1 //===- Parsing, selection, and construction of pass pipelines -------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 /// \file 9 /// 10 /// This file provides the implementation of the PassBuilder based on our 11 /// static pass registry as well as related functionality. It also provides 12 /// helpers to aid in analyzing, debugging, and testing passes and pass 13 /// pipelines. 14 /// 15 //===----------------------------------------------------------------------===// 16 17 #include "llvm/Passes/PassBuilder.h" 18 #include "llvm/ADT/StringSwitch.h" 19 #include "llvm/Analysis/AliasAnalysisEvaluator.h" 20 #include "llvm/Analysis/AliasSetTracker.h" 21 #include "llvm/Analysis/AssumptionCache.h" 22 #include "llvm/Analysis/BasicAliasAnalysis.h" 23 #include "llvm/Analysis/BlockFrequencyInfo.h" 24 #include "llvm/Analysis/BranchProbabilityInfo.h" 25 #include "llvm/Analysis/CFGPrinter.h" 26 #include "llvm/Analysis/CFLAndersAliasAnalysis.h" 27 #include "llvm/Analysis/CFLSteensAliasAnalysis.h" 28 #include "llvm/Analysis/CGSCCPassManager.h" 29 #include "llvm/Analysis/CallGraph.h" 30 #include "llvm/Analysis/DDG.h" 31 #include "llvm/Analysis/DDGPrinter.h" 32 #include "llvm/Analysis/Delinearization.h" 33 #include "llvm/Analysis/DemandedBits.h" 34 #include "llvm/Analysis/DependenceAnalysis.h" 35 #include "llvm/Analysis/DivergenceAnalysis.h" 36 #include "llvm/Analysis/DominanceFrontier.h" 37 #include "llvm/Analysis/FunctionPropertiesAnalysis.h" 38 #include "llvm/Analysis/GlobalsModRef.h" 39 #include "llvm/Analysis/IRSimilarityIdentifier.h" 40 #include "llvm/Analysis/IVUsers.h" 41 #include "llvm/Analysis/InlineAdvisor.h" 42 #include "llvm/Analysis/InlineSizeEstimatorAnalysis.h" 43 #include "llvm/Analysis/InstCount.h" 44 #include "llvm/Analysis/LazyCallGraph.h" 45 #include "llvm/Analysis/LazyValueInfo.h" 46 #include "llvm/Analysis/Lint.h" 47 #include "llvm/Analysis/LoopAccessAnalysis.h" 48 #include "llvm/Analysis/LoopCacheAnalysis.h" 49 #include "llvm/Analysis/LoopInfo.h" 50 #include "llvm/Analysis/LoopNestAnalysis.h" 51 #include "llvm/Analysis/MemDerefPrinter.h" 52 #include "llvm/Analysis/MemoryDependenceAnalysis.h" 53 #include "llvm/Analysis/MemorySSA.h" 54 #include "llvm/Analysis/ModuleDebugInfoPrinter.h" 55 #include "llvm/Analysis/ModuleSummaryAnalysis.h" 56 #include "llvm/Analysis/MustExecute.h" 57 #include "llvm/Analysis/ObjCARCAliasAnalysis.h" 58 #include "llvm/Analysis/OptimizationRemarkEmitter.h" 59 #include "llvm/Analysis/PhiValues.h" 60 #include "llvm/Analysis/PostDominators.h" 61 #include "llvm/Analysis/ProfileSummaryInfo.h" 62 #include "llvm/Analysis/RegionInfo.h" 63 #include "llvm/Analysis/ScalarEvolution.h" 64 #include "llvm/Analysis/ScalarEvolutionAliasAnalysis.h" 65 #include "llvm/Analysis/ScopedNoAliasAA.h" 66 #include "llvm/Analysis/StackLifetime.h" 67 #include "llvm/Analysis/StackSafetyAnalysis.h" 68 #include "llvm/Analysis/TargetLibraryInfo.h" 69 #include "llvm/Analysis/TargetTransformInfo.h" 70 #include "llvm/Analysis/TypeBasedAliasAnalysis.h" 71 #include "llvm/IR/Dominators.h" 72 #include "llvm/IR/IRPrintingPasses.h" 73 #include "llvm/IR/PassManager.h" 74 #include "llvm/IR/PrintPasses.h" 75 #include "llvm/IR/SafepointIRVerifier.h" 76 #include "llvm/IR/Verifier.h" 77 #include "llvm/Support/CommandLine.h" 78 #include "llvm/Support/Debug.h" 79 #include "llvm/Support/ErrorHandling.h" 80 #include "llvm/Support/FormatVariadic.h" 81 #include "llvm/Support/Regex.h" 82 #include "llvm/Target/TargetMachine.h" 83 #include "llvm/Transforms/AggressiveInstCombine/AggressiveInstCombine.h" 84 #include "llvm/Transforms/Coroutines/CoroCleanup.h" 85 #include "llvm/Transforms/Coroutines/CoroEarly.h" 86 #include "llvm/Transforms/Coroutines/CoroElide.h" 87 #include "llvm/Transforms/Coroutines/CoroSplit.h" 88 #include "llvm/Transforms/IPO/AlwaysInliner.h" 89 #include "llvm/Transforms/IPO/Annotation2Metadata.h" 90 #include "llvm/Transforms/IPO/ArgumentPromotion.h" 91 #include "llvm/Transforms/IPO/Attributor.h" 92 #include "llvm/Transforms/IPO/BlockExtractor.h" 93 #include "llvm/Transforms/IPO/CalledValuePropagation.h" 94 #include "llvm/Transforms/IPO/ConstantMerge.h" 95 #include "llvm/Transforms/IPO/CrossDSOCFI.h" 96 #include "llvm/Transforms/IPO/DeadArgumentElimination.h" 97 #include "llvm/Transforms/IPO/ElimAvailExtern.h" 98 #include "llvm/Transforms/IPO/ForceFunctionAttrs.h" 99 #include "llvm/Transforms/IPO/FunctionAttrs.h" 100 #include "llvm/Transforms/IPO/FunctionImport.h" 101 #include "llvm/Transforms/IPO/GlobalDCE.h" 102 #include "llvm/Transforms/IPO/GlobalOpt.h" 103 #include "llvm/Transforms/IPO/GlobalSplit.h" 104 #include "llvm/Transforms/IPO/HotColdSplitting.h" 105 #include "llvm/Transforms/IPO/IROutliner.h" 106 #include "llvm/Transforms/IPO/InferFunctionAttrs.h" 107 #include "llvm/Transforms/IPO/Inliner.h" 108 #include "llvm/Transforms/IPO/Internalize.h" 109 #include "llvm/Transforms/IPO/LoopExtractor.h" 110 #include "llvm/Transforms/IPO/LowerTypeTests.h" 111 #include "llvm/Transforms/IPO/MergeFunctions.h" 112 #include "llvm/Transforms/IPO/OpenMPOpt.h" 113 #include "llvm/Transforms/IPO/PartialInlining.h" 114 #include "llvm/Transforms/IPO/SCCP.h" 115 #include "llvm/Transforms/IPO/SampleProfile.h" 116 #include "llvm/Transforms/IPO/SampleProfileProbe.h" 117 #include "llvm/Transforms/IPO/StripDeadPrototypes.h" 118 #include "llvm/Transforms/IPO/StripSymbols.h" 119 #include "llvm/Transforms/IPO/SyntheticCountsPropagation.h" 120 #include "llvm/Transforms/IPO/WholeProgramDevirt.h" 121 #include "llvm/Transforms/InstCombine/InstCombine.h" 122 #include "llvm/Transforms/Instrumentation.h" 123 #include "llvm/Transforms/Instrumentation/AddressSanitizer.h" 124 #include "llvm/Transforms/Instrumentation/BoundsChecking.h" 125 #include "llvm/Transforms/Instrumentation/CGProfile.h" 126 #include "llvm/Transforms/Instrumentation/ControlHeightReduction.h" 127 #include "llvm/Transforms/Instrumentation/DataFlowSanitizer.h" 128 #include "llvm/Transforms/Instrumentation/GCOVProfiler.h" 129 #include "llvm/Transforms/Instrumentation/HWAddressSanitizer.h" 130 #include "llvm/Transforms/Instrumentation/InstrOrderFile.h" 131 #include "llvm/Transforms/Instrumentation/InstrProfiling.h" 132 #include "llvm/Transforms/Instrumentation/MemProfiler.h" 133 #include "llvm/Transforms/Instrumentation/MemorySanitizer.h" 134 #include "llvm/Transforms/Instrumentation/PGOInstrumentation.h" 135 #include "llvm/Transforms/Instrumentation/PoisonChecking.h" 136 #include "llvm/Transforms/Instrumentation/SanitizerCoverage.h" 137 #include "llvm/Transforms/Instrumentation/ThreadSanitizer.h" 138 #include "llvm/Transforms/ObjCARC.h" 139 #include "llvm/Transforms/Scalar/ADCE.h" 140 #include "llvm/Transforms/Scalar/AlignmentFromAssumptions.h" 141 #include "llvm/Transforms/Scalar/AnnotationRemarks.h" 142 #include "llvm/Transforms/Scalar/BDCE.h" 143 #include "llvm/Transforms/Scalar/CallSiteSplitting.h" 144 #include "llvm/Transforms/Scalar/ConstantHoisting.h" 145 #include "llvm/Transforms/Scalar/ConstraintElimination.h" 146 #include "llvm/Transforms/Scalar/CorrelatedValuePropagation.h" 147 #include "llvm/Transforms/Scalar/DCE.h" 148 #include "llvm/Transforms/Scalar/DeadStoreElimination.h" 149 #include "llvm/Transforms/Scalar/DivRemPairs.h" 150 #include "llvm/Transforms/Scalar/EarlyCSE.h" 151 #include "llvm/Transforms/Scalar/Float2Int.h" 152 #include "llvm/Transforms/Scalar/GVN.h" 153 #include "llvm/Transforms/Scalar/GuardWidening.h" 154 #include "llvm/Transforms/Scalar/IVUsersPrinter.h" 155 #include "llvm/Transforms/Scalar/IndVarSimplify.h" 156 #include "llvm/Transforms/Scalar/InductiveRangeCheckElimination.h" 157 #include "llvm/Transforms/Scalar/InferAddressSpaces.h" 158 #include "llvm/Transforms/Scalar/InstSimplifyPass.h" 159 #include "llvm/Transforms/Scalar/JumpThreading.h" 160 #include "llvm/Transforms/Scalar/LICM.h" 161 #include "llvm/Transforms/Scalar/LoopAccessAnalysisPrinter.h" 162 #include "llvm/Transforms/Scalar/LoopBoundSplit.h" 163 #include "llvm/Transforms/Scalar/LoopDataPrefetch.h" 164 #include "llvm/Transforms/Scalar/LoopDeletion.h" 165 #include "llvm/Transforms/Scalar/LoopDistribute.h" 166 #include "llvm/Transforms/Scalar/LoopFlatten.h" 167 #include "llvm/Transforms/Scalar/LoopFuse.h" 168 #include "llvm/Transforms/Scalar/LoopIdiomRecognize.h" 169 #include "llvm/Transforms/Scalar/LoopInstSimplify.h" 170 #include "llvm/Transforms/Scalar/LoopInterchange.h" 171 #include "llvm/Transforms/Scalar/LoopLoadElimination.h" 172 #include "llvm/Transforms/Scalar/LoopPassManager.h" 173 #include "llvm/Transforms/Scalar/LoopPredication.h" 174 #include "llvm/Transforms/Scalar/LoopReroll.h" 175 #include "llvm/Transforms/Scalar/LoopRotation.h" 176 #include "llvm/Transforms/Scalar/LoopSimplifyCFG.h" 177 #include "llvm/Transforms/Scalar/LoopSink.h" 178 #include "llvm/Transforms/Scalar/LoopStrengthReduce.h" 179 #include "llvm/Transforms/Scalar/LoopUnrollAndJamPass.h" 180 #include "llvm/Transforms/Scalar/LoopUnrollPass.h" 181 #include "llvm/Transforms/Scalar/LoopVersioningLICM.h" 182 #include "llvm/Transforms/Scalar/LowerAtomic.h" 183 #include "llvm/Transforms/Scalar/LowerConstantIntrinsics.h" 184 #include "llvm/Transforms/Scalar/LowerExpectIntrinsic.h" 185 #include "llvm/Transforms/Scalar/LowerGuardIntrinsic.h" 186 #include "llvm/Transforms/Scalar/LowerMatrixIntrinsics.h" 187 #include "llvm/Transforms/Scalar/LowerWidenableCondition.h" 188 #include "llvm/Transforms/Scalar/MakeGuardsExplicit.h" 189 #include "llvm/Transforms/Scalar/MemCpyOptimizer.h" 190 #include "llvm/Transforms/Scalar/MergeICmps.h" 191 #include "llvm/Transforms/Scalar/MergedLoadStoreMotion.h" 192 #include "llvm/Transforms/Scalar/NaryReassociate.h" 193 #include "llvm/Transforms/Scalar/NewGVN.h" 194 #include "llvm/Transforms/Scalar/PartiallyInlineLibCalls.h" 195 #include "llvm/Transforms/Scalar/Reassociate.h" 196 #include "llvm/Transforms/Scalar/Reg2Mem.h" 197 #include "llvm/Transforms/Scalar/RewriteStatepointsForGC.h" 198 #include "llvm/Transforms/Scalar/SCCP.h" 199 #include "llvm/Transforms/Scalar/SROA.h" 200 #include "llvm/Transforms/Scalar/ScalarizeMaskedMemIntrin.h" 201 #include "llvm/Transforms/Scalar/Scalarizer.h" 202 #include "llvm/Transforms/Scalar/SeparateConstOffsetFromGEP.h" 203 #include "llvm/Transforms/Scalar/SimpleLoopUnswitch.h" 204 #include "llvm/Transforms/Scalar/SimplifyCFG.h" 205 #include "llvm/Transforms/Scalar/Sink.h" 206 #include "llvm/Transforms/Scalar/SpeculativeExecution.h" 207 #include "llvm/Transforms/Scalar/StraightLineStrengthReduce.h" 208 #include "llvm/Transforms/Scalar/StructurizeCFG.h" 209 #include "llvm/Transforms/Scalar/TailRecursionElimination.h" 210 #include "llvm/Transforms/Scalar/WarnMissedTransforms.h" 211 #include "llvm/Transforms/Utils/AddDiscriminators.h" 212 #include "llvm/Transforms/Utils/AssumeBundleBuilder.h" 213 #include "llvm/Transforms/Utils/BreakCriticalEdges.h" 214 #include "llvm/Transforms/Utils/CanonicalizeAliases.h" 215 #include "llvm/Transforms/Utils/CanonicalizeFreezeInLoops.h" 216 #include "llvm/Transforms/Utils/EntryExitInstrumenter.h" 217 #include "llvm/Transforms/Utils/FixIrreducible.h" 218 #include "llvm/Transforms/Utils/HelloWorld.h" 219 #include "llvm/Transforms/Utils/InjectTLIMappings.h" 220 #include "llvm/Transforms/Utils/InstructionNamer.h" 221 #include "llvm/Transforms/Utils/LCSSA.h" 222 #include "llvm/Transforms/Utils/LibCallsShrinkWrap.h" 223 #include "llvm/Transforms/Utils/LoopSimplify.h" 224 #include "llvm/Transforms/Utils/LoopVersioning.h" 225 #include "llvm/Transforms/Utils/LowerInvoke.h" 226 #include "llvm/Transforms/Utils/LowerSwitch.h" 227 #include "llvm/Transforms/Utils/Mem2Reg.h" 228 #include "llvm/Transforms/Utils/MetaRenamer.h" 229 #include "llvm/Transforms/Utils/NameAnonGlobals.h" 230 #include "llvm/Transforms/Utils/RelLookupTableConverter.h" 231 #include "llvm/Transforms/Utils/StripGCRelocates.h" 232 #include "llvm/Transforms/Utils/StripNonLineTableDebugInfo.h" 233 #include "llvm/Transforms/Utils/SymbolRewriter.h" 234 #include "llvm/Transforms/Utils/UnifyFunctionExitNodes.h" 235 #include "llvm/Transforms/Utils/UnifyLoopExits.h" 236 #include "llvm/Transforms/Vectorize/LoadStoreVectorizer.h" 237 #include "llvm/Transforms/Vectorize/LoopVectorize.h" 238 #include "llvm/Transforms/Vectorize/SLPVectorizer.h" 239 #include "llvm/Transforms/Vectorize/VectorCombine.h" 240 241 using namespace llvm; 242 243 static cl::opt<InliningAdvisorMode> UseInlineAdvisor( 244 "enable-ml-inliner", cl::init(InliningAdvisorMode::Default), cl::Hidden, 245 cl::desc("Enable ML policy for inliner. Currently trained for -Oz only"), 246 cl::values(clEnumValN(InliningAdvisorMode::Default, "default", 247 "Heuristics-based inliner version."), 248 clEnumValN(InliningAdvisorMode::Development, "development", 249 "Use development mode (runtime-loadable model)."), 250 clEnumValN(InliningAdvisorMode::Release, "release", 251 "Use release mode (AOT-compiled model)."))); 252 253 static cl::opt<bool> EnableSyntheticCounts( 254 "enable-npm-synthetic-counts", cl::init(false), cl::Hidden, cl::ZeroOrMore, 255 cl::desc("Run synthetic function entry count generation " 256 "pass")); 257 258 static const Regex DefaultAliasRegex( 259 "^(default|thinlto-pre-link|thinlto|lto-pre-link|lto)<(O[0123sz])>$"); 260 261 /// Flag to enable inline deferral during PGO. 262 static cl::opt<bool> 263 EnablePGOInlineDeferral("enable-npm-pgo-inline-deferral", cl::init(true), 264 cl::Hidden, 265 cl::desc("Enable inline deferral during PGO")); 266 267 static cl::opt<bool> EnableMemProfiler("enable-mem-prof", cl::init(false), 268 cl::Hidden, cl::ZeroOrMore, 269 cl::desc("Enable memory profiler")); 270 271 static cl::opt<bool> PerformMandatoryInliningsFirst( 272 "mandatory-inlining-first", cl::init(true), cl::Hidden, cl::ZeroOrMore, 273 cl::desc("Perform mandatory inlinings module-wide, before performing " 274 "inlining.")); 275 276 static cl::opt<bool> EnableO3NonTrivialUnswitching( 277 "enable-npm-O3-nontrivial-unswitch", cl::init(true), cl::Hidden, 278 cl::ZeroOrMore, cl::desc("Enable non-trivial loop unswitching for -O3")); 279 280 PipelineTuningOptions::PipelineTuningOptions() { 281 LoopInterleaving = true; 282 LoopVectorization = true; 283 SLPVectorization = false; 284 LoopUnrolling = true; 285 ForgetAllSCEVInLoopUnroll = ForgetSCEVInLoopUnroll; 286 LicmMssaOptCap = SetLicmMssaOptCap; 287 LicmMssaNoAccForPromotionCap = SetLicmMssaNoAccForPromotionCap; 288 CallGraphProfile = true; 289 MergeFunctions = false; 290 } 291 292 namespace llvm { 293 extern cl::opt<unsigned> MaxDevirtIterations; 294 extern cl::opt<bool> EnableConstraintElimination; 295 extern cl::opt<bool> EnableFunctionSpecialization; 296 extern cl::opt<bool> EnableGVNHoist; 297 extern cl::opt<bool> EnableGVNSink; 298 extern cl::opt<bool> EnableHotColdSplit; 299 extern cl::opt<bool> EnableIROutliner; 300 extern cl::opt<bool> EnableOrderFileInstrumentation; 301 extern cl::opt<bool> EnableCHR; 302 extern cl::opt<bool> EnableLoopInterchange; 303 extern cl::opt<bool> EnableUnrollAndJam; 304 extern cl::opt<bool> EnableLoopFlatten; 305 extern cl::opt<bool> RunNewGVN; 306 extern cl::opt<bool> RunPartialInlining; 307 extern cl::opt<bool> ExtraVectorizerPasses; 308 309 extern cl::opt<bool> FlattenedProfileUsed; 310 311 extern cl::opt<AttributorRunOption> AttributorRun; 312 extern cl::opt<bool> EnableKnowledgeRetention; 313 314 extern cl::opt<bool> EnableMatrix; 315 316 extern cl::opt<bool> DisablePreInliner; 317 extern cl::opt<int> PreInlineThreshold; 318 } // namespace llvm 319 320 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O0 = { 321 /*SpeedLevel*/ 0, 322 /*SizeLevel*/ 0}; 323 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O1 = { 324 /*SpeedLevel*/ 1, 325 /*SizeLevel*/ 0}; 326 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O2 = { 327 /*SpeedLevel*/ 2, 328 /*SizeLevel*/ 0}; 329 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::O3 = { 330 /*SpeedLevel*/ 3, 331 /*SizeLevel*/ 0}; 332 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::Os = { 333 /*SpeedLevel*/ 2, 334 /*SizeLevel*/ 1}; 335 const PassBuilder::OptimizationLevel PassBuilder::OptimizationLevel::Oz = { 336 /*SpeedLevel*/ 2, 337 /*SizeLevel*/ 2}; 338 339 namespace { 340 341 // The following passes/analyses have custom names, otherwise their name will 342 // include `(anonymous namespace)`. These are special since they are only for 343 // testing purposes and don't live in a header file. 344 345 /// No-op module pass which does nothing. 