1 //===- Parsing, selection, and construction of pass pipelines -------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 /// \file 9 /// 10 /// This file provides the implementation of the PassBuilder based on our 11 /// static pass registry as well as related functionality. It also provides 12 /// helpers to aid in analyzing, debugging, and testing passes and pass 13 /// pipelines. 14 /// 15 //===----------------------------------------------------------------------===// 16 17 #include "llvm/Passes/PassBuilder.h" 18 #include "llvm/ADT/StringSwitch.h" 19 #include "llvm/Analysis/AliasAnalysisEvaluator.h" 20 #include "llvm/Analysis/AliasSetTracker.h" 21 #include "llvm/Analysis/AssumptionCache.h" 22 #include "llvm/Analysis/BasicAliasAnalysis.h" 23 #include "llvm/Analysis/BlockFrequencyInfo.h" 24 #include "llvm/Analysis/BranchProbabilityInfo.h" 25 #include "llvm/Analysis/CFGPrinter.h" 26 #include "llvm/Analysis/CFLAndersAliasAnalysis.h" 27 #include "llvm/Analysis/CFLSteensAliasAnalysis.h" 28 #include "llvm/Analysis/CGSCCPassManager.h" 29 #include "llvm/Analysis/CallGraph.h" 30 #include "llvm/Analysis/DDG.h" 31 #include "llvm/Analysis/DDGPrinter.h" 32 #include "llvm/Analysis/Delinearization.h" 33 #include "llvm/Analysis/DemandedBits.h" 34 #include "llvm/Analysis/DependenceAnalysis.h" 35 #include "llvm/Analysis/DivergenceAnalysis.h" 36 #include "llvm/Analysis/DominanceFrontier.h" 37 #include "llvm/Analysis/FunctionPropertiesAnalysis.h" 38 #include "llvm/Analysis/GlobalsModRef.h" 39 #include "llvm/Analysis/IRSimilarityIdentifier.h" 40 #include "llvm/Analysis/IVUsers.h" 41 #include "llvm/Analysis/InlineAdvisor.h" 42 #include "llvm/Analysis/InlineSizeEstimatorAnalysis.h" 43 #include "llvm/Analysis/InstCount.h" 44 #include "llvm/Analysis/LazyCallGraph.h" 45 #include "llvm/Analysis/LazyValueInfo.h" 46 #include "llvm/Analysis/Lint.h" 47 #include "llvm/Analysis/LoopAccessAnalysis.h" 48 #include "llvm/Analysis/LoopCacheAnalysis.h" 49 #include "llvm/Analysis/LoopInfo.h" 50 #include "llvm/Analysis/LoopNestAnalysis.h" 51 #include "llvm/Analysis/MemDerefPrinter.h" 52 #include "llvm/Analysis/MemoryDependenceAnalysis.h" 53 #include "llvm/Analysis/MemorySSA.h" 54 #include "llvm/Analysis/ModuleDebugInfoPrinter.h" 55 #include "llvm/Analysis/ModuleSummaryAnalysis.h" 56 #include "llvm/Analysis/MustExecute.h" 57 #include "llvm/Analysis/ObjCARCAliasAnalysis.h" 58 #include "llvm/Analysis/OptimizationRemarkEmitter.h" 59 #include "llvm/Analysis/PhiValues.h" 60 #include "llvm/Analysis/PostDominators.h" 61 #include "llvm/Analysis/ProfileSummaryInfo.h" 62 #include "llvm/Analysis/RegionInfo.h" 63 #include "llvm/Analysis/ScalarEvolution.h" 64 #include "llvm/Analysis/ScalarEvolutionAliasAnalysis.h" 65 #include "llvm/Analysis/ScopedNoAliasAA.h" 66 #include "llvm/Analysis/StackLifetime.h" 67 #include "llvm/Analysis/StackSafetyAnalysis.h" 68 #include "llvm/Analysis/TargetLibraryInfo.h" 69 #include "llvm/Analysis/TargetTransformInfo.h" 70 #include "llvm/Analysis/TypeBasedAliasAnalysis.h" 71 #include "llvm/IR/Dominators.h" 72 #include "llvm/IR/IRPrintingPasses.h" 73 #include "llvm/IR/PassManager.h" 74 #include "llvm/IR/PrintPasses.h" 75 #include "llvm/IR/SafepointIRVerifier.h" 76 #include "llvm/IR/Verifier.h" 77 #include "llvm/Support/CommandLine.h" 78 #include "llvm/Support/Debug.h" 79 #include "llvm/Support/ErrorHandling.h" 80 #include "llvm/Support/FormatVariadic.h" 81 #include "llvm/Support/Regex.h" 82 #include "llvm/Target/TargetMachine.h" 83 #include "llvm/Transforms/AggressiveInstCombine/AggressiveInstCombine.h" 84 #include "llvm/Transforms/Coroutines/CoroCleanup.h" 85 #include "llvm/Transforms/Coroutines/CoroEarly.h" 86 #include "llvm/Transforms/Coroutines/CoroElide.h" 87 #include "llvm/Transforms/Coroutines/CoroSplit.h" 88 #include "llvm/Transforms/IPO/AlwaysInliner.h" 89 #include "llvm/Transforms/IPO/Annotation2Metadata.h" 90 #include "llvm/Transforms/IPO/ArgumentPromotion.h" 91 #include "llvm/Transforms/IPO/Attributor.h" 92 #include "llvm/Transforms/IPO/BlockExtractor.h" 93 #include "llvm/Transforms/IPO/CalledValuePropagation.h" 94 #include "llvm/Transforms/IPO/ConstantMerge.h" 95 #include "llvm/Transforms/IPO/CrossDSOCFI.h" 96 #include "llvm/Transforms/IPO/DeadArgumentElimination.h" 97 #include "llvm/Transforms/IPO/ElimAvailExtern.h" 98 #include "llvm/Transforms/IPO/ForceFunctionAttrs.h" 99 #include "llvm/Transforms/IPO/FunctionAttrs.h" 100 #include "llvm/Transforms/IPO/FunctionImport.h" 101 #include "llvm/Transforms/IPO/GlobalDCE.h" 102 #include "llvm/Transforms/IPO/GlobalOpt.h" 103 #include "llvm/Transforms/IPO/GlobalSplit.h" 104 #include "llvm/Transforms/IPO/HotColdSplitting.h" 105 #include "llvm/Transforms/IPO/IROutliner.h" 106 #include "llvm/Transforms/IPO/InferFunctionAttrs.h" 107 #include "llvm/Transforms/IPO/Inliner.h" 108 #include "llvm/Transforms/IPO/Internalize.h" 109 #include "llvm/Transforms/IPO/LoopExtractor.h" 110 #include "llvm/Transforms/IPO/LowerTypeTests.h" 111 #include "llvm/Transforms/IPO/MergeFunctions.h" 112 #include "llvm/Transforms/IPO/OpenMPOpt.h" 113 #include "llvm/Transforms/IPO/PartialInlining.h" 114 #include "llvm/Transforms/IPO/SCCP.h" 115 #include "llvm/Transforms/IPO/SampleProfile.h" 116 #include "llvm/Transforms/IPO/SampleProfileProbe.h" 117 #include "llvm/Transforms/IPO/StripDeadPrototypes.h" 118 #include "llvm/Transforms/IPO/StripSymbols.h" 119 #include "llvm/Transforms/IPO/SyntheticCountsPropagation.h" 120 #include "llvm/Transforms/IPO/WholeProgramDevirt.h" 121 #include "llvm/Transforms/InstCombine/InstCombine.h" 122 #include "llvm/Transforms/Instrumentation.h" 123 #include "llvm/Transforms/Instrumentation/AddressSanitizer.h" 124 #include "llvm/Transforms/Instrumentation/BoundsChecking.h" 125 #include "llvm/Transforms/Instrumentation/CGProfile.h" 126 #include "llvm/Transforms/Instrumentation/ControlHeightReduction.h" 127 #include "llvm/Transforms/Instrumentation/DataFlowSanitizer.h" 128 #include "llvm/Transforms/Instrumentation/GCOVProfiler.h" 129 #include "llvm/Transforms/Instrumentation/HWAddressSanitizer.h" 130 #include "llvm/Transforms/Instrumentation/InstrOrderFile.h" 131 #include "llvm/Transforms/Instrumentation/InstrProfiling.h" 132 #include "llvm/Transforms/Instrumentation/MemProfiler.h" 133 #include "llvm/Transforms/Instrumentation/MemorySanitizer.h" 134 #include "llvm/Transforms/Instrumentation/PGOInstrumentation.h" 135 #include "llvm/Transforms/Instrumentation/PoisonChecking.h" 136 #include "llvm/Transforms/Instrumentation/SanitizerCoverage.h" 137 #include "llvm/Transforms/Instrumentation/ThreadSanitizer.h" 138 #include "llvm/Transforms/ObjCARC.h" 139 #include "llvm/Transforms/Scalar/ADCE.h" 140 #include "llvm/Transforms/Scalar/AlignmentFromAssumptions.h" 141 #include "llvm/Transforms/Scalar/AnnotationRemarks.h" 142 #include "llvm/Transforms/Scalar/BDCE.h" 143 #include "llvm/Transforms/Scalar/CallSiteSplitting.h" 144 #include "llvm/Transforms/Scalar/ConstantHoisting.h" 145 #include "llvm/Transforms/Scalar/ConstraintElimination.h" 146 #include "llvm/Transforms/Scalar/CorrelatedValuePropagation.h" 147 #include "llvm/Transforms/Scalar/DCE.h" 148 #include "llvm/Transforms/Scalar/DeadStoreElimination.h" 149 #include "llvm/Transforms/Scalar/DivRemPairs.h" 150 #include "llvm/Transforms/Scalar/EarlyCSE.h" 151 #include "llvm/Transforms/Scalar/Float2Int.h" 152 #include "llvm/Transforms/Scalar/GVN.h" 153 #include "llvm/Transforms/Scalar/GuardWidening.h" 154 #include "llvm/Transforms/Scalar/IVUsersPrinter.h" 155 #include "llvm/Transforms/Scalar/IndVarSimplify.h" 156 #include "llvm/Transforms/Scalar/InductiveRangeCheckElimination.h" 157 #include "llvm/Transforms/Scalar/InferAddressSpaces.h" 158 #include "llvm/Transforms/Scalar/InstSimplifyPass.h" 159 #include "llvm/Transforms/Scalar/JumpThreading.h" 160 #include "llvm/Transforms/Scalar/LICM.h" 161 #include "llvm/Transforms/Scalar/LoopAccessAnalysisPrinter.h" 162 #include "llvm/Transforms/Scalar/LoopDataPrefetch.h" 163 #include "llvm/Transforms/Scalar/LoopDeletion.h" 164 #include "llvm/Transforms/Scalar/LoopDistribute.h" 165 #include "llvm/Transforms/Scalar/LoopFlatten.h" 166 #include "llvm/Transforms/Scalar/LoopFuse.h" 167 #include "llvm/Transforms/Scalar/LoopIdiomRecognize.h" 168 #include "llvm/Transforms/Scalar/LoopInstSimplify.h" 169 #include "llvm/Transforms/Scalar/LoopInterchange.h" 170 #include "llvm/Transforms/Scalar/LoopLoadElimination.h" 171 #include "llvm/Transforms/Scalar/LoopPassManager.h" 172 #include "llvm/Transforms/Scalar/LoopPredication.h" 173 #include "llvm/Transforms/Scalar/LoopReroll.h" 174 #include "llvm/Transforms/Scalar/LoopRotation.h" 175 #include "llvm/Transforms/Scalar/LoopSimplifyCFG.h" 176 #include "llvm/Transforms/Scalar/LoopSink.h" 177 #include "llvm/Transforms/Scalar/LoopStrengthReduce.h" 178 #include "llvm/Transforms/Scalar/LoopUnrollAndJamPass.h" 179 #include "llvm/Transforms/Scalar/LoopUnrollPass.h" 180 #include "llvm/Transforms/Scalar/LoopVersioningLICM.h" 181 #include "llvm/Transforms/Scalar/LowerAtomic.h" 182 #include "llvm/Transforms/Scalar/LowerConstantIntrinsics.h" 183 #include "llvm/Transforms/Scalar/LowerExpectIntrinsic.h" 184 #include "llvm/Transforms/Scalar/LowerGuardIntrinsic.h" 185 #include "llvm/Transforms/Scalar/LowerMatrixIntrinsics.h" 186 #include "llvm/Transforms/Scalar/LowerWidenableCondition.h" 187 #include "llvm/Transforms/Scalar/MakeGuardsExplicit.h" 188 #include "llvm/Transforms/Scalar/MemCpyOptimizer.h" 189 #include "llvm/Transforms/Scalar/MergeICmps.h" 190 #include "llvm/Transforms/Scalar/MergedLoadStoreMotion.h" 191 #include "llvm/Transforms/Scalar/NaryReassociate.h" 192 #include "llvm/Transforms/Scalar/NewGVN.h" 193 #include "llvm/Transforms/Scalar/PartiallyInlineLibCalls.h" 194 #include "llvm/Transforms/Scalar/Reassociate.h" 195 #include "llvm/Transforms/Scalar/Reg2Mem.h" 196 #include "llvm/Transforms/Scalar/RewriteStatepointsForGC.h" 197 #include "llvm/Transforms/Scalar/SCCP.h" 198 #include "llvm/Transforms/Scalar/SROA.h" 199 #include "llvm/Transforms/Scalar/ScalarizeMaskedMemIntrin.h" 200 #include "llvm/Transforms/Scalar/Scalarizer.h" 201 #include "llvm/Transforms/Scalar/SeparateConstOffsetFromGEP.h" 202 #include "llvm/Transforms/Scalar/SimpleLoopUnswitch.h" 203 #include "llvm/Transforms/Scalar/SimplifyCFG.h" 204 #include "llvm/Transforms/Scalar/Sink.h" 205 #include "llvm/Transforms/Scalar/SpeculateAroundPHIs.h" 206 #include "llvm/Transforms/Scalar/SpeculativeExecution.h" 207 #include "llvm/Transforms/Scalar/StraightLineStrengthReduce.h" 208 #include "llvm/Transforms/Scalar/StructurizeCFG.h" 209 #include "llvm/Transforms/Scalar/TailRecursionElimination.h" 210 #include "llvm/Transforms/Scalar/WarnMissedTransforms.h" 211 #include "llvm/Transforms/Utils/AddDiscriminators.h" 212 #include "llvm/Transforms/Utils/AssumeBundleBuilder.h" 213 #include "llvm/Transforms/Utils/BreakCriticalEdges.h" 214 #include "llvm/Transforms/Utils/CanonicalizeAliases.h" 215 #include "llvm/Transforms/Utils/CanonicalizeFreezeInLoops.h" 216 #include "llvm/Transforms/Utils/EntryExitInstrumenter.h" 217 #include "llvm/Transforms/Utils/FixIrreducible.h" 218 #include "llvm/Transforms/Utils/HelloWorld.h" 219 #include "llvm/Transforms/Utils/InjectTLIMappings.h" 220 #include "llvm/Transforms/Utils/InstructionNamer.h" 221 #include "llvm/Transforms/Utils/LCSSA.h" 222 #include "llvm/Transforms/Utils/LibCallsShrinkWrap.h" 223 #include "llvm/Transforms/Utils/LoopSimplify.h" 224 #include "llvm/Transforms/Utils/LoopVersioning.h" 225 #include "llvm/Transforms/Utils/LowerInvoke.h" 226 #include "llvm/Transforms/Utils/LowerSwitch.h" 227 #include "llvm/Transforms/Utils/Mem2Reg.h" 228 #include "llvm/Transforms/Utils/MetaRenamer.h" 229 #include "llvm/Transforms/Utils/NameAnonGlobals.h" 230 #include "llvm/Transforms/Utils/RelLookupTableConverter.h" 231 #include "llvm/Transforms/Utils/StripGCRelocates.h" 232 #include "llvm/Transforms/Utils/StripNonLineTableDebugInfo.h" 233 #include "llvm/Transforms/Utils/SymbolRewriter.h" 234 #include "llvm/Transforms/Utils/UnifyFunctionExitNodes.h" 235 #include "llvm/Transforms/Utils/UnifyLoopExits.h" 236 #include "llvm/Transforms/Vectorize/LoadStoreVectorizer.h" 237 #include "llvm/Transforms/Vectorize/LoopVectorize.h" 238 #include "llvm/Transforms/Vectorize/SLPVectorizer.h" 239 #include "llvm/Transforms/Vectorize/VectorCombine.h" 240 241 using namespace llvm; 242 243 static cl::opt<InliningAdvisorMode> UseInlineAdvisor( 244 "enable-ml-inliner", cl::init(InliningAdvisorMode::Default), cl::Hidden, 245 cl::desc("Enable ML policy for inliner. Currently trained for -Oz only"), 246 cl::values(clEnumValN(InliningAdvisorMode::Default, "default", 247 "Heuristics-based inliner version."), 248 clEnumValN(InliningAdvisorMode::Development, "development", 249 "Use development mode (runtime-loadable model)."), 250 clEnumValN(InliningAdvisorMode::Release, "release", 251 "Use release mode (AOT-compiled model)."))); 252 253 static cl::opt<bool> EnableSyntheticCounts( 254 "enable-npm-synthetic-counts", cl::init(false), cl::Hidden, cl::ZeroOrMore, 255 cl::desc("Run synthetic function entry count generation " 256 "pass")); 257 258 static const Regex DefaultAliasRegex( 259 "^(default|thinlto-pre-link|thinlto|lto-pre-link|lto)<(O[0123sz])>$"); 260 261 /// Flag to enable inline deferral during PGO. 262 static cl::opt<bool> 263 EnablePGOInlineDeferral("enable-npm-pgo-inline-deferral", cl::init(true), 264 cl::Hidden, 265 cl::desc("Enable inline deferral during PGO")); 266 267 static cl::opt<bool> EnableMemProfiler("enable-mem-prof", cl::init(false), 268 cl::Hidden, cl::ZeroOrMore, 269 cl::desc("Enable memory profiler")); 270 271 static cl::opt<bool> PerformMandatoryInliningsFirst( 272 "mandatory-inlining-first", cl::init(true), cl::Hidden, cl::ZeroOrMore, 273 cl::desc("Perform mandatory inlinings module-wide, before performing " 274 "inlining.")); 275 276 static cl::opt<bool> EnableO3NonTrivialUnswitching( 277 "enable-npm-O3-nontrivial-unswitch", cl::init(true), cl::Hidden, 278 cl::ZeroOrMore, cl::desc("Enable non-trivial loop unswitching for -O3")); 279 280 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> EnableGVNHoist; 297 extern cl::opt<bool> EnableGVNSink; 298 extern cl::opt<bool> EnableHotColdSplit; 299 extern cl::opt<bool> EnableIROutliner; 300 extern cl::opt<bool> EnableOrderFileInstrumentation; 301 extern cl::opt<bool> EnableCHR; 302 extern cl::opt<bool> EnableLoopInterchange; 303 extern cl::opt<bool> EnableUnrollAndJam; 304 extern cl::opt<bool> EnableLoopFlatten; 305 extern cl::opt<bool> RunNewGVN; 306 extern cl::opt<bool> RunPartialInlining; 307 extern cl::opt<bool> ExtraVectorizerPasses; 308 309 extern cl::opt<bool> FlattenedProfileUsed; 310 311 extern cl::opt<AttributorRunOption> AttributorRun; 312 extern cl::opt<bool> EnableKnowledgeRetention; 313 314 extern cl::opt<bool> EnableMatrix; 315 316 extern cl::opt<bool> DisablePreInliner; 317 extern cl::opt<int> PreInlineThreshold; 318 } // namespace llvm 319 320 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O0 = { 321 /*SpeedLevel*/ 0, 322 /*SizeLevel*/ 0}; 323 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O1 = { 324 /*SpeedLevel*/ 1, 325 /*SizeLevel*/ 0}; 326 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O2 = { 327 /*SpeedLevel*/ 2, 328 /*SizeLevel*/ 0}; 329 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O3 = { 330 /*SpeedLevel*/ 3, 331 /*SizeLevel*/ 0}; 332 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::Os = { 333 /*SpeedLevel*/ 2, 334 /*SizeLevel*/ 1}; 335 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::Oz = { 336 /*SpeedLevel*/ 2, 337 /*SizeLevel*/ 2}; 338 339 namespace { 340 341 // The following passes/analyses have custom names, otherwise their name will 342 // include `(anonymous namespace)`. These are special since they are only for 343 // testing purposes and don't live in a header file. 344 345 /// No-op module pass which does nothing. 