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