346 struct NoOpModulePass : PassInfoMixin<NoOpModulePass> { 347 PreservedAnalyses run(Module &M, ModuleAnalysisManager &) { 348 return PreservedAnalyses::all(); 349 } 350 351 static StringRef name() { return "NoOpModulePass"; } 352 }; 353 354 /// No-op module analysis. 355 class NoOpModuleAnalysis : public AnalysisInfoMixin<NoOpModuleAnalysis> { 356 friend AnalysisInfoMixin<NoOpModuleAnalysis>; 357 static AnalysisKey Key; 358 359 public: 360 struct Result {}; 361 Result run(Module &, ModuleAnalysisManager &) { return Result(); } 362 static StringRef name() { return "NoOpModuleAnalysis"; } 363 }; 364 365 /// No-op CGSCC pass which does nothing. 366 struct NoOpCGSCCPass : PassInfoMixin<NoOpCGSCCPass> { 367 PreservedAnalyses run(LazyCallGraph::SCC &C, CGSCCAnalysisManager &, 368 LazyCallGraph &, CGSCCUpdateResult &UR) { 369 return PreservedAnalyses::all(); 370 } 371 static StringRef name() { return "NoOpCGSCCPass"; } 372 }; 373 374 /// No-op CGSCC analysis. 375 class NoOpCGSCCAnalysis : public AnalysisInfoMixin<NoOpCGSCCAnalysis> { 376 friend AnalysisInfoMixin<NoOpCGSCCAnalysis>; 377 static AnalysisKey Key; 378 379 public: 380 struct Result {}; 381 Result run(LazyCallGraph::SCC &, CGSCCAnalysisManager &, LazyCallGraph &G) { 382 return Result(); 383 } 384 static StringRef name() { return "NoOpCGSCCAnalysis"; } 385 }; 386 387 /// No-op function pass which does nothing. 388 struct NoOpFunctionPass : PassInfoMixin<NoOpFunctionPass> { 389 PreservedAnalyses run(Function &F, FunctionAnalysisManager &) { 390 return PreservedAnalyses::all(); 391 } 392 static StringRef name() { return "NoOpFunctionPass"; } 393 }; 394 395 /// No-op function analysis. 396 class NoOpFunctionAnalysis : public AnalysisInfoMixin<NoOpFunctionAnalysis> { 397 friend AnalysisInfoMixin<NoOpFunctionAnalysis>; 398 static AnalysisKey Key; 399 400 public: 401 struct Result {}; 402 Result run(Function &, FunctionAnalysisManager &) { return Result(); } 403 static StringRef name() { return "NoOpFunctionAnalysis"; } 404 }; 405 406 /// No-op loop pass which does nothing. 407 struct NoOpLoopPass : PassInfoMixin<NoOpLoopPass> { 408 PreservedAnalyses run(Loop &L, LoopAnalysisManager &, 409 LoopStandardAnalysisResults &, LPMUpdater &) { 410 return PreservedAnalyses::all(); 411 } 412 static StringRef name() { return "NoOpLoopPass"; } 413 }; 414 415 /// No-op loop analysis. 416 class NoOpLoopAnalysis : public AnalysisInfoMixin<NoOpLoopAnalysis> { 417 friend AnalysisInfoMixin<NoOpLoopAnalysis>; 418 static AnalysisKey Key; 419 420 public: 421 struct Result {}; 422 Result run(Loop &, LoopAnalysisManager &, LoopStandardAnalysisResults &) { 423 return Result(); 424 } 425 static StringRef name() { return "NoOpLoopAnalysis"; } 426 }; 427 428 AnalysisKey NoOpModuleAnalysis::Key; 429 AnalysisKey NoOpCGSCCAnalysis::Key; 430 AnalysisKey NoOpFunctionAnalysis::Key; 431 AnalysisKey NoOpLoopAnalysis::Key; 432 433 /// Whether or not we should populate a PassInstrumentationCallbacks's class to 434 /// pass name map. 435 /// 436 /// This is for optimization purposes so we don't populate it if we never use 437 /// it. This should be updated if new pass instrumentation wants to use the map. 438 /// We currently only use this for --print-before/after. 439 bool shouldPopulateClassToPassNames() { 440 return !printBeforePasses().empty() || !printAfterPasses().empty(); 441 } 442 443 } // namespace 444 445 PassBuilder::PassBuilder(TargetMachine *TM, PipelineTuningOptions PTO, 446 Optional<PGOOptions> PGOOpt, 447 PassInstrumentationCallbacks *PIC) 448 : TM(TM), PTO(PTO), PGOOpt(PGOOpt), PIC(PIC) { 449 if (TM) 450 TM->registerPassBuilderCallbacks(*this); 451 if (PIC && shouldPopulateClassToPassNames()) { 452 #define MODULE_PASS(NAME, CREATE_PASS) \ 453 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 454 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 455 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 456 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 457 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 458 #define FUNCTION_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 459 PIC->addClassToPassName(CLASS, NAME); 460 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 461 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 462 #define LOOP_PASS(NAME, CREATE_PASS) \ 463 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 464 #define LOOP_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 465 PIC->addClassToPassName(CLASS, NAME); 466 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 467 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 468 #define CGSCC_PASS(NAME, CREATE_PASS) \ 469 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 470 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 471 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 472 #include "PassRegistry.def" 473 } 474 } 475 476 void PassBuilder::invokePeepholeEPCallbacks( 477 FunctionPassManager &FPM, PassBuilder::OptimizationLevel Level) { 478 for (auto &C : PeepholeEPCallbacks) 479 C(FPM, Level); 480 } 481 482 void PassBuilder::registerModuleAnalyses(ModuleAnalysisManager &MAM) { 483 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 484 MAM.registerPass([&] { return CREATE_PASS; }); 485 #include "PassRegistry.def" 486 487 for (auto &C : ModuleAnalysisRegistrationCallbacks) 488 C(MAM); 489 } 490 491 void PassBuilder::registerCGSCCAnalyses(CGSCCAnalysisManager &CGAM) { 492 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 493 CGAM.registerPass([&] { return CREATE_PASS; }); 494 #include "PassRegistry.def" 495 496 for (auto &C : CGSCCAnalysisRegistrationCallbacks) 497 C(CGAM); 498 } 499 500 void PassBuilder::registerFunctionAnalyses(FunctionAnalysisManager &FAM) { 501 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 502 FAM.registerPass([&] { return CREATE_PASS; }); 503 #include "PassRegistry.def" 504 505 for (auto &C : FunctionAnalysisRegistrationCallbacks) 506 C(FAM); 507 } 508 509 void PassBuilder::registerLoopAnalyses(LoopAnalysisManager &LAM) { 510 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 511 LAM.registerPass([&] { return CREATE_PASS; }); 512 #include "PassRegistry.def" 513 514 for (auto &C : LoopAnalysisRegistrationCallbacks) 515 C(LAM); 516 } 517 518 // Helper to add AnnotationRemarksPass. 519 static void addAnnotationRemarksPass(ModulePassManager &MPM) { 520 FunctionPassManager FPM; 521 FPM.addPass(AnnotationRemarksPass()); 522 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 523 } 524 525 // Helper to check if the current compilation phase is preparing for LTO 526 static bool isLTOPreLink(ThinOrFullLTOPhase Phase) { 527 return Phase == ThinOrFullLTOPhase::ThinLTOPreLink || 528 Phase == ThinOrFullLTOPhase::FullLTOPreLink; 529 } 530 531 // TODO: Investigate the cost/benefit of tail call elimination on debugging. 532 FunctionPassManager 533 PassBuilder::buildO1FunctionSimplificationPipeline(OptimizationLevel Level, 534 ThinOrFullLTOPhase Phase) { 535 536 FunctionPassManager FPM; 537 538 // Form SSA out of local memory accesses after breaking apart aggregates into 539 // scalars. 540 FPM.addPass(SROA()); 541 542 // Catch trivial redundancies 543 FPM.addPass(EarlyCSEPass(true /* Enable mem-ssa. */)); 544 545 // Hoisting of scalars and load expressions. 546 FPM.addPass(SimplifyCFGPass()); 547 FPM.addPass(InstCombinePass()); 548 549 FPM.addPass(LibCallsShrinkWrapPass()); 550 551 invokePeepholeEPCallbacks(FPM, Level); 552 553 FPM.addPass(SimplifyCFGPass()); 554 555 // Form canonically associated expression trees, and simplify the trees using 556 // basic mathematical properties. For example, this will form (nearly) 557 // minimal multiplication trees. 558 FPM.addPass(ReassociatePass()); 559 560 // Add the primary loop simplification pipeline. 561 // FIXME: Currently this is split into two loop pass pipelines because we run 562 // some function passes in between them. These can and should be removed 563 // and/or replaced by scheduling the loop pass equivalents in the correct 564 // positions. But those equivalent passes aren't powerful enough yet. 565 // Specifically, `SimplifyCFGPass` and `InstCombinePass` are currently still 566 // used. We have `LoopSimplifyCFGPass` which isn't yet powerful enough yet to 567 // fully replace `SimplifyCFGPass`, and the closest to the other we have is 568 // `LoopInstSimplify`. 569 LoopPassManager LPM1, LPM2; 570 571 // Simplify the loop body. We do this initially to clean up after other loop 572 // passes run, either when iterating on a loop or on inner loops with 573 // implications on the outer loop. 574 LPM1.addPass(LoopInstSimplifyPass()); 575 LPM1.addPass(LoopSimplifyCFGPass()); 576 577 // Try to remove as much code from the loop header as possible, 578 // to reduce amount of IR that will have to be duplicated. 579 // TODO: Investigate promotion cap for O1. 580 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 581 582 LPM1.addPass(LoopRotatePass(/* Disable header duplication */ true, 583 isLTOPreLink(Phase))); 584 // TODO: Investigate promotion cap for O1. 585 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 586 LPM1.addPass(SimpleLoopUnswitchPass()); 587 588 LPM2.addPass(LoopIdiomRecognizePass()); 589 LPM2.addPass(IndVarSimplifyPass()); 590 591 for (auto &C : LateLoopOptimizationsEPCallbacks) 592 C(LPM2, Level); 593 594 LPM2.addPass(LoopDeletionPass()); 595 596 if (EnableLoopInterchange) 597 LPM2.addPass(LoopInterchangePass()); 598 599 // Do not enable unrolling in PreLinkThinLTO phase during sample PGO 600 // because it changes IR to makes profile annotation in back compile 601 // inaccurate. The normal unroller doesn't pay attention to forced full unroll 602 // attributes so we need to make sure and allow the full unroll pass to pay 603 // attention to it. 604 if (Phase != ThinOrFullLTOPhase::ThinLTOPreLink || !PGOOpt || 605 PGOOpt->Action != PGOOptions::SampleUse) 606 LPM2.addPass(LoopFullUnrollPass(Level.getSpeedupLevel(), 607 /* OnlyWhenForced= */ !PTO.LoopUnrolling, 608 PTO.ForgetAllSCEVInLoopUnroll)); 609 610 for (auto &C : LoopOptimizerEndEPCallbacks) 611 C(LPM2, Level); 612 613 // We provide the opt remark emitter pass for LICM to use. We only need to do 614 // this once as it is immutable. 615 FPM.addPass( 616 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 617 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM1), 618 EnableMSSALoopDependency, 619 /*UseBlockFrequencyInfo=*/true)); 620 FPM.addPass(SimplifyCFGPass()); 621 FPM.addPass(InstCombinePass()); 622 if (EnableLoopFlatten) 623 FPM.addPass(createFunctionToLoopPassAdaptor(LoopFlattenPass())); 624 // The loop passes in LPM2 (LoopFullUnrollPass) do not preserve MemorySSA. 625 // *All* loop passes must preserve it, in order to be able to use it. 626 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM2), 627 /*UseMemorySSA=*/false, 628 /*UseBlockFrequencyInfo=*/false)); 629 630 // Delete small array after loop unroll. 631 FPM.addPass(SROA()); 632 633 // Specially optimize memory movement as it doesn't look like dataflow in SSA. 634 FPM.addPass(MemCpyOptPass()); 635 636 // Sparse conditional constant propagation. 637 // FIXME: It isn't clear why we do this *after* loop passes rather than 638 // before... 639 FPM.addPass(SCCPPass()); 640 641 // Delete dead bit computations (instcombine runs after to fold away the dead 642 // computations, and then ADCE will run later to exploit any new DCE 643 // opportunities that creates). 644 FPM.addPass(BDCEPass()); 645 646 // Run instcombine after redundancy and dead bit elimination to exploit 647 // opportunities opened up by them. 648 FPM.addPass(InstCombinePass()); 649 invokePeepholeEPCallbacks(FPM, Level); 650 651 FPM.addPass(CoroElidePass()); 652 653 for (auto &C : ScalarOptimizerLateEPCallbacks) 654 C(FPM, Level); 655 656 // Finally, do an expensive DCE pass to catch all the dead code exposed by 657 // the simplifications and basic cleanup after all the simplifications. 658 // TODO: Investigate if this is too expensive. 659 FPM.addPass(ADCEPass()); 660 FPM.addPass(SimplifyCFGPass()); 661 FPM.addPass(InstCombinePass()); 662 invokePeepholeEPCallbacks(FPM, Level); 663 664 return FPM; 665 } 666 667 FunctionPassManager 668 PassBuilder::buildFunctionSimplificationPipeline(OptimizationLevel Level, 669 ThinOrFullLTOPhase Phase) { 670 assert(Level != OptimizationLevel::O0 && "Must request optimizations!"); 671 672 // The O1 pipeline has a separate pipeline creation function to simplify 673 // construction readability. 674 if (Level.getSpeedupLevel() == 1) 675 return buildO1FunctionSimplificationPipeline(Level, Phase); 676 677 FunctionPassManager FPM; 678 679 // Form SSA out of local memory accesses after breaking apart aggregates into 680 // scalars. 681 FPM.addPass(SROA()); 682 683 // Catch trivial redundancies 684 FPM.addPass(EarlyCSEPass(true /* Enable mem-ssa. */)); 685 if (EnableKnowledgeRetention) 686 FPM.addPass(AssumeSimplifyPass()); 687 688 // Hoisting of scalars and load expressions. 689 if (EnableGVNHoist) 690 FPM.addPass(GVNHoistPass()); 691 692 // Global value numbering based sinking. 693 if (EnableGVNSink) { 694 FPM.addPass(GVNSinkPass()); 695 FPM.addPass(SimplifyCFGPass()); 696 } 697 698 if (EnableConstraintElimination) 699 FPM.addPass(ConstraintEliminationPass()); 700 701 // Speculative execution if the target has divergent branches; otherwise nop. 702 FPM.addPass(SpeculativeExecutionPass(/* OnlyIfDivergentTarget =*/true)); 703 704 // Optimize based on known information about branches, and cleanup afterward. 705 FPM.addPass(JumpThreadingPass()); 706 FPM.addPass(CorrelatedValuePropagationPass()); 707 708 FPM.addPass(SimplifyCFGPass()); 709 if (Level == OptimizationLevel::O3) 710 FPM.addPass(AggressiveInstCombinePass()); 711 FPM.addPass(InstCombinePass()); 712 713 if (!Level.isOptimizingForSize()) 714 FPM.addPass(LibCallsShrinkWrapPass()); 715 716 invokePeepholeEPCallbacks(FPM, Level); 717 718 // For PGO use pipeline, try to optimize memory intrinsics such as memcpy 719 // using the size value profile. Don't perform this when optimizing for size. 720 if (PGOOpt && PGOOpt->Action == PGOOptions::IRUse && 721 !Level.isOptimizingForSize()) 722 FPM.addPass(PGOMemOPSizeOpt()); 723 724 FPM.addPass(TailCallElimPass()); 725 FPM.addPass(SimplifyCFGPass()); 726 727 // Form canonically associated expression trees, and simplify the trees using 728 // basic mathematical properties. For example, this will form (nearly) 729 // minimal multiplication trees. 730 FPM.addPass(ReassociatePass()); 731 732 // Add the primary loop simplification pipeline. 733 // FIXME: Currently this is split into two loop pass pipelines because we run 734 // some function passes in between them. These can and should be removed 735 // and/or replaced by scheduling the loop pass equivalents in the correct 736 // positions. But those equivalent passes aren't powerful enough yet. 737 // Specifically, `SimplifyCFGPass` and `InstCombinePass` are currently still 738 // used. We have `LoopSimplifyCFGPass` which isn't yet powerful enough yet to 739 // fully replace `SimplifyCFGPass`, and the closest to the other we have is 740 // `LoopInstSimplify`. 