346 struct NoOpModulePass : PassInfoMixin<NoOpModulePass> { 347 PreservedAnalyses run(Module &M, ModuleAnalysisManager &) { 348 return PreservedAnalyses::all(); 349 } 350 351 static StringRef name() { return "NoOpModulePass"; } 352 }; 353 354 /// No-op module analysis. 355 class NoOpModuleAnalysis : public AnalysisInfoMixin<NoOpModuleAnalysis> { 356 friend AnalysisInfoMixin<NoOpModuleAnalysis>; 357 static AnalysisKey Key; 358 359 public: 360 struct Result {}; 361 Result run(Module &, ModuleAnalysisManager &) { return Result(); } 362 static StringRef name() { return "NoOpModuleAnalysis"; } 363 }; 364 365 /// No-op CGSCC pass which does nothing. 366 struct NoOpCGSCCPass : PassInfoMixin<NoOpCGSCCPass> { 367 PreservedAnalyses run(LazyCallGraph::SCC &C, CGSCCAnalysisManager &, 368 LazyCallGraph &, CGSCCUpdateResult &UR) { 369 return PreservedAnalyses::all(); 370 } 371 static StringRef name() { return "NoOpCGSCCPass"; } 372 }; 373 374 /// No-op CGSCC analysis. 375 class NoOpCGSCCAnalysis : public AnalysisInfoMixin<NoOpCGSCCAnalysis> { 376 friend AnalysisInfoMixin<NoOpCGSCCAnalysis>; 377 static AnalysisKey Key; 378 379 public: 380 struct Result {}; 381 Result run(LazyCallGraph::SCC &, CGSCCAnalysisManager &, LazyCallGraph &G) { 382 return Result(); 383 } 384 static StringRef name() { return "NoOpCGSCCAnalysis"; } 385 }; 386 387 /// No-op function pass which does nothing. 388 struct NoOpFunctionPass : PassInfoMixin<NoOpFunctionPass> { 389 PreservedAnalyses run(Function &F, FunctionAnalysisManager &) { 390 return PreservedAnalyses::all(); 391 } 392 static StringRef name() { return "NoOpFunctionPass"; } 393 }; 394 395 /// No-op function analysis. 396 class NoOpFunctionAnalysis : public AnalysisInfoMixin<NoOpFunctionAnalysis> { 397 friend AnalysisInfoMixin<NoOpFunctionAnalysis>; 398 static AnalysisKey Key; 399 400 public: 401 struct Result {}; 402 Result run(Function &, FunctionAnalysisManager &) { return Result(); } 403 static StringRef name() { return "NoOpFunctionAnalysis"; } 404 }; 405 406 /// No-op loop pass which does nothing. 407 struct NoOpLoopPass : PassInfoMixin<NoOpLoopPass> { 408 PreservedAnalyses run(Loop &L, LoopAnalysisManager &, 409 LoopStandardAnalysisResults &, LPMUpdater &) { 410 return PreservedAnalyses::all(); 411 } 412 static StringRef name() { return "NoOpLoopPass"; } 413 }; 414 415 /// No-op loop analysis. 416 class NoOpLoopAnalysis : public AnalysisInfoMixin<NoOpLoopAnalysis> { 417 friend AnalysisInfoMixin<NoOpLoopAnalysis>; 418 static AnalysisKey Key; 419 420 public: 421 struct Result {}; 422 Result run(Loop &, LoopAnalysisManager &, LoopStandardAnalysisResults &) { 423 return Result(); 424 } 425 static StringRef name() { return "NoOpLoopAnalysis"; } 426 }; 427 428 AnalysisKey NoOpModuleAnalysis::Key; 429 AnalysisKey NoOpCGSCCAnalysis::Key; 430 AnalysisKey NoOpFunctionAnalysis::Key; 431 AnalysisKey NoOpLoopAnalysis::Key; 432 433 /// Whether or not we should populate a PassInstrumentationCallbacks's class to 434 /// pass name map. 435 /// 436 /// This is for optimization purposes so we don't populate it if we never use 437 /// it. This should be updated if new pass instrumentation wants to use the map. 438 /// We currently only use this for --print-before/after. 439 bool shouldPopulateClassToPassNames() { 440 return !printBeforePasses().empty() || !printAfterPasses().empty(); 441 } 442 443 } // namespace 444 445 PassBuilder::PassBuilder(TargetMachine *TM, PipelineTuningOptions PTO, 446 Optional<PGOOptions> PGOOpt, 447 PassInstrumentationCallbacks *PIC) 448 : TM(TM), PTO(PTO), PGOOpt(PGOOpt), PIC(PIC) { 449 if (TM) 450 TM->registerPassBuilderCallbacks(*this); 451 if (PIC && shouldPopulateClassToPassNames()) { 452 #define MODULE_PASS(NAME, CREATE_PASS) \ 453 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 454 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 455 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 456 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 457 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 458 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 459 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 460 #define LOOP_PASS(NAME, CREATE_PASS) \ 461 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 462 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 463 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 464 #define CGSCC_PASS(NAME, CREATE_PASS) \ 465 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 466 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 467 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 468 #include "PassRegistry.def" 469 } 470 } 471 472 void PassBuilder::invokePeepholeEPCallbacks( 473 FunctionPassManager &FPM, PassBuilder::OptimizationLevel Level) { 474 for (auto &C : PeepholeEPCallbacks) 475 C(FPM, Level); 476 } 477 478 void PassBuilder::registerModuleAnalyses(ModuleAnalysisManager &MAM) { 479 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 480 MAM.registerPass([&] { return CREATE_PASS; }); 481 #include "PassRegistry.def" 482 483 for (auto &C : ModuleAnalysisRegistrationCallbacks) 484 C(MAM); 485 } 486 487 void PassBuilder::registerCGSCCAnalyses(CGSCCAnalysisManager &CGAM) { 488 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 489 CGAM.registerPass([&] { return CREATE_PASS; }); 490 #include "PassRegistry.def" 491 492 for (auto &C : CGSCCAnalysisRegistrationCallbacks) 493 C(CGAM); 494 } 495 496 void PassBuilder::registerFunctionAnalyses(FunctionAnalysisManager &FAM) { 497 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 498 FAM.registerPass([&] { return CREATE_PASS; }); 499 #include "PassRegistry.def" 500 501 for (auto &C : FunctionAnalysisRegistrationCallbacks) 502 C(FAM); 503 } 504 505 void PassBuilder::registerLoopAnalyses(LoopAnalysisManager &LAM) { 506 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 507 LAM.registerPass([&] { return CREATE_PASS; }); 508 #include "PassRegistry.def" 509 510 for (auto &C : LoopAnalysisRegistrationCallbacks) 511 C(LAM); 512 } 513 514 // Helper to add AnnotationRemarksPass. 515 static void addAnnotationRemarksPass(ModulePassManager &MPM) { 516 FunctionPassManager FPM; 517 FPM.addPass(AnnotationRemarksPass()); 518 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 519 } 520 521 // Helper to check if the current compilation phase is preparing for LTO 522 static bool isLTOPreLink(ThinOrFullLTOPhase Phase) { 523 return Phase == ThinOrFullLTOPhase::ThinLTOPreLink || 524 Phase == ThinOrFullLTOPhase::FullLTOPreLink; 525 } 526 527 // TODO: Investigate the cost/benefit of tail call elimination on debugging. 528 FunctionPassManager 529 PassBuilder::buildO1FunctionSimplificationPipeline(OptimizationLevel Level, 530 ThinOrFullLTOPhase Phase) { 531 532 FunctionPassManager FPM; 533 534 // Form SSA out of local memory accesses after breaking apart aggregates into 535 // scalars. 536 FPM.addPass(SROA()); 537 538 // Catch trivial redundancies 539 FPM.addPass(EarlyCSEPass(true /* Enable mem-ssa. */)); 540 541 // Hoisting of scalars and load expressions. 542 FPM.addPass(SimplifyCFGPass()); 543 FPM.addPass(InstCombinePass()); 544 545 FPM.addPass(LibCallsShrinkWrapPass()); 546 547 invokePeepholeEPCallbacks(FPM, Level); 548 549 FPM.addPass(SimplifyCFGPass()); 550 551 // Form canonically associated expression trees, and simplify the trees using 552 // basic mathematical properties. For example, this will form (nearly) 553 // minimal multiplication trees. 554 FPM.addPass(ReassociatePass()); 555 556 // Add the primary loop simplification pipeline. 557 // FIXME: Currently this is split into two loop pass pipelines because we run 558 // some function passes in between them. These can and should be removed 559 // and/or replaced by scheduling the loop pass equivalents in the correct 560 // positions. But those equivalent passes aren't powerful enough yet. 561 // Specifically, `SimplifyCFGPass` and `InstCombinePass` are currently still 562 // used. We have `LoopSimplifyCFGPass` which isn't yet powerful enough yet to 563 // fully replace `SimplifyCFGPass`, and the closest to the other we have is 564 // `LoopInstSimplify`. 565 LoopPassManager LPM1, LPM2; 566 567 // Simplify the loop body. We do this initially to clean up after other loop 568 // passes run, either when iterating on a loop or on inner loops with 569 // implications on the outer loop. 570 LPM1.addPass(LoopInstSimplifyPass()); 571 LPM1.addPass(LoopSimplifyCFGPass()); 572 573 // Try to remove as much code from the loop header as possible, 574 // to reduce amount of IR that will have to be duplicated. 575 // TODO: Investigate promotion cap for O1. 576 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 577 578 LPM1.addPass(LoopRotatePass(/* Disable header duplication */ true, 579 isLTOPreLink(Phase))); 580 // TODO: Investigate promotion cap for O1. 581 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 582 LPM1.addPass(SimpleLoopUnswitchPass()); 583 584 LPM2.addPass(LoopIdiomRecognizePass()); 585 LPM2.addPass(IndVarSimplifyPass()); 586 587 for (auto &C : LateLoopOptimizationsEPCallbacks) 588 C(LPM2, Level); 589 590 LPM2.addPass(LoopDeletionPass()); 591 592 if (EnableLoopInterchange) 593 LPM2.addPass(LoopInterchangePass()); 594 595 // Do not enable unrolling in PreLinkThinLTO phase during sample PGO 596 // because it changes IR to makes profile annotation in back compile 597 // inaccurate. The normal unroller doesn't pay attention to forced full unroll 598 // attributes so we need to make sure and allow the full unroll pass to pay 599 // attention to it. 600 if (Phase != ThinOrFullLTOPhase::ThinLTOPreLink || !PGOOpt || 601 PGOOpt->Action != PGOOptions::SampleUse) 602 LPM2.addPass(LoopFullUnrollPass(Level.getSpeedupLevel(), 603 /* OnlyWhenForced= */ !PTO.LoopUnrolling, 604 PTO.ForgetAllSCEVInLoopUnroll)); 605 606 for (auto &C : LoopOptimizerEndEPCallbacks) 607 C(LPM2, Level); 608 609 // We provide the opt remark emitter pass for LICM to use. We only need to do 610 // this once as it is immutable. 611 FPM.addPass( 612 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 613 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM1), 614 EnableMSSALoopDependency, 615 /*UseBlockFrequencyInfo=*/true)); 616 FPM.addPass(SimplifyCFGPass()); 617 FPM.addPass(InstCombinePass()); 618 if (EnableLoopFlatten) 619 FPM.addPass(createFunctionToLoopPassAdaptor(LoopFlattenPass())); 620 // The loop passes in LPM2 (LoopFullUnrollPass) do not preserve MemorySSA. 621 // *All* loop passes must preserve it, in order to be able to use it. 622 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM2), 623 /*UseMemorySSA=*/false, 624 /*UseBlockFrequencyInfo=*/false)); 625 626 // Delete small array after loop unroll. 627 FPM.addPass(SROA()); 628 629 // Specially optimize memory movement as it doesn't look like dataflow in SSA. 630 FPM.addPass(MemCpyOptPass()); 631 632 // Sparse conditional constant propagation. 633 // FIXME: It isn't clear why we do this *after* loop passes rather than 634 // before... 635 FPM.addPass(SCCPPass()); 636 637 // Delete dead bit computations (instcombine runs after to fold away the dead 638 // computations, and then ADCE will run later to exploit any new DCE 639 // opportunities that creates). 640 FPM.addPass(BDCEPass()); 641 642 // Run instcombine after redundancy and dead bit elimination to exploit 643 // opportunities opened up by them. 644 FPM.addPass(InstCombinePass()); 645 invokePeepholeEPCallbacks(FPM, Level); 646 647 if (PTO.Coroutines) 648 FPM.addPass(CoroElidePass()); 649 650 for (auto &C : ScalarOptimizerLateEPCallbacks) 651 C(FPM, Level); 652 653 // Finally, do an expensive DCE pass to catch all the dead code exposed by 654 // the simplifications and basic cleanup after all the simplifications. 655 // TODO: Investigate if this is too expensive. 656 FPM.addPass(ADCEPass()); 657 FPM.addPass(SimplifyCFGPass()); 658 FPM.addPass(InstCombinePass()); 659 invokePeepholeEPCallbacks(FPM, Level); 660 661 return FPM; 662 } 663 664 FunctionPassManager 665 PassBuilder::buildFunctionSimplificationPipeline(OptimizationLevel Level, 666 ThinOrFullLTOPhase Phase) { 667 assert(Level != OptimizationLevel::O0 && "Must request optimizations!"); 668 669 // The O1 pipeline has a separate pipeline creation function to simplify 670 // construction readability. 671 if (Level.getSpeedupLevel() == 1) 672 return buildO1FunctionSimplificationPipeline(Level, Phase); 673 674 FunctionPassManager FPM; 675 676 // Form SSA out of local memory accesses after breaking apart aggregates into 677 // scalars. 678 FPM.addPass(SROA()); 679 680 // Catch trivial redundancies 681 FPM.addPass(EarlyCSEPass(true /* Enable mem-ssa. */)); 682 if (EnableKnowledgeRetention) 683 FPM.addPass(AssumeSimplifyPass()); 684 685 // Hoisting of scalars and load expressions. 686 if (EnableGVNHoist) 687 FPM.addPass(GVNHoistPass()); 688 689 // Global value numbering based sinking. 690 if (EnableGVNSink) { 691 FPM.addPass(GVNSinkPass()); 692 FPM.addPass(SimplifyCFGPass()); 693 } 694 695 if (EnableConstraintElimination) 696 FPM.addPass(ConstraintEliminationPass()); 697 698 // Speculative execution if the target has divergent branches; otherwise nop. 699 FPM.addPass(SpeculativeExecutionPass(/* OnlyIfDivergentTarget =*/true)); 700 701 // Optimize based on known information about branches, and cleanup afterward. 