); 1302 FPM.addPass(WarnMissedTransformationsPass()); 1303 FPM.addPass(InstCombinePass()); 1304 FPM.addPass( 1305 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 1306 FPM.addPass(createFunctionToLoopPassAdaptor( 1307 LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap), 1308 EnableMSSALoopDependency, /*UseBlockFrequencyInfo=*/true)); 1309 } 1310 1311 // Now that we've vectorized and unrolled loops, we may have more refined 1312 // alignment information, try to re-derive it here. 1313 FPM.addPass(AlignmentFromAssumptionsPass()); 1314 1315 if (IsLTO) 1316 FPM.addPass(InstCombinePass()); 1317 } 1318 1319 ModulePassManager 1320 PassBuilder::buildModuleOptimizationPipeline(OptimizationLevel Level, 1321 bool LTOPreLink) { 1322 ModulePassManager MPM; 1323 1324 // Optimize globals now that the module is fully simplified. 1325 MPM.addPass(GlobalOptPass()); 1326 MPM.addPass(GlobalDCEPass()); 1327 1328 // Run partial inlining pass to partially inline functions that have 1329 // large bodies. 1330 if (RunPartialInlining) 1331 MPM.addPass(PartialInlinerPass()); 1332 1333 // Remove avail extern fns and globals definitions since we aren't compiling 1334 // an object file for later LTO. For LTO we want to preserve these so they 1335 // are eligible for inlining at link-time. Note if they are unreferenced they 1336 // will be removed by GlobalDCE later, so this only impacts referenced 1337 // available externally globals. Eventually they will be suppressed during 1338 // codegen, but eliminating here enables more opportunity for GlobalDCE as it 1339 // may make globals referenced by available external functions dead and saves 1340 // running remaining passes on the eliminated functions. These should be 1341 // preserved during prelinking for link-time inlining decisions. 1342 if (!LTOPreLink) 1343 MPM.addPass(EliminateAvailableExternallyPass()); 1344 1345 if (EnableOrderFileInstrumentation) 1346 MPM.addPass(InstrOrderFilePass()); 1347 1348 // Do RPO function attribute inference across the module to forward-propagate 1349 // attributes where applicable. 1350 // FIXME: Is this really an optimization rather than a canonicalization? 1351 MPM.addPass(ReversePostOrderFunctionAttrsPass()); 1352 1353 // Do a post inline PGO instrumentation and use pass. This is a context 1354 // sensitive PGO pass. We don't want to do this in LTOPreLink phrase as 1355 // cross-module inline has not been done yet. The context sensitive 1356 // instrumentation is after all the inlines are done. 1357 if (!LTOPreLink && PGOOpt) { 1358 if (PGOOpt->CSAction == PGOOptions::CSIRInstr) 1359 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ true, 1360 /* IsCS */ true, PGOOpt->CSProfileGenFile, 1361 PGOOpt->ProfileRemappingFile); 1362 else if (PGOOpt->CSAction == PGOOptions::CSIRUse) 1363 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ false, 1364 /* IsCS */ true, PGOOpt->ProfileFile, 1365 PGOOpt->ProfileRemappingFile); 1366 } 1367 1368 // Re-require GloblasAA here prior to function passes. This is particularly 1369 // useful as the above will have inlined, DCE'ed, and function-attr 1370 // propagated everything. We should at this point have a reasonably minimal 1371 // and richly annotated call graph. By computing aliasing and mod/ref 1372 // information for all local globals here, the late loop passes and notably 1373 // the vectorizer will be able to use them to help recognize vectorizable 1374 // memory operations. 1375 MPM.addPass(RequireAnalysisPass<GlobalsAA, Module>()); 1376 1377 FunctionPassManager OptimizePM; 1378 OptimizePM.addPass(Float2IntPass()); 1379 OptimizePM.addPass(LowerConstantIntrinsicsPass()); 1380 1381 if (EnableMatrix) { 1382 OptimizePM.addPass(LowerMatrixIntrinsicsPass()); 1383 OptimizePM.addPass(EarlyCSEPass()); 1384 } 1385 1386 // FIXME: We need to run some loop optimizations to re-rotate loops after 1387 // simplify-cfg and others undo their rotation. 1388 1389 // Optimize the loop execution. These passes operate on entire loop nests 1390 // rather than on each loop in an inside-out manner, and so they are actually 1391 // function passes. 1392 1393 for (auto &C : VectorizerStartEPCallbacks) 1394 C(OptimizePM, Level); 1395 1396 // First rotate loops that may have been un-rotated by prior passes. 1397 // Disable header duplication at -Oz. 1398 OptimizePM.addPass(createFunctionToLoopPassAdaptor( 1399 LoopRotatePass(Level != OptimizationLevel::Oz, LTOPreLink), 1400 EnableMSSALoopDependency, 1401 /*UseBlockFrequencyInfo=*/false)); 1402 1403 // Distribute loops to allow partial vectorization. I.e. isolate dependences 1404 // into separate loop that would otherwise inhibit vectorization. This is 1405 // currently only performed for loops marked with the metadata 1406 // llvm.loop.distribute=true or when -enable-loop-distribute is specified. 1407 OptimizePM.addPass(LoopDistributePass()); 1408 1409 // Populates the VFABI attribute with the scalar-to-vector mappings 1410 // from the TargetLibraryInfo. 1411 OptimizePM.addPass(InjectTLIMappings()); 1412 1413 addVectorPasses(Level, OptimizePM, /* IsLTO */ false); 1414 1415 // Split out cold code. Splitting is done late to avoid hiding context from 1416 // other optimizations and inadvertently regressing performance. The tradeoff 1417 // is that this has a higher code size cost than splitting early. 1418 if (EnableHotColdSplit && !LTOPreLink) 1419 MPM.addPass(HotColdSplittingPass()); 1420 1421 // Search the code for similar regions of code. If enough similar regions can 1422 // be found where extracting the regions into their own function will decrease 1423 // the size of the program, we extract the regions, a deduplicate the 1424 // structurally similar regions. 1425 if (EnableIROutliner) 1426 MPM.addPass(IROutlinerPass()); 1427 1428 // Merge functions if requested. 1429 if (PTO.MergeFunctions) 1430 MPM.addPass(MergeFunctionsPass()); 1431 1432 // LoopSink pass sinks instructions hoisted by LICM, which serves as a 1433 // canonicalization pass that enables other optimizations. As a result, 1434 // LoopSink pass needs to be a very late IR pass to avoid undoing LICM 1435 // result too early. 1436 OptimizePM.addPass(LoopSinkPass()); 1437 1438 // And finally clean up LCSSA form before generating code. 1439 OptimizePM.addPass(InstSimplifyPass()); 1440 1441 // This hoists/decomposes div/rem ops. It should run after other sink/hoist 1442 // passes to avoid re-sinking, but before SimplifyCFG because it can allow 1443 // flattening of blocks. 1444 OptimizePM.addPass(DivRemPairsPass()); 1445 1446 // LoopSink (and other loop passes since the last simplifyCFG) might have 1447 // resulted in single-entry-single-exit or empty blocks. Clean up the CFG. 1448 OptimizePM.addPass(SimplifyCFGPass()); 1449 1450 if (PTO.Coroutines) 1451 OptimizePM.addPass(CoroCleanupPass()); 1452 1453 // Add the core optimizing pipeline. 1454 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(OptimizePM))); 1455 1456 for (auto &C : OptimizerLastEPCallbacks) 1457 C(MPM, Level); 1458 1459 if (PTO.CallGraphProfile) 1460 MPM.addPass(CGProfilePass()); 1461 1462 // Now we need to do some global optimization transforms. 1463 // FIXME: It would seem like these should come first in the optimization 1464 // pipeline and maybe be the bottom of the canonicalization pipeline? Weird 1465 // ordering here. 1466 MPM.addPass(GlobalDCEPass()); 1467 MPM.addPass(ConstantMergePass()); 1468 1469 // TODO: Relative look table converter pass caused an issue when full lto is 1470 // enabled. See https://reviews.llvm.org/D94355 for more details. 1471 // Until the issue fixed, disable this pass during pre-linking phase. 1472 if (!LTOPreLink) 1473 MPM.addPass(RelLookupTableConverterPass()); 1474 1475 return MPM; 1476 } 1477 1478 ModulePassManager 1479 PassBuilder::buildPerModuleDefaultPipeline(OptimizationLevel Level, 1480 bool LTOPreLink) { 1481 assert(Level != OptimizationLevel::O0 && 1482 "Must request optimizations for the default pipeline!"); 1483 1484 ModulePassManager MPM; 1485 1486 // Convert @llvm.global.annotations to !annotation metadata. 1487 MPM.addPass(Annotation2MetadataPass()); 1488 1489 // Force any function attributes we want the rest of the pipeline to observe. 1490 MPM.addPass(ForceFunctionAttrsPass()); 1491 1492 // Apply module pipeline start EP callback. 1493 for (auto &C : PipelineStartEPCallbacks) 1494 C(MPM, Level); 1495 1496 if (PGOOpt && PGOOpt->DebugInfoForProfiling) 1497 MPM.addPass(createModuleToFunctionPassAdaptor(AddDiscriminatorsPass())); 1498 1499 // Add the core simplification pipeline. 1500 MPM.addPass(buildModuleSimplificationPipeline( 1501 Level, LTOPreLink ? ThinOrFullLTOPhase::FullLTOPreLink 1502 : ThinOrFullLTOPhase::None)); 1503 1504 // Now add the optimization pipeline. 1505 MPM.addPass(buildModuleOptimizationPipeline(Level, LTOPreLink)); 1506 1507 if (PGOOpt && PGOOpt->PseudoProbeForProfiling) 1508 MPM.addPass(PseudoProbeUpdatePass()); 1509 1510 // Emit annotation remarks. 