741 LoopPassManager LPM1, LPM2; 742 743 // Simplify the loop body. We do this initially to clean up after other loop 744 // passes run, either when iterating on a loop or on inner loops with 745 // implications on the outer loop. 746 LPM1.addPass(LoopInstSimplifyPass()); 747 LPM1.addPass(LoopSimplifyCFGPass()); 748 749 // Try to remove as much code from the loop header as possible, 750 // to reduce amount of IR that will have to be duplicated. 751 // TODO: Investigate promotion cap for O1. 752 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 753 754 // Disable header duplication in loop rotation at -Oz. 755 LPM1.addPass( 756 LoopRotatePass(Level != OptimizationLevel::Oz, isLTOPreLink(Phase))); 757 // TODO: Investigate promotion cap for O1. 758 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 759 LPM1.addPass( 760 SimpleLoopUnswitchPass(/* NonTrivial */ Level == OptimizationLevel::O3 && 761 EnableO3NonTrivialUnswitching)); 762 LPM2.addPass(LoopIdiomRecognizePass()); 763 LPM2.addPass(IndVarSimplifyPass()); 764 765 for (auto &C : LateLoopOptimizationsEPCallbacks) 766 C(LPM2, Level); 767 768 LPM2.addPass(LoopDeletionPass()); 769 770 if (EnableLoopInterchange) 771 LPM2.addPass(LoopInterchangePass()); 772 773 // Do not enable unrolling in PreLinkThinLTO phase during sample PGO 774 // because it changes IR to makes profile annotation in back compile 775 // inaccurate. The normal unroller doesn't pay attention to forced full unroll 776 // attributes so we need to make sure and allow the full unroll pass to pay 777 // attention to it. 778 if (Phase != ThinOrFullLTOPhase::ThinLTOPreLink || !PGOOpt || 779 PGOOpt->Action != PGOOptions::SampleUse) 780 LPM2.addPass(LoopFullUnrollPass(Level.getSpeedupLevel(), 781 /* OnlyWhenForced= */ !PTO.LoopUnrolling, 782 PTO.ForgetAllSCEVInLoopUnroll)); 783 784 for (auto &C : LoopOptimizerEndEPCallbacks) 785 C(LPM2, Level); 786 787 // We provide the opt remark emitter pass for LICM to use. We only need to do 788 // this once as it is immutable. 789 FPM.addPass( 790 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 791 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM1), 792 EnableMSSALoopDependency, 793 /*UseBlockFrequencyInfo=*/true)); 794 FPM.addPass(SimplifyCFGPass()); 795 FPM.addPass(InstCombinePass()); 796 if (EnableLoopFlatten) 797 FPM.addPass(createFunctionToLoopPassAdaptor(LoopFlattenPass())); 798 // The loop passes in LPM2 (LoopIdiomRecognizePass, IndVarSimplifyPass, 799 // LoopDeletionPass and LoopFullUnrollPass) do not preserve MemorySSA. 800 // *All* loop passes must preserve it, in order to be able to use it. 801 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM2), 802 /*UseMemorySSA=*/false, 803 /*UseBlockFrequencyInfo=*/false)); 804 805 // Delete small array after loop unroll. 806 FPM.addPass(SROA()); 807 808 // Eliminate redundancies. 809 FPM.addPass(MergedLoadStoreMotionPass()); 810 if (RunNewGVN) 811 FPM.addPass(NewGVNPass()); 812 else 813 FPM.addPass(GVN()); 814 815 // Sparse conditional constant propagation. 816 // FIXME: It isn't clear why we do this *after* loop passes rather than 817 // before... 818 FPM.addPass(SCCPPass()); 819 820 // Delete dead bit computations (instcombine runs after to fold away the dead 821 // computations, and then ADCE will run later to exploit any new DCE 822 // opportunities that creates). 823 FPM.addPass(BDCEPass()); 824 825 // Run instcombine after redundancy and dead bit elimination to exploit 826 // opportunities opened up by them. 827 FPM.addPass(InstCombinePass()); 828 invokePeepholeEPCallbacks(FPM, Level); 829 830 // Re-consider control flow based optimizations after redundancy elimination, 831 // redo DCE, etc. 832 FPM.addPass(JumpThreadingPass()); 833 FPM.addPass(CorrelatedValuePropagationPass()); 834 835 // Finally, do an expensive DCE pass to catch all the dead code exposed by 836 // the simplifications and basic cleanup after all the simplifications. 837 // TODO: Investigate if this is too expensive. 838 FPM.addPass(ADCEPass()); 839 840 // Specially optimize memory movement as it doesn't look like dataflow in SSA. 841 FPM.addPass(MemCpyOptPass()); 842 843 FPM.addPass(DSEPass()); 844 FPM.addPass(createFunctionToLoopPassAdaptor( 845 LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap), 846 EnableMSSALoopDependency, /*UseBlockFrequencyInfo=*/true)); 847 848 FPM.addPass(CoroElidePass()); 849 850 for (auto &C : ScalarOptimizerLateEPCallbacks) 851 C(FPM, Level); 852 853 FPM.addPass(SimplifyCFGPass( 854 SimplifyCFGOptions().hoistCommonInsts(true).sinkCommonInsts(true))); 855 FPM.addPass(InstCombinePass()); 856 invokePeepholeEPCallbacks(FPM, Level); 857 858 if (EnableCHR && Level == OptimizationLevel::O3 && PGOOpt && 859 (PGOOpt->Action == PGOOptions::IRUse || 860 PGOOpt->Action == PGOOptions::SampleUse)) 861 FPM.addPass(ControlHeightReductionPass()); 862 863 return FPM; 864 } 865 866 void PassBuilder::addRequiredLTOPreLinkPasses(ModulePassManager &MPM) { 867 MPM.addPass(CanonicalizeAliasesPass()); 868 MPM.addPass(NameAnonGlobalPass()); 869 } 870 871 void PassBuilder::addPGOInstrPasses(ModulePassManager &MPM, 872 PassBuilder::OptimizationLevel Level, 873 bool RunProfileGen, bool IsCS, 874 std::string ProfileFile, 875 std::string ProfileRemappingFile) { 876 assert(Level != OptimizationLevel::O0 && "Not expecting O0 here!"); 877 if (!IsCS && !DisablePreInliner) { 878 InlineParams IP; 879 880 IP.DefaultThreshold = PreInlineThreshold; 881 882 // FIXME: The hint threshold has the same value used by the regular inliner 883 // when not optimzing for size. This should probably be lowered after 884 // performance testing. 885 // FIXME: this comment is cargo culted from the old pass manager, revisit). 886 IP.HintThreshold = Level.isOptimizingForSize() ? PreInlineThreshold : 325; 887 ModuleInlinerWrapperPass MIWP(IP); 888 CGSCCPassManager &CGPipeline = MIWP.getPM(); 889 890 FunctionPassManager FPM; 891 FPM.addPass(SROA()); 892 FPM.addPass(EarlyCSEPass()); // Catch trivial redundancies. 893 FPM.addPass(SimplifyCFGPass()); // Merge & remove basic blocks. 894 FPM.addPass(InstCombinePass()); // Combine silly sequences. 895 invokePeepholeEPCallbacks(FPM, Level); 896 897 CGPipeline.addPass(createCGSCCToFunctionPassAdaptor(std::move(FPM))); 898 899 MPM.addPass(std::move(MIWP)); 900 901 // Delete anything that is now dead to make sure that we don't instrument 902 // dead code. Instrumentation can end up keeping dead code around and 903 // dramatically increase code size. 904 MPM.addPass(GlobalDCEPass()); 905 } 906 907 if (!RunProfileGen) { 908 assert(!ProfileFile.empty() && "Profile use expecting a profile file!"); 909 MPM.addPass(PGOInstrumentationUse(ProfileFile, ProfileRemappingFile, IsCS)); 910 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 911 // RequireAnalysisPass for PSI before subsequent non-module passes. 912 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 913 return; 914 } 915 916 // Perform PGO instrumentation. 917 MPM.addPass(PGOInstrumentationGen(IsCS)); 918 919 FunctionPassManager FPM; 920 // Disable header duplication in loop rotation at -Oz. 921 FPM.addPass(createFunctionToLoopPassAdaptor( 922 LoopRotatePass(Level != OptimizationLevel::Oz), EnableMSSALoopDependency, 923 /*UseBlockFrequencyInfo=*/false)); 924 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 925 926 // Add the profile lowering pass. 927 InstrProfOptions Options; 928 if (!ProfileFile.empty()) 929 Options.InstrProfileOutput = ProfileFile; 930 // Do counter promotion at Level greater than O0. 931 Options.DoCounterPromotion = true; 932 Options.UseBFIInPromotion = IsCS; 933 MPM.addPass(InstrProfiling(Options, IsCS)); 934 } 935 936 void PassBuilder::addPGOInstrPassesForO0(ModulePassManager &MPM, 937 bool RunProfileGen, bool IsCS, 938 std::string ProfileFile, 939 std::string ProfileRemappingFile) { 940 if (!RunProfileGen) { 941 assert(!ProfileFile.empty() && "Profile use expecting a profile file!"); 942 MPM.addPass(PGOInstrumentationUse(ProfileFile, ProfileRemappingFile, IsCS)); 943 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 944 // RequireAnalysisPass for PSI before subsequent non-module passes. 945 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 946 return; 947 } 948 949 // Perform PGO instrumentation. 950 MPM.addPass(PGOInstrumentationGen(IsCS)); 951 // Add the profile lowering pass. 952 InstrProfOptions Options; 953 if (!ProfileFile.empty()) 954 Options.InstrProfileOutput = ProfileFile; 955 // Do not do counter promotion at O0. 956 Options.DoCounterPromotion = false; 957 Options.UseBFIInPromotion = IsCS; 958 MPM.addPass(InstrProfiling(Options, IsCS)); 959 } 960 961 static InlineParams 962 getInlineParamsFromOptLevel(PassBuilder::OptimizationLevel Level) { 963 return getInlineParams(Level.getSpeedupLevel(), Level.getSizeLevel()); 964 } 965 966 ModuleInlinerWrapperPass 967 PassBuilder::buildInlinerPipeline(OptimizationLevel Level, 968 ThinOrFullLTOPhase Phase) { 969 InlineParams IP = getInlineParamsFromOptLevel(Level); 970 if (Phase == ThinOrFullLTOPhase::ThinLTOPreLink && PGOOpt && 971 PGOOpt->Action == PGOOptions::SampleUse) 972 IP.HotCallSiteThreshold = 0; 973 974 if (PGOOpt) 975 IP.EnableDeferral = EnablePGOInlineDeferral; 976 977 ModuleInlinerWrapperPass MIWP(IP, PerformMandatoryInliningsFirst, 978 UseInlineAdvisor, MaxDevirtIterations); 979 980 // Require the GlobalsAA analysis for the module so we can query it within 981 // the CGSCC pipeline. 982 MIWP.addModulePass(RequireAnalysisPass<GlobalsAA, Module>()); 983 // Invalidate AAManager so it can be recreated and pick up the newly available 984 // GlobalsAA. 985 MIWP.addModulePass( 986 createModuleToFunctionPassAdaptor(InvalidateAnalysisPass<AAManager>())); 987 988 // Require the ProfileSummaryAnalysis for the module so we can query it within 989 // the inliner pass. 990 MIWP.addModulePass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 991 992 // Now begin the main postorder CGSCC pipeline. 993 // FIXME: The current CGSCC pipeline has its origins in the legacy pass 994 // manager and trying to emulate its precise behavior. Much of this doesn't 995 // make a lot of sense and we should revisit the core CGSCC structure. 996 CGSCCPassManager &MainCGPipeline = MIWP.getPM(); 997 998 // Note: historically, the PruneEH pass was run first to deduce nounwind and 999 // generally clean up exception handling overhead. It isn't clear this is 1000 // valuable as the inliner doesn't currently care whether it is inlining an 1001 // invoke or a call. 1002 1003 if (AttributorRun & AttributorRunOption::CGSCC) 1004 MainCGPipeline.addPass(AttributorCGSCCPass()); 1005 1006 // Now deduce any function attributes based in the current code. 1007 MainCGPipeline.addPass(PostOrderFunctionAttrsPass()); 1008 1009 // When at O3 add argument promotion to the pass pipeline. 1010 // FIXME: It isn't at all clear why this should be limited to O3. 1011 if (Level == OptimizationLevel::O3) 1012 MainCGPipeline.addPass(ArgumentPromotionPass()); 1013 1014 // Try to perform OpenMP specific optimizations. This is a (quick!) no-op if 1015 // there are no OpenMP runtime calls present in the module. 1016 if (Level == OptimizationLevel::O2 || Level == OptimizationLevel::O3) 1017 MainCGPipeline.addPass(OpenMPOptCGSCCPass()); 1018 1019 for (auto &C : CGSCCOptimizerLateEPCallbacks) 1020 C(MainCGPipeline, Level); 1021 1022 // Lastly, add the core function simplification pipeline nested inside the 1023 // CGSCC walk. 1024 MainCGPipeline.addPass(createCGSCCToFunctionPassAdaptor( 1025 buildFunctionSimplificationPipeline(Level, Phase))); 1026 1027 MainCGPipeline.addPass(CoroSplitPass(Level != OptimizationLevel::O0)); 1028 1029 return MIWP; 1030 } 1031 1032 ModulePassManager 1033 PassBuilder::buildModuleSimplificationPipeline(OptimizationLevel Level, 1034 ThinOrFullLTOPhase Phase) { 1035 ModulePassManager MPM; 1036 1037 // Place pseudo probe instrumentation as the first pass of the pipeline to 1038 // minimize the impact of optimization changes. 1039 if (PGOOpt && PGOOpt->PseudoProbeForProfiling && 1040 Phase != ThinOrFullLTOPhase::ThinLTOPostLink) 1041 MPM.addPass(SampleProfileProbePass(TM)); 1042 1043 bool HasSampleProfile = PGOOpt && (PGOOpt->Action == PGOOptions::SampleUse); 1044 1045 // In ThinLTO mode, when flattened profile is used, all the available 1046 // profile information will be annotated in PreLink phase so there is 1047 // no need to load the profile again in PostLink. 1048 bool LoadSampleProfile = 1049 HasSampleProfile && 1050 !(FlattenedProfileUsed && Phase == ThinOrFullLTOPhase::ThinLTOPostLink); 1051 1052 // During the ThinLTO backend phase we perform early indirect call promotion 1053 // here, before globalopt. Otherwise imported available_externally functions 1054 // look unreferenced and are removed. If we are going to load the sample 1055 // profile then defer until later. 1056 // TODO: See if we can move later and consolidate with the location where 1057 // we perform ICP when we are loading a sample profile. 1058 // TODO: We pass HasSampleProfile (whether there was a sample profile file 1059 // passed to the compile) to the SamplePGO flag of ICP. This is used to 1060 // determine whether the new direct calls are annotated with prof metadata. 1061 // Ideally this should be determined from whether the IR is annotated with 1062 // sample profile, and not whether the a sample profile was provided on the 1063 // command line. E.g. for flattened profiles where we will not be reloading 1064 // the sample profile in the ThinLTO backend, we ideally shouldn't have to 1065 // provide the sample profile file. 1066 if (Phase == ThinOrFullLTOPhase::ThinLTOPostLink && !LoadSampleProfile) 1067 MPM.addPass(PGOIndirectCallPromotion(true /* InLTO */, HasSampleProfile)); 1068 1069 // Do basic inference of function attributes from known properties of system 1070 // libraries and other oracles. 1071 MPM.addPass(InferFunctionAttrsPass()); 1072 1073 // Create an early function pass manager to cleanup the output of the 1074 // frontend. 1075 FunctionPassManager EarlyFPM; 1076 // Lower llvm.expect to metadata before attempting transforms. 1077 // Compare/branch metadata may alter the behavior of passes like SimplifyCFG. 1078 EarlyFPM.addPass(LowerExpectIntrinsicPass()); 1079 EarlyFPM.addPass(SimplifyCFGPass()); 1080 EarlyFPM.addPass(SROA()); 1081 EarlyFPM.addPass(EarlyCSEPass()); 1082 EarlyFPM.addPass(CoroEarlyPass()); 1083 if (Level == OptimizationLevel::O3) 1084 EarlyFPM.addPass(CallSiteSplittingPass()); 1085 1086 // In SamplePGO ThinLTO backend, we need instcombine before profile annotation 1087 // to convert bitcast to direct calls so that they can be inlined during the 1088 // profile annotation prepration step. 1089 // More details about SamplePGO design can be found in: 1090 // https://research.google.com/pubs/pub45290.html 1091 // FIXME: revisit how SampleProfileLoad/Inliner/ICP is structured. 1092 if (LoadSampleProfile) 1093 EarlyFPM.addPass(InstCombinePass()); 1094 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(EarlyFPM))); 1095 1096 if (LoadSampleProfile) { 1097 // Annotate sample profile right after early FPM to ensure freshness of 1098 // the debug info. 1099 MPM.addPass(SampleProfileLoaderPass(PGOOpt->ProfileFile, 1100 PGOOpt->ProfileRemappingFile, Phase)); 1101 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 1102 // RequireAnalysisPass for PSI before subsequent non-module passes. 