702 FPM.addPass(JumpThreadingPass()); 703 FPM.addPass(CorrelatedValuePropagationPass()); 704 705 FPM.addPass(SimplifyCFGPass()); 706 if (Level == OptimizationLevel::O3) 707 FPM.addPass(AggressiveInstCombinePass()); 708 FPM.addPass(InstCombinePass()); 709 710 if (!Level.isOptimizingForSize()) 711 FPM.addPass(LibCallsShrinkWrapPass()); 712 713 invokePeepholeEPCallbacks(FPM, Level); 714 715 // For PGO use pipeline, try to optimize memory intrinsics such as memcpy 716 // using the size value profile. Don't perform this when optimizing for size. 717 if (PGOOpt && PGOOpt->Action == PGOOptions::IRUse && 718 !Level.isOptimizingForSize()) 719 FPM.addPass(PGOMemOPSizeOpt()); 720 721 FPM.addPass(TailCallElimPass()); 722 FPM.addPass(SimplifyCFGPass()); 723 724 // Form canonically associated expression trees, and simplify the trees using 725 // basic mathematical properties. For example, this will form (nearly) 726 // minimal multiplication trees. 727 FPM.addPass(ReassociatePass()); 728 729 // Add the primary loop simplification pipeline. 730 // FIXME: Currently this is split into two loop pass pipelines because we run 731 // some function passes in between them. These can and should be removed 732 // and/or replaced by scheduling the loop pass equivalents in the correct 733 // positions. But those equivalent passes aren't powerful enough yet. 734 // Specifically, `SimplifyCFGPass` and `InstCombinePass` are currently still 735 // used. We have `LoopSimplifyCFGPass` which isn't yet powerful enough yet to 736 // fully replace `SimplifyCFGPass`, and the closest to the other we have is 737 // `LoopInstSimplify`. 738 LoopPassManager LPM1, LPM2; 739 740 // Simplify the loop body. We do this initially to clean up after other loop 741 // passes run, either when iterating on a loop or on inner loops with 742 // implications on the outer loop. 743 LPM1.addPass(LoopInstSimplifyPass()); 744 LPM1.addPass(LoopSimplifyCFGPass()); 745 746 // Try to remove as much code from the loop header as possible, 747 // to reduce amount of IR that will have to be duplicated. 748 // TODO: Investigate promotion cap for O1. 749 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 750 751 // Disable header duplication in loop rotation at -Oz. 752 LPM1.addPass( 753 LoopRotatePass(Level != OptimizationLevel::Oz, isLTOPreLink(Phase))); 754 // TODO: Investigate promotion cap for O1. 755 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 756 LPM1.addPass( 757 SimpleLoopUnswitchPass(/* NonTrivial */ Level == OptimizationLevel::O3 && 758 EnableO3NonTrivialUnswitching)); 759 LPM2.addPass(LoopIdiomRecognizePass()); 760 LPM2.addPass(IndVarSimplifyPass()); 761 762 for (auto &C : LateLoopOptimizationsEPCallbacks) 763 C(LPM2, Level); 764 765 LPM2.addPass(LoopDeletionPass()); 766 767 if (EnableLoopInterchange) 768 LPM2.addPass(LoopInterchangePass()); 769 770 // Do not enable unrolling in PreLinkThinLTO phase during sample PGO 771 // because it changes IR to makes profile annotation in back compile 772 // inaccurate. The normal unroller doesn't pay attention to forced full unroll 773 // attributes so we need to make sure and allow the full unroll pass to pay 774 // attention to it. 775 if (Phase != ThinOrFullLTOPhase::ThinLTOPreLink || !PGOOpt || 776 PGOOpt->Action != PGOOptions::SampleUse) 777 LPM2.addPass(LoopFullUnrollPass(Level.getSpeedupLevel(), 778 /* OnlyWhenForced= */ !PTO.LoopUnrolling, 779 PTO.ForgetAllSCEVInLoopUnroll)); 780 781 for (auto &C : LoopOptimizerEndEPCallbacks) 782 C(LPM2, Level); 783 784 // We provide the opt remark emitter pass for LICM to use. We only need to do 785 // this once as it is immutable. 786 FPM.addPass( 787 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 788 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM1), 789 EnableMSSALoopDependency, 790 /*UseBlockFrequencyInfo=*/true)); 791 FPM.addPass(SimplifyCFGPass()); 792 FPM.addPass(InstCombinePass()); 793 if (EnableLoopFlatten) 794 FPM.addPass(createFunctionToLoopPassAdaptor(LoopFlattenPass())); 795 // The loop passes in LPM2 (LoopIdiomRecognizePass, IndVarSimplifyPass, 796 // LoopDeletionPass and LoopFullUnrollPass) do not preserve MemorySSA. 797 // *All* loop passes must preserve it, in order to be able to use it. 798 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM2), 799 /*UseMemorySSA=*/false, 800 /*UseBlockFrequencyInfo=*/false)); 801 802 // Delete small array after loop unroll. 803 FPM.addPass(SROA()); 804 805 // Eliminate redundancies. 806 FPM.addPass(MergedLoadStoreMotionPass()); 807 if (RunNewGVN) 808 FPM.addPass(NewGVNPass()); 809 else 810 FPM.addPass(GVN()); 811 812 // Sparse conditional constant propagation. 813 // FIXME: It isn't clear why we do this *after* loop passes rather than 814 // before... 815 FPM.addPass(SCCPPass()); 816 817 // Delete dead bit computations (instcombine runs after to fold away the dead 818 // computations, and then ADCE will run later to exploit any new DCE 819 // opportunities that creates). 820 FPM.addPass(BDCEPass()); 821 822 // Run instcombine after redundancy and dead bit elimination to exploit 823 // opportunities opened up by them. 824 FPM.addPass(InstCombinePass()); 825 invokePeepholeEPCallbacks(FPM, Level); 826 827 // Re-consider control flow based optimizations after redundancy elimination, 828 // redo DCE, etc. 829 FPM.addPass(JumpThreadingPass()); 830 FPM.addPass(CorrelatedValuePropagationPass()); 831 832 // Finally, do an expensive DCE pass to catch all the dead code exposed by 833 // the simplifications and basic cleanup after all the simplifications. 834 // TODO: Investigate if this is too expensive. 835 FPM.addPass(ADCEPass()); 836 837 // Specially optimize memory movement as it doesn't look like dataflow in SSA. 838 FPM.addPass(MemCpyOptPass()); 839 840 FPM.addPass(DSEPass()); 841 FPM.addPass(createFunctionToLoopPassAdaptor( 842 LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap), 843 EnableMSSALoopDependency, /*UseBlockFrequencyInfo=*/true)); 844 845 if (PTO.Coroutines) 846 FPM.addPass(CoroElidePass()); 847 848 for (auto &C : ScalarOptimizerLateEPCallbacks) 849 C(FPM, Level); 850 851 FPM.addPass(SimplifyCFGPass( 852 SimplifyCFGOptions().hoistCommonInsts(true).sinkCommonInsts(true))); 853 FPM.addPass(InstCombinePass()); 854 invokePeepholeEPCallbacks(FPM, Level); 855 856 if (EnableCHR && Level == OptimizationLevel::O3 && PGOOpt && 857 (PGOOpt->Action == PGOOptions::IRUse || 858 PGOOpt->Action == PGOOptions::SampleUse)) 859 FPM.addPass(ControlHeightReductionPass()); 860 861 return FPM; 862 } 863 864 void PassBuilder::addRequiredLTOPreLinkPasses(ModulePassManager &MPM) { 865 MPM.addPass(CanonicalizeAliasesPass()); 866 MPM.addPass(NameAnonGlobalPass()); 867 } 868 869 void PassBuilder::addPGOInstrPasses(ModulePassManager &MPM, 870 PassBuilder::OptimizationLevel Level, 871 bool RunProfileGen, bool IsCS, 872 std::string ProfileFile, 873 std::string ProfileRemappingFile) { 874 assert(Level != OptimizationLevel::O0 && "Not expecting O0 here!"); 875 if (!IsCS && !DisablePreInliner) { 876 InlineParams IP; 877 878 IP.DefaultThreshold = PreInlineThreshold; 879 880 // FIXME: The hint threshold has the same value used by the regular inliner 881 // when not optimzing for size. This should probably be lowered after 882 // performance testing. 883 // FIXME: this comment is cargo culted from the old pass manager, revisit). 884 IP.HintThreshold = Level.isOptimizingForSize() ? PreInlineThreshold : 325; 885 ModuleInlinerWrapperPass MIWP(IP); 886 CGSCCPassManager &CGPipeline = MIWP.getPM(); 887 888 FunctionPassManager FPM; 889 FPM.addPass(SROA()); 890 FPM.addPass(EarlyCSEPass()); // Catch trivial redundancies. 891 FPM.addPass(SimplifyCFGPass()); // Merge & remove basic blocks. 892 FPM.addPass(InstCombinePass()); // Combine silly sequences. 893 invokePeepholeEPCallbacks(FPM, Level); 894 895 CGPipeline.addPass(createCGSCCToFunctionPassAdaptor(std::move(FPM))); 896 897 MPM.addPass(std::move(MIWP)); 898 899 // Delete anything that is now dead to make sure that we don't instrument 900 // dead code. Instrumentation can end up keeping dead code around and 901 // dramatically increase code size. 902 MPM.addPass(GlobalDCEPass()); 903 } 904 905 if (!RunProfileGen) { 906 assert(!ProfileFile.empty() && "Profile use expecting a profile file!"); 907 MPM.addPass(PGOInstrumentationUse(ProfileFile, ProfileRemappingFile, IsCS)); 908 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 909 // RequireAnalysisPass for PSI before subsequent non-module passes. 910 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 911 return; 912 } 913 914 // Perform PGO instrumentation. 915 MPM.addPass(PGOInstrumentationGen(IsCS)); 916 917 FunctionPassManager FPM; 918 // Disable header duplication in loop rotation at -Oz. 919 FPM.addPass(createFunctionToLoopPassAdaptor( 920 LoopRotatePass(Level != OptimizationLevel::Oz), EnableMSSALoopDependency, 921 /*UseBlockFrequencyInfo=*/false)); 922 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 923 924 // Add the profile lowering pass. 925 InstrProfOptions Options; 926 if (!ProfileFile.empty()) 927 Options.InstrProfileOutput = ProfileFile; 928 // Do counter promotion at Level greater than O0. 929 Options.DoCounterPromotion = true; 930 Options.UseBFIInPromotion = IsCS; 931 MPM.addPass(InstrProfiling(Options, IsCS)); 932 } 933 934 void PassBuilder::addPGOInstrPassesForO0(ModulePassManager &MPM, 935 bool RunProfileGen, bool IsCS, 936 std::string ProfileFile, 937 std::string ProfileRemappingFile) { 938 if (!RunProfileGen) { 939 assert(!ProfileFile.empty() && "Profile use expecting a profile file!"); 940 MPM.addPass(PGOInstrumentationUse(ProfileFile, ProfileRemappingFile, IsCS)); 941 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 942 // RequireAnalysisPass for PSI before subsequent non-module passes. 943 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 944 return; 945 } 946 947 // Perform PGO instrumentation. 948 MPM.addPass(PGOInstrumentationGen(IsCS)); 949 // Add the profile lowering pass. 950 InstrProfOptions Options; 951 if (!ProfileFile.empty()) 952 Options.InstrProfileOutput = ProfileFile; 953 // Do not do counter promotion at O0. 954 Options.DoCounterPromotion = false; 955 Options.UseBFIInPromotion = IsCS; 956 MPM.addPass(InstrProfiling(Options, IsCS)); 957 } 958 959 static InlineParams 960 getInlineParamsFromOptLevel(PassBuilder::OptimizationLevel Level) { 961 return getInlineParams(Level.getSpeedupLevel(), Level.getSizeLevel()); 962 } 963 964 ModuleInlinerWrapperPass 965 PassBuilder::buildInlinerPipeline(OptimizationLevel Level, 966 ThinOrFullLTOPhase Phase) { 967 InlineParams IP = getInlineParamsFromOptLevel(Level); 968 if (Phase == ThinOrFullLTOPhase::ThinLTOPreLink && PGOOpt && 969 PGOOpt->Action == PGOOptions::SampleUse) 970 IP.HotCallSiteThreshold = 0; 971 972 if (PGOOpt) 973 IP.EnableDeferral = EnablePGOInlineDeferral; 974 975 ModuleInlinerWrapperPass MIWP(IP, PerformMandatoryInliningsFirst, 976 UseInlineAdvisor, MaxDevirtIterations); 977 978 // Require the GlobalsAA analysis for the module so we can query it within 979 // the CGSCC pipeline. 980 MIWP.addModulePass(RequireAnalysisPass<GlobalsAA, Module>()); 981 // Invalidate AAManager so it can be recreated and pick up the newly available 982 // GlobalsAA. 983 MIWP.addModulePass( 984 createModuleToFunctionPassAdaptor(InvalidateAnalysisPass<AAManager>())); 985 986 // Require the ProfileSummaryAnalysis for the module so we can query it within 987 // the inliner pass. 988 MIWP.addModulePass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 989 990 // Now begin the main postorder CGSCC pipeline. 991 // FIXME: The current CGSCC pipeline has its origins in the legacy pass 992 // manager and trying to emulate its precise behavior. Much of this doesn't 993 // make a lot of sense and we should revisit the core CGSCC structure. 994 CGSCCPassManager &MainCGPipeline = MIWP.getPM(); 995 996 // Note: historically, the PruneEH pass was run first to deduce nounwind and 997 // generally clean up exception handling overhead. It isn't clear this is 998 // valuable as the inliner doesn't currently care whether it is inlining an 999 // invoke or a call. 1000 1001 if (AttributorRun & AttributorRunOption::CGSCC) 1002 MainCGPipeline.addPass(AttributorCGSCCPass()); 1003 1004 if (PTO.Coroutines) 1005 MainCGPipeline.addPass(CoroSplitPass(Level != OptimizationLevel::O0)); 1006 1007 // Now deduce any function attributes based in the current code. 1008 MainCGPipeline.addPass(PostOrderFunctionAttrsPass()); 1009 1010 // When at O3 add argument promotion to the pass pipeline. 1011 // FIXME: It isn't at all clear why this should be limited to O3. 1012 if (Level == OptimizationLevel::O3) 1013 MainCGPipeline.addPass(ArgumentPromotionPass()); 1014 1015 // Try to perform OpenMP specific optimizations. This is a (quick!) no-op if 1016 // there are no OpenMP runtime calls present in the module. 1017 if (Level == OptimizationLevel::O2 || Level == OptimizationLevel::O3) 1018 MainCGPipeline.addPass(OpenMPOptCGSCCPass()); 1019 1020 for (auto &C : CGSCCOptimizerLateEPCallbacks) 1021 C(MainCGPipeline, Level); 1022 1023 // Lastly, add the core function simplification pipeline nested inside the 1024 // CGSCC walk. 1025 MainCGPipeline.addPass(createCGSCCToFunctionPassAdaptor( 1026 buildFunctionSimplificationPipeline(Level, Phase))); 1027 1028 return MIWP; 1029 } 1030 1031 ModulePassManager 1032 PassBuilder::buildModuleSimplificationPipeline(OptimizationLevel Level, 1033 ThinOrFullLTOPhase Phase) { 1034 ModulePassManager MPM; 1035 1036 // Place pseudo probe instrumentation as the first pass of the pipeline to 1037 // minimize the impact of optimization changes. 1038 if (PGOOpt && PGOOpt->PseudoProbeForProfiling && 1039 Phase != ThinOrFullLTOPhase::ThinLTOPostLink) 1040 MPM.addPass(SampleProfileProbePass(TM)); 1041 1042 bool HasSampleProfile = PGOOpt && (PGOOpt->Action == PGOOptions::SampleUse); 1043 1044 // In ThinLTO mode, when flattened profile is used, all the available 1045 // profile information will be annotated in PreLink phase so there is 1046 // no need to load the profile again in PostLink. 1047 bool LoadSampleProfile = 1048 HasSampleProfile && 1049 !(FlattenedProfileUsed && Phase == ThinOrFullLTOPhase::ThinLTOPostLink); 1050 1051 // During the ThinLTO backend phase we perform early indirect call promotion 1052 // here, before globalopt. Otherwise imported available_externally functions 1053 // look unreferenced and are removed. If we are going to load the sample 1054 // profile then defer until later. 1055 // TODO: See if we can move later and consolidate with the location where 1056 // we perform ICP when we are loading a sample profile. 1057 // TODO: We pass HasSampleProfile (whether there was a sample profile file 1058 // passed to the compile) to the SamplePGO flag of ICP. This is used to 1059 // determine whether the new direct calls are annotated with prof metadata. 