1511 addAnnotationRemarksPass(MPM); 1512 1513 if (LTOPreLink) 1514 addRequiredLTOPreLinkPasses(MPM); 1515 1516 return MPM; 1517 } 1518 1519 ModulePassManager 1520 PassBuilder::buildThinLTOPreLinkDefaultPipeline(OptimizationLevel Level) { 1521 assert(Level != OptimizationLevel::O0 && 1522 "Must request optimizations for the default pipeline!"); 1523 1524 ModulePassManager MPM; 1525 1526 // Convert @llvm.global.annotations to !annotation metadata. 1527 MPM.addPass(Annotation2MetadataPass()); 1528 1529 // Force any function attributes we want the rest of the pipeline to observe. 1530 MPM.addPass(ForceFunctionAttrsPass()); 1531 1532 if (PGOOpt && PGOOpt->DebugInfoForProfiling) 1533 MPM.addPass(createModuleToFunctionPassAdaptor(AddDiscriminatorsPass())); 1534 1535 // Apply module pipeline start EP callback. 1536 for (auto &C : PipelineStartEPCallbacks) 1537 C(MPM, Level); 1538 1539 // If we are planning to perform ThinLTO later, we don't bloat the code with 1540 // unrolling/vectorization/... now. Just simplify the module as much as we 1541 // can. 1542 MPM.addPass(buildModuleSimplificationPipeline( 1543 Level, ThinOrFullLTOPhase::ThinLTOPreLink)); 1544 1545 // Run partial inlining pass to partially inline functions that have 1546 // large bodies. 1547 // FIXME: It isn't clear whether this is really the right place to run this 1548 // in ThinLTO. Because there is another canonicalization and simplification 1549 // phase that will run after the thin link, running this here ends up with 1550 // less information than will be available later and it may grow functions in 1551 // ways that aren't beneficial. 1552 if (RunPartialInlining) 1553 MPM.addPass(PartialInlinerPass()); 1554 1555 // Reduce the size of the IR as much as possible. 1556 MPM.addPass(GlobalOptPass()); 1557 1558 // Module simplification splits coroutines, but does not fully clean up 1559 // coroutine intrinsics. To ensure ThinLTO optimization passes don't trip up 1560 // on these, we schedule the cleanup here. 1561 if (PTO.Coroutines) 1562 MPM.addPass(createModuleToFunctionPassAdaptor(CoroCleanupPass())); 1563 1564 if (PGOOpt && PGOOpt->PseudoProbeForProfiling) 1565 MPM.addPass(PseudoProbeUpdatePass()); 1566 1567 // Handle OptimizerLastEPCallbacks added by clang on PreLink. Actual 1568 // optimization is going to be done in PostLink stage, but clang can't 1569 // add callbacks there in case of in-process ThinLTO called by linker. 1570 for (auto &C : OptimizerLastEPCallbacks) 1571 C(MPM, Level); 1572 1573 // Emit annotation remarks. 1574 addAnnotationRemarksPass(MPM); 1575 1576 addRequiredLTOPreLinkPasses(MPM); 1577 1578 return MPM; 1579 } 1580 1581 ModulePassManager PassBuilder::buildThinLTODefaultPipeline( 1582 OptimizationLevel Level, const ModuleSummaryIndex *ImportSummary) { 1583 ModulePassManager MPM; 1584 1585 // Convert @llvm.global.annotations to !annotation metadata. 1586 MPM.addPass(Annotation2MetadataPass()); 1587 1588 if (ImportSummary) { 1589 // These passes import type identifier resolutions for whole-program 1590 // devirtualization and CFI. They must run early because other passes may 1591 // disturb the specific instruction patterns that these passes look for, 1592 // creating dependencies on resolutions that may not appear in the summary. 1593 // 1594 // For example, GVN may transform the pattern assume(type.test) appearing in 1595 // two basic blocks into assume(phi(type.test, type.test)), which would 1596 // transform a dependency on a WPD resolution into a dependency on a type 1597 // identifier resolution for CFI. 1598 // 1599 // Also, WPD has access to more precise information than ICP and can 1600 // devirtualize more effectively, so it should operate on the IR first. 1601 // 1602 // The WPD and LowerTypeTest passes need to run at -O0 to lower type 1603 // metadata and intrinsics. 1604 MPM.addPass(WholeProgramDevirtPass(nullptr, ImportSummary)); 1605 MPM.addPass(LowerTypeTestsPass(nullptr, ImportSummary)); 1606 } 1607 1608 if (Level == OptimizationLevel::O0) { 1609 // Run a second time to clean up any type tests left behind by WPD for use 1610 // in ICP. 1611 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1612 // Drop available_externally and unreferenced globals. This is necessary 1613 // with ThinLTO in order to avoid leaving undefined references to dead 1614 // globals in the object file. 1615 MPM.addPass(EliminateAvailableExternallyPass()); 1616 MPM.addPass(GlobalDCEPass()); 1617 return MPM; 1618 } 1619 1620 // Force any function attributes we want the rest of the pipeline to observe. 1621 MPM.addPass(ForceFunctionAttrsPass()); 1622 1623 // Add the core simplification pipeline. 1624 MPM.addPass(buildModuleSimplificationPipeline( 1625 Level, ThinOrFullLTOPhase::ThinLTOPostLink)); 1626 1627 // Now add the optimization pipeline. 1628 MPM.addPass(buildModuleOptimizationPipeline(Level)); 1629 1630 // Emit annotation remarks. 1631 addAnnotationRemarksPass(MPM); 1632 1633 return MPM; 1634 } 1635 1636 ModulePassManager 1637 PassBuilder::buildLTOPreLinkDefaultPipeline(OptimizationLevel Level) { 1638 assert(Level != OptimizationLevel::O0 && 1639 "Must request optimizations for the default pipeline!"); 1640 // FIXME: We should use a customized pre-link pipeline! 1641 return buildPerModuleDefaultPipeline(Level, 1642 /* LTOPreLink */ true); 1643 } 1644 1645 ModulePassManager 1646 PassBuilder::buildLTODefaultPipeline(OptimizationLevel Level, 1647 ModuleSummaryIndex *ExportSummary) { 1648 ModulePassManager MPM; 1649 1650 // Convert @llvm.global.annotations to !annotation metadata. 1651 MPM.addPass(Annotation2MetadataPass()); 1652 1653 if (Level == OptimizationLevel::O0) { 1654 // The WPD and LowerTypeTest passes need to run at -O0 to lower type 1655 // metadata and intrinsics. 1656 MPM.addPass(WholeProgramDevirtPass(ExportSummary, nullptr)); 1657 MPM.addPass(LowerTypeTestsPass(ExportSummary, nullptr)); 1658 // Run a second time to clean up any type tests left behind by WPD for use 1659 // in ICP. 1660 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1661 1662 // Emit annotation remarks. 1663 addAnnotationRemarksPass(MPM); 1664 1665 return MPM; 1666 } 1667 1668 if (PGOOpt && PGOOpt->Action == PGOOptions::SampleUse) { 1669 // Load sample profile before running the LTO optimization pipeline. 1670 MPM.addPass(SampleProfileLoaderPass(PGOOpt->ProfileFile, 1671 PGOOpt->ProfileRemappingFile, 1672 ThinOrFullLTOPhase::FullLTOPostLink)); 1673 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 1674 // RequireAnalysisPass for PSI before subsequent non-module passes. 1675 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 1676 } 1677 1678 // Remove unused virtual tables to improve the quality of code generated by 1679 // whole-program devirtualization and bitset lowering. 1680 MPM.addPass(GlobalDCEPass()); 1681 1682 // Force any function attributes we want the rest of the pipeline to observe. 1683 MPM.addPass(ForceFunctionAttrsPass()); 1684 1685 // Do basic inference of function attributes from known properties of system 1686 // libraries and other oracles. 1687 MPM.addPass(InferFunctionAttrsPass()); 1688 1689 if (Level.getSpeedupLevel() > 1) { 1690 FunctionPassManager EarlyFPM; 1691 EarlyFPM.addPass(CallSiteSplittingPass()); 1692 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(EarlyFPM))); 1693 1694 // Indirect call promotion. This should promote all the targets that are 1695 // left by the earlier promotion pass that promotes intra-module targets. 1696 // This two-step promotion is to save the compile time. For LTO, it should 1697 // produce the same result as if we only do promotion here. 1698 MPM.addPass(PGOIndirectCallPromotion( 1699 true /* InLTO */, PGOOpt && PGOOpt->Action == PGOOptions::SampleUse)); 1700 1701 if (EnableFunctionSpecialization) 1702 MPM.addPass(FunctionSpecializationPass()); 1703 // Propagate constants at call sites into the functions they call. This 1704 // opens opportunities for globalopt (and inlining) by substituting function 1705 // pointers passed as arguments to direct uses of functions. 1706 MPM.addPass(IPSCCPPass()); 1707 1708 // Attach metadata to indirect call sites indicating the set of functions 1709 // they may target at run-time. This should follow IPSCCP. 1710 MPM.addPass(CalledValuePropagationPass()); 1711 } 1712 1713 // Now deduce any function attributes based in the current code. 1714 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor( 1715 PostOrderFunctionAttrsPass())); 1716 1717 // Do RPO function attribute inference across the module to forward-propagate 1718 // attributes where applicable. 1719 // FIXME: Is this really an optimization rather than a canonicalization? 1720 MPM.addPass(ReversePostOrderFunctionAttrsPass()); 1721 1722 // Use in-range annotations on GEP indices to split globals where beneficial. 1723 MPM.addPass(GlobalSplitPass()); 1724 1725 // Run whole program optimization of virtual call when the list of callees 1726 // is fixed. 