1103 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 1104 // Do not invoke ICP in the LTOPrelink phase as it makes it hard 1105 // for the profile annotation to be accurate in the LTO backend. 1106 if (Phase != ThinOrFullLTOPhase::ThinLTOPreLink && 1107 Phase != ThinOrFullLTOPhase::FullLTOPreLink) 1108 // We perform early indirect call promotion here, before globalopt. 1109 // This is important for the ThinLTO backend phase because otherwise 1110 // imported available_externally functions look unreferenced and are 1111 // removed. 1112 MPM.addPass( 1113 PGOIndirectCallPromotion(true /* IsInLTO */, true /* SamplePGO */)); 1114 } 1115 1116 // Try to perform OpenMP specific optimizations on the module. This is a 1117 // (quick!) no-op if there are no OpenMP runtime calls present in the module. 1118 if (Level != OptimizationLevel::O0) 1119 MPM.addPass(OpenMPOptPass()); 1120 1121 if (AttributorRun & AttributorRunOption::MODULE) 1122 MPM.addPass(AttributorPass()); 1123 1124 // Lower type metadata and the type.test intrinsic in the ThinLTO 1125 // post link pipeline after ICP. This is to enable usage of the type 1126 // tests in ICP sequences. 1127 if (Phase == ThinOrFullLTOPhase::ThinLTOPostLink) 1128 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1129 1130 for (auto &C : PipelineEarlySimplificationEPCallbacks) 1131 C(MPM, Level); 1132 1133 // Specialize functions with IPSCCP. 1134 if (EnableFunctionSpecialization) 1135 MPM.addPass(FunctionSpecializationPass()); 1136 1137 // Interprocedural constant propagation now that basic cleanup has occurred 1138 // and prior to optimizing globals. 1139 // FIXME: This position in the pipeline hasn't been carefully considered in 1140 // years, it should be re-analyzed. 1141 MPM.addPass(IPSCCPPass()); 1142 1143 // Attach metadata to indirect call sites indicating the set of functions 1144 // they may target at run-time. This should follow IPSCCP. 1145 MPM.addPass(CalledValuePropagationPass()); 1146 1147 // Optimize globals to try and fold them into constants. 1148 MPM.addPass(GlobalOptPass()); 1149 1150 // Promote any localized globals to SSA registers. 1151 // FIXME: Should this instead by a run of SROA? 1152 // FIXME: We should probably run instcombine and simplifycfg afterward to 1153 // delete control flows that are dead once globals have been folded to 1154 // constants. 1155 MPM.addPass(createModuleToFunctionPassAdaptor(PromotePass())); 1156 1157 // Remove any dead arguments exposed by cleanups and constant folding 1158 // globals. 1159 MPM.addPass(DeadArgumentEliminationPass()); 1160 1161 // Create a small function pass pipeline to cleanup after all the global 1162 // optimizations. 1163 FunctionPassManager GlobalCleanupPM; 1164 GlobalCleanupPM.addPass(InstCombinePass()); 1165 invokePeepholeEPCallbacks(GlobalCleanupPM, Level); 1166 1167 GlobalCleanupPM.addPass(SimplifyCFGPass()); 1168 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(GlobalCleanupPM))); 1169 1170 // Add all the requested passes for instrumentation PGO, if requested. 1171 if (PGOOpt && Phase != ThinOrFullLTOPhase::ThinLTOPostLink && 1172 (PGOOpt->Action == PGOOptions::IRInstr || 1173 PGOOpt->Action == PGOOptions::IRUse)) { 1174 addPGOInstrPasses(MPM, Level, 1175 /* RunProfileGen */ PGOOpt->Action == PGOOptions::IRInstr, 1176 /* IsCS */ false, PGOOpt->ProfileFile, 1177 PGOOpt->ProfileRemappingFile); 1178 MPM.addPass(PGOIndirectCallPromotion(false, false)); 1179 } 1180 if (PGOOpt && Phase != ThinOrFullLTOPhase::ThinLTOPostLink && 1181 PGOOpt->CSAction == PGOOptions::CSIRInstr) 1182 MPM.addPass(PGOInstrumentationGenCreateVar(PGOOpt->CSProfileGenFile)); 1183 1184 // Synthesize function entry counts for non-PGO compilation. 1185 if (EnableSyntheticCounts && !PGOOpt) 1186 MPM.addPass(SyntheticCountsPropagation()); 1187 1188 MPM.addPass(buildInlinerPipeline(Level, Phase)); 1189 1190 if (EnableMemProfiler && Phase != ThinOrFullLTOPhase::ThinLTOPreLink) { 1191 MPM.addPass(createModuleToFunctionPassAdaptor(MemProfilerPass())); 1192 MPM.addPass(ModuleMemProfilerPass()); 1193 } 1194 1195 return MPM; 1196 } 1197 1198 /// TODO: Should LTO cause any differences to this set of passes? 1199 void PassBuilder::addVectorPasses(OptimizationLevel Level, 1200 FunctionPassManager &FPM, bool IsFullLTO) { 1201 FPM.addPass(LoopVectorizePass( 1202 LoopVectorizeOptions(!PTO.LoopInterleaving, !PTO.LoopVectorization))); 1203 1204 if (IsFullLTO) { 1205 // The vectorizer may have significantly shortened a loop body; unroll 1206 // again. Unroll small loops to hide loop backedge latency and saturate any 1207 // parallel execution resources of an out-of-order processor. We also then 1208 // need to clean up redundancies and loop invariant code. 1209 // FIXME: It would be really good to use a loop-integrated instruction 1210 // combiner for cleanup here so that the unrolling and LICM can be pipelined 1211 // across the loop nests. 1212 // We do UnrollAndJam in a separate LPM to ensure it happens before unroll 1213 if (EnableUnrollAndJam && PTO.LoopUnrolling) 1214 FPM.addPass(createFunctionToLoopPassAdaptor( 1215 LoopUnrollAndJamPass(Level.getSpeedupLevel()))); 1216 FPM.addPass(LoopUnrollPass(LoopUnrollOptions( 1217 Level.getSpeedupLevel(), /*OnlyWhenForced=*/!PTO.LoopUnrolling, 1218 PTO.ForgetAllSCEVInLoopUnroll))); 1219 FPM.addPass(WarnMissedTransformationsPass()); 1220 } 1221 1222 if (!IsFullLTO) { 1223 // Eliminate loads by forwarding stores from the previous iteration to loads 1224 // of the current iteration. 1225 FPM.addPass(LoopLoadEliminationPass()); 1226 } 1227 // Cleanup after the loop optimization passes. 1228 FPM.addPass(InstCombinePass()); 1229 1230 if (Level.getSpeedupLevel() > 1 && ExtraVectorizerPasses) { 1231 // At higher optimization levels, try to clean up any runtime overlap and 1232 // alignment checks inserted by the vectorizer. We want to track correlated 1233 // runtime checks for two inner loops in the same outer loop, fold any 1234 // common computations, hoist loop-invariant aspects out of any outer loop, 1235 // and unswitch the runtime checks if possible. Once hoisted, we may have 1236 // dead (or speculatable) control flows or more combining opportunities. 1237 FPM.addPass(EarlyCSEPass()); 1238 FPM.addPass(CorrelatedValuePropagationPass()); 1239 FPM.addPass(InstCombinePass()); 1240 LoopPassManager LPM; 1241 LPM.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 1242 LPM.addPass(SimpleLoopUnswitchPass(/* NonTrivial */ Level == 1243 OptimizationLevel::O3)); 1244 FPM.addPass( 1245 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 1246 FPM.addPass(createFunctionToLoopPassAdaptor( 1247 std::move(LPM), EnableMSSALoopDependency, 1248 /*UseBlockFrequencyInfo=*/true)); 1249 FPM.addPass(SimplifyCFGPass()); 1250 FPM.addPass(InstCombinePass()); 1251 } 1252 1253 // Now that we've formed fast to execute loop structures, we do further 1254 // optimizations. These are run afterward as they might block doing complex 1255 // analyses and transforms such as what are needed for loop vectorization. 1256 1257 // Cleanup after loop vectorization, etc. Simplification passes like CVP and 1258 // GVN, loop transforms, and others have already run, so it's now better to 1259 // convert to more optimized IR using more aggressive simplify CFG options. 1260 // The extra sinking transform can create larger basic blocks, so do this 1261 // before SLP vectorization. 1262 FPM.addPass(SimplifyCFGPass(SimplifyCFGOptions() 1263 .forwardSwitchCondToPhi(true) 1264 .convertSwitchToLookupTable(true) 1265 .needCanonicalLoops(false) 1266 .hoistCommonInsts(true) 1267 .sinkCommonInsts(true))); 1268 1269 if (IsFullLTO) { 1270 FPM.addPass(SCCPPass()); 1271 FPM.addPass(InstCombinePass()); 1272 FPM.addPass(BDCEPass()); 1273 } 1274 1275 // Optimize parallel scalar instruction chains into SIMD instructions. 1276 if (PTO.SLPVectorization) { 1277 FPM.addPass(SLPVectorizerPass()); 1278 if (Level.getSpeedupLevel() > 1 && ExtraVectorizerPasses) { 1279 FPM.addPass(EarlyCSEPass()); 1280 } 1281 } 1282 // Enhance/cleanup vector code. 1283 FPM.addPass(VectorCombinePass()); 1284 1285 if (!IsFullLTO) { 1286 FPM.addPass(InstCombinePass()); 1287 // Unroll small loops to hide loop backedge latency and saturate any 1288 // parallel execution resources of an out-of-order processor. We also then 1289 // need to clean up redundancies and loop invariant code. 1290 // FIXME: It would be really good to use a loop-integrated instruction 1291 // combiner for cleanup here so that the unrolling and LICM can be pipelined 1292 // across the loop nests. 1293 // We do UnrollAndJam in a separate LPM to ensure it happens before unroll 1294 if (EnableUnrollAndJam && PTO.LoopUnrolling) { 1295 FPM.addPass(createFunctionToLoopPassAdaptor( 1296 LoopUnrollAndJamPass(Level.getSpeedupLevel()))); 1297 } 1298 FPM.addPass(LoopUnrollPass(LoopUnrollOptions( 1299 Level.getSpeedupLevel(), /*OnlyWhenForced=*/!PTO.LoopUnrolling, 1300 PTO.ForgetAllSCEVInLoopUnroll))); 1301 FPM.addPass(WarnMissedTransformationsPass()); 1302 FPM.addPass(InstCombinePass()); 1303 FPM.addPass( 1304 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 1305 FPM.addPass(createFunctionToLoopPassAdaptor( 1306 LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap), 1307 EnableMSSALoopDependency, /*UseBlockFrequencyInfo=*/true)); 1308 } 1309 1310 // Now that we've vectorized and unrolled loops, we may have more refined 1311 // alignment information, try to re-derive it here. 1312 FPM.addPass(AlignmentFromAssumptionsPass()); 1313 1314 if (IsFullLTO) 1315 FPM.addPass(InstCombinePass()); 1316 } 1317 1318 ModulePassManager 1319 PassBuilder::buildModuleOptimizationPipeline(OptimizationLevel Level, 1320 bool LTOPreLink) { 1321 ModulePassManager MPM; 1322 1323 // Optimize globals now that the module is fully simplified. 1324 MPM.addPass(GlobalOptPass()); 1325 MPM.addPass(GlobalDCEPass()); 1326 1327 // Run partial inlining pass to partially inline functions that have 1328 // large bodies. 1329 if (RunPartialInlining) 1330 MPM.addPass(PartialInlinerPass()); 1331 1332 // Remove avail extern fns and globals definitions since we aren't compiling 1333 // an object file for later LTO. For LTO we want to preserve these so they 1334 // are eligible for inlining at link-time. Note if they are unreferenced they 1335 // will be removed by GlobalDCE later, so this only impacts referenced 1336 // available externally globals. Eventually they will be suppressed during 1337 // codegen, but eliminating here enables more opportunity for GlobalDCE as it 1338 // may make globals referenced by available external functions dead and saves 1339 // running remaining passes on the eliminated functions. These should be 1340 // preserved during prelinking for link-time inlining decisions. 1341 if (!LTOPreLink) 1342 MPM.addPass(EliminateAvailableExternallyPass()); 1343 1344 if (EnableOrderFileInstrumentation) 1345 MPM.addPass(InstrOrderFilePass()); 1346 1347 // Do RPO function attribute inference across the module to forward-propagate 1348 // attributes where applicable. 1349 // FIXME: Is this really an optimization rather than a canonicalization? 1350 MPM.addPass(ReversePostOrderFunctionAttrsPass()); 1351 1352 // Do a post inline PGO instrumentation and use pass. This is a context 1353 // sensitive PGO pass. We don't want to do this in LTOPreLink phrase as 1354 // cross-module inline has not been done yet. The context sensitive 1355 // instrumentation is after all the inlines are done. 1356 if (!LTOPreLink && PGOOpt) { 1357 if (PGOOpt->CSAction == PGOOptions::CSIRInstr) 1358 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ true, 1359 /* IsCS */ true, PGOOpt->CSProfileGenFile, 1360 PGOOpt->ProfileRemappingFile); 1361 else if (PGOOpt->CSAction == PGOOptions::CSIRUse) 1362 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ false, 1363 /* IsCS */ true, PGOOpt->ProfileFile, 1364 PGOOpt->ProfileRemappingFile); 1365 } 1366 1367 // Re-require GloblasAA here prior to function passes. This is particularly 1368 // useful as the above will have inlined, DCE'ed, and function-attr 1369 // propagated everything. We should at this point have a reasonably minimal 1370 // and richly annotated call graph. By computing aliasing and mod/ref 1371 // information for all local globals here, the late loop passes and notably 1372 // the vectorizer will be able to use them to help recognize vectorizable 1373 // memory operations. 1374 MPM.addPass(RequireAnalysisPass<GlobalsAA, Module>()); 1375 1376 FunctionPassManager OptimizePM; 1377 OptimizePM.addPass(Float2IntPass()); 1378 OptimizePM.addPass(LowerConstantIntrinsicsPass()); 1379 1380 if (EnableMatrix) { 1381 OptimizePM.addPass(LowerMatrixIntrinsicsPass()); 1382 OptimizePM.addPass(EarlyCSEPass()); 1383 } 1384 1385 // FIXME: We need to run some loop optimizations to re-rotate loops after 1386 // simplifycfg and others undo their rotation. 1387 1388 // Optimize the loop execution. These passes operate on entire loop nests 1389 // rather than on each loop in an inside-out manner, and so they are actually 1390 // function passes. 1391 1392 for (auto &C : VectorizerStartEPCallbacks) 1393 C(OptimizePM, Level); 1394 1395 // First rotate loops that may have been un-rotated by prior passes. 1396 // Disable header duplication at -Oz. 1397 OptimizePM.addPass(createFunctionToLoopPassAdaptor( 1398 LoopRotatePass(Level != OptimizationLevel::Oz, LTOPreLink), 1399 EnableMSSALoopDependency, 1400 /*UseBlockFrequencyInfo=*/false)); 1401 1402 // Distribute loops to allow partial vectorization. I.e. isolate dependences 1403 // into separate loop that would otherwise inhibit vectorization. This is 1404 // currently only performed for loops marked with the metadata 1405 // llvm.loop.distribute=true or when -enable-loop-distribute is specified. 1406 OptimizePM.addPass(LoopDistributePass()); 1407 1408 // Populates the VFABI attribute with the scalar-to-vector mappings 1409 // from the TargetLibraryInfo. 1410 OptimizePM.addPass(InjectTLIMappings()); 1411 1412 addVectorPasses(Level, OptimizePM, /* IsFullLTO */ false); 1413 1414 // Split out cold code. Splitting is done late to avoid hiding context from 1415 // other optimizations and inadvertently regressing performance. The tradeoff 1416 // is that this has a higher code size cost than splitting early. 1417 if (EnableHotColdSplit && !LTOPreLink) 1418 MPM.addPass(HotColdSplittingPass()); 1419 1420 // Search the code for similar regions of code. If enough similar regions can 1421 // be found where extracting the regions into their own function will decrease 1422 // the size of the program, we extract the regions, a deduplicate the 1423 // structurally similar regions. 1424 if (EnableIROutliner) 1425 MPM.addPass(IROutlinerPass()); 1426 1427 // Merge functions if requested. 1428 if (PTO.MergeFunctions) 1429 MPM.addPass(MergeFunctionsPass()); 1430 1431 // LoopSink pass sinks instructions hoisted by LICM, which serves as a 1432 // canonicalization pass that enables other optimizations. As a result, 1433 // LoopSink pass needs to be a very late IR pass to avoid undoing LICM 1434 // result too early. 1435 OptimizePM.addPass(LoopSinkPass()); 1436 1437 // And finally clean up LCSSA form before generating code. 1438 OptimizePM.addPass(InstSimplifyPass()); 1439 1440 // This hoists/decomposes div/rem ops. It should run after other sink/hoist 1441 // passes to avoid re-sinking, but before SimplifyCFG because it can allow 1442 // flattening of blocks. 1443 OptimizePM.addPass(DivRemPairsPass()); 1444 1445 // LoopSink (and other loop passes since the last simplifyCFG) might have 1446 // resulted in single-entry-single-exit or empty blocks. Clean up the CFG. 1447 OptimizePM.addPass(SimplifyCFGPass()); 1448 1449 OptimizePM.addPass(CoroCleanupPass()); 1450 1451 // Add the core optimizing pipeline. 