1060 // Ideally this should be determined from whether the IR is annotated with 1061 // sample profile, and not whether the a sample profile was provided on the 1062 // command line. E.g. for flattened profiles where we will not be reloading 1063 // the sample profile in the ThinLTO backend, we ideally shouldn't have to 1064 // provide the sample profile file. 1065 if (Phase == ThinOrFullLTOPhase::ThinLTOPostLink && !LoadSampleProfile) 1066 MPM.addPass(PGOIndirectCallPromotion(true /* InLTO */, HasSampleProfile)); 1067 1068 // Do basic inference of function attributes from known properties of system 1069 // libraries and other oracles. 1070 MPM.addPass(InferFunctionAttrsPass()); 1071 1072 // Create an early function pass manager to cleanup the output of the 1073 // frontend. 1074 FunctionPassManager EarlyFPM; 1075 // Lower llvm.expect to metadata before attempting transforms. 1076 // Compare/branch metadata may alter the behavior of passes like SimplifyCFG. 1077 EarlyFPM.addPass(LowerExpectIntrinsicPass()); 1078 EarlyFPM.addPass(SimplifyCFGPass()); 1079 EarlyFPM.addPass(SROA()); 1080 EarlyFPM.addPass(EarlyCSEPass()); 1081 if (PTO.Coroutines) 1082 EarlyFPM.addPass(CoroEarlyPass()); 1083 if (Level == OptimizationLevel::O3) 1084 EarlyFPM.addPass(CallSiteSplittingPass()); 1085 1086 // In SamplePGO ThinLTO backend, we need instcombine before profile annotation 1087 // to convert bitcast to direct calls so that they can be inlined during the 1088 // profile annotation prepration step. 1089 // More details about SamplePGO design can be found in: 1090 // https://research.google.com/pubs/pub45290.html 1091 // FIXME: revisit how SampleProfileLoad/Inliner/ICP is structured. 1092 if (LoadSampleProfile) 1093 EarlyFPM.addPass(InstCombinePass()); 1094 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(EarlyFPM))); 1095 1096 if (LoadSampleProfile) { 1097 // Annotate sample profile right after early FPM to ensure freshness of 1098 // the debug info. 1099 MPM.addPass(SampleProfileLoaderPass(PGOOpt->ProfileFile, 1100 PGOOpt->ProfileRemappingFile, Phase)); 1101 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 1102 // RequireAnalysisPass for PSI before subsequent non-module passes. 1103 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 1104 // Do not invoke ICP in the LTOPrelink phase as it makes it hard 1105 // for the profile annotation to be accurate in the LTO backend. 1106 if (Phase != ThinOrFullLTOPhase::ThinLTOPreLink && 1107 Phase != ThinOrFullLTOPhase::FullLTOPreLink) 1108 // We perform early indirect call promotion here, before globalopt. 1109 // This is important for the ThinLTO backend phase because otherwise 1110 // imported available_externally functions look unreferenced and are 1111 // removed. 1112 MPM.addPass( 1113 PGOIndirectCallPromotion(true /* IsInLTO */, true /* SamplePGO */)); 1114 } 1115 1116 // Try to perform OpenMP specific optimizations on the module. This is a 1117 // (quick!) no-op if there are no OpenMP runtime calls present in the module. 1118 if (Level == OptimizationLevel::O2 || Level == OptimizationLevel::O3) 1119 MPM.addPass(OpenMPOptPass()); 1120 1121 if (AttributorRun & AttributorRunOption::MODULE) 1122 MPM.addPass(AttributorPass()); 1123 1124 // Lower type metadata and the type.test intrinsic in the ThinLTO 1125 // post link pipeline after ICP. This is to enable usage of the type 1126 // tests in ICP sequences. 1127 if (Phase == ThinOrFullLTOPhase::ThinLTOPostLink) 1128 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1129 1130 for (auto &C : PipelineEarlySimplificationEPCallbacks) 1131 C(MPM, Level); 1132 1133 // Interprocedural constant propagation now that basic cleanup has occurred 1134 // and prior to optimizing globals. 1135 // FIXME: This position in the pipeline hasn't been carefully considered in 1136 // years, it should be re-analyzed. 1137 MPM.addPass(IPSCCPPass()); 1138 1139 // Attach metadata to indirect call sites indicating the set of functions 1140 // they may target at run-time. This should follow IPSCCP. 1141 MPM.addPass(CalledValuePropagationPass()); 1142 1143 // Optimize globals to try and fold them into constants. 1144 MPM.addPass(GlobalOptPass()); 1145 1146 // Promote any localized globals to SSA registers. 1147 // FIXME: Should this instead by a run of SROA? 1148 // FIXME: We should probably run instcombine and simplify-cfg afterward to 1149 // delete control flows that are dead once globals have been folded to 1150 // constants. 1151 MPM.addPass(createModuleToFunctionPassAdaptor(PromotePass())); 1152 1153 // Remove any dead arguments exposed by cleanups and constant folding 1154 // globals. 1155 MPM.addPass(DeadArgumentEliminationPass()); 1156 1157 // Create a small function pass pipeline to cleanup after all the global 1158 // optimizations. 1159 FunctionPassManager GlobalCleanupPM; 1160 GlobalCleanupPM.addPass(InstCombinePass()); 1161 invokePeepholeEPCallbacks(GlobalCleanupPM, Level); 1162 1163 GlobalCleanupPM.addPass(SimplifyCFGPass()); 1164 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(GlobalCleanupPM))); 1165 1166 // Add all the requested passes for instrumentation PGO, if requested. 1167 if (PGOOpt && Phase != ThinOrFullLTOPhase::ThinLTOPostLink && 1168 (PGOOpt->Action == PGOOptions::IRInstr || 1169 PGOOpt->Action == PGOOptions::IRUse)) { 1170 addPGOInstrPasses(MPM, Level, 1171 /* RunProfileGen */ PGOOpt->Action == PGOOptions::IRInstr, 1172 /* IsCS */ false, PGOOpt->ProfileFile, 1173 PGOOpt->ProfileRemappingFile); 1174 MPM.addPass(PGOIndirectCallPromotion(false, false)); 1175 } 1176 if (PGOOpt && Phase != ThinOrFullLTOPhase::ThinLTOPostLink && 1177 PGOOpt->CSAction == PGOOptions::CSIRInstr) 1178 MPM.addPass(PGOInstrumentationGenCreateVar(PGOOpt->CSProfileGenFile)); 1179 1180 // Synthesize function entry counts for non-PGO compilation. 1181 if (EnableSyntheticCounts && !PGOOpt) 1182 MPM.addPass(SyntheticCountsPropagation()); 1183 1184 MPM.addPass(buildInlinerPipeline(Level, Phase)); 1185 1186 if (EnableMemProfiler && Phase != ThinOrFullLTOPhase::ThinLTOPreLink) { 1187 MPM.addPass(createModuleToFunctionPassAdaptor(MemProfilerPass())); 1188 MPM.addPass(ModuleMemProfilerPass()); 1189 } 1190 1191 return MPM; 1192 } 1193 1194 /// TODO: Should LTO cause any differences to this set of passes? 1195 void PassBuilder::addVectorPasses(OptimizationLevel Level, 1196 FunctionPassManager &FPM, bool IsLTO) { 1197 FPM.addPass(LoopVectorizePass( 1198 LoopVectorizeOptions(!PTO.LoopInterleaving, !PTO.LoopVectorization))); 1199 1200 if (IsLTO) { 1201 // The vectorizer may have significantly shortened a loop body; unroll 1202 // again. Unroll small loops to hide loop backedge latency and saturate any 1203 // parallel execution resources of an out-of-order processor. We also then 1204 // need to clean up redundancies and loop invariant code. 1205 // FIXME: It would be really good to use a loop-integrated instruction 1206 // combiner for cleanup here so that the unrolling and LICM can be pipelined 1207 // across the loop nests. 1208 // We do UnrollAndJam in a separate LPM to ensure it happens before unroll 1209 if (EnableUnrollAndJam && PTO.LoopUnrolling) 1210 FPM.addPass(createFunctionToLoopPassAdaptor( 1211 LoopUnrollAndJamPass(Level.getSpeedupLevel()))); 1212 FPM.addPass(LoopUnrollPass(LoopUnrollOptions( 1213 Level.getSpeedupLevel(), /*OnlyWhenForced=*/!PTO.LoopUnrolling, 1214 PTO.ForgetAllSCEVInLoopUnroll))); 1215 FPM.addPass(WarnMissedTransformationsPass()); 1216 } 1217 1218 if (!IsLTO) { 1219 // Eliminate loads by forwarding stores from the previous iteration to loads 1220 // of the current iteration. 1221 FPM.addPass(LoopLoadEliminationPass()); 1222 } 1223 // Cleanup after the loop optimization passes. 1224 FPM.addPass(InstCombinePass()); 1225 1226 if (Level.getSpeedupLevel() > 1 && ExtraVectorizerPasses) { 1227 // At higher optimization levels, try to clean up any runtime overlap and 1228 // alignment checks inserted by the vectorizer. We want to track correlated 1229 // runtime checks for two inner loops in the same outer loop, fold any 1230 // common computations, hoist loop-invariant aspects out of any outer loop, 1231 // and unswitch the runtime checks if possible. Once hoisted, we may have 1232 // dead (or speculatable) control flows or more combining opportunities. 1233 FPM.addPass(EarlyCSEPass()); 1234 FPM.addPass(CorrelatedValuePropagationPass()); 1235 FPM.addPass(InstCombinePass()); 1236 LoopPassManager LPM; 1237 LPM.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 1238 LPM.addPass(SimpleLoopUnswitchPass(/* NonTrivial */ Level == 1239 OptimizationLevel::O3)); 1240 FPM.addPass( 1241 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 1242 FPM.addPass(createFunctionToLoopPassAdaptor( 1243 std::move(LPM), EnableMSSALoopDependency, 1244 /*UseBlockFrequencyInfo=*/true)); 1245 FPM.addPass(SimplifyCFGPass()); 1246 FPM.addPass(InstCombinePass()); 1247 } 1248 1249 // Now that we've formed fast to execute loop structures, we do further 1250 // optimizations. These are run afterward as they might block doing complex 1251 // analyses and transforms such as what are needed for loop vectorization. 1252 1253 // Cleanup after loop vectorization, etc. Simplification passes like CVP and 1254 // GVN, loop transforms, and others have already run, so it's now better to 1255 // convert to more optimized IR using more aggressive simplify CFG options. 1256 // The extra sinking transform can create larger basic blocks, so do this 1257 // before SLP vectorization. 1258 FPM.addPass(SimplifyCFGPass(SimplifyCFGOptions() 1259 .forwardSwitchCondToPhi(true) 1260 .convertSwitchToLookupTable(true) 1261 .needCanonicalLoops(false) 1262 .hoistCommonInsts(true) 1263 .sinkCommonInsts(true))); 1264 1265 if (IsLTO) { 1266 FPM.addPass(SCCPPass()); 1267 FPM.addPass(InstCombinePass()); 1268 FPM.addPass(BDCEPass()); 1269 } 1270 1271 // Optimize parallel scalar instruction chains into SIMD instructions. 1272 if (PTO.SLPVectorization) { 1273 FPM.addPass(SLPVectorizerPass()); 1274 if (Level.getSpeedupLevel() > 1 && ExtraVectorizerPasses) { 1275 FPM.addPass(EarlyCSEPass()); 1276 } 1277 } 1278 // Enhance/cleanup vector code. 1279 FPM.addPass(VectorCombinePass()); 1280 1281 if (!IsLTO) { 1282 FPM.addPass(InstCombinePass()); 1283 // Unroll small loops to hide loop backedge latency and saturate any 1284 // parallel execution resources of an out-of-order processor. We also then 1285 // need to clean up redundancies and loop invariant code. 1286 // FIXME: It would be really good to use a loop-integrated instruction 1287 // combiner for cleanup here so that the unrolling and LICM can be pipelined 1288 // across the loop nests. 1289 // We do UnrollAndJam in a separate LPM to ensure it happens before unroll 1290 if (EnableUnrollAndJam && PTO.LoopUnrolling) { 1291 FPM.addPass(createFunctionToLoopPassAdaptor( 1292 LoopUnrollAndJamPass(Level.getSpeedupLevel()))); 1293 } 1294 FPM.addPass(LoopUnrollPass(LoopUnrollOptions( 1295 Level.getSpeedupLevel(), /*OnlyWhenForced=*/!PTO.LoopUnrolling, 1296 PTO.ForgetAllSCEVInLoopUnroll))); 1297 FPM.addPass(WarnMissedTransformationsPass()); 1298 FPM.addPass(InstCombinePass()); 1299 FPM.addPass( 1300 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 1301 FPM.addPass(createFunctionToLoopPassAdaptor( 1302 LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap), 1303 EnableMSSALoopDependency, /*UseBlockFrequencyInfo=*/true)); 1304 } 1305 1306 // Now that we've vectorized and unrolled loops, we may have more refined 1307 // alignment information, try to re-derive it here. 1308 FPM.addPass(AlignmentFromAssumptionsPass()); 1309 1310 if (IsLTO) 1311 FPM.addPass(InstCombinePass()); 1312 } 1313 1314 ModulePassManager 1315 PassBuilder::buildModuleOptimizationPipeline(OptimizationLevel Level, 1316 bool LTOPreLink) { 1317 ModulePassManager MPM; 1318 1319 // Optimize globals now that the module is fully simplified. 1320 MPM.addPass(GlobalOptPass()); 1321 MPM.addPass(GlobalDCEPass()); 1322 1323 // Run partial inlining pass to partially inline functions that have 1324 // large bodies. 1325 if (RunPartialInlining) 1326 MPM.addPass(PartialInlinerPass()); 1327 1328 // Remove avail extern fns and globals definitions since we aren't compiling 1329 // an object file for later LTO. For LTO we want to preserve these so they 1330 // are eligible for inlining at link-time. Note if they are unreferenced they 1331 // will be removed by GlobalDCE later, so this only impacts referenced 1332 // available externally globals. Eventually they will be suppressed during 1333 // codegen, but eliminating here enables more opportunity for GlobalDCE as it 1334 // may make globals referenced by available external functions dead and saves 1335 // running remaining passes on the eliminated functions. These should be 1336 // preserved during prelinking for link-time inlining decisions. 1337 if (!LTOPreLink) 1338 MPM.addPass(EliminateAvailableExternallyPass()); 1339 1340 if (EnableOrderFileInstrumentation) 1341 MPM.addPass(InstrOrderFilePass()); 1342 1343 // Do RPO function attribute inference across the module to forward-propagate 1344 // attributes where applicable. 1345 // FIXME: Is this really an optimization rather than a canonicalization? 1346 MPM.addPass(ReversePostOrderFunctionAttrsPass()); 1347 1348 // Do a post inline PGO instrumentation and use pass. This is a context 1349 // sensitive PGO pass. We don't want to do this in LTOPreLink phrase as 1350 // cross-module inline has not been done yet. The context sensitive 1351 // instrumentation is after all the inlines are done. 1352 if (!LTOPreLink && PGOOpt) { 1353 if (PGOOpt->CSAction == PGOOptions::CSIRInstr) 1354 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ true, 1355 /* IsCS */ true, PGOOpt->CSProfileGenFile, 1356 PGOOpt->ProfileRemappingFile); 1357 else if (PGOOpt->CSAction == PGOOptions::CSIRUse) 1358 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ false, 1359 /* IsCS */ true, PGOOpt->ProfileFile, 1360 PGOOpt->ProfileRemappingFile); 1361 } 1362 1363 // Re-require GloblasAA here prior to function passes. This is particularly 1364 // useful as the above will have inlined, DCE'ed, and function-attr 1365 // propagated everything. We should at this point have a reasonably minimal 1366 // and richly annotated call graph. By computing aliasing and mod/ref 1367 // information for all local globals here, the late loop passes and notably 1368 // the vectorizer will be able to use them to help recognize vectorizable 1369 // memory operations. 