1727 MPM.addPass(WholeProgramDevirtPass(ExportSummary, nullptr)); 1728 1729 // Stop here at -O1. 1730 if (Level == OptimizationLevel::O1) { 1731 // The LowerTypeTestsPass needs to run to lower type metadata and the 1732 // type.test intrinsics. The pass does nothing if CFI is disabled. 1733 MPM.addPass(LowerTypeTestsPass(ExportSummary, nullptr)); 1734 // Run a second time to clean up any type tests left behind by WPD for use 1735 // in ICP (which is performed earlier than this in the regular LTO 1736 // pipeline). 1737 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1738 1739 // Emit annotation remarks. 1740 addAnnotationRemarksPass(MPM); 1741 1742 return MPM; 1743 } 1744 1745 // Optimize globals to try and fold them into constants. 1746 MPM.addPass(GlobalOptPass()); 1747 1748 // Promote any localized globals to SSA registers. 1749 MPM.addPass(createModuleToFunctionPassAdaptor(PromotePass())); 1750 1751 // Linking modules together can lead to duplicate global constant, only 1752 // keep one copy of each constant. 1753 MPM.addPass(ConstantMergePass()); 1754 1755 // Remove unused arguments from functions. 1756 MPM.addPass(DeadArgumentEliminationPass()); 1757 1758 // Reduce the code after globalopt and ipsccp. Both can open up significant 1759 // simplification opportunities, and both can propagate functions through 1760 // function pointers. When this happens, we often have to resolve varargs 1761 // calls, etc, so let instcombine do this. 1762 FunctionPassManager PeepholeFPM; 1763 if (Level == OptimizationLevel::O3) 1764 PeepholeFPM.addPass(AggressiveInstCombinePass()); 1765 PeepholeFPM.addPass(InstCombinePass()); 1766 invokePeepholeEPCallbacks(PeepholeFPM, Level); 1767 1768 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(PeepholeFPM))); 1769 1770 // Note: historically, the PruneEH pass was run first to deduce nounwind and 1771 // generally clean up exception handling overhead. It isn't clear this is 1772 // valuable as the inliner doesn't currently care whether it is inlining an 1773 // invoke or a call. 1774 // Run the inliner now. 1775 MPM.addPass(ModuleInlinerWrapperPass(getInlineParamsFromOptLevel(Level))); 1776 1777 // Optimize globals again after we ran the inliner. 1778 MPM.addPass(GlobalOptPass()); 1779 1780 // Garbage collect dead functions. 1781 // FIXME: Add ArgumentPromotion pass after once it's ported. 1782 MPM.addPass(GlobalDCEPass()); 1783 1784 FunctionPassManager FPM; 1785 // The IPO Passes may leave cruft around. Clean up after them. 1786 FPM.addPass(InstCombinePass()); 1787 invokePeepholeEPCallbacks(FPM, Level); 1788 1789 FPM.addPass(JumpThreadingPass(/*InsertFreezeWhenUnfoldingSelect*/ true)); 1790 1791 // Do a post inline PGO instrumentation and use pass. This is a context 1792 // sensitive PGO pass. 1793 if (PGOOpt) { 1794 if (PGOOpt->CSAction == PGOOptions::CSIRInstr) 1795 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ true, 1796 /* IsCS */ true, PGOOpt->CSProfileGenFile, 1797 PGOOpt->ProfileRemappingFile); 1798 else if (PGOOpt->CSAction == PGOOptions::CSIRUse) 1799 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ false, 1800 /* IsCS */ true, PGOOpt->ProfileFile, 1801 PGOOpt->ProfileRemappingFile); 1802 } 1803 1804 // Break up allocas 1805 FPM.addPass(SROA()); 1806 1807 // LTO provides additional opportunities for tailcall elimination due to 1808 // link-time inlining, and visibility of nocapture attribute. 1809 FPM.addPass(TailCallElimPass()); 1810 1811 // Run a few AA driver optimizations here and now to cleanup the code. 1812 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 1813 1814 MPM.addPass( 1815 createModuleToPostOrderCGSCCPassAdaptor(PostOrderFunctionAttrsPass())); 1816 1817 // Require the GlobalsAA analysis for the module so we can query it within 1818 // MainFPM. 1819 MPM.addPass(RequireAnalysisPass<GlobalsAA, Module>()); 1820 // Invalidate AAManager so it can be recreated and pick up the newly available 1821 // GlobalsAA. 1822 MPM.addPass( 1823 createModuleToFunctionPassAdaptor(InvalidateAnalysisPass<AAManager>())); 1824 1825 FunctionPassManager MainFPM; 1826 MainFPM.addPass(createFunctionToLoopPassAdaptor( 1827 LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap), 1828 EnableMSSALoopDependency, /*UseBlockFrequencyInfo=*/true)); 1829 1830 if (RunNewGVN) 1831 MainFPM.addPass(NewGVNPass()); 1832 else 1833 MainFPM.addPass(GVN()); 1834 1835 // Remove dead memcpy()'s. 1836 MainFPM.addPass(MemCpyOptPass()); 1837 1838 // Nuke dead stores. 1839 MainFPM.addPass(DSEPass()); 1840 MainFPM.addPass(MergedLoadStoreMotionPass()); 1841 1842 // More loops are countable; try to optimize them. 1843 if (EnableLoopFlatten && Level.getSpeedupLevel() > 1) 1844 MainFPM.addPass(createFunctionToLoopPassAdaptor(LoopFlattenPass())); 1845 1846 if (EnableConstraintElimination) 1847 MainFPM.addPass(ConstraintEliminationPass()); 1848 1849 LoopPassManager LPM; 1850 LPM.addPass(IndVarSimplifyPass()); 1851 LPM.addPass(LoopDeletionPass()); 1852 // FIXME: Add loop interchange. 1853 1854 // Unroll small loops and perform peeling. 1855 LPM.addPass(LoopFullUnrollPass(Level.getSpeedupLevel(), 1856 /* OnlyWhenForced= */ !PTO.LoopUnrolling, 1857 PTO.ForgetAllSCEVInLoopUnroll)); 1858 // The loop passes in LPM (LoopFullUnrollPass) do not preserve MemorySSA. 1859 // *All* loop passes must preserve it, in order to be able to use it. 1860 MainFPM.addPass(createFunctionToLoopPassAdaptor( 1861 std::move(LPM), /*UseMemorySSA=*/false, /*UseBlockFrequencyInfo=*/true)); 1862 1863 MainFPM.addPass(LoopDistributePass()); 1864 1865 addVectorPasses(Level, MainFPM, /* IsLTO */ true); 1866 1867 invokePeepholeEPCallbacks(MainFPM, Level); 1868 MainFPM.addPass(JumpThreadingPass(/*InsertFreezeWhenUnfoldingSelect*/ true)); 1869 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(MainFPM))); 1870 1871 // Create a function that performs CFI checks for cross-DSO calls with 1872 // targets in the current module. 1873 MPM.addPass(CrossDSOCFIPass()); 1874 1875 // Lower type metadata and the type.test intrinsic. This pass supports 1876 // clang's control flow integrity mechanisms (-fsanitize=cfi*) and needs 1877 // to be run at link time if CFI is enabled. This pass does nothing if 1878 // CFI is disabled. 1879 MPM.addPass(LowerTypeTestsPass(ExportSummary, nullptr)); 1880 // Run a second time to clean up any type tests left behind by WPD for use 1881 // in ICP (which is performed earlier than this in the regular LTO pipeline). 1882 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1883 1884 // Enable splitting late in the FullLTO post-link pipeline. This is done in 1885 // the same stage in the old pass manager (\ref addLateLTOOptimizationPasses). 1886 if (EnableHotColdSplit) 1887 MPM.addPass(HotColdSplittingPass()); 1888 1889 // Add late LTO optimization passes. 1890 // Delete basic blocks, which optimization passes may have killed. 1891 MPM.addPass(createModuleToFunctionPassAdaptor( 1892 SimplifyCFGPass(SimplifyCFGOptions().hoistCommonInsts(true)))); 1893 1894 // Drop bodies of available eternally objects to improve GlobalDCE. 1895 MPM.addPass(EliminateAvailableExternallyPass()); 1896 1897 // Now that we have optimized the program, discard unreachable functions. 1898 MPM.addPass(GlobalDCEPass()); 1899 1900 if (PTO.MergeFunctions) 1901 MPM.addPass(MergeFunctionsPass()); 1902 1903 // Emit annotation remarks. 1904 addAnnotationRemarksPass(MPM); 1905 1906 return MPM; 1907 } 1908 1909 ModulePassManager PassBuilder::buildO0DefaultPipeline(OptimizationLevel Level, 1910 bool LTOPreLink) { 1911 assert(Level == OptimizationLevel::O0 && 1912 "buildO0DefaultPipeline should only be used with O0"); 1913 1914 ModulePassManager MPM; 1915 1916 if (PGOOpt && (PGOOpt->Action == PGOOptions::IRInstr || 1917 PGOOpt->Action == PGOOptions::IRUse)) 1918 addPGOInstrPassesForO0( 1919 MPM, 1920 /* RunProfileGen */ (PGOOpt->Action == PGOOptions::IRInstr), 1921 /* IsCS */ false, PGOOpt->ProfileFile, PGOOpt->ProfileRemappingFile); 1922 1923 for (auto &C : PipelineStartEPCallbacks) 1924 C(MPM, Level); 1925 for (auto &C : PipelineEarlySimplificationEPCallbacks) 1926 C(MPM, Level); 1927 1928 // Build a minimal pipeline based on the semantics required by LLVM, 1929 // which is just that always inlining occurs. Further, disable generating 1930 // lifetime intrinsics to avoid enabling further optimizations during 1931 // code generation. 1932 // However, we need to insert lifetime intrinsics to avoid invalid access 1933 // caused by multithreaded coroutines. 