1452 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(OptimizePM))); 1453 1454 for (auto &C : OptimizerLastEPCallbacks) 1455 C(MPM, Level); 1456 1457 if (PTO.CallGraphProfile) 1458 MPM.addPass(CGProfilePass()); 1459 1460 // Now we need to do some global optimization transforms. 1461 // FIXME: It would seem like these should come first in the optimization 1462 // pipeline and maybe be the bottom of the canonicalization pipeline? Weird 1463 // ordering here. 1464 MPM.addPass(GlobalDCEPass()); 1465 MPM.addPass(ConstantMergePass()); 1466 1467 // TODO: Relative look table converter pass caused an issue when full lto is 1468 // enabled. See https://reviews.llvm.org/D94355 for more details. 1469 // Until the issue fixed, disable this pass during pre-linking phase. 1470 if (!LTOPreLink) 1471 MPM.addPass(RelLookupTableConverterPass()); 1472 1473 return MPM; 1474 } 1475 1476 ModulePassManager 1477 PassBuilder::buildPerModuleDefaultPipeline(OptimizationLevel Level, 1478 bool LTOPreLink) { 1479 assert(Level != OptimizationLevel::O0 && 1480 "Must request optimizations for the default pipeline!"); 1481 1482 ModulePassManager MPM; 1483 1484 // Convert @llvm.global.annotations to !annotation metadata. 1485 MPM.addPass(Annotation2MetadataPass()); 1486 1487 // Force any function attributes we want the rest of the pipeline to observe. 1488 MPM.addPass(ForceFunctionAttrsPass()); 1489 1490 // Apply module pipeline start EP callback. 1491 for (auto &C : PipelineStartEPCallbacks) 1492 C(MPM, Level); 1493 1494 if (PGOOpt && PGOOpt->DebugInfoForProfiling) 1495 MPM.addPass(createModuleToFunctionPassAdaptor(AddDiscriminatorsPass())); 1496 1497 // Add the core simplification pipeline. 1498 MPM.addPass(buildModuleSimplificationPipeline( 1499 Level, LTOPreLink ? ThinOrFullLTOPhase::FullLTOPreLink 1500 : ThinOrFullLTOPhase::None)); 1501 1502 // Now add the optimization pipeline. 1503 MPM.addPass(buildModuleOptimizationPipeline(Level, LTOPreLink)); 1504 1505 if (PGOOpt && PGOOpt->PseudoProbeForProfiling) 1506 MPM.addPass(PseudoProbeUpdatePass()); 1507 1508 // Emit annotation remarks. 1509 addAnnotationRemarksPass(MPM); 1510 1511 if (LTOPreLink) 1512 addRequiredLTOPreLinkPasses(MPM); 1513 1514 return MPM; 1515 } 1516 1517 ModulePassManager 1518 PassBuilder::buildThinLTOPreLinkDefaultPipeline(OptimizationLevel Level) { 1519 assert(Level != OptimizationLevel::O0 && 1520 "Must request optimizations for the default pipeline!"); 1521 1522 ModulePassManager MPM; 1523 1524 // Convert @llvm.global.annotations to !annotation metadata. 1525 MPM.addPass(Annotation2MetadataPass()); 1526 1527 // Force any function attributes we want the rest of the pipeline to observe. 1528 MPM.addPass(ForceFunctionAttrsPass()); 1529 1530 if (PGOOpt && PGOOpt->DebugInfoForProfiling) 1531 MPM.addPass(createModuleToFunctionPassAdaptor(AddDiscriminatorsPass())); 1532 1533 // Apply module pipeline start EP callback. 1534 for (auto &C : PipelineStartEPCallbacks) 1535 C(MPM, Level); 1536 1537 // If we are planning to perform ThinLTO later, we don't bloat the code with 1538 // unrolling/vectorization/... now. Just simplify the module as much as we 1539 // can. 1540 MPM.addPass(buildModuleSimplificationPipeline( 1541 Level, ThinOrFullLTOPhase::ThinLTOPreLink)); 1542 1543 // Run partial inlining pass to partially inline functions that have 1544 // large bodies. 1545 // FIXME: It isn't clear whether this is really the right place to run this 1546 // in ThinLTO. Because there is another canonicalization and simplification 1547 // phase that will run after the thin link, running this here ends up with 1548 // less information than will be available later and it may grow functions in 1549 // ways that aren't beneficial. 1550 if (RunPartialInlining) 1551 MPM.addPass(PartialInlinerPass()); 1552 1553 // Reduce the size of the IR as much as possible. 1554 MPM.addPass(GlobalOptPass()); 1555 1556 // Module simplification splits coroutines, but does not fully clean up 1557 // coroutine intrinsics. To ensure ThinLTO optimization passes don't trip up 1558 // on these, we schedule the cleanup here. 1559 MPM.addPass(createModuleToFunctionPassAdaptor(CoroCleanupPass())); 1560 1561 if (PGOOpt && PGOOpt->PseudoProbeForProfiling) 1562 MPM.addPass(PseudoProbeUpdatePass()); 1563 1564 // Handle OptimizerLastEPCallbacks added by clang on PreLink. Actual 1565 // optimization is going to be done in PostLink stage, but clang can't 1566 // add callbacks there in case of in-process ThinLTO called by linker. 1567 for (auto &C : OptimizerLastEPCallbacks) 1568 C(MPM, Level); 1569 1570 // Emit annotation remarks. 1571 addAnnotationRemarksPass(MPM); 1572 1573 addRequiredLTOPreLinkPasses(MPM); 1574 1575 return MPM; 1576 } 1577 1578 ModulePassManager PassBuilder::buildThinLTODefaultPipeline( 1579 OptimizationLevel Level, const ModuleSummaryIndex *ImportSummary) { 1580 ModulePassManager MPM; 1581 1582 // Convert @llvm.global.annotations to !annotation metadata. 1583 MPM.addPass(Annotation2MetadataPass()); 1584 1585 if (ImportSummary) { 1586 // These passes import type identifier resolutions for whole-program 1587 // devirtualization and CFI. They must run early because other passes may 1588 // disturb the specific instruction patterns that these passes look for, 1589 // creating dependencies on resolutions that may not appear in the summary. 1590 // 1591 // For example, GVN may transform the pattern assume(type.test) appearing in 1592 // two basic blocks into assume(phi(type.test, type.test)), which would 1593 // transform a dependency on a WPD resolution into a dependency on a type 1594 // identifier resolution for CFI. 1595 // 1596 // Also, WPD has access to more precise information than ICP and can 1597 // devirtualize more effectively, so it should operate on the IR first. 1598 // 1599 // The WPD and LowerTypeTest passes need to run at -O0 to lower type 1600 // metadata and intrinsics. 1601 MPM.addPass(WholeProgramDevirtPass(nullptr, ImportSummary)); 1602 MPM.addPass(LowerTypeTestsPass(nullptr, ImportSummary)); 1603 } 1604 1605 if (Level == OptimizationLevel::O0) { 1606 // Run a second time to clean up any type tests left behind by WPD for use 1607 // in ICP. 1608 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1609 // Drop available_externally and unreferenced globals. This is necessary 1610 // with ThinLTO in order to avoid leaving undefined references to dead 1611 // globals in the object file. 1612 MPM.addPass(EliminateAvailableExternallyPass()); 1613 MPM.addPass(GlobalDCEPass()); 1614 return MPM; 1615 } 1616 1617 // Force any function attributes we want the rest of the pipeline to observe. 1618 MPM.addPass(ForceFunctionAttrsPass()); 1619 1620 // Add the core simplification pipeline. 1621 MPM.addPass(buildModuleSimplificationPipeline( 1622 Level, ThinOrFullLTOPhase::ThinLTOPostLink)); 1623 1624 // Now add the optimization pipeline. 1625 MPM.addPass(buildModuleOptimizationPipeline(Level)); 1626 1627 // Emit annotation remarks. 1628 addAnnotationRemarksPass(MPM); 1629 1630 return MPM; 1631 } 1632 1633 ModulePassManager 1634 PassBuilder::buildLTOPreLinkDefaultPipeline(OptimizationLevel Level) { 1635 assert(Level != OptimizationLevel::O0 && 1636 "Must request optimizations for the default pipeline!"); 1637 // FIXME: We should use a customized pre-link pipeline! 1638 return buildPerModuleDefaultPipeline(Level, 1639 /* LTOPreLink */ true); 1640 } 1641 1642 ModulePassManager 1643 PassBuilder::buildLTODefaultPipeline(OptimizationLevel Level, 1644 ModuleSummaryIndex *ExportSummary) { 1645 ModulePassManager MPM; 1646 1647 // Convert @llvm.global.annotations to !annotation metadata. 1648 MPM.addPass(Annotation2MetadataPass()); 1649 1650 // Create a function that performs CFI checks for cross-DSO calls with targets 1651 // in the current module. 1652 MPM.addPass(CrossDSOCFIPass()); 1653 1654 if (Level == OptimizationLevel::O0) { 1655 // The WPD and LowerTypeTest passes need to run at -O0 to lower type 1656 // metadata and intrinsics. 1657 MPM.addPass(WholeProgramDevirtPass(ExportSummary, nullptr)); 1658 MPM.addPass(LowerTypeTestsPass(ExportSummary, nullptr)); 1659 // Run a second time to clean up any type tests left behind by WPD for use 1660 // in ICP. 1661 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1662 1663 // Emit annotation remarks. 1664 addAnnotationRemarksPass(MPM); 1665 1666 return MPM; 1667 } 1668 1669 if (PGOOpt && PGOOpt->Action == PGOOptions::SampleUse) { 1670 // Load sample profile before running the LTO optimization pipeline. 1671 MPM.addPass(SampleProfileLoaderPass(PGOOpt->ProfileFile, 1672 PGOOpt->ProfileRemappingFile, 1673 ThinOrFullLTOPhase::FullLTOPostLink)); 1674 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 1675 // RequireAnalysisPass for PSI before subsequent non-module passes. 1676 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 1677 } 1678 1679 // Remove unused virtual tables to improve the quality of code generated by 1680 // whole-program devirtualization and bitset lowering. 1681 MPM.addPass(GlobalDCEPass()); 1682 1683 // Force any function attributes we want the rest of the pipeline to observe. 1684 MPM.addPass(ForceFunctionAttrsPass()); 1685 1686 // Do basic inference of function attributes from known properties of system 1687 // libraries and other oracles. 1688 MPM.addPass(InferFunctionAttrsPass()); 1689 1690 if (Level.getSpeedupLevel() > 1) { 1691 FunctionPassManager EarlyFPM; 1692 EarlyFPM.addPass(CallSiteSplittingPass()); 1693 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(EarlyFPM))); 1694 1695 // Indirect call promotion. This should promote all the targets that are 1696 // left by the earlier promotion pass that promotes intra-module targets. 1697 // This two-step promotion is to save the compile time. For LTO, it should 1698 // produce the same result as if we only do promotion here. 1699 MPM.addPass(PGOIndirectCallPromotion( 1700 true /* InLTO */, PGOOpt && PGOOpt->Action == PGOOptions::SampleUse)); 1701 1702 if (EnableFunctionSpecialization) 1703 MPM.addPass(FunctionSpecializationPass()); 1704 // Propagate constants at call sites into the functions they call. This 1705 // opens opportunities for globalopt (and inlining) by substituting function 1706 // pointers passed as arguments to direct uses of functions. 1707 MPM.addPass(IPSCCPPass()); 1708 1709 // Attach metadata to indirect call sites indicating the set of functions 1710 // they may target at run-time. This should follow IPSCCP. 1711 MPM.addPass(CalledValuePropagationPass()); 1712 } 1713 1714 // Now deduce any function attributes based in the current code. 1715 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor( 1716 PostOrderFunctionAttrsPass())); 1717 1718 // Do RPO function attribute inference across the module to forward-propagate 1719 // attributes where applicable. 1720 // FIXME: Is this really an optimization rather than a canonicalization? 1721 MPM.addPass(ReversePostOrderFunctionAttrsPass()); 1722 1723 // Use in-range annotations on GEP indices to split globals where beneficial. 1724 MPM.addPass(GlobalSplitPass()); 1725 1726 // Run whole program optimization of virtual call when the list of callees 1727 // is fixed. 1728 MPM.addPass(WholeProgramDevirtPass(ExportSummary, nullptr)); 1729 1730 // Stop here at -O1. 1731 if (Level == OptimizationLevel::O1) { 1732 // The LowerTypeTestsPass needs to run to lower type metadata and the 1733 // type.test intrinsics. The pass does nothing if CFI is disabled. 1734 MPM.addPass(LowerTypeTestsPass(ExportSummary, nullptr)); 1735 // Run a second time to clean up any type tests left behind by WPD for use 1736 // in ICP (which is performed earlier than this in the regular LTO 1737 // pipeline). 1738 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1739 1740 // Emit annotation remarks. 1741 addAnnotationRemarksPass(MPM); 1742 1743 return MPM; 1744 } 1745 1746 // Optimize globals to try and fold them into constants. 1747 MPM.addPass(GlobalOptPass()); 1748 1749 // Promote any localized globals to SSA registers. 1750 MPM.addPass(createModuleToFunctionPassAdaptor(PromotePass())); 1751 1752 // Linking modules together can lead to duplicate global constant, only 1753 // keep one copy of each constant. 1754 MPM.addPass(ConstantMergePass()); 1755 1756 // Remove unused arguments from functions. 1757 MPM.addPass(DeadArgumentEliminationPass()); 1758 1759 // Reduce the code after globalopt and ipsccp. Both can open up significant 1760 // simplification opportunities, and both can propagate functions through 1761 // function pointers. When this happens, we often have to resolve varargs 1762 // calls, etc, so let instcombine do this. 1763 FunctionPassManager PeepholeFPM; 1764 if (Level == OptimizationLevel::O3) 1765 PeepholeFPM.addPass(AggressiveInstCombinePass()); 1766 PeepholeFPM.addPass(InstCombinePass()); 1767 invokePeepholeEPCallbacks(PeepholeFPM, Level); 1768 1769 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(PeepholeFPM))); 1770 1771 // Note: historically, the PruneEH pass was run first to deduce nounwind and 1772 // generally clean up exception handling overhead. It isn't clear this is 1773 // valuable as the inliner doesn't currently care whether it is inlining an 1774 // invoke or a call. 1775 // Run the inliner now. 1776 MPM.addPass(ModuleInlinerWrapperPass(getInlineParamsFromOptLevel(Level))); 1777 1778 // Optimize globals again after we ran the inliner. 1779 MPM.addPass(GlobalOptPass()); 1780 1781 // Garbage collect dead functions. 1782 // FIXME: Add ArgumentPromotion pass after once it's ported. 1783 MPM.addPass(GlobalDCEPass()); 1784 1785 FunctionPassManager FPM; 1786 // The IPO Passes may leave cruft around. Clean up after them. 1787 FPM.addPass(InstCombinePass()); 1788 invokePeepholeEPCallbacks(FPM, Level); 1789 1790 FPM.addPass(JumpThreadingPass(/*InsertFreezeWhenUnfoldingSelect*/ true)); 1791 1792 // Do a post inline PGO instrumentation and use pass. This is a context 1793 // sensitive PGO pass. 1794 if (PGOOpt) { 1795 if (PGOOpt->CSAction == PGOOptions::CSIRInstr) 1796 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ true, 1797 /* IsCS */ true, PGOOpt->CSProfileGenFile, 1798 PGOOpt->ProfileRemappingFile); 1799 else if (PGOOpt->CSAction == PGOOptions::CSIRUse) 1800 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ false, 1801 /* IsCS */ true, PGOOpt->ProfileFile, 1802 PGOOpt->ProfileRemappingFile); 1803 } 1804 1805 // Break up allocas 1806 FPM.addPass(SROA()); 1807 1808 // LTO provides additional opportunities for tailcall elimination due to 1809 // link-time inlining, and visibility of nocapture attribute. 1810 FPM.addPass(TailCallElimPass()); 1811 1812 // Run a few AA driver optimizations here and now to cleanup the code. 1813 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 1814 1815 MPM.addPass( 1816 createModuleToPostOrderCGSCCPassAdaptor(PostOrderFunctionAttrsPass())); 1817 1818 // Require the GlobalsAA analysis for the module so we can query it within 1819 // MainFPM. 1820 MPM.addPass(RequireAnalysisPass<GlobalsAA, Module>()); 1821 // Invalidate AAManager so it can be recreated and pick up the newly available 1822 // GlobalsAA. 1823 MPM.addPass( 1824 createModuleToFunctionPassAdaptor(InvalidateAnalysisPass<AAManager>())); 1825 1826 FunctionPassManager MainFPM; 1827 MainFPM.addPass(createFunctionToLoopPassAdaptor( 1828 LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap), 1829 EnableMSSALoopDependency, /*UseBlockFrequencyInfo=*/true)); 1830 1831 if (RunNewGVN) 1832 MainFPM.addPass(NewGVNPass()); 1833 else 1834 MainFPM.addPass(GVN()); 1835 1836 // Remove dead memcpy()'s. 1837 MainFPM.addPass(MemCpyOptPass()); 1838 1839 // Nuke dead stores. 