1370 MPM.addPass(RequireAnalysisPass<GlobalsAA, Module>()); 1371 1372 FunctionPassManager OptimizePM; 1373 OptimizePM.addPass(Float2IntPass()); 1374 OptimizePM.addPass(LowerConstantIntrinsicsPass()); 1375 1376 if (EnableMatrix) { 1377 OptimizePM.addPass(LowerMatrixIntrinsicsPass()); 1378 OptimizePM.addPass(EarlyCSEPass()); 1379 } 1380 1381 // FIXME: We need to run some loop optimizations to re-rotate loops after 1382 // simplify-cfg and others undo their rotation. 1383 1384 // Optimize the loop execution. These passes operate on entire loop nests 1385 // rather than on each loop in an inside-out manner, and so they are actually 1386 // function passes. 1387 1388 for (auto &C : VectorizerStartEPCallbacks) 1389 C(OptimizePM, Level); 1390 1391 // First rotate loops that may have been un-rotated by prior passes. 1392 // Disable header duplication at -Oz. 1393 OptimizePM.addPass(createFunctionToLoopPassAdaptor( 1394 LoopRotatePass(Level != OptimizationLevel::Oz, LTOPreLink), 1395 EnableMSSALoopDependency, 1396 /*UseBlockFrequencyInfo=*/false)); 1397 1398 // Distribute loops to allow partial vectorization. I.e. isolate dependences 1399 // into separate loop that would otherwise inhibit vectorization. This is 1400 // currently only performed for loops marked with the metadata 1401 // llvm.loop.distribute=true or when -enable-loop-distribute is specified. 1402 OptimizePM.addPass(LoopDistributePass()); 1403 1404 // Populates the VFABI attribute with the scalar-to-vector mappings 1405 // from the TargetLibraryInfo. 1406 OptimizePM.addPass(InjectTLIMappings()); 1407 1408 addVectorPasses(Level, OptimizePM, /* IsLTO */ false); 1409 1410 // Split out cold code. Splitting is done late to avoid hiding context from 1411 // other optimizations and inadvertently regressing performance. The tradeoff 1412 // is that this has a higher code size cost than splitting early. 1413 if (EnableHotColdSplit && !LTOPreLink) 1414 MPM.addPass(HotColdSplittingPass()); 1415 1416 // Search the code for similar regions of code. If enough similar regions can 1417 // be found where extracting the regions into their own function will decrease 1418 // the size of the program, we extract the regions, a deduplicate the 1419 // structurally similar regions. 1420 if (EnableIROutliner) 1421 MPM.addPass(IROutlinerPass()); 1422 1423 // Merge functions if requested. 1424 if (PTO.MergeFunctions) 1425 MPM.addPass(MergeFunctionsPass()); 1426 1427 // LoopSink pass sinks instructions hoisted by LICM, which serves as a 1428 // canonicalization pass that enables other optimizations. As a result, 1429 // LoopSink pass needs to be a very late IR pass to avoid undoing LICM 1430 // result too early. 1431 OptimizePM.addPass(LoopSinkPass()); 1432 1433 // And finally clean up LCSSA form before generating code. 1434 OptimizePM.addPass(InstSimplifyPass()); 1435 1436 // This hoists/decomposes div/rem ops. It should run after other sink/hoist 1437 // passes to avoid re-sinking, but before SimplifyCFG because it can allow 1438 // flattening of blocks. 1439 OptimizePM.addPass(DivRemPairsPass()); 1440 1441 // LoopSink (and other loop passes since the last simplifyCFG) might have 1442 // resulted in single-entry-single-exit or empty blocks. Clean up the CFG. 1443 OptimizePM.addPass(SimplifyCFGPass()); 1444 1445 // Optimize PHIs by speculating around them when profitable. Note that this 1446 // pass needs to be run after any PRE or similar pass as it is essentially 1447 // inserting redundancies into the program. This even includes SimplifyCFG. 1448 OptimizePM.addPass(SpeculateAroundPHIsPass()); 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 // Propagate constants at call sites into the functions they call. This 1701 // opens opportunities for globalopt (and inlining) by substituting function 1702 // pointers passed as arguments to direct uses of functions. 1703 MPM.addPass(IPSCCPPass()); 1704 1705 // Attach metadata to indirect call sites indicating the set of functions 1706 // they may target at run-time. This should follow IPSCCP. 1707 MPM.addPass(CalledValuePropagationPass()); 1708 } 1709 1710 // Now deduce any function attributes based in the current code. 1711 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor( 1712 PostOrderFunctionAttrsPass())); 1713 1714 // Do RPO function attribute inference across the module to forward-propagate 1715 // attributes where applicable. 1716 // FIXME: Is this really an optimization rather than a canonicalization? 1717 MPM.addPass(ReversePostOrderFunctionAttrsPass()); 1718 1719 // Use in-range annotations on GEP indices to split globals where beneficial. 1720 MPM.addPass(GlobalSplitPass()); 1721 1722 // Run whole program optimization of virtual call when the list of callees 1723 // is fixed. 1724 MPM.addPass(WholeProgramDevirtPass(ExportSummary, nullptr)); 1725 1726 // Stop here at -O1. 1727 if (Level == OptimizationLevel::O1) { 1728 // The LowerTypeTestsPass needs to run to lower type metadata and the 1729 // type.test intrinsics. The pass does nothing if CFI is disabled. 1730 MPM.addPass(LowerTypeTestsPass(ExportSummary, nullptr)); 1731 // Run a second time to clean up any type tests left behind by WPD for use 1732 // in ICP (which is performed earlier than this in the regular LTO 1733 // pipeline). 1734 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1735 1736 // Emit annotation remarks. 1737 addAnnotationRemarksPass(MPM); 1738 1739 return MPM; 1740 } 1741 1742 // Optimize globals to try and fold them into constants. 1743 MPM.addPass(GlobalOptPass()); 1744 1745 // Promote any localized globals to SSA registers. 1746 MPM.addPass(createModuleToFunctionPassAdaptor(PromotePass())); 1747 1748 // Linking modules together can lead to duplicate global constant, only 1749 // keep one copy of each constant. 1750 MPM.addPass(ConstantMergePass()); 1751 1752 // Remove unused arguments from functions. 1753 MPM.addPass(DeadArgumentEliminationPass()); 1754 1755 // Reduce the code after globalopt and ipsccp. Both can open up significant 1756 // simplification opportunities, and both can propagate functions through 1757 // function pointers. When this happens, we often have to resolve varargs 1758 // calls, etc, so let instcombine do this. 1759 FunctionPassManager PeepholeFPM; 1760 if (Level == OptimizationLevel::O3) 1761 PeepholeFPM.addPass(AggressiveInstCombinePass()); 1762 PeepholeFPM.addPass(InstCombinePass()); 1763 invokePeepholeEPCallbacks(PeepholeFPM, Level); 1764 1765 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(PeepholeFPM))); 1766 1767 // Note: historically, the PruneEH pass was run first to deduce nounwind and 1768 // generally clean up exception handling overhead. It isn't clear this is 1769 // valuable as the inliner doesn't currently care whether it is inlining an 1770 // invoke or a call. 1771 // Run the inliner now. 1772 MPM.addPass(ModuleInlinerWrapperPass(getInlineParamsFromOptLevel(Level))); 1773 1774 // Optimize globals again after we ran the inliner. 1775 MPM.addPass(GlobalOptPass()); 1776 1777 // Garbage collect dead functions. 1778 // FIXME: Add ArgumentPromotion pass after once it's ported. 1779 MPM.addPass(GlobalDCEPass()); 1780 1781 FunctionPassManager FPM; 1782 // The IPO Passes may leave cruft around. Clean up after them. 1783 FPM.addPass(InstCombinePass()); 1784 invokePeepholeEPCallbacks(FPM, Level); 1785 1786 FPM.addPass(JumpThreadingPass(/*InsertFreezeWhenUnfoldingSelect*/ true)); 1787 1788 // Do a post inline PGO instrumentation and use pass. This is a context 1789 // sensitive PGO pass. 1790 if (PGOOpt) { 1791 if (PGOOpt->CSAction == PGOOptions::CSIRInstr) 1792 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ true, 1793 /* IsCS */ true, PGOOpt->CSProfileGenFile, 1794 PGOOpt->ProfileRemappingFile); 1795 else if (PGOOpt->CSAction == PGOOptions::CSIRUse) 1796 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ false, 1797 /* IsCS */ true, PGOOpt->ProfileFile, 1798 PGOOpt->ProfileRemappingFile); 1799 } 1800 1801 // Break up allocas 1802 FPM.addPass(SROA()); 1803 1804 // LTO provides additional opportunities for tailcall elimination due to 1805 // link-time inlining, and visibility of nocapture attribute. 1806 FPM.addPass(TailCallElimPass()); 1807 1808 // Run a few AA driver optimizations here and now to cleanup the code. 1809 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 1810 1811 MPM.addPass( 1812 createModuleToPostOrderCGSCCPassAdaptor(PostOrderFunctionAttrsPass())); 1813 1814 // Require the GlobalsAA analysis for the module so we can query it within 1815 // MainFPM. 1816 MPM.addPass(RequireAnalysisPass<GlobalsAA, Module>()); 1817 // Invalidate AAManager so it can be recreated and pick up the newly available 1818 // GlobalsAA. 1819 MPM.addPass( 1820 createModuleToFunctionPassAdaptor(InvalidateAnalysisPass<AAManager>())); 1821 1822 FunctionPassManager MainFPM; 1823 MainFPM.addPass(createFunctionToLoopPassAdaptor( 1824 LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap), 1825 EnableMSSALoopDependency, /*UseBlockFrequencyInfo=*/true)); 1826 1827 if (RunNewGVN) 1828 MainFPM.addPass(NewGVNPass()); 1829 else 1830 MainFPM.addPass(GVN()); 1831 1832 // Remove dead memcpy()'s. 1833 MainFPM.addPass(MemCpyOptPass()); 1834 1835 // Nuke dead stores. 1836 MainFPM.addPass(DSEPass()); 1837 MainFPM.addPass(MergedLoadStoreMotionPass()); 1838 1839 // More loops are countable; try to optimize them. 1840 if (EnableLoopFlatten && Level.getSpeedupLevel() > 1) 1841 MainFPM.addPass(createFunctionToLoopPassAdaptor(LoopFlattenPass())); 1842 1843 if (EnableConstraintElimination) 1844 MainFPM.addPass(ConstraintEliminationPass()); 1845 1846 LoopPassManager LPM; 1847 LPM.addPass(IndVarSimplifyPass()); 1848 LPM.addPass(LoopDeletionPass()); 1849 // FIXME: Add loop interchange. 1850 1851 // Unroll small loops and perform peeling. 1852 LPM.addPass(LoopFullUnrollPass(Level.getSpeedupLevel(), 1853 /* OnlyWhenForced= */ !PTO.LoopUnrolling, 1854 PTO.ForgetAllSCEVInLoopUnroll)); 1855 // The loop passes in LPM (LoopFullUnrollPass) do not preserve MemorySSA. 1856 // *All* loop passes must preserve it, in order to be able to use it. 1857 MainFPM.addPass(createFunctionToLoopPassAdaptor( 1858 std::move(LPM), /*UseMemorySSA=*/false, /*UseBlockFrequencyInfo=*/true)); 1859 1860 MainFPM.addPass(LoopDistributePass()); 1861 1862 addVectorPasses(Level, MainFPM, /* IsLTO */ true); 1863 1864 invokePeepholeEPCallbacks(MainFPM, Level); 1865 MainFPM.addPass(JumpThreadingPass(/*InsertFreezeWhenUnfoldingSelect*/ true)); 1866 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(MainFPM))); 1867 1868 // Create a function that performs CFI checks for cross-DSO calls with 1869 // targets in the current module. 1870 MPM.addPass(CrossDSOCFIPass()); 1871 1872 // Lower type metadata and the type.test intrinsic. This pass supports 1873 // clang's control flow integrity mechanisms (-fsanitize=cfi*) and needs 1874 // to be run at link time if CFI is enabled. This pass does nothing if 1875 // CFI is disabled. 1876 MPM.addPass(LowerTypeTestsPass(ExportSummary, nullptr)); 1877 // Run a second time to clean up any type tests left behind by WPD for use 1878 // in ICP (which is performed earlier than this in the regular LTO pipeline). 1879 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1880 1881 // Enable splitting late in the FullLTO post-link pipeline. This is done in 1882 // the same stage in the old pass manager (\ref addLateLTOOptimizationPasses). 1883 if (EnableHotColdSplit) 1884 MPM.addPass(HotColdSplittingPass()); 1885 1886 // Add late LTO optimization passes. 1887 // Delete basic blocks, which optimization passes may have killed. 1888 MPM.addPass(createModuleToFunctionPassAdaptor( 1889 SimplifyCFGPass(SimplifyCFGOptions().hoistCommonInsts(true)))); 1890 1891 // Drop bodies of available eternally objects to improve GlobalDCE. 1892 MPM.addPass(EliminateAvailableExternallyPass()); 1893 1894 // Now that we have optimized the program, discard unreachable functions. 1895 MPM.addPass(GlobalDCEPass()); 1896 1897 if (PTO.MergeFunctions) 1898 MPM.addPass(MergeFunctionsPass()); 1899 1900 // Emit annotation remarks. 1901 addAnnotationRemarksPass(MPM); 1902 1903 return MPM; 1904 } 1905 1906 ModulePassManager PassBuilder::buildO0DefaultPipeline(OptimizationLevel Level, 1907 bool LTOPreLink) { 1908 assert(Level == OptimizationLevel::O0 && 1909 "buildO0DefaultPipeline should only be used with O0"); 1910 1911 ModulePassManager MPM; 1912 1913 if (PGOOpt && (PGOOpt->Action == PGOOptions::IRInstr || 1914 PGOOpt->Action == PGOOptions::IRUse)) 1915 addPGOInstrPassesForO0( 1916 MPM, 1917 /* RunProfileGen */ (PGOOpt->Action == PGOOptions::IRInstr), 1918 /* IsCS */ false, PGOOpt->ProfileFile, PGOOpt->ProfileRemappingFile); 1919 1920 for (auto &C : PipelineStartEPCallbacks) 1921 C(MPM, Level); 1922 for (auto &C : PipelineEarlySimplificationEPCallbacks) 1923 C(MPM, Level); 1924 1925 // Build a minimal pipeline based on the semantics required by LLVM, 1926 // which is just that always inlining occurs. Further, disable generating 1927 // lifetime intrinsics to avoid enabling further optimizations during 1928 // code generation. 1929 // However, we need to insert lifetime intrinsics to avoid invalid access 1930 // caused by multithreaded coroutines. 1931 MPM.addPass(AlwaysInlinerPass( 1932 /*InsertLifetimeIntrinsics=*/PTO.Coroutines)); 1933 1934 if (PTO.MergeFunctions) 1935 MPM.addPass(MergeFunctionsPass()); 1936 1937 if (EnableMatrix) 1938 MPM.addPass( 1939 createModuleToFunctionPassAdaptor(LowerMatrixIntrinsicsPass(true))); 1940 1941 if (!CGSCCOptimizerLateEPCallbacks.empty()) { 1942 CGSCCPassManager CGPM; 1943 for (auto &C : CGSCCOptimizerLateEPCallbacks) 1944 C(CGPM, Level); 1945 if (!CGPM.isEmpty()) 1946 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPM))); 1947 } 1948 if (!LateLoopOptimizationsEPCallbacks.empty()) { 1949 LoopPassManager LPM; 1950 for (auto &C : LateLoopOptimizationsEPCallbacks) 1951 C(LPM, Level); 1952 if (!LPM.isEmpty()) { 1953 MPM.addPass(createModuleToFunctionPassAdaptor( 1954 createFunctionToLoopPassAdaptor(std::move(LPM)))); 1955 } 1956 } 1957 if (!LoopOptimizerEndEPCallbacks.empty()) { 1958 LoopPassManager LPM; 1959 for (auto &C : LoopOptimizerEndEPCallbacks) 1960 C(LPM, Level); 1961 if (!LPM.isEmpty()) { 1962 MPM.addPass(createModuleToFunctionPassAdaptor( 1963 createFunctionToLoopPassAdaptor(std::move(LPM)))); 1964 } 1965 } 1966 if (!ScalarOptimizerLateEPCallbacks.empty()) { 1967 FunctionPassManager FPM; 1968 for (auto &C : ScalarOptimizerLateEPCallbacks) 1969 C(FPM, Level); 1970 if (!FPM.isEmpty()) 1971 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 1972 } 1973 if (!VectorizerStartEPCallbacks.empty()) { 1974 FunctionPassManager FPM; 1975 for (auto &C : VectorizerStartEPCallbacks) 1976 C(FPM, Level); 1977 if (!FPM.isEmpty()) 1978 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 1979 } 1980 1981 if (PTO.Coroutines) { 1982 MPM.addPass(createModuleToFunctionPassAdaptor(CoroEarlyPass())); 1983 1984 CGSCCPassManager CGPM; 1985 CGPM.addPass(CoroSplitPass()); 1986 CGPM.addPass(createCGSCCToFunctionPassAdaptor(CoroElidePass())); 1987 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPM))); 1988 1989 MPM.addPass(createModuleToFunctionPassAdaptor(CoroCleanupPass())); 1990 } 1991 1992 for (auto &C : OptimizerLastEPCallbacks) 1993 C(MPM, Level); 1994 1995 if (LTOPreLink) 1996 addRequiredLTOPreLinkPasses(MPM); 1997 1998 return MPM; 1999 } 2000 2001 AAManager PassBuilder::buildDefaultAAPipeline() { 2002 AAManager AA; 2003 2004 // The order in which these are registered determines their priority when 2005 // being queried. 