1934 MPM.addPass(AlwaysInlinerPass( 1935 /*InsertLifetimeIntrinsics=*/PTO.Coroutines)); 1936 1937 if (PTO.MergeFunctions) 1938 MPM.addPass(MergeFunctionsPass()); 1939 1940 if (EnableMatrix) 1941 MPM.addPass( 1942 createModuleToFunctionPassAdaptor(LowerMatrixIntrinsicsPass(true))); 1943 1944 if (!CGSCCOptimizerLateEPCallbacks.empty()) { 1945 CGSCCPassManager CGPM; 1946 for (auto &C : CGSCCOptimizerLateEPCallbacks) 1947 C(CGPM, Level); 1948 if (!CGPM.isEmpty()) 1949 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPM))); 1950 } 1951 if (!LateLoopOptimizationsEPCallbacks.empty()) { 1952 LoopPassManager LPM; 1953 for (auto &C : LateLoopOptimizationsEPCallbacks) 1954 C(LPM, Level); 1955 if (!LPM.isEmpty()) { 1956 MPM.addPass(createModuleToFunctionPassAdaptor( 1957 createFunctionToLoopPassAdaptor(std::move(LPM)))); 1958 } 1959 } 1960 if (!LoopOptimizerEndEPCallbacks.empty()) { 1961 LoopPassManager LPM; 1962 for (auto &C : LoopOptimizerEndEPCallbacks) 1963 C(LPM, Level); 1964 if (!LPM.isEmpty()) { 1965 MPM.addPass(createModuleToFunctionPassAdaptor( 1966 createFunctionToLoopPassAdaptor(std::move(LPM)))); 1967 } 1968 } 1969 if (!ScalarOptimizerLateEPCallbacks.empty()) { 1970 FunctionPassManager FPM; 1971 for (auto &C : ScalarOptimizerLateEPCallbacks) 1972 C(FPM, Level); 1973 if (!FPM.isEmpty()) 1974 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 1975 } 1976 if (!VectorizerStartEPCallbacks.empty()) { 1977 FunctionPassManager FPM; 1978 for (auto &C : VectorizerStartEPCallbacks) 1979 C(FPM, Level); 1980 if (!FPM.isEmpty()) 1981 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 1982 } 1983 1984 if (PTO.Coroutines) { 1985 MPM.addPass(createModuleToFunctionPassAdaptor(CoroEarlyPass())); 1986 1987 CGSCCPassManager CGPM; 1988 CGPM.addPass(CoroSplitPass()); 1989 CGPM.addPass(createCGSCCToFunctionPassAdaptor(CoroElidePass())); 1990 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPM))); 1991 1992 MPM.addPass(createModuleToFunctionPassAdaptor(CoroCleanupPass())); 1993 } 1994 1995 for (auto &C : OptimizerLastEPCallbacks) 1996 C(MPM, Level); 1997 1998 if (LTOPreLink) 1999 addRequiredLTOPreLinkPasses(MPM); 2000 2001 return MPM; 2002 } 2003 2004 AAManager PassBuilder::buildDefaultAAPipeline() { 2005 AAManager AA; 2006 2007 // The order in which these are registered determines their priority when 2008 // being queried. 2009 2010 // First we register the basic alias analysis that provides the majority of 2011 // per-function local AA logic. This is a stateless, on-demand local set of 2012 // AA techniques. 2013 AA.registerFunctionAnalysis<BasicAA>(); 2014 2015 // Next we query fast, specialized alias analyses that wrap IR-embedded 2016 // information about aliasing. 2017 AA.registerFunctionAnalysis<ScopedNoAliasAA>(); 2018 AA.registerFunctionAnalysis<TypeBasedAA>(); 2019 2020 // Add support for querying global aliasing information when available. 2021 // Because the `AAManager` is a function analysis and `GlobalsAA` is a module 2022 // analysis, all that the `AAManager` can do is query for any *cached* 2023 // results from `GlobalsAA` through a readonly proxy. 2024 AA.registerModuleAnalysis<GlobalsAA>(); 2025 2026 // Add target-specific alias analyses. 2027 if (TM) 2028 TM->registerDefaultAliasAnalyses(AA); 2029 2030 return AA; 2031 } 2032 2033 static Optional<int> parseRepeatPassName(StringRef Name) { 2034 if (!Name.consume_front("repeat<") || !Name.consume_back(">")) 2035 return None; 2036 int Count; 2037 if (Name.getAsInteger(0, Count) || Count <= 0) 2038 return None; 2039 return Count; 2040 } 2041 2042 static Optional<int> parseDevirtPassName(StringRef Name) { 2043 if (!Name.consume_front("devirt<") || !Name.consume_back(">")) 2044 return None; 2045 int Count; 2046 if (Name.getAsInteger(0, Count) || Count < 0) 2047 return None; 2048 return Count; 2049 } 2050 2051 static bool checkParametrizedPassName(StringRef Name, StringRef PassName) { 2052 if (!Name.consume_front(PassName)) 2053 return false; 2054 // normal pass name w/o parameters == default parameters 2055 if (Name.empty()) 2056 return true; 2057 return Name.startswith("<") && Name.endswith(">"); 2058 } 2059 2060 namespace { 2061 2062 /// This performs customized parsing of pass name with parameters. 2063 /// 2064 /// We do not need parametrization of passes in textual pipeline very often, 2065 /// yet on a rare occasion ability to specify parameters right there can be 2066 /// useful. 2067 /// 2068 /// \p Name - parameterized specification of a pass from a textual pipeline 2069 /// is a string in a form of : 2070 /// PassName '<' parameter-list '>' 2071 /// 2072 /// Parameter list is being parsed by the parser callable argument, \p Parser, 2073 /// It takes a string-ref of parameters and returns either StringError or a 2074 /// parameter list in a form of a custom parameters type, all wrapped into 2075 /// Expected<> template class. 2076 /// 2077 template <typename ParametersParseCallableT> 2078 auto parsePassParameters(ParametersParseCallableT &&Parser, StringRef Name, 2079 StringRef PassName) -> decltype(Parser(StringRef{})) { 2080 using ParametersT = typename decltype(Parser(StringRef{}))::value_type; 2081 2082 StringRef Params = Name; 2083 if (!Params.consume_front(PassName)) { 2084 assert(false && 2085 "unable to strip pass name from parametrized pass specification"); 2086 } 2087 if (Params.empty()) 2088 return ParametersT{}; 2089 if (!Params.consume_front("<") || !Params.consume_back(">")) { 2090 assert(false && "invalid format for parametrized pass name"); 2091 } 2092 2093 Expected<ParametersT> Result = Parser(Params); 2094 assert((Result || Result.template errorIsA<StringError>()) && 2095 "Pass parameter parser can only return StringErrors."); 2096 return Result; 2097 } 2098 2099 /// Parser of parameters for LoopUnroll pass. 2100 Expected<LoopUnrollOptions> parseLoopUnrollOptions(StringRef Params) { 2101 LoopUnrollOptions UnrollOpts; 2102 while (!Params.empty()) { 2103 StringRef ParamName; 2104 std::tie(ParamName, Params) = Params.split(';'); 2105 int OptLevel = StringSwitch<int>(ParamName) 2106 .Case("O0", 0) 2107 .Case("O1", 1) 2108 .Case("O2", 2) 2109 .Case("O3", 3) 2110 .Default(-1); 2111 if (OptLevel >= 0) { 2112 UnrollOpts.setOptLevel(OptLevel); 2113 continue; 2114 } 2115 if (ParamName.consume_front("full-unroll-max=")) { 2116 int Count; 2117 if (ParamName.getAsInteger(0, Count)) 2118 return make_error<StringError>( 2119 formatv("invalid LoopUnrollPass parameter '{0}' ", ParamName).str(), 2120 inconvertibleErrorCode()); 2121 UnrollOpts.setFullUnrollMaxCount(Count); 2122 continue; 2123 } 2124 2125 bool Enable = !ParamName.consume_front("no-"); 2126 if (ParamName == "partial") { 2127 UnrollOpts.setPartial(Enable); 2128 } else if (ParamName == "peeling") { 2129 UnrollOpts.setPeeling(Enable); 2130 } else if (ParamName == "profile-peeling") { 2131 UnrollOpts.setProfileBasedPeeling(Enable); 2132 } else if (ParamName == "runtime") { 2133 UnrollOpts.setRuntime(Enable); 2134 } else if (ParamName == "upperbound") { 2135 UnrollOpts.setUpperBound(Enable); 2136 } else { 2137 return make_error<StringError>( 2138 formatv("invalid LoopUnrollPass parameter '{0}' ", ParamName).str(), 2139 inconvertibleErrorCode()); 2140 } 2141 } 2142 return UnrollOpts; 2143 } 2144 2145 Expected<MemorySanitizerOptions> parseMSanPassOptions(StringRef Params) { 2146 MemorySanitizerOptions Result; 2147 while (!Params.empty()) { 2148 StringRef ParamName; 2149 std::tie(ParamName, Params) = Params.split(';'); 2150 2151 if (ParamName == "recover") { 2152 Result.Recover = true; 2153 } else if (ParamName == "kernel") { 2154 Result.Kernel = true; 2155 } else if (ParamName.consume_front("track-origins=")) { 2156 if (ParamName.getAsInteger(0, Result.TrackOrigins)) 2157 return make_error<StringError>( 2158 formatv("invalid argument to MemorySanitizer pass track-origins " 2159 "parameter: '{0}' ", 2160 ParamName) 2161 .str(), 2162 inconvertibleErrorCode()); 2163 } else { 2164 return make_error<StringError>( 2165 formatv("invalid MemorySanitizer pass parameter '{0}' ", ParamName) 2166 .str(), 2167 inconvertibleErrorCode()); 2168 } 2169 } 2170 return Result; 2171 } 2172 2173 /// Parser of parameters for SimplifyCFG pass. 2174 Expected<SimplifyCFGOptions> parseSimplifyCFGOptions(StringRef Params) { 2175 SimplifyCFGOptions Result; 2176 while (!Params.empty()) { 2177 StringRef ParamName; 2178 std::tie(ParamName, Params) = Params.split(';'); 2179 2180 bool Enable = !ParamName.consume_front("no-"); 2181 if (ParamName == "forward-switch-cond") { 2182 Result.forwardSwitchCondToPhi(Enable); 2183 } else if (ParamName == "switch-to-lookup") { 2184 Result.convertSwitchToLookupTable(Enable); 2185 } else if (ParamName == "keep-loops") { 2186 Result.needCanonicalLoops(Enable); 2187 } else if (ParamName == "hoist-common-insts") { 2188 Result.hoistCommonInsts(Enable); 2189 } else if (ParamName == "sink-common-insts") { 2190 Result.sinkCommonInsts(Enable); 2191 } else if (Enable && ParamName.consume_front("bonus-inst-threshold=")) { 2192 APInt BonusInstThreshold; 2193 if (ParamName.getAsInteger(0, BonusInstThreshold)) 2194 return make_error<StringError>( 2195 formatv("invalid argument to SimplifyCFG pass bonus-threshold " 2196 "parameter: '{0}' ", 2197 ParamName).str(), 2198 inconvertibleErrorCode()); 2199 Result.bonusInstThreshold(BonusInstThreshold.getSExtValue()); 2200 } else { 2201 return make_error<StringError>( 2202 formatv("invalid SimplifyCFG pass parameter '{0}' ", ParamName).str(), 2203 inconvertibleErrorCode()); 2204 } 2205 } 2206 return Result; 2207 } 2208 2209 /// Parser of parameters for LoopVectorize pass. 2210 Expected<LoopVectorizeOptions> parseLoopVectorizeOptions(StringRef Params) { 2211 LoopVectorizeOptions Opts; 2212 while (!Params.empty()) { 2213 StringRef ParamName; 2214 std::tie(ParamName, Params) = Params.split(';'); 2215 2216 bool Enable = !ParamName.consume_front("no-"); 2217 if (ParamName == "interleave-forced-only") { 2218 Opts.setInterleaveOnlyWhenForced(Enable); 2219 } else if (ParamName == "vectorize-forced-only") { 2220 Opts.setVectorizeOnlyWhenForced(Enable); 2221 } else { 2222 return make_error<StringError>( 2223 formatv("invalid LoopVectorize parameter '{0}' ", ParamName).str(), 2224 inconvertibleErrorCode()); 2225 } 2226 } 2227 return Opts; 2228 } 2229 2230 Expected<bool> parseLoopUnswitchOptions(StringRef Params) { 2231 bool Result = false; 2232 while (!Params.empty()) { 2233 StringRef ParamName; 2234 std::tie(ParamName, Params) = Params.split(';'); 2235 2236 bool Enable = !ParamName.consume_front("no-"); 2237 if (ParamName == "nontrivial") { 2238 Result = Enable; 2239 } else { 2240 return make_error<StringError>( 2241 formatv("invalid LoopUnswitch pass parameter '{0}' ", ParamName) 2242 .str(), 2243 inconvertibleErrorCode()); 2244 } 2245 } 2246 return Result; 2247 } 2248 2249 Expected<bool> parseMergedLoadStoreMotionOptions(StringRef Params) { 2250 bool Result = false; 2251 while (!Params.empty()) { 2252 StringRef ParamName; 2253 std::tie(ParamName, Params) = Params.split(';'); 2254 2255 bool Enable = !ParamName.consume_front("no-"); 2256 if (ParamName == "split-footer-bb") { 2257 Result = Enable; 2258 } else { 2259 return make_error<StringError>( 2260 formatv("invalid MergedLoadStoreMotion pass parameter '{0}' ", 2261 ParamName) 2262 .str(), 2263 inconvertibleErrorCode()); 2264 } 2265 } 2266 return Result; 2267 } 2268 2269 Expected<GVNOptions> parseGVNOptions(StringRef Params) { 2270 GVNOptions Result; 2271 while (!Params.empty()) { 2272 StringRef ParamName; 2273 std::tie(ParamName, Params) = Params.split(';'); 2274 2275 bool Enable = !ParamName.consume_front("no-"); 2276 if (ParamName == "pre") { 2277 Result.setPRE(Enable); 2278 } else if (ParamName == "load-pre") { 2279 Result.setLoadPRE(Enable); 2280 } else if (ParamName == "split-backedge-load-pre") { 2281 Result.setLoadPRESplitBackedge(Enable); 2282 } else if (ParamName == "memdep") { 2283 Result.setMemDep(Enable); 2284 } else { 2285 return make_error<StringError>( 2286 formatv("invalid GVN pass parameter '{0}' ", ParamName).str(), 2287 inconvertibleErrorCode()); 2288 } 2289 } 2290 return Result; 2291 } 2292 2293 Expected<StackLifetime::LivenessType> 2294 parseStackLifetimeOptions(StringRef Params) { 2295 StackLifetime::LivenessType Result = StackLifetime::LivenessType::May; 2296 while (!Params.empty()) { 2297 StringRef ParamName; 2298 std::tie(ParamName, Params) = Params.split(';'); 2299 2300 if (ParamName == "may") { 2301 Result = StackLifetime::LivenessType::May; 2302 } else if (ParamName == "must") { 2303 Result = StackLifetime::LivenessType::Must; 2304 } else { 2305 return make_error<StringError>( 2306 formatv("invalid StackLifetime parameter '{0}' ", ParamName).str(), 2307 inconvertibleErrorCode()); 2308 } 2309 } 2310 return Result; 2311 } 2312 2313 } // namespace 2314 2315 /// Tests whether a pass name starts with a valid prefix for a default pipeline 2316 /// alias. 2317 static bool startsWithDefaultPipelineAliasPrefix(StringRef Name) { 2318 return Name.startswith("default") || Name.startswith("thinlto") || 2319 Name.startswith("lto"); 2320 } 2321 2322 /// Tests whether registered callbacks will accept a given pass name. 2323 /// 2324 /// When parsing a pipeline text, the type of the outermost pipeline may be 2325 /// omitted, in which case the type is automatically determined from the first 2326 /// pass name in the text. This may be a name that is handled through one of the 2327 /// callbacks. We check this through the oridinary parsing callbacks by setting 2328 /// up a dummy PassManager in order to not force the client to also handle this 2329 /// type of query. 2330 template <typename PassManagerT, typename CallbacksT> 2331 static bool callbacksAcceptPassName(StringRef Name, CallbacksT &Callbacks) { 2332 if (!Callbacks.empty()) { 2333 PassManagerT DummyPM; 2334 for (auto &CB : Callbacks) 2335 if (CB(Name, DummyPM, {})) 2336 return true; 2337 } 2338 return false; 2339 } 2340 2341 template <typename CallbacksT> 2342 static bool isModulePassName(StringRef Name, CallbacksT &Callbacks) { 2343 // Manually handle aliases for pre-configured pipeline fragments. 2344 if (startsWithDefaultPipelineAliasPrefix(Name)) 2345 return DefaultAliasRegex.match(Name); 2346 2347 // Explicitly handle pass manager names. 2348 if (Name == "module") 2349 return true; 2350 if (Name == "cgscc") 2351 return true; 2352 if (Name == "function") 2353 return true; 2354 2355 // Explicitly handle custom-parsed pass names. 2356 if (parseRepeatPassName(Name)) 2357 return true; 2358 2359 #define MODULE_PASS(NAME, CREATE_PASS) \ 2360 if (Name == NAME) \ 2361 return true; 2362 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 2363 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2364 return true; 2365 #include "PassRegistry.def" 2366 2367 return callbacksAcceptPassName<ModulePassManager>(Name, Callbacks); 2368 } 2369 2370 template <typename CallbacksT> 2371 static bool isCGSCCPassName(StringRef Name, CallbacksT &Callbacks) { 2372 // Explicitly handle pass manager names. 2373 if (Name == "cgscc") 2374 return true; 2375 if (Name == "function") 2376 return true; 2377 2378 // Explicitly handle custom-parsed pass names. 2379 if (parseRepeatPassName(Name)) 2380 return true; 2381 if (parseDevirtPassName(Name)) 2382 return true; 2383 2384 #define CGSCC_PASS(NAME, CREATE_PASS) \ 2385 if (Name == NAME) \ 2386 return true; 2387 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 2388 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2389 return true; 2390 #include "PassRegistry.def" 2391 2392 return callbacksAcceptPassName<CGSCCPassManager>(Name, Callbacks); 2393 } 2394 2395 template <typename CallbacksT> 2396 static bool isFunctionPassName(StringRef Name, CallbacksT &Callbacks) { 2397 // Explicitly handle pass manager names. 2398 if (Name == "function") 2399 return true; 2400 if (Name == "loop" || Name == "loop-mssa") 2401 return true; 2402 2403 // Explicitly handle custom-parsed pass names. 2404 if (parseRepeatPassName(Name)) 2405 return true; 2406 2407 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2408 if (Name == NAME) \ 2409 return true; 2410 #define FUNCTION_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2411 if (checkParametrizedPassName(Name, NAME)) \ 2412 return true; 2413 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 2414 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2415 return true; 2416 #include "PassRegistry.def" 2417 2418 return callbacksAcceptPassName<FunctionPassManager>(Name, Callbacks); 2419 } 2420 2421 template <typename CallbacksT> 2422 static bool isLoopPassName(StringRef Name, CallbacksT &Callbacks) { 2423 // Explicitly handle pass manager names. 2424 if (Name == "loop" || Name == "loop-mssa") 2425 return true; 2426 2427 // Explicitly handle custom-parsed pass names. 2428 if (parseRepeatPassName(Name)) 2429 return true; 2430 2431 #define LOOP_PASS(NAME, CREATE_PASS) \ 2432 if (Name == NAME) \ 2433 return true; 2434 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2435 if (checkParametrizedPassName(Name, NAME)) \ 2436 return true; 2437 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 2438 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2439 return true; 2440 #include "PassRegistry.def" 2441 2442 return callbacksAcceptPassName<LoopPassManager>(Name, Callbacks); 2443 } 2444 2445 Optional<std::vector<PassBuilder::PipelineElement>> 2446 PassBuilder::parsePipelineText(StringRef Text) { 2447 std::vector<PipelineElement> ResultPipeline; 2448 2449 SmallVector<std::vector<PipelineElement> *, 4> PipelineStack = { 2450 &ResultPipeline}; 2451 for (;;) { 2452 std::vector<PipelineElement> &Pipeline = *PipelineStack.back(); 2453 size_t Pos = Text.find_first_of(",()"); 2454 Pipeline.push_back({Text.substr(0, Pos), {}}); 2455 2456 // If we have a single terminating name, we're done. 2457 if (Pos == Text.npos) 2458 break; 2459 2460 char Sep = Text[Pos]; 2461 Text = Text.substr(Pos + 1); 2462 if (Sep == ',') 2463 // Just a name ending in a comma, continue. 2464 continue; 2465 2466 if (Sep == '(') { 2467 // Push the inner pipeline onto the stack to continue processing. 2468 PipelineStack.push_back(&Pipeline.back().InnerPipeline); 2469 continue; 2470 } 2471 2472 assert(Sep == ')' && "Bogus separator!"); 2473 // When handling the close parenthesis, we greedily consume them to avoid 2474 // empty strings in the pipeline. 2475 do { 2476 // If we try to pop the outer pipeline we have unbalanced parentheses. 