1840 MainFPM.addPass(DSEPass()); 1841 MainFPM.addPass(MergedLoadStoreMotionPass()); 1842 1843 // More loops are countable; try to optimize them. 1844 if (EnableLoopFlatten && Level.getSpeedupLevel() > 1) 1845 MainFPM.addPass(createFunctionToLoopPassAdaptor(LoopFlattenPass())); 1846 1847 if (EnableConstraintElimination) 1848 MainFPM.addPass(ConstraintEliminationPass()); 1849 1850 LoopPassManager LPM; 1851 LPM.addPass(IndVarSimplifyPass()); 1852 LPM.addPass(LoopDeletionPass()); 1853 // FIXME: Add loop interchange. 1854 1855 // Unroll small loops and perform peeling. 1856 LPM.addPass(LoopFullUnrollPass(Level.getSpeedupLevel(), 1857 /* OnlyWhenForced= */ !PTO.LoopUnrolling, 1858 PTO.ForgetAllSCEVInLoopUnroll)); 1859 // The loop passes in LPM (LoopFullUnrollPass) do not preserve MemorySSA. 1860 // *All* loop passes must preserve it, in order to be able to use it. 1861 MainFPM.addPass(createFunctionToLoopPassAdaptor( 1862 std::move(LPM), /*UseMemorySSA=*/false, /*UseBlockFrequencyInfo=*/true)); 1863 1864 MainFPM.addPass(LoopDistributePass()); 1865 1866 addVectorPasses(Level, MainFPM, /* IsFullLTO */ true); 1867 1868 invokePeepholeEPCallbacks(MainFPM, Level); 1869 MainFPM.addPass(JumpThreadingPass(/*InsertFreezeWhenUnfoldingSelect*/ true)); 1870 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(MainFPM))); 1871 1872 // Lower type metadata and the type.test intrinsic. This pass supports 1873 // clang's control flow integrity mechanisms (-fsanitize=cfi*) and needs 1874 // to be run at link time if CFI is enabled. This pass does nothing if 1875 // CFI is disabled. 1876 MPM.addPass(LowerTypeTestsPass(ExportSummary, nullptr)); 1877 // Run a second time to clean up any type tests left behind by WPD for use 1878 // in ICP (which is performed earlier than this in the regular LTO pipeline). 1879 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1880 1881 // Enable splitting late in the FullLTO post-link pipeline. This is done in 1882 // the same stage in the old pass manager (\ref addLateLTOOptimizationPasses). 1883 if (EnableHotColdSplit) 1884 MPM.addPass(HotColdSplittingPass()); 1885 1886 // Add late LTO optimization passes. 1887 // Delete basic blocks, which optimization passes may have killed. 1888 MPM.addPass(createModuleToFunctionPassAdaptor( 1889 SimplifyCFGPass(SimplifyCFGOptions().hoistCommonInsts(true)))); 1890 1891 // Drop bodies of available eternally objects to improve GlobalDCE. 1892 MPM.addPass(EliminateAvailableExternallyPass()); 1893 1894 // Now that we have optimized the program, discard unreachable functions. 1895 MPM.addPass(GlobalDCEPass()); 1896 1897 if (PTO.MergeFunctions) 1898 MPM.addPass(MergeFunctionsPass()); 1899 1900 // Emit annotation remarks. 1901 addAnnotationRemarksPass(MPM); 1902 1903 return MPM; 1904 } 1905 1906 ModulePassManager PassBuilder::buildO0DefaultPipeline(OptimizationLevel Level, 1907 bool LTOPreLink) { 1908 assert(Level == OptimizationLevel::O0 && 1909 "buildO0DefaultPipeline should only be used with O0"); 1910 1911 ModulePassManager MPM; 1912 1913 if (PGOOpt && (PGOOpt->Action == PGOOptions::IRInstr || 1914 PGOOpt->Action == PGOOptions::IRUse)) 1915 addPGOInstrPassesForO0( 1916 MPM, 1917 /* RunProfileGen */ (PGOOpt->Action == PGOOptions::IRInstr), 1918 /* IsCS */ false, PGOOpt->ProfileFile, PGOOpt->ProfileRemappingFile); 1919 1920 for (auto &C : PipelineStartEPCallbacks) 1921 C(MPM, Level); 1922 1923 if (PGOOpt && PGOOpt->DebugInfoForProfiling) 1924 MPM.addPass(createModuleToFunctionPassAdaptor(AddDiscriminatorsPass())); 1925 1926 for (auto &C : PipelineEarlySimplificationEPCallbacks) 1927 C(MPM, Level); 1928 1929 // Build a minimal pipeline based on the semantics required by LLVM, 1930 // which is just that always inlining occurs. Further, disable generating 1931 // lifetime intrinsics to avoid enabling further optimizations during 1932 // code generation. 1933 MPM.addPass(AlwaysInlinerPass( 1934 /*InsertLifetimeIntrinsics=*/false)); 1935 1936 if (PTO.MergeFunctions) 1937 MPM.addPass(MergeFunctionsPass()); 1938 1939 if (EnableMatrix) 1940 MPM.addPass( 1941 createModuleToFunctionPassAdaptor(LowerMatrixIntrinsicsPass(true))); 1942 1943 if (!CGSCCOptimizerLateEPCallbacks.empty()) { 1944 CGSCCPassManager CGPM; 1945 for (auto &C : CGSCCOptimizerLateEPCallbacks) 1946 C(CGPM, Level); 1947 if (!CGPM.isEmpty()) 1948 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPM))); 1949 } 1950 if (!LateLoopOptimizationsEPCallbacks.empty()) { 1951 LoopPassManager LPM; 1952 for (auto &C : LateLoopOptimizationsEPCallbacks) 1953 C(LPM, Level); 1954 if (!LPM.isEmpty()) { 1955 MPM.addPass(createModuleToFunctionPassAdaptor( 1956 createFunctionToLoopPassAdaptor(std::move(LPM)))); 1957 } 1958 } 1959 if (!LoopOptimizerEndEPCallbacks.empty()) { 1960 LoopPassManager LPM; 1961 for (auto &C : LoopOptimizerEndEPCallbacks) 1962 C(LPM, Level); 1963 if (!LPM.isEmpty()) { 1964 MPM.addPass(createModuleToFunctionPassAdaptor( 1965 createFunctionToLoopPassAdaptor(std::move(LPM)))); 1966 } 1967 } 1968 if (!ScalarOptimizerLateEPCallbacks.empty()) { 1969 FunctionPassManager FPM; 1970 for (auto &C : ScalarOptimizerLateEPCallbacks) 1971 C(FPM, Level); 1972 if (!FPM.isEmpty()) 1973 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 1974 } 1975 if (!VectorizerStartEPCallbacks.empty()) { 1976 FunctionPassManager FPM; 1977 for (auto &C : VectorizerStartEPCallbacks) 1978 C(FPM, Level); 1979 if (!FPM.isEmpty()) 1980 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 1981 } 1982 1983 MPM.addPass(createModuleToFunctionPassAdaptor(CoroEarlyPass())); 1984 CGSCCPassManager CGPM; 1985 CGPM.addPass(CoroSplitPass()); 1986 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPM))); 1987 MPM.addPass(createModuleToFunctionPassAdaptor(CoroCleanupPass())); 1988 1989 for (auto &C : OptimizerLastEPCallbacks) 1990 C(MPM, Level); 1991 1992 if (LTOPreLink) 1993 addRequiredLTOPreLinkPasses(MPM); 1994 1995 return MPM; 1996 } 1997 1998 AAManager PassBuilder::buildDefaultAAPipeline() { 1999 AAManager AA; 2000 2001 // The order in which these are registered determines their priority when 2002 // being queried. 2003 2004 // First we register the basic alias analysis that provides the majority of 2005 // per-function local AA logic. This is a stateless, on-demand local set of 2006 // AA techniques. 2007 AA.registerFunctionAnalysis<BasicAA>(); 2008 2009 // Next we query fast, specialized alias analyses that wrap IR-embedded 2010 // information about aliasing. 2011 AA.registerFunctionAnalysis<ScopedNoAliasAA>(); 2012 AA.registerFunctionAnalysis<TypeBasedAA>(); 2013 2014 // Add support for querying global aliasing information when available. 2015 // Because the `AAManager` is a function analysis and `GlobalsAA` is a module 2016 // analysis, all that the `AAManager` can do is query for any *cached* 2017 // results from `GlobalsAA` through a readonly proxy. 2018 AA.registerModuleAnalysis<GlobalsAA>(); 2019 2020 // Add target-specific alias analyses. 2021 if (TM) 2022 TM->registerDefaultAliasAnalyses(AA); 2023 2024 return AA; 2025 } 2026 2027 static Optional<int> parseRepeatPassName(StringRef Name) { 2028 if (!Name.consume_front("repeat<") || !Name.consume_back(">")) 2029 return None; 2030 int Count; 2031 if (Name.getAsInteger(0, Count) || Count <= 0) 2032 return None; 2033 return Count; 2034 } 2035 2036 static Optional<int> parseDevirtPassName(StringRef Name) { 2037 if (!Name.consume_front("devirt<") || !Name.consume_back(">")) 2038 return None; 2039 int Count; 2040 if (Name.getAsInteger(0, Count) || Count < 0) 2041 return None; 2042 return Count; 2043 } 2044 2045 static bool checkParametrizedPassName(StringRef Name, StringRef PassName) { 2046 if (!Name.consume_front(PassName)) 2047 return false; 2048 // normal pass name w/o parameters == default parameters 2049 if (Name.empty()) 2050 return true; 2051 return Name.startswith("<") && Name.endswith(">"); 2052 } 2053 2054 namespace { 2055 2056 /// This performs customized parsing of pass name with parameters. 2057 /// 2058 /// We do not need parametrization of passes in textual pipeline very often, 2059 /// yet on a rare occasion ability to specify parameters right there can be 2060 /// useful. 2061 /// 2062 /// \p Name - parameterized specification of a pass from a textual pipeline 2063 /// is a string in a form of : 2064 /// PassName '<' parameter-list '>' 2065 /// 2066 /// Parameter list is being parsed by the parser callable argument, \p Parser, 2067 /// It takes a string-ref of parameters and returns either StringError or a 2068 /// parameter list in a form of a custom parameters type, all wrapped into 2069 /// Expected<> template class. 2070 /// 2071 template <typename ParametersParseCallableT> 2072 auto parsePassParameters(ParametersParseCallableT &&Parser, StringRef Name, 2073 StringRef PassName) -> decltype(Parser(StringRef{})) { 2074 using ParametersT = typename decltype(Parser(StringRef{}))::value_type; 2075 2076 StringRef Params = Name; 2077 if (!Params.consume_front(PassName)) { 2078 assert(false && 2079 "unable to strip pass name from parametrized pass specification"); 2080 } 2081 if (!Params.empty() && 2082 (!Params.consume_front("<") || !Params.consume_back(">"))) { 2083 assert(false && "invalid format for parametrized pass name"); 2084 } 2085 2086 Expected<ParametersT> Result = Parser(Params); 2087 assert((Result || Result.template errorIsA<StringError>()) && 2088 "Pass parameter parser can only return StringErrors."); 2089 return Result; 2090 } 2091 2092 /// Parser of parameters for LoopUnroll pass. 2093 Expected<LoopUnrollOptions> parseLoopUnrollOptions(StringRef Params) { 2094 LoopUnrollOptions UnrollOpts; 2095 while (!Params.empty()) { 2096 StringRef ParamName; 2097 std::tie(ParamName, Params) = Params.split(';'); 2098 int OptLevel = StringSwitch<int>(ParamName) 2099 .Case("O0", 0) 2100 .Case("O1", 1) 2101 .Case("O2", 2) 2102 .Case("O3", 3) 2103 .Default(-1); 2104 if (OptLevel >= 0) { 2105 UnrollOpts.setOptLevel(OptLevel); 2106 continue; 2107 } 2108 if (ParamName.consume_front("full-unroll-max=")) { 2109 int Count; 2110 if (ParamName.getAsInteger(0, Count)) 2111 return make_error<StringError>( 2112 formatv("invalid LoopUnrollPass parameter '{0}' ", ParamName).str(), 2113 inconvertibleErrorCode()); 2114 UnrollOpts.setFullUnrollMaxCount(Count); 2115 continue; 2116 } 2117 2118 bool Enable = !ParamName.consume_front("no-"); 2119 if (ParamName == "partial") { 2120 UnrollOpts.setPartial(Enable); 2121 } else if (ParamName == "peeling") { 2122 UnrollOpts.setPeeling(Enable); 2123 } else if (ParamName == "profile-peeling") { 2124 UnrollOpts.setProfileBasedPeeling(Enable); 2125 } else if (ParamName == "runtime") { 2126 UnrollOpts.setRuntime(Enable); 2127 } else if (ParamName == "upperbound") { 2128 UnrollOpts.setUpperBound(Enable); 2129 } else { 2130 return make_error<StringError>( 2131 formatv("invalid LoopUnrollPass parameter '{0}' ", ParamName).str(), 2132 inconvertibleErrorCode()); 2133 } 2134 } 2135 return UnrollOpts; 2136 } 2137 2138 Expected<MemorySanitizerOptions> parseMSanPassOptions(StringRef Params) { 2139 MemorySanitizerOptions Result; 2140 while (!Params.empty()) { 2141 StringRef ParamName; 2142 std::tie(ParamName, Params) = Params.split(';'); 2143 2144 if (ParamName == "recover") { 2145 Result.Recover = true; 2146 } else if (ParamName == "kernel") { 2147 Result.Kernel = true; 2148 } else if (ParamName.consume_front("track-origins=")) { 2149 if (ParamName.getAsInteger(0, Result.TrackOrigins)) 2150 return make_error<StringError>( 2151 formatv("invalid argument to MemorySanitizer pass track-origins " 2152 "parameter: '{0}' ", 2153 ParamName) 2154 .str(), 2155 inconvertibleErrorCode()); 2156 } else { 2157 return make_error<StringError>( 2158 formatv("invalid MemorySanitizer pass parameter '{0}' ", ParamName) 2159 .str(), 2160 inconvertibleErrorCode()); 2161 } 2162 } 2163 return Result; 2164 } 2165 2166 /// Parser of parameters for SimplifyCFG pass. 2167 Expected<SimplifyCFGOptions> parseSimplifyCFGOptions(StringRef Params) { 2168 SimplifyCFGOptions Result; 2169 while (!Params.empty()) { 2170 StringRef ParamName; 2171 std::tie(ParamName, Params) = Params.split(';'); 2172 2173 bool Enable = !ParamName.consume_front("no-"); 2174 if (ParamName == "forward-switch-cond") { 2175 Result.forwardSwitchCondToPhi(Enable); 2176 } else if (ParamName == "switch-to-lookup") { 2177 Result.convertSwitchToLookupTable(Enable); 2178 } else if (ParamName == "keep-loops") { 2179 Result.needCanonicalLoops(Enable); 2180 } else if (ParamName == "hoist-common-insts") { 2181 Result.hoistCommonInsts(Enable); 2182 } else if (ParamName == "sink-common-insts") { 2183 Result.sinkCommonInsts(Enable); 2184 } else if (Enable && ParamName.consume_front("bonus-inst-threshold=")) { 2185 APInt BonusInstThreshold; 2186 if (ParamName.getAsInteger(0, BonusInstThreshold)) 2187 return make_error<StringError>( 2188 formatv("invalid argument to SimplifyCFG pass bonus-threshold " 2189 "parameter: '{0}' ", 2190 ParamName).str(), 2191 inconvertibleErrorCode()); 2192 Result.bonusInstThreshold(BonusInstThreshold.getSExtValue()); 2193 } else { 2194 return make_error<StringError>( 2195 formatv("invalid SimplifyCFG pass parameter '{0}' ", ParamName).str(), 2196 inconvertibleErrorCode()); 2197 } 2198 } 2199 return Result; 2200 } 2201 2202 /// Parser of parameters for LoopVectorize pass. 2203 Expected<LoopVectorizeOptions> parseLoopVectorizeOptions(StringRef Params) { 2204 LoopVectorizeOptions Opts; 2205 while (!Params.empty()) { 2206 StringRef ParamName; 2207 std::tie(ParamName, Params) = Params.split(';'); 2208 2209 bool Enable = !ParamName.consume_front("no-"); 2210 if (ParamName == "interleave-forced-only") { 2211 Opts.setInterleaveOnlyWhenForced(Enable); 2212 } else if (ParamName == "vectorize-forced-only") { 2213 Opts.setVectorizeOnlyWhenForced(Enable); 2214 } else { 2215 return make_error<StringError>( 2216 formatv("invalid LoopVectorize parameter '{0}' ", ParamName).str(), 2217 inconvertibleErrorCode()); 2218 } 2219 } 2220 return Opts; 2221 } 2222 2223 Expected<std::pair<bool, bool>> parseLoopUnswitchOptions(StringRef Params) { 2224 std::pair<bool, bool> Result = {false, true}; 2225 while (!Params.empty()) { 2226 StringRef ParamName; 2227 std::tie(ParamName, Params) = Params.split(';'); 2228 2229 bool Enable = !ParamName.consume_front("no-"); 2230 if (ParamName == "nontrivial") { 2231 Result.first = Enable; 2232 } else if (ParamName == "trivial") { 2233 Result.second = Enable; 2234 } else { 2235 return make_error<StringError>( 2236 formatv("invalid LoopUnswitch pass parameter '{0}' ", ParamName) 2237 .str(), 2238 inconvertibleErrorCode()); 2239 } 2240 } 2241 return Result; 2242 } 2243 2244 Expected<bool> parseMergedLoadStoreMotionOptions(StringRef Params) { 2245 bool Result = false; 2246 while (!Params.empty()) { 2247 StringRef ParamName; 2248 std::tie(ParamName, Params) = Params.split(';'); 2249 2250 bool Enable = !ParamName.consume_front("no-"); 2251 if (ParamName == "split-footer-bb") { 2252 Result = Enable; 2253 } else { 2254 return make_error<StringError>( 2255 formatv("invalid MergedLoadStoreMotion pass parameter '{0}' ", 2256 ParamName) 2257 .str(), 2258 inconvertibleErrorCode()); 2259 } 2260 } 2261 return Result; 2262 } 2263 2264 Expected<GVNOptions> parseGVNOptions(StringRef Params) { 2265 GVNOptions Result; 2266 while (!Params.empty()) { 2267 StringRef ParamName; 2268 std::tie(ParamName, Params) = Params.split(';'); 2269 2270 bool Enable = !ParamName.consume_front("no-"); 2271 if (ParamName == "pre") { 2272 Result.setPRE(Enable); 2273 } else if (ParamName == "load-pre") { 2274 Result.setLoadPRE(Enable); 2275 } else if (ParamName == "split-backedge-load-pre") { 2276 Result.setLoadPRESplitBackedge(Enable); 2277 } else if (ParamName == "memdep") { 2278 Result.setMemDep(Enable); 2279 } else { 2280 return make_error<StringError>( 2281 formatv("invalid GVN pass parameter '{0}' ", ParamName).str(), 2282 inconvertibleErrorCode()); 2283 } 2284 } 2285 return Result; 2286 } 2287 2288 Expected<StackLifetime::LivenessType> 2289 parseStackLifetimeOptions(StringRef Params) { 2290 StackLifetime::LivenessType Result = StackLifetime::LivenessType::May; 2291 while (!Params.empty()) { 2292 StringRef ParamName; 2293 std::tie(ParamName, Params) = Params.split(';'); 2294 2295 if (ParamName == "may") { 2296 Result = StackLifetime::LivenessType::May; 2297 } else if (ParamName == "must") { 2298 Result = StackLifetime::LivenessType::Must; 2299 } else { 2300 return make_error<StringError>( 2301 formatv("invalid StackLifetime parameter '{0}' ", ParamName).str(), 2302 inconvertibleErrorCode()); 2303 } 2304 } 2305 return Result; 2306 } 2307 2308 } // namespace 2309 2310 /// Tests whether a pass name starts with a valid prefix for a default pipeline 2311 /// alias. 