2006 2007 // First we register the basic alias analysis that provides the majority of 2008 // per-function local AA logic. This is a stateless, on-demand local set of 2009 // AA techniques. 2010 AA.registerFunctionAnalysis<BasicAA>(); 2011 2012 // Next we query fast, specialized alias analyses that wrap IR-embedded 2013 // information about aliasing. 2014 AA.registerFunctionAnalysis<ScopedNoAliasAA>(); 2015 AA.registerFunctionAnalysis<TypeBasedAA>(); 2016 2017 // Add support for querying global aliasing information when available. 2018 // Because the `AAManager` is a function analysis and `GlobalsAA` is a module 2019 // analysis, all that the `AAManager` can do is query for any *cached* 2020 // results from `GlobalsAA` through a readonly proxy. 2021 AA.registerModuleAnalysis<GlobalsAA>(); 2022 2023 // Add target-specific alias analyses. 2024 if (TM) 2025 TM->registerDefaultAliasAnalyses(AA); 2026 2027 return AA; 2028 } 2029 2030 static Optional<int> parseRepeatPassName(StringRef Name) { 2031 if (!Name.consume_front("repeat<") || !Name.consume_back(">")) 2032 return None; 2033 int Count; 2034 if (Name.getAsInteger(0, Count) || Count <= 0) 2035 return None; 2036 return Count; 2037 } 2038 2039 static Optional<int> parseDevirtPassName(StringRef Name) { 2040 if (!Name.consume_front("devirt<") || !Name.consume_back(">")) 2041 return None; 2042 int Count; 2043 if (Name.getAsInteger(0, Count) || Count < 0) 2044 return None; 2045 return Count; 2046 } 2047 2048 static bool checkParametrizedPassName(StringRef Name, StringRef PassName) { 2049 if (!Name.consume_front(PassName)) 2050 return false; 2051 // normal pass name w/o parameters == default parameters 2052 if (Name.empty()) 2053 return true; 2054 return Name.startswith("<") && Name.endswith(">"); 2055 } 2056 2057 namespace { 2058 2059 /// This performs customized parsing of pass name with parameters. 2060 /// 2061 /// We do not need parametrization of passes in textual pipeline very often, 2062 /// yet on a rare occasion ability to specify parameters right there can be 2063 /// useful. 2064 /// 2065 /// \p Name - parameterized specification of a pass from a textual pipeline 2066 /// is a string in a form of : 2067 /// PassName '<' parameter-list '>' 2068 /// 2069 /// Parameter list is being parsed by the parser callable argument, \p Parser, 2070 /// It takes a string-ref of parameters and returns either StringError or a 2071 /// parameter list in a form of a custom parameters type, all wrapped into 2072 /// Expected<> template class. 2073 /// 2074 template <typename ParametersParseCallableT> 2075 auto parsePassParameters(ParametersParseCallableT &&Parser, StringRef Name, 2076 StringRef PassName) -> decltype(Parser(StringRef{})) { 2077 using ParametersT = typename decltype(Parser(StringRef{}))::value_type; 2078 2079 StringRef Params = Name; 2080 if (!Params.consume_front(PassName)) { 2081 assert(false && 2082 "unable to strip pass name from parametrized pass specification"); 2083 } 2084 if (Params.empty()) 2085 return ParametersT{}; 2086 if (!Params.consume_front("<") || !Params.consume_back(">")) { 2087 assert(false && "invalid format for parametrized pass name"); 2088 } 2089 2090 Expected<ParametersT> Result = Parser(Params); 2091 assert((Result || Result.template errorIsA<StringError>()) && 2092 "Pass parameter parser can only return StringErrors."); 2093 return Result; 2094 } 2095 2096 /// Parser of parameters for LoopUnroll pass. 2097 Expected<LoopUnrollOptions> parseLoopUnrollOptions(StringRef Params) { 2098 LoopUnrollOptions UnrollOpts; 2099 while (!Params.empty()) { 2100 StringRef ParamName; 2101 std::tie(ParamName, Params) = Params.split(';'); 2102 int OptLevel = StringSwitch<int>(ParamName) 2103 .Case("O0", 0) 2104 .Case("O1", 1) 2105 .Case("O2", 2) 2106 .Case("O3", 3) 2107 .Default(-1); 2108 if (OptLevel >= 0) { 2109 UnrollOpts.setOptLevel(OptLevel); 2110 continue; 2111 } 2112 if (ParamName.consume_front("full-unroll-max=")) { 2113 int Count; 2114 if (ParamName.getAsInteger(0, Count)) 2115 return make_error<StringError>( 2116 formatv("invalid LoopUnrollPass parameter '{0}' ", ParamName).str(), 2117 inconvertibleErrorCode()); 2118 UnrollOpts.setFullUnrollMaxCount(Count); 2119 continue; 2120 } 2121 2122 bool Enable = !ParamName.consume_front("no-"); 2123 if (ParamName == "partial") { 2124 UnrollOpts.setPartial(Enable); 2125 } else if (ParamName == "peeling") { 2126 UnrollOpts.setPeeling(Enable); 2127 } else if (ParamName == "profile-peeling") { 2128 UnrollOpts.setProfileBasedPeeling(Enable); 2129 } else if (ParamName == "runtime") { 2130 UnrollOpts.setRuntime(Enable); 2131 } else if (ParamName == "upperbound") { 2132 UnrollOpts.setUpperBound(Enable); 2133 } else { 2134 return make_error<StringError>( 2135 formatv("invalid LoopUnrollPass parameter '{0}' ", ParamName).str(), 2136 inconvertibleErrorCode()); 2137 } 2138 } 2139 return UnrollOpts; 2140 } 2141 2142 Expected<MemorySanitizerOptions> parseMSanPassOptions(StringRef Params) { 2143 MemorySanitizerOptions Result; 2144 while (!Params.empty()) { 2145 StringRef ParamName; 2146 std::tie(ParamName, Params) = Params.split(';'); 2147 2148 if (ParamName == "recover") { 2149 Result.Recover = true; 2150 } else if (ParamName == "kernel") { 2151 Result.Kernel = true; 2152 } else if (ParamName.consume_front("track-origins=")) { 2153 if (ParamName.getAsInteger(0, Result.TrackOrigins)) 2154 return make_error<StringError>( 2155 formatv("invalid argument to MemorySanitizer pass track-origins " 2156 "parameter: '{0}' ", 2157 ParamName) 2158 .str(), 2159 inconvertibleErrorCode()); 2160 } else { 2161 return make_error<StringError>( 2162 formatv("invalid MemorySanitizer pass parameter '{0}' ", ParamName) 2163 .str(), 2164 inconvertibleErrorCode()); 2165 } 2166 } 2167 return Result; 2168 } 2169 2170 /// Parser of parameters for SimplifyCFG pass. 2171 Expected<SimplifyCFGOptions> parseSimplifyCFGOptions(StringRef Params) { 2172 SimplifyCFGOptions Result; 2173 while (!Params.empty()) { 2174 StringRef ParamName; 2175 std::tie(ParamName, Params) = Params.split(';'); 2176 2177 bool Enable = !ParamName.consume_front("no-"); 2178 if (ParamName == "forward-switch-cond") { 2179 Result.forwardSwitchCondToPhi(Enable); 2180 } else if (ParamName == "switch-to-lookup") { 2181 Result.convertSwitchToLookupTable(Enable); 2182 } else if (ParamName == "keep-loops") { 2183 Result.needCanonicalLoops(Enable); 2184 } else if (ParamName == "hoist-common-insts") { 2185 Result.hoistCommonInsts(Enable); 2186 } else if (ParamName == "sink-common-insts") { 2187 Result.sinkCommonInsts(Enable); 2188 } else if (Enable && ParamName.consume_front("bonus-inst-threshold=")) { 2189 APInt BonusInstThreshold; 2190 if (ParamName.getAsInteger(0, BonusInstThreshold)) 2191 return make_error<StringError>( 2192 formatv("invalid argument to SimplifyCFG pass bonus-threshold " 2193 "parameter: '{0}' ", 2194 ParamName).str(), 2195 inconvertibleErrorCode()); 2196 Result.bonusInstThreshold(BonusInstThreshold.getSExtValue()); 2197 } else { 2198 return make_error<StringError>( 2199 formatv("invalid SimplifyCFG pass parameter '{0}' ", ParamName).str(), 2200 inconvertibleErrorCode()); 2201 } 2202 } 2203 return Result; 2204 } 2205 2206 /// Parser of parameters for LoopVectorize pass. 2207 Expected<LoopVectorizeOptions> parseLoopVectorizeOptions(StringRef Params) { 2208 LoopVectorizeOptions Opts; 2209 while (!Params.empty()) { 2210 StringRef ParamName; 2211 std::tie(ParamName, Params) = Params.split(';'); 2212 2213 bool Enable = !ParamName.consume_front("no-"); 2214 if (ParamName == "interleave-forced-only") { 2215 Opts.setInterleaveOnlyWhenForced(Enable); 2216 } else if (ParamName == "vectorize-forced-only") { 2217 Opts.setVectorizeOnlyWhenForced(Enable); 2218 } else { 2219 return make_error<StringError>( 2220 formatv("invalid LoopVectorize parameter '{0}' ", ParamName).str(), 2221 inconvertibleErrorCode()); 2222 } 2223 } 2224 return Opts; 2225 } 2226 2227 Expected<bool> parseLoopUnswitchOptions(StringRef Params) { 2228 bool Result = false; 2229 while (!Params.empty()) { 2230 StringRef ParamName; 2231 std::tie(ParamName, Params) = Params.split(';'); 2232 2233 bool Enable = !ParamName.consume_front("no-"); 2234 if (ParamName == "nontrivial") { 2235 Result = Enable; 2236 } else { 2237 return make_error<StringError>( 2238 formatv("invalid LoopUnswitch pass parameter '{0}' ", ParamName) 2239 .str(), 2240 inconvertibleErrorCode()); 2241 } 2242 } 2243 return Result; 2244 } 2245 2246 Expected<bool> parseMergedLoadStoreMotionOptions(StringRef Params) { 2247 bool Result = false; 2248 while (!Params.empty()) { 2249 StringRef ParamName; 2250 std::tie(ParamName, Params) = Params.split(';'); 2251 2252 bool Enable = !ParamName.consume_front("no-"); 2253 if (ParamName == "split-footer-bb") { 2254 Result = Enable; 2255 } else { 2256 return make_error<StringError>( 2257 formatv("invalid MergedLoadStoreMotion pass parameter '{0}' ", 2258 ParamName) 2259 .str(), 2260 inconvertibleErrorCode()); 2261 } 2262 } 2263 return Result; 2264 } 2265 2266 Expected<GVNOptions> parseGVNOptions(StringRef Params) { 2267 GVNOptions Result; 2268 while (!Params.empty()) { 2269 StringRef ParamName; 2270 std::tie(ParamName, Params) = Params.split(';'); 2271 2272 bool Enable = !ParamName.consume_front("no-"); 2273 if (ParamName == "pre") { 2274 Result.setPRE(Enable); 2275 } else if (ParamName == "load-pre") { 2276 Result.setLoadPRE(Enable); 2277 } else if (ParamName == "split-backedge-load-pre") { 2278 Result.setLoadPRESplitBackedge(Enable); 2279 } else if (ParamName == "memdep") { 2280 Result.setMemDep(Enable); 2281 } else { 2282 return make_error<StringError>( 2283 formatv("invalid GVN pass parameter '{0}' ", ParamName).str(), 2284 inconvertibleErrorCode()); 2285 } 2286 } 2287 return Result; 2288 } 2289 2290 Expected<StackLifetime::LivenessType> 2291 parseStackLifetimeOptions(StringRef Params) { 2292 StackLifetime::LivenessType Result = StackLifetime::LivenessType::May; 2293 while (!Params.empty()) { 2294 StringRef ParamName; 2295 std::tie(ParamName, Params) = Params.split(';'); 2296 2297 if (ParamName == "may") { 2298 Result = StackLifetime::LivenessType::May; 2299 } else if (ParamName == "must") { 2300 Result = StackLifetime::LivenessType::Must; 2301 } else { 2302 return make_error<StringError>( 2303 formatv("invalid StackLifetime parameter '{0}' ", ParamName).str(), 2304 inconvertibleErrorCode()); 2305 } 2306 } 2307 return Result; 2308 } 2309 2310 } // namespace 2311 2312 /// Tests whether a pass name starts with a valid prefix for a default pipeline 2313 /// alias. 2314 static bool startsWithDefaultPipelineAliasPrefix(StringRef Name) { 2315 return Name.startswith("default") || Name.startswith("thinlto") || 2316 Name.startswith("lto"); 2317 } 2318 2319 /// Tests whether registered callbacks will accept a given pass name. 2320 /// 2321 /// When parsing a pipeline text, the type of the outermost pipeline may be 2322 /// omitted, in which case the type is automatically determined from the first 2323 /// pass name in the text. This may be a name that is handled through one of the 2324 /// callbacks. We check this through the oridinary parsing callbacks by setting 2325 /// up a dummy PassManager in order to not force the client to also handle this 2326 /// type of query. 2327 template <typename PassManagerT, typename CallbacksT> 2328 static bool callbacksAcceptPassName(StringRef Name, CallbacksT &Callbacks) { 2329 if (!Callbacks.empty()) { 2330 PassManagerT DummyPM; 2331 for (auto &CB : Callbacks) 2332 if (CB(Name, DummyPM, {})) 2333 return true; 2334 } 2335 return false; 2336 } 2337 2338 template <typename CallbacksT> 2339 static bool isModulePassName(StringRef Name, CallbacksT &Callbacks) { 2340 // Manually handle aliases for pre-configured pipeline fragments. 2341 if (startsWithDefaultPipelineAliasPrefix(Name)) 2342 return DefaultAliasRegex.match(Name); 2343 2344 // Explicitly handle pass manager names. 2345 if (Name == "module") 2346 return true; 2347 if (Name == "cgscc") 2348 return true; 2349 if (Name == "function") 2350 return true; 2351 2352 // Explicitly handle custom-parsed pass names. 2353 if (parseRepeatPassName(Name)) 2354 return true; 2355 2356 #define MODULE_PASS(NAME, CREATE_PASS) \ 2357 if (Name == NAME) \ 2358 return true; 2359 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 2360 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2361 return true; 2362 #include "PassRegistry.def" 2363 2364 return callbacksAcceptPassName<ModulePassManager>(Name, Callbacks); 2365 } 2366 2367 template <typename CallbacksT> 2368 static bool isCGSCCPassName(StringRef Name, CallbacksT &Callbacks) { 2369 // Explicitly handle pass manager names. 2370 if (Name == "cgscc") 2371 return true; 2372 if (Name == "function") 2373 return true; 2374 2375 // Explicitly handle custom-parsed pass names. 2376 if (parseRepeatPassName(Name)) 2377 return true; 2378 if (parseDevirtPassName(Name)) 2379 return true; 2380 2381 #define CGSCC_PASS(NAME, CREATE_PASS) \ 2382 if (Name == NAME) \ 2383 return true; 2384 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 2385 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2386 return true; 2387 #include "PassRegistry.def" 2388 2389 return callbacksAcceptPassName<CGSCCPassManager>(Name, Callbacks); 2390 } 2391 2392 template <typename CallbacksT> 2393 static bool isFunctionPassName(StringRef Name, CallbacksT &Callbacks) { 2394 // Explicitly handle pass manager names. 2395 if (Name == "function") 2396 return true; 2397 if (Name == "loop" || Name == "loop-mssa") 2398 return true; 2399 2400 // Explicitly handle custom-parsed pass names. 