2477 if (PipelineStack.size() == 1) 2478 return None; 2479 2480 PipelineStack.pop_back(); 2481 } while (Text.consume_front(")")); 2482 2483 // Check if we've finished parsing. 2484 if (Text.empty()) 2485 break; 2486 2487 // Otherwise, the end of an inner pipeline always has to be followed by 2488 // a comma, and then we can continue. 2489 if (!Text.consume_front(",")) 2490 return None; 2491 } 2492 2493 if (PipelineStack.size() > 1) 2494 // Unbalanced paretheses. 2495 return None; 2496 2497 assert(PipelineStack.back() == &ResultPipeline && 2498 "Wrong pipeline at the bottom of the stack!"); 2499 return {std::move(ResultPipeline)}; 2500 } 2501 2502 Error PassBuilder::parseModulePass(ModulePassManager &MPM, 2503 const PipelineElement &E) { 2504 auto &Name = E.Name; 2505 auto &InnerPipeline = E.InnerPipeline; 2506 2507 // First handle complex passes like the pass managers which carry pipelines. 2508 if (!InnerPipeline.empty()) { 2509 if (Name == "module") { 2510 ModulePassManager NestedMPM; 2511 if (auto Err = parseModulePassPipeline(NestedMPM, InnerPipeline)) 2512 return Err; 2513 MPM.addPass(std::move(NestedMPM)); 2514 return Error::success(); 2515 } 2516 if (Name == "cgscc") { 2517 CGSCCPassManager CGPM; 2518 if (auto Err = parseCGSCCPassPipeline(CGPM, InnerPipeline)) 2519 return Err; 2520 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPM))); 2521 return Error::success(); 2522 } 2523 if (Name == "function") { 2524 FunctionPassManager FPM; 2525 if (auto Err = parseFunctionPassPipeline(FPM, InnerPipeline)) 2526 return Err; 2527 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 2528 return Error::success(); 2529 } 2530 if (auto Count = parseRepeatPassName(Name)) { 2531 ModulePassManager NestedMPM; 2532 if (auto Err = parseModulePassPipeline(NestedMPM, InnerPipeline)) 2533 return Err; 2534 MPM.addPass(createRepeatedPass(*Count, std::move(NestedMPM))); 2535 return Error::success(); 2536 } 2537 2538 for (auto &C : ModulePipelineParsingCallbacks) 2539 if (C(Name, MPM, InnerPipeline)) 2540 return Error::success(); 2541 2542 // Normal passes can't have pipelines. 2543 return make_error<StringError>( 2544 formatv("invalid use of '{0}' pass as module pipeline", Name).str(), 2545 inconvertibleErrorCode()); 2546 ; 2547 } 2548 2549 // Manually handle aliases for pre-configured pipeline fragments. 2550 if (startsWithDefaultPipelineAliasPrefix(Name)) { 2551 SmallVector<StringRef, 3> Matches; 2552 if (!DefaultAliasRegex.match(Name, &Matches)) 2553 return make_error<StringError>( 2554 formatv("unknown default pipeline alias '{0}'", Name).str(), 2555 inconvertibleErrorCode()); 2556 2557 assert(Matches.size() == 3 && "Must capture two matched strings!"); 2558 2559 OptimizationLevel L = StringSwitch<OptimizationLevel>(Matches[2]) 2560 .Case("O0", OptimizationLevel::O0) 2561 .Case("O1", OptimizationLevel::O1) 2562 .Case("O2", OptimizationLevel::O2) 2563 .Case("O3", OptimizationLevel::O3) 2564 .Case("Os", OptimizationLevel::Os) 2565 .Case("Oz", OptimizationLevel::Oz); 2566 if (L == OptimizationLevel::O0 && Matches[1] != "thinlto" && 2567 Matches[1] != "lto") { 2568 MPM.addPass(buildO0DefaultPipeline(L, Matches[1] == "thinlto-pre-link" || 2569 Matches[1] == "lto-pre-link")); 2570 return Error::success(); 2571 } 2572 2573 // This is consistent with old pass manager invoked via opt, but 2574 // inconsistent with clang. Clang doesn't enable loop vectorization 2575 // but does enable slp vectorization at Oz. 2576 PTO.LoopVectorization = 2577 L.getSpeedupLevel() > 1 && L != OptimizationLevel::Oz; 2578 PTO.SLPVectorization = 2579 L.getSpeedupLevel() > 1 && L != OptimizationLevel::Oz; 2580 2581 if (Matches[1] == "default") { 2582 MPM.addPass(buildPerModuleDefaultPipeline(L)); 2583 } else if (Matches[1] == "thinlto-pre-link") { 2584 MPM.addPass(buildThinLTOPreLinkDefaultPipeline(L)); 2585 } else if (Matches[1] == "thinlto") { 2586 MPM.addPass(buildThinLTODefaultPipeline(L, nullptr)); 2587 } else if (Matches[1] == "lto-pre-link") { 2588 MPM.addPass(buildLTOPreLinkDefaultPipeline(L)); 2589 } else { 2590 assert(Matches[1] == "lto" && "Not one of the matched options!"); 2591 MPM.addPass(buildLTODefaultPipeline(L, nullptr)); 2592 } 2593 return Error::success(); 2594 } 2595 2596 // Finally expand the basic registered passes from the .inc file. 2597 #define MODULE_PASS(NAME, CREATE_PASS) \ 2598 if (Name == NAME) { \ 2599 MPM.addPass(CREATE_PASS); \ 2600 return Error::success(); \ 2601 } 2602 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 2603 if (Name == "require<" NAME ">") { \ 2604 MPM.addPass( \ 2605 RequireAnalysisPass< \ 2606 std::remove_reference<decltype(CREATE_PASS)>::type, Module>()); \ 2607 return Error::success(); \ 2608 } \ 2609 if (Name == "invalidate<" NAME ">") { \ 2610 MPM.addPass(InvalidateAnalysisPass< \ 2611 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2612 return Error::success(); \ 2613 } 2614 #define CGSCC_PASS(NAME, CREATE_PASS) \ 2615 if (Name == NAME) { \ 2616 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(CREATE_PASS)); \ 2617 return Error::success(); \ 2618 } 2619 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2620 if (Name == NAME) { \ 2621 MPM.addPass(createModuleToFunctionPassAdaptor(CREATE_PASS)); \ 2622 return Error::success(); \ 2623 } 2624 #define FUNCTION_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2625 if (checkParametrizedPassName(Name, NAME)) { \ 2626 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2627 if (!Params) \ 2628 return Params.takeError(); \ 2629 MPM.addPass(createModuleToFunctionPassAdaptor(CREATE_PASS(Params.get()))); \ 2630 return Error::success(); \ 2631 } 2632 #define LOOP_PASS(NAME, CREATE_PASS) \ 2633 if (Name == NAME) { \ 2634 MPM.addPass(createModuleToFunctionPassAdaptor( \ 2635 createFunctionToLoopPassAdaptor(CREATE_PASS, false, false))); \ 2636 return Error::success(); \ 2637 } 2638 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2639 if (checkParametrizedPassName(Name, NAME)) { \ 2640 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2641 if (!Params) \ 2642 return Params.takeError(); \ 2643 MPM.addPass( \ 2644 createModuleToFunctionPassAdaptor(createFunctionToLoopPassAdaptor( \ 2645 CREATE_PASS(Params.get()), false, false))); \ 2646 return Error::success(); \ 2647 } 2648 #include "PassRegistry.def" 2649 2650 for (auto &C : ModulePipelineParsingCallbacks) 2651 if (C(Name, MPM, InnerPipeline)) 2652 return Error::success(); 2653 return make_error<StringError>( 2654 formatv("unknown module pass '{0}'", Name).str(), 2655 inconvertibleErrorCode()); 2656 } 2657 2658 Error PassBuilder::parseCGSCCPass(CGSCCPassManager &CGPM, 2659 const PipelineElement &E) { 2660 auto &Name = E.Name; 2661 auto &InnerPipeline = E.InnerPipeline; 2662 2663 // First handle complex passes like the pass managers which carry pipelines. 2664 if (!InnerPipeline.empty()) { 2665 if (Name == "cgscc") { 2666 CGSCCPassManager NestedCGPM; 2667 if (auto Err = parseCGSCCPassPipeline(NestedCGPM, InnerPipeline)) 2668 return Err; 2669 // Add the nested pass manager with the appropriate adaptor. 2670 CGPM.addPass(std::move(NestedCGPM)); 2671 return Error::success(); 2672 } 2673 if (Name == "function") { 2674 FunctionPassManager FPM; 2675 if (auto Err = parseFunctionPassPipeline(FPM, InnerPipeline)) 2676 return Err; 2677 // Add the nested pass manager with the appropriate adaptor. 2678 CGPM.addPass(createCGSCCToFunctionPassAdaptor(std::move(FPM))); 2679 return Error::success(); 2680 } 2681 if (auto Count = parseRepeatPassName(Name)) { 2682 CGSCCPassManager NestedCGPM; 2683 if (auto Err = parseCGSCCPassPipeline(NestedCGPM, InnerPipeline)) 2684 return Err; 2685 CGPM.addPass(createRepeatedPass(*Count, std::move(NestedCGPM))); 2686 return Error::success(); 2687 } 2688 if (auto MaxRepetitions = parseDevirtPassName(Name)) { 2689 CGSCCPassManager NestedCGPM; 2690 if (auto Err = parseCGSCCPassPipeline(NestedCGPM, InnerPipeline)) 2691 return Err; 2692 CGPM.addPass( 2693 createDevirtSCCRepeatedPass(std::move(NestedCGPM), *MaxRepetitions)); 2694 return Error::success(); 2695 } 2696 2697 for (auto &C : CGSCCPipelineParsingCallbacks) 2698 if (C(Name, CGPM, InnerPipeline)) 2699 return Error::success(); 2700 2701 // Normal passes can't have pipelines. 