2312 static bool startsWithDefaultPipelineAliasPrefix(StringRef Name) { 2313 return Name.startswith("default") || Name.startswith("thinlto") || 2314 Name.startswith("lto"); 2315 } 2316 2317 /// Tests whether registered callbacks will accept a given pass name. 2318 /// 2319 /// When parsing a pipeline text, the type of the outermost pipeline may be 2320 /// omitted, in which case the type is automatically determined from the first 2321 /// pass name in the text. This may be a name that is handled through one of the 2322 /// callbacks. We check this through the oridinary parsing callbacks by setting 2323 /// up a dummy PassManager in order to not force the client to also handle this 2324 /// type of query. 2325 template <typename PassManagerT, typename CallbacksT> 2326 static bool callbacksAcceptPassName(StringRef Name, CallbacksT &Callbacks) { 2327 if (!Callbacks.empty()) { 2328 PassManagerT DummyPM; 2329 for (auto &CB : Callbacks) 2330 if (CB(Name, DummyPM, {})) 2331 return true; 2332 } 2333 return false; 2334 } 2335 2336 template <typename CallbacksT> 2337 static bool isModulePassName(StringRef Name, CallbacksT &Callbacks) { 2338 // Manually handle aliases for pre-configured pipeline fragments. 2339 if (startsWithDefaultPipelineAliasPrefix(Name)) 2340 return DefaultAliasRegex.match(Name); 2341 2342 // Explicitly handle pass manager names. 2343 if (Name == "module") 2344 return true; 2345 if (Name == "cgscc") 2346 return true; 2347 if (Name == "function") 2348 return true; 2349 2350 // Explicitly handle custom-parsed pass names. 2351 if (parseRepeatPassName(Name)) 2352 return true; 2353 2354 #define MODULE_PASS(NAME, CREATE_PASS) \ 2355 if (Name == NAME) \ 2356 return true; 2357 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 2358 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2359 return true; 2360 #include "PassRegistry.def" 2361 2362 return callbacksAcceptPassName<ModulePassManager>(Name, Callbacks); 2363 } 2364 2365 template <typename CallbacksT> 2366 static bool isCGSCCPassName(StringRef Name, CallbacksT &Callbacks) { 2367 // Explicitly handle pass manager names. 2368 if (Name == "cgscc") 2369 return true; 2370 if (Name == "function") 2371 return true; 2372 2373 // Explicitly handle custom-parsed pass names. 2374 if (parseRepeatPassName(Name)) 2375 return true; 2376 if (parseDevirtPassName(Name)) 2377 return true; 2378 2379 #define CGSCC_PASS(NAME, CREATE_PASS) \ 2380 if (Name == NAME) \ 2381 return true; 2382 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 2383 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2384 return true; 2385 #include "PassRegistry.def" 2386 2387 return callbacksAcceptPassName<CGSCCPassManager>(Name, Callbacks); 2388 } 2389 2390 template <typename CallbacksT> 2391 static bool isFunctionPassName(StringRef Name, CallbacksT &Callbacks) { 2392 // Explicitly handle pass manager names. 2393 if (Name == "function") 2394 return true; 2395 if (Name == "loop" || Name == "loop-mssa") 2396 return true; 2397 2398 // Explicitly handle custom-parsed pass names. 2399 if (parseRepeatPassName(Name)) 2400 return true; 2401 2402 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2403 if (Name == NAME) \ 2404 return true; 2405 #define FUNCTION_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 2406 if (checkParametrizedPassName(Name, NAME)) \ 2407 return true; 2408 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 2409 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2410 return true; 2411 #include "PassRegistry.def" 2412 2413 return callbacksAcceptPassName<FunctionPassManager>(Name, Callbacks); 2414 } 2415 2416 template <typename CallbacksT> 2417 static bool isLoopPassName(StringRef Name, CallbacksT &Callbacks) { 2418 // Explicitly handle pass manager names. 2419 if (Name == "loop" || Name == "loop-mssa") 2420 return true; 2421 2422 // Explicitly handle custom-parsed pass names. 2423 if (parseRepeatPassName(Name)) 2424 return true; 2425 2426 #define LOOP_PASS(NAME, CREATE_PASS) \ 2427 if (Name == NAME) \ 2428 return true; 2429 #define LOOP_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 2430 if (checkParametrizedPassName(Name, NAME)) \ 2431 return true; 2432 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 2433 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2434 return true; 2435 #include "PassRegistry.def" 2436 2437 return callbacksAcceptPassName<LoopPassManager>(Name, Callbacks); 2438 } 2439 2440 Optional<std::vector<PassBuilder::PipelineElement>> 2441 PassBuilder::parsePipelineText(StringRef Text) { 2442 std::vector<PipelineElement> ResultPipeline; 2443 2444 SmallVector<std::vector<PipelineElement> *, 4> PipelineStack = { 2445 &ResultPipeline}; 2446 for (;;) { 2447 std::vector<PipelineElement> &Pipeline = *PipelineStack.back(); 2448 size_t Pos = Text.find_first_of(",()"); 2449 Pipeline.push_back({Text.substr(0, Pos), {}}); 2450 2451 // If we have a single terminating name, we're done. 2452 if (Pos == Text.npos) 2453 break; 2454 2455 char Sep = Text[Pos]; 2456 Text = Text.substr(Pos + 1); 2457 if (Sep == ',') 2458 // Just a name ending in a comma, continue. 2459 continue; 2460 2461 if (Sep == '(') { 2462 // Push the inner pipeline onto the stack to continue processing. 2463 PipelineStack.push_back(&Pipeline.back().InnerPipeline); 2464 continue; 2465 } 2466 2467 assert(Sep == ')' && "Bogus separator!"); 2468 // When handling the close parenthesis, we greedily consume them to avoid 2469 // empty strings in the pipeline. 2470 do { 2471 // If we try to pop the outer pipeline we have unbalanced parentheses. 2472 if (PipelineStack.size() == 1) 2473 return None; 2474 2475 PipelineStack.pop_back(); 2476 } while (Text.consume_front(")")); 2477 2478 // Check if we've finished parsing. 2479 if (Text.empty()) 2480 break; 2481 2482 // Otherwise, the end of an inner pipeline always has to be followed by 2483 // a comma, and then we can continue. 2484 if (!Text.consume_front(",")) 2485 return None; 2486 } 2487 2488 if (PipelineStack.size() > 1) 2489 // Unbalanced paretheses. 2490 return None; 2491 2492 assert(PipelineStack.back() == &ResultPipeline && 2493 "Wrong pipeline at the bottom of the stack!"); 2494 return {std::move(ResultPipeline)}; 2495 } 2496 2497 Error PassBuilder::parseModulePass(ModulePassManager &MPM, 2498 const PipelineElement &E) { 2499 auto &Name = E.Name; 2500 auto &InnerPipeline = E.InnerPipeline; 2501 2502 // First handle complex passes like the pass managers which carry pipelines. 2503 if (!InnerPipeline.empty()) { 2504 if (Name == "module") { 2505 ModulePassManager NestedMPM; 2506 if (auto Err = parseModulePassPipeline(NestedMPM, InnerPipeline)) 2507 return Err; 2508 MPM.addPass(std::move(NestedMPM)); 2509 return Error::success(); 2510 } 2511 if (Name == "cgscc") { 2512 CGSCCPassManager CGPM; 2513 if (auto Err = parseCGSCCPassPipeline(CGPM, InnerPipeline)) 2514 return Err; 2515 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPM))); 2516 return Error::success(); 2517 } 2518 if (Name == "function") { 2519 FunctionPassManager FPM; 2520 if (auto Err = parseFunctionPassPipeline(FPM, InnerPipeline)) 2521 return Err; 2522 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 2523 return Error::success(); 2524 } 2525 if (auto Count = parseRepeatPassName(Name)) { 2526 ModulePassManager NestedMPM; 2527 if (auto Err = parseModulePassPipeline(NestedMPM, InnerPipeline)) 2528 return Err; 2529 MPM.addPass(createRepeatedPass(*Count, std::move(NestedMPM))); 2530 return Error::success(); 2531 } 2532 2533 for (auto &C : ModulePipelineParsingCallbacks) 2534 if (C(Name, MPM, InnerPipeline)) 2535 return Error::success(); 2536 2537 // Normal passes can't have pipelines. 2538 return make_error<StringError>( 2539 formatv("invalid use of '{0}' pass as module pipeline", Name).str(), 2540 inconvertibleErrorCode()); 2541 ; 2542 } 2543 2544 // Manually handle aliases for pre-configured pipeline fragments. 2545 if (startsWithDefaultPipelineAliasPrefix(Name)) { 2546 SmallVector<StringRef, 3> Matches; 2547 if (!DefaultAliasRegex.match(Name, &Matches)) 2548 return make_error<StringError>( 2549 formatv("unknown default pipeline alias '{0}'", Name).str(), 2550 inconvertibleErrorCode()); 2551 2552 assert(Matches.size() == 3 && "Must capture two matched strings!"); 2553 2554 OptimizationLevel L = StringSwitch<OptimizationLevel>(Matches[2]) 2555 .Case("O0", OptimizationLevel::O0) 2556 .Case("O1", OptimizationLevel::O1) 2557 .Case("O2", OptimizationLevel::O2) 2558 .Case("O3", OptimizationLevel::O3) 2559 .Case("Os", OptimizationLevel::Os) 2560 .Case("Oz", OptimizationLevel::Oz); 2561 if (L == OptimizationLevel::O0 && Matches[1] != "thinlto" && 2562 Matches[1] != "lto") { 2563 MPM.addPass(buildO0DefaultPipeline(L, Matches[1] == "thinlto-pre-link" || 2564 Matches[1] == "lto-pre-link")); 2565 return Error::success(); 2566 } 2567 2568 // This is consistent with old pass manager invoked via opt, but 2569 // inconsistent with clang. Clang doesn't enable loop vectorization 2570 // but does enable slp vectorization at Oz. 2571 PTO.LoopVectorization = 2572 L.getSpeedupLevel() > 1 && L != OptimizationLevel::Oz; 2573 PTO.SLPVectorization = 2574 L.getSpeedupLevel() > 1 && L != OptimizationLevel::Oz; 2575 2576 if (Matches[1] == "default") { 2577 MPM.addPass(buildPerModuleDefaultPipeline(L)); 2578 } else if (Matches[1] == "thinlto-pre-link") { 2579 MPM.addPass(buildThinLTOPreLinkDefaultPipeline(L)); 2580 } else if (Matches[1] == "thinlto") { 2581 MPM.addPass(buildThinLTODefaultPipeline(L, nullptr)); 2582 } else if (Matches[1] == "lto-pre-link") { 2583 MPM.addPass(buildLTOPreLinkDefaultPipeline(L)); 2584 } else { 2585 assert(Matches[1] == "lto" && "Not one of the matched options!"); 2586 MPM.addPass(buildLTODefaultPipeline(L, nullptr)); 2587 } 2588 return Error::success(); 2589 } 2590 2591 // Finally expand the basic registered passes from the .inc file. 2592 #define MODULE_PASS(NAME, CREATE_PASS) \ 2593 if (Name == NAME) { \ 2594 MPM.addPass(CREATE_PASS); \ 2595 return Error::success(); \ 2596 } 2597 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 2598 if (Name == "require<" NAME ">") { \ 2599 MPM.addPass( \ 2600 RequireAnalysisPass< \ 2601 std::remove_reference<decltype(CREATE_PASS)>::type, Module>()); \ 2602 return Error::success(); \ 2603 } \ 2604 if (Name == "invalidate<" NAME ">") { \ 2605 MPM.addPass(InvalidateAnalysisPass< \ 2606 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2607 return Error::success(); \ 2608 } 2609 #define CGSCC_PASS(NAME, CREATE_PASS) \ 2610 if (Name == NAME) { \ 2611 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(CREATE_PASS)); \ 2612 return Error::success(); \ 2613 } 2614 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2615 if (Name == NAME) { \ 2616 MPM.addPass(createModuleToFunctionPassAdaptor(CREATE_PASS)); \ 2617 return Error::success(); \ 2618 } 2619 #define FUNCTION_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 2620 if (checkParametrizedPassName(Name, NAME)) { \ 2621 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2622 if (!Params) \ 2623 return Params.takeError(); \ 2624 MPM.addPass(createModuleToFunctionPassAdaptor(CREATE_PASS(Params.get()))); \ 2625 return Error::success(); \ 2626 } 2627 #define LOOP_PASS(NAME, CREATE_PASS) \ 2628 if (Name == NAME) { \ 2629 MPM.addPass(createModuleToFunctionPassAdaptor( \ 2630 createFunctionToLoopPassAdaptor(CREATE_PASS, false, false))); \ 2631 return Error::success(); \ 2632 } 2633 #define LOOP_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 2634 if (checkParametrizedPassName(Name, NAME)) { \ 2635 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2636 if (!Params) \ 2637 return Params.takeError(); \ 2638 MPM.addPass( \ 2639 createModuleToFunctionPassAdaptor(createFunctionToLoopPassAdaptor( \ 2640 CREATE_PASS(Params.get()), false, false))); \ 2641 return Error::success(); \ 2642 } 2643 #include "PassRegistry.def" 2644 2645 for (auto &C : ModulePipelineParsingCallbacks) 2646 if (C(Name, MPM, InnerPipeline)) 2647 return Error::success(); 2648 return make_error<StringError>( 2649 formatv("unknown module pass '{0}'", Name).str(), 2650 inconvertibleErrorCode()); 2651 } 2652 2653 Error PassBuilder::parseCGSCCPass(CGSCCPassManager &CGPM, 2654 const PipelineElement &E) { 2655 auto &Name = E.Name; 2656 auto &InnerPipeline = E.InnerPipeline; 2657 2658 // First handle complex passes like the pass managers which carry pipelines. 2659 if (!InnerPipeline.empty()) { 2660 if (Name == "cgscc") { 2661 CGSCCPassManager NestedCGPM; 2662 if (auto Err = parseCGSCCPassPipeline(NestedCGPM, InnerPipeline)) 2663 return Err; 2664 // Add the nested pass manager with the appropriate adaptor. 2665 CGPM.addPass(std::move(NestedCGPM)); 2666 return Error::success(); 2667 } 2668 if (Name == "function") { 2669 FunctionPassManager FPM; 2670 if (auto Err = parseFunctionPassPipeline(FPM, InnerPipeline)) 2671 return Err; 2672 // Add the nested pass manager with the appropriate adaptor. 2673 CGPM.addPass(createCGSCCToFunctionPassAdaptor(std::move(FPM))); 2674 return Error::success(); 2675 } 2676 if (auto Count = parseRepeatPassName(Name)) { 2677 CGSCCPassManager NestedCGPM; 2678 if (auto Err = parseCGSCCPassPipeline(NestedCGPM, InnerPipeline)) 2679 return Err; 2680 CGPM.addPass(createRepeatedPass(*Count, std::move(NestedCGPM))); 2681 return Error::success(); 2682 } 2683 if (auto MaxRepetitions = parseDevirtPassName(Name)) { 2684 CGSCCPassManager NestedCGPM; 2685 if (auto Err = parseCGSCCPassPipeline(NestedCGPM, InnerPipeline)) 2686 return Err; 2687 CGPM.addPass( 2688 createDevirtSCCRepeatedPass(std::move(NestedCGPM), *MaxRepetitions)); 2689 return Error::success(); 2690 } 2691 2692 for (auto &C : CGSCCPipelineParsingCallbacks) 2693 if (C(Name, CGPM, InnerPipeline)) 2694 return Error::success(); 2695 2696 // Normal passes can't have pipelines. 