2401 if (parseRepeatPassName(Name)) 2402 return true; 2403 2404 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2405 if (Name == NAME) \ 2406 return true; 2407 #define FUNCTION_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2408 if (checkParametrizedPassName(Name, NAME)) \ 2409 return true; 2410 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 2411 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2412 return true; 2413 #include "PassRegistry.def" 2414 2415 return callbacksAcceptPassName<FunctionPassManager>(Name, Callbacks); 2416 } 2417 2418 template <typename CallbacksT> 2419 static bool isLoopPassName(StringRef Name, CallbacksT &Callbacks) { 2420 // Explicitly handle pass manager names. 2421 if (Name == "loop" || Name == "loop-mssa") 2422 return true; 2423 2424 // Explicitly handle custom-parsed pass names. 2425 if (parseRepeatPassName(Name)) 2426 return true; 2427 2428 #define LOOP_PASS(NAME, CREATE_PASS) \ 2429 if (Name == NAME) \ 2430 return true; 2431 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2432 if (checkParametrizedPassName(Name, NAME)) \ 2433 return true; 2434 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 2435 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2436 return true; 2437 #include "PassRegistry.def" 2438 2439 return callbacksAcceptPassName<LoopPassManager>(Name, Callbacks); 2440 } 2441 2442 Optional<std::vector<PassBuilder::PipelineElement>> 2443 PassBuilder::parsePipelineText(StringRef Text) { 2444 std::vector<PipelineElement> ResultPipeline; 2445 2446 SmallVector<std::vector<PipelineElement> *, 4> PipelineStack = { 2447 &ResultPipeline}; 2448 for (;;) { 2449 std::vector<PipelineElement> &Pipeline = *PipelineStack.back(); 2450 size_t Pos = Text.find_first_of(",()"); 2451 Pipeline.push_back({Text.substr(0, Pos), {}}); 2452 2453 // If we have a single terminating name, we're done. 2454 if (Pos == Text.npos) 2455 break; 2456 2457 char Sep = Text[Pos]; 2458 Text = Text.substr(Pos + 1); 2459 if (Sep == ',') 2460 // Just a name ending in a comma, continue. 2461 continue; 2462 2463 if (Sep == '(') { 2464 // Push the inner pipeline onto the stack to continue processing. 2465 PipelineStack.push_back(&Pipeline.back().InnerPipeline); 2466 continue; 2467 } 2468 2469 assert(Sep == ')' && "Bogus separator!"); 2470 // When handling the close parenthesis, we greedily consume them to avoid 2471 // empty strings in the pipeline. 2472 do { 2473 // If we try to pop the outer pipeline we have unbalanced parentheses. 2474 if (PipelineStack.size() == 1) 2475 return None; 2476 2477 PipelineStack.pop_back(); 2478 } while (Text.consume_front(")")); 2479 2480 // Check if we've finished parsing. 2481 if (Text.empty()) 2482 break; 2483 2484 // Otherwise, the end of an inner pipeline always has to be followed by 2485 // a comma, and then we can continue. 2486 if (!Text.consume_front(",")) 2487 return None; 2488 } 2489 2490 if (PipelineStack.size() > 1) 2491 // Unbalanced paretheses. 2492 return None; 2493 2494 assert(PipelineStack.back() == &ResultPipeline && 2495 "Wrong pipeline at the bottom of the stack!"); 2496 return {std::move(ResultPipeline)}; 2497 } 2498 2499 Error PassBuilder::parseModulePass(ModulePassManager &MPM, 2500 const PipelineElement &E) { 2501 auto &Name = E.Name; 2502 auto &InnerPipeline = E.InnerPipeline; 2503 2504 // First handle complex passes like the pass managers which carry pipelines. 2505 if (!InnerPipeline.empty()) { 2506 if (Name == "module") { 2507 ModulePassManager NestedMPM; 2508 if (auto Err = parseModulePassPipeline(NestedMPM, InnerPipeline)) 2509 return Err; 2510 MPM.addPass(std::move(NestedMPM)); 2511 return Error::success(); 2512 } 2513 if (Name == "cgscc") { 2514 CGSCCPassManager CGPM; 2515 if (auto Err = parseCGSCCPassPipeline(CGPM, InnerPipeline)) 2516 return Err; 2517 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPM))); 2518 return Error::success(); 2519 } 2520 if (Name == "function") { 2521 FunctionPassManager FPM; 2522 if (auto Err = parseFunctionPassPipeline(FPM, InnerPipeline)) 2523 return Err; 2524 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 2525 return Error::success(); 2526 } 2527 if (auto Count = parseRepeatPassName(Name)) { 2528 ModulePassManager NestedMPM; 2529 if (auto Err = parseModulePassPipeline(NestedMPM, InnerPipeline)) 2530 return Err; 2531 MPM.addPass(createRepeatedPass(*Count, std::move(NestedMPM))); 2532 return Error::success(); 2533 } 2534 2535 for (auto &C : ModulePipelineParsingCallbacks) 2536 if (C(Name, MPM, InnerPipeline)) 2537 return Error::success(); 2538 2539 // Normal passes can't have pipelines. 2540 return make_error<StringError>( 2541 formatv("invalid use of '{0}' pass as module pipeline", Name).str(), 2542 inconvertibleErrorCode()); 2543 ; 2544 } 2545 2546 // Manually handle aliases for pre-configured pipeline fragments. 2547 if (startsWithDefaultPipelineAliasPrefix(Name)) { 2548 SmallVector<StringRef, 3> Matches; 2549 if (!DefaultAliasRegex.match(Name, &Matches)) 2550 return make_error<StringError>( 2551 formatv("unknown default pipeline alias '{0}'", Name).str(), 2552 inconvertibleErrorCode()); 2553 2554 assert(Matches.size() == 3 && "Must capture two matched strings!"); 2555 2556 OptimizationLevel L = StringSwitch<OptimizationLevel>(Matches[2]) 2557 .Case("O0", OptimizationLevel::O0) 2558 .Case("O1", OptimizationLevel::O1) 2559 .Case("O2", OptimizationLevel::O2) 2560 .Case("O3", OptimizationLevel::O3) 2561 .Case("Os", OptimizationLevel::Os) 2562 .Case("Oz", OptimizationLevel::Oz); 2563 if (L == OptimizationLevel::O0 && Matches[1] != "thinlto" && 2564 Matches[1] != "lto") { 2565 MPM.addPass(buildO0DefaultPipeline(L, Matches[1] == "thinlto-pre-link" || 2566 Matches[1] == "lto-pre-link")); 2567 return Error::success(); 2568 } 2569 2570 // This is consistent with old pass manager invoked via opt, but 2571 // inconsistent with clang. Clang doesn't enable loop vectorization 2572 // but does enable slp vectorization at Oz. 2573 PTO.LoopVectorization = 2574 L.getSpeedupLevel() > 1 && L != OptimizationLevel::Oz; 2575 PTO.SLPVectorization = 2576 L.getSpeedupLevel() > 1 && L != OptimizationLevel::Oz; 2577 2578 if (Matches[1] == "default") { 2579 MPM.addPass(buildPerModuleDefaultPipeline(L)); 2580 } else if (Matches[1] == "thinlto-pre-link") { 2581 MPM.addPass(buildThinLTOPreLinkDefaultPipeline(L)); 2582 } else if (Matches[1] == "thinlto") { 2583 MPM.addPass(buildThinLTODefaultPipeline(L, nullptr)); 2584 } else if (Matches[1] == "lto-pre-link") { 2585 MPM.addPass(buildLTOPreLinkDefaultPipeline(L)); 2586 } else { 2587 assert(Matches[1] == "lto" && "Not one of the matched options!"); 2588 MPM.addPass(buildLTODefaultPipeline(L, nullptr)); 2589 } 2590 return Error::success(); 2591 } 2592 2593 // Finally expand the basic registered passes from the .inc file. 2594 #define MODULE_PASS(NAME, CREATE_PASS) \ 2595 if (Name == NAME) { \ 2596 MPM.addPass(CREATE_PASS); \ 2597 return Error::success(); \ 2598 } 2599 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 2600 if (Name == "require<" NAME ">") { \ 2601 MPM.addPass( \ 2602 RequireAnalysisPass< \ 2603 std::remove_reference<decltype(CREATE_PASS)>::type, Module>()); \ 2604 return Error::success(); \ 2605 } \ 2606 if (Name == "invalidate<" NAME ">") { \ 2607 MPM.addPass(InvalidateAnalysisPass< \ 2608 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2609 return Error::success(); \ 2610 } 2611 #define CGSCC_PASS(NAME, CREATE_PASS) \ 2612 if (Name == NAME) { \ 2613 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(CREATE_PASS)); \ 2614 return Error::success(); \ 2615 } 2616 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2617 if (Name == NAME) { \ 2618 MPM.addPass(createModuleToFunctionPassAdaptor(CREATE_PASS)); \ 2619 return Error::success(); \ 2620 } 2621 #define FUNCTION_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2622 if (checkParametrizedPassName(Name, NAME)) { \ 2623 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2624 if (!Params) \ 2625 return Params.takeError(); \ 2626 MPM.addPass(createModuleToFunctionPassAdaptor(CREATE_PASS(Params.get()))); \ 2627 return Error::success(); \ 2628 } 2629 #define LOOP_PASS(NAME, CREATE_PASS) \ 2630 if (Name == NAME) { \ 2631 MPM.addPass(createModuleToFunctionPassAdaptor( \ 2632 createFunctionToLoopPassAdaptor(CREATE_PASS, false, false))); \ 2633 return Error::success(); \ 2634 } 2635 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2636 if (checkParametrizedPassName(Name, NAME)) { \ 2637 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2638 if (!Params) \ 2639 return Params.takeError(); \ 2640 MPM.addPass( \ 2641 createModuleToFunctionPassAdaptor(createFunctionToLoopPassAdaptor( \ 2642 CREATE_PASS(Params.get()), false, false))); \ 2643 return Error::success(); \ 2644 } 2645 #include "PassRegistry.def" 2646 2647 for (auto &C : ModulePipelineParsingCallbacks) 2648 if (C(Name, MPM, InnerPipeline)) 2649 return Error::success(); 2650 return make_error<StringError>( 2651 formatv("unknown module pass '{0}'", Name).str(), 2652 inconvertibleErrorCode()); 2653 } 2654 2655 Error PassBuilder::parseCGSCCPass(CGSCCPassManager &CGPM, 2656 const PipelineElement &E) { 2657 auto &Name = E.Name; 2658 auto &InnerPipeline = E.InnerPipeline; 2659 2660 // First handle complex passes like the pass managers which carry pipelines. 2661 if (!InnerPipeline.empty()) { 2662 if (Name == "cgscc") { 2663 CGSCCPassManager NestedCGPM; 2664 if (auto Err = parseCGSCCPassPipeline(NestedCGPM, InnerPipeline)) 2665 return Err; 2666 // Add the nested pass manager with the appropriate adaptor. 2667 CGPM.addPass(std::move(NestedCGPM)); 2668 return Error::success(); 2669 } 2670 if (Name == "function") { 2671 FunctionPassManager FPM; 2672 if (auto Err = parseFunctionPassPipeline(FPM, InnerPipeline)) 2673 return Err; 2674 // Add the nested pass manager with the appropriate adaptor. 2675 CGPM.addPass(createCGSCCToFunctionPassAdaptor(std::move(FPM))); 2676 return Error::success(); 2677 } 2678 if (auto Count = parseRepeatPassName(Name)) { 2679 CGSCCPassManager NestedCGPM; 2680 if (auto Err = parseCGSCCPassPipeline(NestedCGPM, InnerPipeline)) 2681 return Err; 2682 CGPM.addPass(createRepeatedPass(*Count, std::move(NestedCGPM))); 2683 return Error::success(); 2684 } 2685 if (auto MaxRepetitions = parseDevirtPassName(Name)) { 2686 CGSCCPassManager NestedCGPM; 2687 if (auto Err = parseCGSCCPassPipeline(NestedCGPM, InnerPipeline)) 2688 return Err; 2689 CGPM.addPass( 2690 createDevirtSCCRepeatedPass(std::move(NestedCGPM), *MaxRepetitions)); 2691 return Error::success(); 2692 } 2693 2694 for (auto &C : CGSCCPipelineParsingCallbacks) 2695 if (C(Name, CGPM, InnerPipeline)) 2696 return Error::success(); 2697 2698 // Normal passes can't have pipelines. 2699 return make_error<StringError>( 2700 formatv("invalid use of '{0}' pass as cgscc pipeline", Name).str(), 2701 inconvertibleErrorCode()); 2702 } 2703 2704 // Now expand the basic registered passes from the .inc file. 2705 #define CGSCC_PASS(NAME, CREATE_PASS) \ 2706 if (Name == NAME) { \ 2707 CGPM.addPass(CREATE_PASS); \ 2708 return Error::success(); \ 2709 } 2710 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 2711 if (Name == "require<" NAME ">") { \ 2712 CGPM.addPass(RequireAnalysisPass< \ 2713 std::remove_reference<decltype(CREATE_PASS)>::type, \ 2714 LazyCallGraph::SCC, CGSCCAnalysisManager, LazyCallGraph &, \ 2715 CGSCCUpdateResult &>()); \ 2716 return Error::success(); \ 2717 } \ 2718 if (Name == "invalidate<" NAME ">") { \ 2719 CGPM.addPass(InvalidateAnalysisPass< \ 2720 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2721 return Error::success(); \ 2722 } 2723 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2724 if (Name == NAME) { \ 2725 CGPM.addPass(createCGSCCToFunctionPassAdaptor(CREATE_PASS)); \ 2726 return Error::success(); \ 2727 } 2728 #define FUNCTION_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2729 if (checkParametrizedPassName(Name, NAME)) { \ 2730 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2731 if (!Params) \ 2732 return Params.takeError(); \ 2733 CGPM.addPass(createCGSCCToFunctionPassAdaptor(CREATE_PASS(Params.get()))); \ 2734 return Error::success(); \ 2735 } 2736 #define LOOP_PASS(NAME, CREATE_PASS) \ 2737 if (Name == NAME) { \ 2738 CGPM.addPass(createCGSCCToFunctionPassAdaptor( \ 2739 createFunctionToLoopPassAdaptor(CREATE_PASS, false, false))); \ 2740 return Error::success(); \ 2741 } 2742 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2743 if (checkParametrizedPassName(Name, NAME)) { \ 2744 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2745 if (!Params) \ 2746 return Params.takeError(); \ 2747 CGPM.addPass( \ 2748 createCGSCCToFunctionPassAdaptor(createFunctionToLoopPassAdaptor( \ 2749 CREATE_PASS(Params.get()), false, false))); \ 2750 return Error::success(); \ 2751 } 2752 #include "PassRegistry.def" 2753 2754 for (auto &C : CGSCCPipelineParsingCallbacks) 2755 if (C(Name, CGPM, InnerPipeline)) 2756 return Error::success(); 2757 return make_error<StringError>( 2758 formatv("unknown cgscc pass '{0}'", Name).str(), 2759 inconvertibleErrorCode()); 2760 } 2761 2762 Error PassBuilder::parseFunctionPass(FunctionPassManager &FPM, 2763 const PipelineElement &E) { 2764 auto &Name = E.Name; 2765 auto &InnerPipeline = E.InnerPipeline; 2766 2767 // First handle complex passes like the pass managers which carry pipelines. 2768 if (!InnerPipeline.empty()) { 2769 if (Name == "function") { 2770 FunctionPassManager NestedFPM; 2771 if (auto Err = parseFunctionPassPipeline(NestedFPM, InnerPipeline)) 2772 return Err; 2773 // Add the nested pass manager with the appropriate adaptor. 2774 FPM.addPass(std::move(NestedFPM)); 2775 return Error::success(); 2776 } 2777 if (Name == "loop" || Name == "loop-mssa") { 2778 LoopPassManager LPM; 2779 if (auto Err = parseLoopPassPipeline(LPM, InnerPipeline)) 2780 return Err; 2781 // Add the nested pass manager with the appropriate adaptor. 2782 bool UseMemorySSA = (Name == "loop-mssa"); 2783 bool UseBFI = llvm::any_of( 2784 InnerPipeline, [](auto Pipeline) { return Pipeline.Name == "licm"; }); 2785 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM), UseMemorySSA, 2786 UseBFI)); 2787 return Error::success(); 2788 } 2789 if (auto Count = parseRepeatPassName(Name)) { 2790 FunctionPassManager NestedFPM; 2791 if (auto Err = parseFunctionPassPipeline(NestedFPM, InnerPipeline)) 2792 return Err; 2793 FPM.addPass(createRepeatedPass(*Count, std::move(NestedFPM))); 2794 return Error::success(); 2795 } 2796 2797 for (auto &C : FunctionPipelineParsingCallbacks) 2798 if (C(Name, FPM, InnerPipeline)) 2799 return Error::success(); 2800 2801 // Normal passes can't have pipelines. 