2702 return make_error<StringError>( 2703 formatv("invalid use of '{0}' pass as cgscc pipeline", Name).str(), 2704 inconvertibleErrorCode()); 2705 } 2706 2707 // Now expand the basic registered passes from the .inc file. 2708 #define CGSCC_PASS(NAME, CREATE_PASS) \ 2709 if (Name == NAME) { \ 2710 CGPM.addPass(CREATE_PASS); \ 2711 return Error::success(); \ 2712 } 2713 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 2714 if (Name == "require<" NAME ">") { \ 2715 CGPM.addPass(RequireAnalysisPass< \ 2716 std::remove_reference<decltype(CREATE_PASS)>::type, \ 2717 LazyCallGraph::SCC, CGSCCAnalysisManager, LazyCallGraph &, \ 2718 CGSCCUpdateResult &>()); \ 2719 return Error::success(); \ 2720 } \ 2721 if (Name == "invalidate<" NAME ">") { \ 2722 CGPM.addPass(InvalidateAnalysisPass< \ 2723 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2724 return Error::success(); \ 2725 } 2726 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2727 if (Name == NAME) { \ 2728 CGPM.addPass(createCGSCCToFunctionPassAdaptor(CREATE_PASS)); \ 2729 return Error::success(); \ 2730 } 2731 #define FUNCTION_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2732 if (checkParametrizedPassName(Name, NAME)) { \ 2733 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2734 if (!Params) \ 2735 return Params.takeError(); \ 2736 CGPM.addPass(createCGSCCToFunctionPassAdaptor(CREATE_PASS(Params.get()))); \ 2737 return Error::success(); \ 2738 } 2739 #define LOOP_PASS(NAME, CREATE_PASS) \ 2740 if (Name == NAME) { \ 2741 CGPM.addPass(createCGSCCToFunctionPassAdaptor( \ 2742 createFunctionToLoopPassAdaptor(CREATE_PASS, false, false))); \ 2743 return Error::success(); \ 2744 } 2745 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2746 if (checkParametrizedPassName(Name, NAME)) { \ 2747 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2748 if (!Params) \ 2749 return Params.takeError(); \ 2750 CGPM.addPass( \ 2751 createCGSCCToFunctionPassAdaptor(createFunctionToLoopPassAdaptor( \ 2752 CREATE_PASS(Params.get()), false, false))); \ 2753 return Error::success(); \ 2754 } 2755 #include "PassRegistry.def" 2756 2757 for (auto &C : CGSCCPipelineParsingCallbacks) 2758 if (C(Name, CGPM, InnerPipeline)) 2759 return Error::success(); 2760 return make_error<StringError>( 2761 formatv("unknown cgscc pass '{0}'", Name).str(), 2762 inconvertibleErrorCode()); 2763 } 2764 2765 Error PassBuilder::parseFunctionPass(FunctionPassManager &FPM, 2766 const PipelineElement &E) { 2767 auto &Name = E.Name; 2768 auto &InnerPipeline = E.InnerPipeline; 2769 2770 // First handle complex passes like the pass managers which carry pipelines. 2771 if (!InnerPipeline.empty()) { 2772 if (Name == "function") { 2773 FunctionPassManager NestedFPM; 2774 if (auto Err = parseFunctionPassPipeline(NestedFPM, InnerPipeline)) 2775 return Err; 2776 // Add the nested pass manager with the appropriate adaptor. 2777 FPM.addPass(std::move(NestedFPM)); 2778 return Error::success(); 2779 } 2780 if (Name == "loop" || Name == "loop-mssa") { 2781 LoopPassManager LPM; 2782 if (auto Err = parseLoopPassPipeline(LPM, InnerPipeline)) 2783 return Err; 2784 // Add the nested pass manager with the appropriate adaptor. 2785 bool UseMemorySSA = (Name == "loop-mssa"); 2786 bool UseBFI = llvm::any_of( 2787 InnerPipeline, [](auto Pipeline) { return Pipeline.Name == "licm"; }); 2788 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM), UseMemorySSA, 2789 UseBFI)); 2790 return Error::success(); 2791 } 2792 if (auto Count = parseRepeatPassName(Name)) { 2793 FunctionPassManager NestedFPM; 2794 if (auto Err = parseFunctionPassPipeline(NestedFPM, InnerPipeline)) 2795 return Err; 2796 FPM.addPass(createRepeatedPass(*Count, std::move(NestedFPM))); 2797 return Error::success(); 2798 } 2799 2800 for (auto &C : FunctionPipelineParsingCallbacks) 2801 if (C(Name, FPM, InnerPipeline)) 2802 return Error::success(); 2803 2804 // Normal passes can't have pipelines. 2805 return make_error<StringError>( 2806 formatv("invalid use of '{0}' pass as function pipeline", Name).str(), 2807 inconvertibleErrorCode()); 2808 } 2809 2810 // Now expand the basic registered passes from the .inc file. 2811 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2812 if (Name == NAME) { \ 2813 FPM.addPass(CREATE_PASS); \ 2814 return Error::success(); \ 2815 } 2816 #define FUNCTION_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2817 if (checkParametrizedPassName(Name, NAME)) { \ 2818 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2819 if (!Params) \ 2820 return Params.takeError(); \ 2821 FPM.addPass(CREATE_PASS(Params.get())); \ 2822 return Error::success(); \ 2823 } 2824 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 2825 if (Name == "require<" NAME ">") { \ 2826 FPM.addPass( \ 2827 RequireAnalysisPass< \ 2828 std::remove_reference<decltype(CREATE_PASS)>::type, Function>()); \ 2829 return Error::success(); \ 2830 } \ 2831 if (Name == "invalidate<" NAME ">") { \ 2832 FPM.addPass(InvalidateAnalysisPass< \ 2833 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2834 return Error::success(); \ 2835 } 2836 // FIXME: UseMemorySSA is set to false. Maybe we could do things like: 2837 // bool UseMemorySSA = !("canon-freeze" || "loop-predication" || 2838 // "guard-widening"); 2839 // The risk is that it may become obsolete if we're not careful. 2840 #define LOOP_PASS(NAME, CREATE_PASS) \ 2841 if (Name == NAME) { \ 2842 FPM.addPass(createFunctionToLoopPassAdaptor(CREATE_PASS, false, false)); \ 2843 return Error::success(); \ 2844 } 2845 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2846 if (checkParametrizedPassName(Name, NAME)) { \ 2847 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2848 if (!Params) \ 2849 return Params.takeError(); \ 2850 FPM.addPass(createFunctionToLoopPassAdaptor(CREATE_PASS(Params.get()), \ 2851 false, false)); \ 2852 return Error::success(); \ 2853 } 2854 #include "PassRegistry.def" 2855 2856 for (auto &C : FunctionPipelineParsingCallbacks) 2857 if (C(Name, FPM, InnerPipeline)) 2858 return Error::success(); 2859 return make_error<StringError>( 2860 formatv("unknown function pass '{0}'", Name).str(), 2861 inconvertibleErrorCode()); 2862 } 2863 2864 Error PassBuilder::parseLoopPass(LoopPassManager &LPM, 2865 const PipelineElement &E) { 2866 StringRef Name = E.Name; 2867 auto &InnerPipeline = E.InnerPipeline; 2868 2869 // First handle complex passes like the pass managers which carry pipelines. 2870 if (!InnerPipeline.empty()) { 2871 if (Name == "loop") { 2872 LoopPassManager NestedLPM; 2873 if (auto Err = parseLoopPassPipeline(NestedLPM, InnerPipeline)) 2874 return Err; 2875 // Add the nested pass manager with the appropriate adaptor. 2876 LPM.addPass(std::move(NestedLPM)); 2877 return Error::success(); 2878 } 2879 if (auto Count = parseRepeatPassName(Name)) { 2880 LoopPassManager NestedLPM; 2881 if (auto Err = parseLoopPassPipeline(NestedLPM, InnerPipeline)) 2882 return Err; 2883 LPM.addPass(createRepeatedPass(*Count, std::move(NestedLPM))); 2884 return Error::success(); 2885 } 2886 2887 for (auto &C : LoopPipelineParsingCallbacks) 2888 if (C(Name, LPM, InnerPipeline)) 2889 return Error::success(); 2890 2891 // Normal passes can't have pipelines. 2892 return make_error<StringError>( 2893 formatv("invalid use of '{0}' pass as loop pipeline", Name).str(), 2894 inconvertibleErrorCode()); 2895 } 2896 2897 // Now expand the basic registered passes from the .inc file. 2898 #define LOOP_PASS(NAME, CREATE_PASS) \ 2899 if (Name == NAME) { \ 2900 LPM.addPass(CREATE_PASS); \ 2901 return Error::success(); \ 2902 } 2903 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2904 if (checkParametrizedPassName(Name, NAME)) { \ 2905 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2906 if (!Params) \ 2907 return Params.takeError(); \ 2908 LPM.addPass(CREATE_PASS(Params.get())); \ 2909 return Error::success(); \ 2910 } 2911 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 2912 if (Name == "require<" NAME ">") { \ 2913 LPM.addPass(RequireAnalysisPass< \ 2914 std::remove_reference<decltype(CREATE_PASS)>::type, Loop, \ 2915 LoopAnalysisManager, LoopStandardAnalysisResults &, \ 2916 LPMUpdater &>()); \ 2917 return Error::success(); \ 2918 } \ 2919 if (Name == "invalidate<" NAME ">") { \ 2920 LPM.addPass(InvalidateAnalysisPass< \ 2921 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2922 return Error::success(); \ 2923 } 2924 #include "PassRegistry.def" 2925 2926 for (auto &C : LoopPipelineParsingCallbacks) 2927 if (C(Name, LPM, InnerPipeline)) 2928 return Error::success(); 2929 return make_error<StringError>(formatv("unknown loop pass '{0}'", Name).str(), 2930 inconvertibleErrorCode()); 2931 } 2932 2933 bool PassBuilder::parseAAPassName(AAManager &AA, StringRef Name) { 2934 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 2935 if (Name == NAME) { \ 2936 AA.registerModuleAnalysis< \ 2937 std::remove_reference<decltype(CREATE_PASS)>::type>(); \ 2938 return true; \ 2939 } 2940 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 2941 if (Name == NAME) { \ 2942 AA.registerFunctionAnalysis< \ 2943 std::remove_reference<decltype(CREATE_PASS)>::type>(); \ 2944 return true; \ 2945 } 2946 #include "PassRegistry.def" 2947 2948 for (auto &C : AAParsingCallbacks) 2949 if (C(Name, AA)) 2950 return true; 2951 return false; 2952 } 2953 2954 Error PassBuilder::parseLoopPassPipeline(LoopPassManager &LPM, 2955 ArrayRef<PipelineElement> Pipeline) { 2956 for (const auto &Element : Pipeline) { 2957 if (auto Err = parseLoopPass(LPM, Element)) 2958 return Err; 2959 } 2960 return Error::success(); 2961 } 2962 2963 Error PassBuilder::parseFunctionPassPipeline( 2964 FunctionPassManager &FPM, ArrayRef<PipelineElement> Pipeline) { 2965 for (const auto &Element : Pipeline) { 2966 if (auto Err = parseFunctionPass(FPM, Element)) 2967 return Err; 2968 } 2969 return Error::success(); 2970 } 2971 2972 Error PassBuilder::parseCGSCCPassPipeline(CGSCCPassManager &CGPM, 2973 ArrayRef<PipelineElement> Pipeline) { 2974 for (const auto &Element : Pipeline) { 2975 if (auto Err = parseCGSCCPass(CGPM, Element)) 2976 return Err; 2977 } 2978 return Error::success(); 2979 } 2980 2981 void PassBuilder::crossRegisterProxies(LoopAnalysisManager &LAM, 2982 FunctionAnalysisManager &FAM, 2983 CGSCCAnalysisManager &CGAM, 2984 ModuleAnalysisManager &MAM) { 2985 MAM.registerPass([&] { return FunctionAnalysisManagerModuleProxy(FAM); }); 2986 MAM.registerPass([&] { return CGSCCAnalysisManagerModuleProxy(CGAM); }); 2987 CGAM.registerPass([&] { return ModuleAnalysisManagerCGSCCProxy(MAM); }); 2988 FAM.registerPass([&] { return CGSCCAnalysisManagerFunctionProxy(CGAM); }); 2989 FAM.registerPass([&] { return ModuleAnalysisManagerFunctionProxy(MAM); }); 2990 FAM.registerPass([&] { return LoopAnalysisManagerFunctionProxy(LAM); }); 2991 LAM.registerPass([&] { return FunctionAnalysisManagerLoopProxy(FAM); }); 2992 } 2993 2994 Error PassBuilder::parseModulePassPipeline(ModulePassManager &MPM, 2995 ArrayRef<PipelineElement> Pipeline) { 2996 for (const auto &Element : Pipeline) { 2997 if (auto Err = parseModulePass(MPM, Element)) 2998 return Err; 2999 } 3000 return Error::success(); 3001 } 3002 3003 // Primary pass pipeline description parsing routine for a \c ModulePassManager 3004 // FIXME: Should this routine accept a TargetMachine or require the caller to 3005 // pre-populate the analysis managers with target-specific stuff? 3006 Error PassBuilder::parsePassPipeline(ModulePassManager &MPM, 3007 StringRef PipelineText) { 3008 auto Pipeline = parsePipelineText(PipelineText); 3009 if (!Pipeline || Pipeline->empty()) 3010 return make_error<StringError>( 3011 formatv("invalid pipeline '{0}'", PipelineText).str(), 3012 inconvertibleErrorCode()); 3013 3014 // If the first name isn't at the module layer, wrap the pipeline up 3015 // automatically. 3016 StringRef FirstName = Pipeline->front().Name; 3017 3018 if (!isModulePassName(FirstName, ModulePipelineParsingCallbacks)) { 3019 if (isCGSCCPassName(FirstName, CGSCCPipelineParsingCallbacks)) { 3020 Pipeline = {{"cgscc", std::move(*Pipeline)}}; 3021 } else if (isFunctionPassName(FirstName, 3022 FunctionPipelineParsingCallbacks)) { 3023 Pipeline = {{"function", std::move(*Pipeline)}}; 3024 } else if (isLoopPassName(FirstName, LoopPipelineParsingCallbacks)) { 3025 Pipeline = {{"function", {{"loop", std::move(*Pipeline)}}}}; 3026 } else { 3027 for (auto &C : TopLevelPipelineParsingCallbacks) 3028 if (C(MPM, *Pipeline)) 3029 return Error::success(); 3030 3031 // Unknown pass or pipeline name! 3032 auto &InnerPipeline = Pipeline->front().InnerPipeline; 3033 return make_error<StringError>( 3034 formatv("unknown {0} name '{1}'", 3035 (InnerPipeline.empty() ? "pass" : "pipeline"), FirstName) 3036 .str(), 3037 inconvertibleErrorCode()); 3038 } 3039 } 3040 3041 if (auto Err = parseModulePassPipeline(MPM, *Pipeline)) 3042 return Err; 3043 return Error::success(); 3044 } 3045 3046 // Primary pass pipeline description parsing routine for a \c CGSCCPassManager 3047 Error PassBuilder::parsePassPipeline(CGSCCPassManager &CGPM, 3048 StringRef PipelineText) { 3049 auto Pipeline = parsePipelineText(PipelineText); 3050 if (!Pipeline || Pipeline->empty()) 3051 return make_error<StringError>( 3052 formatv("invalid pipeline '{0}'", PipelineText).str(), 3053 inconvertibleErrorCode()); 3054 3055 StringRef FirstName = Pipeline->front().Name; 3056 if (!isCGSCCPassName(FirstName, CGSCCPipelineParsingCallbacks)) 3057 return make_error<StringError>( 3058 formatv("unknown cgscc pass '{0}' in pipeline '{1}'", FirstName, 3059 PipelineText) 3060 .str(), 3061 inconvertibleErrorCode()); 3062 3063 if (auto Err = parseCGSCCPassPipeline(CGPM, *Pipeline)) 3064 return Err; 3065 return Error::success(); 3066 } 3067 3068 // Primary pass pipeline description parsing routine for a \c 3069 // FunctionPassManager 3070 Error PassBuilder::parsePassPipeline(FunctionPassManager &FPM, 3071 StringRef PipelineText) { 3072 auto Pipeline = parsePipelineText(PipelineText); 3073 if (!Pipeline || Pipeline->empty()) 3074 return make_error<StringError>( 3075 formatv("invalid pipeline '{0}'", PipelineText).str(), 3076 inconvertibleErrorCode()); 3077 3078 StringRef FirstName = Pipeline->front().Name; 3079 if (!isFunctionPassName(FirstName, FunctionPipelineParsingCallbacks)) 3080 return make_error<StringError>( 3081 formatv("unknown function pass '{0}' in pipeline '{1}'", FirstName, 3082 PipelineText) 3083 .str(), 3084 inconvertibleErrorCode()); 3085 3086 if (auto Err = parseFunctionPassPipeline(FPM, *Pipeline)) 3087 return Err; 3088 return Error::success(); 3089 } 3090 3091 // Primary pass pipeline description parsing routine for a \c LoopPassManager 3092 Error PassBuilder::parsePassPipeline(LoopPassManager &CGPM, 3093 StringRef PipelineText) { 3094 auto Pipeline = parsePipelineText(PipelineText); 3095 if (!Pipeline || Pipeline->empty()) 3096 return make_error<StringError>( 3097 formatv("invalid pipeline '{0}'", PipelineText).str(), 3098 inconvertibleErrorCode()); 3099 3100 if (auto Err = parseLoopPassPipeline(CGPM, *Pipeline)) 3101 return Err; 3102 3103 return Error::success(); 3104 } 3105 3106 Error PassBuilder::parseAAPipeline(AAManager &AA, StringRef PipelineText) { 3107 // If the pipeline just consists of the word 'default' just replace the AA 3108 // manager with our default one. 3109 if (PipelineText == "default") { 3110 AA = buildDefaultAAPipeline(); 3111 return Error::success(); 3112 } 3113 3114 while (!PipelineText.empty()) { 3115 StringRef Name; 3116 std::tie(Name, PipelineText) = PipelineText.split(','); 3117 if (!parseAAPassName(AA, Name)) 3118 return make_error<StringError>( 3119 formatv("unknown alias analysis name '{0}'", Name).str(), 3120 inconvertibleErrorCode()); 3121 } 3122 3123 return Error::success(); 3124 } 3125 3126 bool PassBuilder::isAAPassName(StringRef PassName) { 3127 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3128 if (PassName == NAME) \ 3129 return true; 3130 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3131 if (PassName == NAME) \ 3132 return true; 3133 #include "PassRegistry.def" 3134 return false; 3135 } 3136 3137 bool PassBuilder::isAnalysisPassName(StringRef PassName) { 3138 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 3139 if (PassName == NAME) \ 3140 return true; 3141 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 3142 if (PassName == NAME) \ 3143 return true; 3144 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 3145 if (PassName == NAME) \ 3146 return true; 3147 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 3148 if (PassName == NAME) \ 3149 return true; 3150 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3151 if (PassName == NAME) \ 3152 return true; 3153 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3154 if (PassName == NAME) \ 3155 return true; 3156 #include "PassRegistry.def" 3157 return false; 3158 } 3159 3160 static void printPassName(StringRef PassName, raw_ostream &OS) { 3161 OS << " " << PassName << "\n"; 3162 } 3163 3164 void PassBuilder::printPassNames(raw_ostream &OS) { 3165 // TODO: print pass descriptions when they are available 3166 3167 OS << "Module passes:\n"; 3168 #define MODULE_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3169 #include "PassRegistry.def" 3170 3171 OS << "Module analyses:\n"; 3172 #define MODULE_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3173 #include "PassRegistry.def" 3174 3175 OS << "Module alias analyses:\n"; 3176 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3177 #include "PassRegistry.def" 3178 3179 OS << "CGSCC passes:\n"; 3180 #define CGSCC_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3181 #include "PassRegistry.def" 3182 3183 OS << "CGSCC analyses:\n"; 3184 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3185 #include "PassRegistry.def" 3186 3187 OS << "Function passes:\n"; 3188 #define FUNCTION_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3189 #include "PassRegistry.def" 3190 3191 OS << "Function analyses:\n"; 3192 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3193 #include "PassRegistry.def" 3194 3195 OS << "Function alias analyses:\n"; 3196 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3197 #include "PassRegistry.def" 3198 3199 OS << "Loop passes:\n"; 3200 #define LOOP_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3201 #include "PassRegistry.def" 3202 3203 OS << "Loop analyses:\n"; 3204 #define LOOP_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3205 #include "PassRegistry.def" 3206 } 3207 3208 void PassBuilder::registerParseTopLevelPipelineCallback( 3209 const std::function<bool(ModulePassManager &, ArrayRef<PipelineElement>)> 3210 &C) { 3211 TopLevelPipelineParsingCallbacks.push_back(C); 3212 } 3213