2802 return make_error<StringError>( 2803 formatv("invalid use of '{0}' pass as function pipeline", Name).str(), 2804 inconvertibleErrorCode()); 2805 } 2806 2807 // Now expand the basic registered passes from the .inc file. 2808 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2809 if (Name == NAME) { \ 2810 FPM.addPass(CREATE_PASS); \ 2811 return Error::success(); \ 2812 } 2813 #define FUNCTION_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2814 if (checkParametrizedPassName(Name, NAME)) { \ 2815 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2816 if (!Params) \ 2817 return Params.takeError(); \ 2818 FPM.addPass(CREATE_PASS(Params.get())); \ 2819 return Error::success(); \ 2820 } 2821 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 2822 if (Name == "require<" NAME ">") { \ 2823 FPM.addPass( \ 2824 RequireAnalysisPass< \ 2825 std::remove_reference<decltype(CREATE_PASS)>::type, Function>()); \ 2826 return Error::success(); \ 2827 } \ 2828 if (Name == "invalidate<" NAME ">") { \ 2829 FPM.addPass(InvalidateAnalysisPass< \ 2830 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2831 return Error::success(); \ 2832 } 2833 // FIXME: UseMemorySSA is set to false. Maybe we could do things like: 2834 // bool UseMemorySSA = !("canon-freeze" || "loop-predication" || 2835 // "guard-widening"); 2836 // The risk is that it may become obsolete if we're not careful. 2837 #define LOOP_PASS(NAME, CREATE_PASS) \ 2838 if (Name == NAME) { \ 2839 FPM.addPass(createFunctionToLoopPassAdaptor(CREATE_PASS, false, false)); \ 2840 return Error::success(); \ 2841 } 2842 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2843 if (checkParametrizedPassName(Name, NAME)) { \ 2844 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2845 if (!Params) \ 2846 return Params.takeError(); \ 2847 FPM.addPass(createFunctionToLoopPassAdaptor(CREATE_PASS(Params.get()), \ 2848 false, false)); \ 2849 return Error::success(); \ 2850 } 2851 #include "PassRegistry.def" 2852 2853 for (auto &C : FunctionPipelineParsingCallbacks) 2854 if (C(Name, FPM, InnerPipeline)) 2855 return Error::success(); 2856 return make_error<StringError>( 2857 formatv("unknown function pass '{0}'", Name).str(), 2858 inconvertibleErrorCode()); 2859 } 2860 2861 Error PassBuilder::parseLoopPass(LoopPassManager &LPM, 2862 const PipelineElement &E) { 2863 StringRef Name = E.Name; 2864 auto &InnerPipeline = E.InnerPipeline; 2865 2866 // First handle complex passes like the pass managers which carry pipelines. 2867 if (!InnerPipeline.empty()) { 2868 if (Name == "loop") { 2869 LoopPassManager NestedLPM; 2870 if (auto Err = parseLoopPassPipeline(NestedLPM, InnerPipeline)) 2871 return Err; 2872 // Add the nested pass manager with the appropriate adaptor. 2873 LPM.addPass(std::move(NestedLPM)); 2874 return Error::success(); 2875 } 2876 if (auto Count = parseRepeatPassName(Name)) { 2877 LoopPassManager NestedLPM; 2878 if (auto Err = parseLoopPassPipeline(NestedLPM, InnerPipeline)) 2879 return Err; 2880 LPM.addPass(createRepeatedPass(*Count, std::move(NestedLPM))); 2881 return Error::success(); 2882 } 2883 2884 for (auto &C : LoopPipelineParsingCallbacks) 2885 if (C(Name, LPM, InnerPipeline)) 2886 return Error::success(); 2887 2888 // Normal passes can't have pipelines. 2889 return make_error<StringError>( 2890 formatv("invalid use of '{0}' pass as loop pipeline", Name).str(), 2891 inconvertibleErrorCode()); 2892 } 2893 2894 // Now expand the basic registered passes from the .inc file. 2895 #define LOOP_PASS(NAME, CREATE_PASS) \ 2896 if (Name == NAME) { \ 2897 LPM.addPass(CREATE_PASS); \ 2898 return Error::success(); \ 2899 } 2900 #define LOOP_PASS_WITH_PARAMS(NAME, CREATE_PASS, PARSER) \ 2901 if (checkParametrizedPassName(Name, NAME)) { \ 2902 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2903 if (!Params) \ 2904 return Params.takeError(); \ 2905 LPM.addPass(CREATE_PASS(Params.get())); \ 2906 return Error::success(); \ 2907 } 2908 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 2909 if (Name == "require<" NAME ">") { \ 2910 LPM.addPass(RequireAnalysisPass< \ 2911 std::remove_reference<decltype(CREATE_PASS)>::type, Loop, \ 2912 LoopAnalysisManager, LoopStandardAnalysisResults &, \ 2913 LPMUpdater &>()); \ 2914 return Error::success(); \ 2915 } \ 2916 if (Name == "invalidate<" NAME ">") { \ 2917 LPM.addPass(InvalidateAnalysisPass< \ 2918 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2919 return Error::success(); \ 2920 } 2921 #include "PassRegistry.def" 2922 2923 for (auto &C : LoopPipelineParsingCallbacks) 2924 if (C(Name, LPM, InnerPipeline)) 2925 return Error::success(); 2926 return make_error<StringError>(formatv("unknown loop pass '{0}'", Name).str(), 2927 inconvertibleErrorCode()); 2928 } 2929 2930 bool PassBuilder::parseAAPassName(AAManager &AA, StringRef Name) { 2931 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 2932 if (Name == NAME) { \ 2933 AA.registerModuleAnalysis< \ 2934 std::remove_reference<decltype(CREATE_PASS)>::type>(); \ 2935 return true; \ 2936 } 2937 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 2938 if (Name == NAME) { \ 2939 AA.registerFunctionAnalysis< \ 2940 std::remove_reference<decltype(CREATE_PASS)>::type>(); \ 2941 return true; \ 2942 } 2943 #include "PassRegistry.def" 2944 2945 for (auto &C : AAParsingCallbacks) 2946 if (C(Name, AA)) 2947 return true; 2948 return false; 2949 } 2950 2951 Error PassBuilder::parseLoopPassPipeline(LoopPassManager &LPM, 2952 ArrayRef<PipelineElement> Pipeline) { 2953 for (const auto &Element : Pipeline) { 2954 if (auto Err = parseLoopPass(LPM, Element)) 2955 return Err; 2956 } 2957 return Error::success(); 2958 } 2959 2960 Error PassBuilder::parseFunctionPassPipeline( 2961 FunctionPassManager &FPM, ArrayRef<PipelineElement> Pipeline) { 2962 for (const auto &Element : Pipeline) { 2963 if (auto Err = parseFunctionPass(FPM, Element)) 2964 return Err; 2965 } 2966 return Error::success(); 2967 } 2968 2969 Error PassBuilder::parseCGSCCPassPipeline(CGSCCPassManager &CGPM, 2970 ArrayRef<PipelineElement> Pipeline) { 2971 for (const auto &Element : Pipeline) { 2972 if (auto Err = parseCGSCCPass(CGPM, Element)) 2973 return Err; 2974 } 2975 return Error::success(); 2976 } 2977 2978 void PassBuilder::crossRegisterProxies(LoopAnalysisManager &LAM, 2979 FunctionAnalysisManager &FAM, 2980 CGSCCAnalysisManager &CGAM, 2981 ModuleAnalysisManager &MAM) { 2982 MAM.registerPass([&] { return FunctionAnalysisManagerModuleProxy(FAM); }); 2983 MAM.registerPass([&] { return CGSCCAnalysisManagerModuleProxy(CGAM); }); 2984 CGAM.registerPass([&] { return ModuleAnalysisManagerCGSCCProxy(MAM); }); 2985 FAM.registerPass([&] { return CGSCCAnalysisManagerFunctionProxy(CGAM); }); 2986 FAM.registerPass([&] { return ModuleAnalysisManagerFunctionProxy(MAM); }); 2987 FAM.registerPass([&] { return LoopAnalysisManagerFunctionProxy(LAM); }); 2988 LAM.registerPass([&] { return FunctionAnalysisManagerLoopProxy(FAM); }); 2989 } 2990 2991 Error PassBuilder::parseModulePassPipeline(ModulePassManager &MPM, 2992 ArrayRef<PipelineElement> Pipeline) { 2993 for (const auto &Element : Pipeline) { 2994 if (auto Err = parseModulePass(MPM, Element)) 2995 return Err; 2996 } 2997 return Error::success(); 2998 } 2999 3000 // Primary pass pipeline description parsing routine for a \c ModulePassManager 3001 // FIXME: Should this routine accept a TargetMachine or require the caller to 3002 // pre-populate the analysis managers with target-specific stuff? 3003 Error PassBuilder::parsePassPipeline(ModulePassManager &MPM, 3004 StringRef PipelineText) { 3005 auto Pipeline = parsePipelineText(PipelineText); 3006 if (!Pipeline || Pipeline->empty()) 3007 return make_error<StringError>( 3008 formatv("invalid pipeline '{0}'", PipelineText).str(), 3009 inconvertibleErrorCode()); 3010 3011 // If the first name isn't at the module layer, wrap the pipeline up 3012 // automatically. 3013 StringRef FirstName = Pipeline->front().Name; 3014 3015 if (!isModulePassName(FirstName, ModulePipelineParsingCallbacks)) { 3016 if (isCGSCCPassName(FirstName, CGSCCPipelineParsingCallbacks)) { 3017 Pipeline = {{"cgscc", std::move(*Pipeline)}}; 3018 } else if (isFunctionPassName(FirstName, 3019 FunctionPipelineParsingCallbacks)) { 3020 Pipeline = {{"function", std::move(*Pipeline)}}; 3021 } else if (isLoopPassName(FirstName, LoopPipelineParsingCallbacks)) { 3022 Pipeline = {{"function", {{"loop", std::move(*Pipeline)}}}}; 3023 } else { 3024 for (auto &C : TopLevelPipelineParsingCallbacks) 3025 if (C(MPM, *Pipeline)) 3026 return Error::success(); 3027 3028 // Unknown pass or pipeline name! 3029 auto &InnerPipeline = Pipeline->front().InnerPipeline; 3030 return make_error<StringError>( 3031 formatv("unknown {0} name '{1}'", 3032 (InnerPipeline.empty() ? "pass" : "pipeline"), FirstName) 3033 .str(), 3034 inconvertibleErrorCode()); 3035 } 3036 } 3037 3038 if (auto Err = parseModulePassPipeline(MPM, *Pipeline)) 3039 return Err; 3040 return Error::success(); 3041 } 3042 3043 // Primary pass pipeline description parsing routine for a \c CGSCCPassManager 3044 Error PassBuilder::parsePassPipeline(CGSCCPassManager &CGPM, 3045 StringRef PipelineText) { 3046 auto Pipeline = parsePipelineText(PipelineText); 3047 if (!Pipeline || Pipeline->empty()) 3048 return make_error<StringError>( 3049 formatv("invalid pipeline '{0}'", PipelineText).str(), 3050 inconvertibleErrorCode()); 3051 3052 StringRef FirstName = Pipeline->front().Name; 3053 if (!isCGSCCPassName(FirstName, CGSCCPipelineParsingCallbacks)) 3054 return make_error<StringError>( 3055 formatv("unknown cgscc pass '{0}' in pipeline '{1}'", FirstName, 3056 PipelineText) 3057 .str(), 3058 inconvertibleErrorCode()); 3059 3060 if (auto Err = parseCGSCCPassPipeline(CGPM, *Pipeline)) 3061 return Err; 3062 return Error::success(); 3063 } 3064 3065 // Primary pass pipeline description parsing routine for a \c 3066 // FunctionPassManager 3067 Error PassBuilder::parsePassPipeline(FunctionPassManager &FPM, 3068 StringRef PipelineText) { 3069 auto Pipeline = parsePipelineText(PipelineText); 3070 if (!Pipeline || Pipeline->empty()) 3071 return make_error<StringError>( 3072 formatv("invalid pipeline '{0}'", PipelineText).str(), 3073 inconvertibleErrorCode()); 3074 3075 StringRef FirstName = Pipeline->front().Name; 3076 if (!isFunctionPassName(FirstName, FunctionPipelineParsingCallbacks)) 3077 return make_error<StringError>( 3078 formatv("unknown function pass '{0}' in pipeline '{1}'", FirstName, 3079 PipelineText) 3080 .str(), 3081 inconvertibleErrorCode()); 3082 3083 if (auto Err = parseFunctionPassPipeline(FPM, *Pipeline)) 3084 return Err; 3085 return Error::success(); 3086 } 3087 3088 // Primary pass pipeline description parsing routine for a \c LoopPassManager 3089 Error PassBuilder::parsePassPipeline(LoopPassManager &CGPM, 3090 StringRef PipelineText) { 3091 auto Pipeline = parsePipelineText(PipelineText); 3092 if (!Pipeline || Pipeline->empty()) 3093 return make_error<StringError>( 3094 formatv("invalid pipeline '{0}'", PipelineText).str(), 3095 inconvertibleErrorCode()); 3096 3097 if (auto Err = parseLoopPassPipeline(CGPM, *Pipeline)) 3098 return Err; 3099 3100 return Error::success(); 3101 } 3102 3103 Error PassBuilder::parseAAPipeline(AAManager &AA, StringRef PipelineText) { 3104 // If the pipeline just consists of the word 'default' just replace the AA 3105 // manager with our default one. 3106 if (PipelineText == "default") { 3107 AA = buildDefaultAAPipeline(); 3108 return Error::success(); 3109 } 3110 3111 while (!PipelineText.empty()) { 3112 StringRef Name; 3113 std::tie(Name, PipelineText) = PipelineText.split(','); 3114 if (!parseAAPassName(AA, Name)) 3115 return make_error<StringError>( 3116 formatv("unknown alias analysis name '{0}'", Name).str(), 3117 inconvertibleErrorCode()); 3118 } 3119 3120 return Error::success(); 3121 } 3122 3123 bool PassBuilder::isAAPassName(StringRef PassName) { 3124 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3125 if (PassName == NAME) \ 3126 return true; 3127 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3128 if (PassName == NAME) \ 3129 return true; 3130 #include "PassRegistry.def" 3131 return false; 3132 } 3133 3134 bool PassBuilder::isAnalysisPassName(StringRef PassName) { 3135 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 3136 if (PassName == NAME) \ 3137 return true; 3138 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 3139 if (PassName == NAME) \ 3140 return true; 3141 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 3142 if (PassName == NAME) \ 3143 return true; 3144 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 3145 if (PassName == NAME) \ 3146 return true; 3147 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3148 if (PassName == NAME) \ 3149 return true; 3150 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3151 if (PassName == NAME) \ 3152 return true; 3153 #include "PassRegistry.def" 3154 return false; 3155 } 3156 3157 static void printPassName(StringRef PassName, raw_ostream &OS) { 3158 OS << " " << PassName << "\n"; 3159 } 3160 3161 void PassBuilder::printPassNames(raw_ostream &OS) { 3162 // TODO: print pass descriptions when they are available 3163 3164 OS << "Module passes:\n"; 3165 #define MODULE_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3166 #include "PassRegistry.def" 3167 3168 OS << "Module analyses:\n"; 3169 #define MODULE_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3170 #include "PassRegistry.def" 3171 3172 OS << "Module alias analyses:\n"; 3173 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3174 #include "PassRegistry.def" 3175 3176 OS << "CGSCC passes:\n"; 3177 #define CGSCC_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3178 #include "PassRegistry.def" 3179 3180 OS << "CGSCC analyses:\n"; 3181 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3182 #include "PassRegistry.def" 3183 3184 OS << "Function passes:\n"; 3185 #define FUNCTION_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3186 #include "PassRegistry.def" 3187 3188 OS << "Function analyses:\n"; 3189 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3190 #include "PassRegistry.def" 3191 3192 OS << "Function alias analyses:\n"; 3193 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3194 #include "PassRegistry.def" 3195 3196 OS << "Loop passes:\n"; 3197 #define LOOP_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3198 #include "PassRegistry.def" 3199 3200 OS << "Loop analyses:\n"; 3201 #define LOOP_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3202 #include "PassRegistry.def" 3203 } 3204 3205 void PassBuilder::registerParseTopLevelPipelineCallback( 3206 const std::function<bool(ModulePassManager &, ArrayRef<PipelineElement>)> 3207 &C) { 3208 TopLevelPipelineParsingCallbacks.push_back(C); 3209 } 3210