2697 return make_error<StringError>( 2698 formatv("invalid use of '{0}' pass as cgscc pipeline", Name).str(), 2699 inconvertibleErrorCode()); 2700 } 2701 2702 // Now expand the basic registered passes from the .inc file. 2703 #define CGSCC_PASS(NAME, CREATE_PASS) \ 2704 if (Name == NAME) { \ 2705 CGPM.addPass(CREATE_PASS); \ 2706 return Error::success(); \ 2707 } 2708 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 2709 if (Name == "require<" NAME ">") { \ 2710 CGPM.addPass(RequireAnalysisPass< \ 2711 std::remove_reference<decltype(CREATE_PASS)>::type, \ 2712 LazyCallGraph::SCC, CGSCCAnalysisManager, LazyCallGraph &, \ 2713 CGSCCUpdateResult &>()); \ 2714 return Error::success(); \ 2715 } \ 2716 if (Name == "invalidate<" NAME ">") { \ 2717 CGPM.addPass(InvalidateAnalysisPass< \ 2718 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2719 return Error::success(); \ 2720 } 2721 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2722 if (Name == NAME) { \ 2723 CGPM.addPass(createCGSCCToFunctionPassAdaptor(CREATE_PASS)); \ 2724 return Error::success(); \ 2725 } 2726 #define FUNCTION_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 2727 if (checkParametrizedPassName(Name, NAME)) { \ 2728 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2729 if (!Params) \ 2730 return Params.takeError(); \ 2731 CGPM.addPass(createCGSCCToFunctionPassAdaptor(CREATE_PASS(Params.get()))); \ 2732 return Error::success(); \ 2733 } 2734 #define LOOP_PASS(NAME, CREATE_PASS) \ 2735 if (Name == NAME) { \ 2736 CGPM.addPass(createCGSCCToFunctionPassAdaptor( \ 2737 createFunctionToLoopPassAdaptor(CREATE_PASS, false, false))); \ 2738 return Error::success(); \ 2739 } 2740 #define LOOP_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 2741 if (checkParametrizedPassName(Name, NAME)) { \ 2742 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2743 if (!Params) \ 2744 return Params.takeError(); \ 2745 CGPM.addPass( \ 2746 createCGSCCToFunctionPassAdaptor(createFunctionToLoopPassAdaptor( \ 2747 CREATE_PASS(Params.get()), false, false))); \ 2748 return Error::success(); \ 2749 } 2750 #include "PassRegistry.def" 2751 2752 for (auto &C : CGSCCPipelineParsingCallbacks) 2753 if (C(Name, CGPM, InnerPipeline)) 2754 return Error::success(); 2755 return make_error<StringError>( 2756 formatv("unknown cgscc pass '{0}'", Name).str(), 2757 inconvertibleErrorCode()); 2758 } 2759 2760 Error PassBuilder::parseFunctionPass(FunctionPassManager &FPM, 2761 const PipelineElement &E) { 2762 auto &Name = E.Name; 2763 auto &InnerPipeline = E.InnerPipeline; 2764 2765 // First handle complex passes like the pass managers which carry pipelines. 2766 if (!InnerPipeline.empty()) { 2767 if (Name == "function") { 2768 FunctionPassManager NestedFPM; 2769 if (auto Err = parseFunctionPassPipeline(NestedFPM, InnerPipeline)) 2770 return Err; 2771 // Add the nested pass manager with the appropriate adaptor. 2772 FPM.addPass(std::move(NestedFPM)); 2773 return Error::success(); 2774 } 2775 if (Name == "loop" || Name == "loop-mssa") { 2776 LoopPassManager LPM; 2777 if (auto Err = parseLoopPassPipeline(LPM, InnerPipeline)) 2778 return Err; 2779 // Add the nested pass manager with the appropriate adaptor. 2780 bool UseMemorySSA = (Name == "loop-mssa"); 2781 bool UseBFI = llvm::any_of( 2782 InnerPipeline, [](auto Pipeline) { return Pipeline.Name == "licm"; }); 2783 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM), UseMemorySSA, 2784 UseBFI)); 2785 return Error::success(); 2786 } 2787 if (auto Count = parseRepeatPassName(Name)) { 2788 FunctionPassManager NestedFPM; 2789 if (auto Err = parseFunctionPassPipeline(NestedFPM, InnerPipeline)) 2790 return Err; 2791 FPM.addPass(createRepeatedPass(*Count, std::move(NestedFPM))); 2792 return Error::success(); 2793 } 2794 2795 for (auto &C : FunctionPipelineParsingCallbacks) 2796 if (C(Name, FPM, InnerPipeline)) 2797 return Error::success(); 2798 2799 // Normal passes can't have pipelines. 2800 return make_error<StringError>( 2801 formatv("invalid use of '{0}' pass as function pipeline", Name).str(), 2802 inconvertibleErrorCode()); 2803 } 2804 2805 // Now expand the basic registered passes from the .inc file. 2806 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2807 if (Name == NAME) { \ 2808 FPM.addPass(CREATE_PASS); \ 2809 return Error::success(); \ 2810 } 2811 #define FUNCTION_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 2812 if (checkParametrizedPassName(Name, NAME)) { \ 2813 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2814 if (!Params) \ 2815 return Params.takeError(); \ 2816 FPM.addPass(CREATE_PASS(Params.get())); \ 2817 return Error::success(); \ 2818 } 2819 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 2820 if (Name == "require<" NAME ">") { \ 2821 FPM.addPass( \ 2822 RequireAnalysisPass< \ 2823 std::remove_reference<decltype(CREATE_PASS)>::type, Function>()); \ 2824 return Error::success(); \ 2825 } \ 2826 if (Name == "invalidate<" NAME ">") { \ 2827 FPM.addPass(InvalidateAnalysisPass< \ 2828 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2829 return Error::success(); \ 2830 } 2831 // FIXME: UseMemorySSA is set to false. Maybe we could do things like: 2832 // bool UseMemorySSA = !("canon-freeze" || "loop-predication" || 2833 // "guard-widening"); 2834 // The risk is that it may become obsolete if we're not careful. 2835 #define LOOP_PASS(NAME, CREATE_PASS) \ 2836 if (Name == NAME) { \ 2837 FPM.addPass(createFunctionToLoopPassAdaptor(CREATE_PASS, false, false)); \ 2838 return Error::success(); \ 2839 } 2840 #define LOOP_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 2841 if (checkParametrizedPassName(Name, NAME)) { \ 2842 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2843 if (!Params) \ 2844 return Params.takeError(); \ 2845 FPM.addPass(createFunctionToLoopPassAdaptor(CREATE_PASS(Params.get()), \ 2846 false, false)); \ 2847 return Error::success(); \ 2848 } 2849 #include "PassRegistry.def" 2850 2851 for (auto &C : FunctionPipelineParsingCallbacks) 2852 if (C(Name, FPM, InnerPipeline)) 2853 return Error::success(); 2854 return make_error<StringError>( 2855 formatv("unknown function pass '{0}'", Name).str(), 2856 inconvertibleErrorCode()); 2857 } 2858 2859 Error PassBuilder::parseLoopPass(LoopPassManager &LPM, 2860 const PipelineElement &E) { 2861 StringRef Name = E.Name; 2862 auto &InnerPipeline = E.InnerPipeline; 2863 2864 // First handle complex passes like the pass managers which carry pipelines. 2865 if (!InnerPipeline.empty()) { 2866 if (Name == "loop") { 2867 LoopPassManager NestedLPM; 2868 if (auto Err = parseLoopPassPipeline(NestedLPM, InnerPipeline)) 2869 return Err; 2870 // Add the nested pass manager with the appropriate adaptor. 2871 LPM.addPass(std::move(NestedLPM)); 2872 return Error::success(); 2873 } 2874 if (auto Count = parseRepeatPassName(Name)) { 2875 LoopPassManager NestedLPM; 2876 if (auto Err = parseLoopPassPipeline(NestedLPM, InnerPipeline)) 2877 return Err; 2878 LPM.addPass(createRepeatedPass(*Count, std::move(NestedLPM))); 2879 return Error::success(); 2880 } 2881 2882 for (auto &C : LoopPipelineParsingCallbacks) 2883 if (C(Name, LPM, InnerPipeline)) 2884 return Error::success(); 2885 2886 // Normal passes can't have pipelines. 2887 return make_error<StringError>( 2888 formatv("invalid use of '{0}' pass as loop pipeline", Name).str(), 2889 inconvertibleErrorCode()); 2890 } 2891 2892 // Now expand the basic registered passes from the .inc file. 2893 #define LOOP_PASS(NAME, CREATE_PASS) \ 2894 if (Name == NAME) { \ 2895 LPM.addPass(CREATE_PASS); \ 2896 return Error::success(); \ 2897 } 2898 #define LOOP_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 2899 if (checkParametrizedPassName(Name, NAME)) { \ 2900 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2901 if (!Params) \ 2902 return Params.takeError(); \ 2903 LPM.addPass(CREATE_PASS(Params.get())); \ 2904 return Error::success(); \ 2905 } 2906 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 2907 if (Name == "require<" NAME ">") { \ 2908 LPM.addPass(RequireAnalysisPass< \ 2909 std::remove_reference<decltype(CREATE_PASS)>::type, Loop, \ 2910 LoopAnalysisManager, LoopStandardAnalysisResults &, \ 2911 LPMUpdater &>()); \ 2912 return Error::success(); \ 2913 } \ 2914 if (Name == "invalidate<" NAME ">") { \ 2915 LPM.addPass(InvalidateAnalysisPass< \ 2916 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2917 return Error::success(); \ 2918 } 2919 #include "PassRegistry.def" 2920 2921 for (auto &C : LoopPipelineParsingCallbacks) 2922 if (C(Name, LPM, InnerPipeline)) 2923 return Error::success(); 2924 return make_error<StringError>(formatv("unknown loop pass '{0}'", Name).str(), 2925 inconvertibleErrorCode()); 2926 } 2927 2928 bool PassBuilder::parseAAPassName(AAManager &AA, StringRef Name) { 2929 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 2930 if (Name == NAME) { \ 2931 AA.registerModuleAnalysis< \ 2932 std::remove_reference<decltype(CREATE_PASS)>::type>(); \ 2933 return true; \ 2934 } 2935 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 2936 if (Name == NAME) { \ 2937 AA.registerFunctionAnalysis< \ 2938 std::remove_reference<decltype(CREATE_PASS)>::type>(); \ 2939 return true; \ 2940 } 2941 #include "PassRegistry.def" 2942 2943 for (auto &C : AAParsingCallbacks) 2944 if (C(Name, AA)) 2945 return true; 2946 return false; 2947 } 2948 2949 Error PassBuilder::parseLoopPassPipeline(LoopPassManager &LPM, 2950 ArrayRef<PipelineElement> Pipeline) { 2951 for (const auto &Element : Pipeline) { 2952 if (auto Err = parseLoopPass(LPM, Element)) 2953 return Err; 2954 } 2955 return Error::success(); 2956 } 2957 2958 Error PassBuilder::parseFunctionPassPipeline( 2959 FunctionPassManager &FPM, ArrayRef<PipelineElement> Pipeline) { 2960 for (const auto &Element : Pipeline) { 2961 if (auto Err = parseFunctionPass(FPM, Element)) 2962 return Err; 2963 } 2964 return Error::success(); 2965 } 2966 2967 Error PassBuilder::parseCGSCCPassPipeline(CGSCCPassManager &CGPM, 2968 ArrayRef<PipelineElement> Pipeline) { 2969 for (const auto &Element : Pipeline) { 2970 if (auto Err = parseCGSCCPass(CGPM, Element)) 2971 return Err; 2972 } 2973 return Error::success(); 2974 } 2975 2976 void PassBuilder::crossRegisterProxies(LoopAnalysisManager &LAM, 2977 FunctionAnalysisManager &FAM, 2978 CGSCCAnalysisManager &CGAM, 2979 ModuleAnalysisManager &MAM) { 2980 MAM.registerPass([&] { return FunctionAnalysisManagerModuleProxy(FAM); }); 2981 MAM.registerPass([&] { return CGSCCAnalysisManagerModuleProxy(CGAM); }); 2982 CGAM.registerPass([&] { return ModuleAnalysisManagerCGSCCProxy(MAM); }); 2983 FAM.registerPass([&] { return CGSCCAnalysisManagerFunctionProxy(CGAM); }); 2984 FAM.registerPass([&] { return ModuleAnalysisManagerFunctionProxy(MAM); }); 2985 FAM.registerPass([&] { return LoopAnalysisManagerFunctionProxy(LAM); }); 2986 LAM.registerPass([&] { return FunctionAnalysisManagerLoopProxy(FAM); }); 2987 } 2988 2989 Error PassBuilder::parseModulePassPipeline(ModulePassManager &MPM, 2990 ArrayRef<PipelineElement> Pipeline) { 2991 for (const auto &Element : Pipeline) { 2992 if (auto Err = parseModulePass(MPM, Element)) 2993 return Err; 2994 } 2995 return Error::success(); 2996 } 2997 2998 // Primary pass pipeline description parsing routine for a \c ModulePassManager 2999 // FIXME: Should this routine accept a TargetMachine or require the caller to 3000 // pre-populate the analysis managers with target-specific stuff? 3001 Error PassBuilder::parsePassPipeline(ModulePassManager &MPM, 3002 StringRef PipelineText) { 3003 auto Pipeline = parsePipelineText(PipelineText); 3004 if (!Pipeline || Pipeline->empty()) 3005 return make_error<StringError>( 3006 formatv("invalid pipeline '{0}'", PipelineText).str(), 3007 inconvertibleErrorCode()); 3008 3009 // If the first name isn't at the module layer, wrap the pipeline up 3010 // automatically. 3011 StringRef FirstName = Pipeline->front().Name; 3012 3013 if (!isModulePassName(FirstName, ModulePipelineParsingCallbacks)) { 3014 if (isCGSCCPassName(FirstName, CGSCCPipelineParsingCallbacks)) { 3015 Pipeline = {{"cgscc", std::move(*Pipeline)}}; 3016 } else if (isFunctionPassName(FirstName, 3017 FunctionPipelineParsingCallbacks)) { 3018 Pipeline = {{"function", std::move(*Pipeline)}}; 3019 } else if (isLoopPassName(FirstName, LoopPipelineParsingCallbacks)) { 3020 Pipeline = {{"function", {{"loop", std::move(*Pipeline)}}}}; 3021 } else { 3022 for (auto &C : TopLevelPipelineParsingCallbacks) 3023 if (C(MPM, *Pipeline)) 3024 return Error::success(); 3025 3026 // Unknown pass or pipeline name! 3027 auto &InnerPipeline = Pipeline->front().InnerPipeline; 3028 return make_error<StringError>( 3029 formatv("unknown {0} name '{1}'", 3030 (InnerPipeline.empty() ? "pass" : "pipeline"), FirstName) 3031 .str(), 3032 inconvertibleErrorCode()); 3033 } 3034 } 3035 3036 if (auto Err = parseModulePassPipeline(MPM, *Pipeline)) 3037 return Err; 3038 return Error::success(); 3039 } 3040 3041 // Primary pass pipeline description parsing routine for a \c CGSCCPassManager 3042 Error PassBuilder::parsePassPipeline(CGSCCPassManager &CGPM, 3043 StringRef PipelineText) { 3044 auto Pipeline = parsePipelineText(PipelineText); 3045 if (!Pipeline || Pipeline->empty()) 3046 return make_error<StringError>( 3047 formatv("invalid pipeline '{0}'", PipelineText).str(), 3048 inconvertibleErrorCode()); 3049 3050 StringRef FirstName = Pipeline->front().Name; 3051 if (!isCGSCCPassName(FirstName, CGSCCPipelineParsingCallbacks)) 3052 return make_error<StringError>( 3053 formatv("unknown cgscc pass '{0}' in pipeline '{1}'", FirstName, 3054 PipelineText) 3055 .str(), 3056 inconvertibleErrorCode()); 3057 3058 if (auto Err = parseCGSCCPassPipeline(CGPM, *Pipeline)) 3059 return Err; 3060 return Error::success(); 3061 } 3062 3063 // Primary pass pipeline description parsing routine for a \c 3064 // FunctionPassManager 3065 Error PassBuilder::parsePassPipeline(FunctionPassManager &FPM, 3066 StringRef PipelineText) { 3067 auto Pipeline = parsePipelineText(PipelineText); 3068 if (!Pipeline || Pipeline->empty()) 3069 return make_error<StringError>( 3070 formatv("invalid pipeline '{0}'", PipelineText).str(), 3071 inconvertibleErrorCode()); 3072 3073 StringRef FirstName = Pipeline->front().Name; 3074 if (!isFunctionPassName(FirstName, FunctionPipelineParsingCallbacks)) 3075 return make_error<StringError>( 3076 formatv("unknown function pass '{0}' in pipeline '{1}'", FirstName, 3077 PipelineText) 3078 .str(), 3079 inconvertibleErrorCode()); 3080 3081 if (auto Err = parseFunctionPassPipeline(FPM, *Pipeline)) 3082 return Err; 3083 return Error::success(); 3084 } 3085 3086 // Primary pass pipeline description parsing routine for a \c LoopPassManager 3087 Error PassBuilder::parsePassPipeline(LoopPassManager &CGPM, 3088 StringRef PipelineText) { 3089 auto Pipeline = parsePipelineText(PipelineText); 3090 if (!Pipeline || Pipeline->empty()) 3091 return make_error<StringError>( 3092 formatv("invalid pipeline '{0}'", PipelineText).str(), 3093 inconvertibleErrorCode()); 3094 3095 if (auto Err = parseLoopPassPipeline(CGPM, *Pipeline)) 3096 return Err; 3097 3098 return Error::success(); 3099 } 3100 3101 Error PassBuilder::parseAAPipeline(AAManager &AA, StringRef PipelineText) { 3102 // If the pipeline just consists of the word 'default' just replace the AA 3103 // manager with our default one. 3104 if (PipelineText == "default") { 3105 AA = buildDefaultAAPipeline(); 3106 return Error::success(); 3107 } 3108 3109 while (!PipelineText.empty()) { 3110 StringRef Name; 3111 std::tie(Name, PipelineText) = PipelineText.split(','); 3112 if (!parseAAPassName(AA, Name)) 3113 return make_error<StringError>( 3114 formatv("unknown alias analysis name '{0}'", Name).str(), 3115 inconvertibleErrorCode()); 3116 } 3117 3118 return Error::success(); 3119 } 3120 3121 bool PassBuilder::isAAPassName(StringRef PassName) { 3122 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3123 if (PassName == NAME) \ 3124 return true; 3125 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3126 if (PassName == NAME) \ 3127 return true; 3128 #include "PassRegistry.def" 3129 return false; 3130 } 3131 3132 bool PassBuilder::isAnalysisPassName(StringRef PassName) { 3133 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 3134 if (PassName == NAME) \ 3135 return true; 3136 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 3137 if (PassName == NAME) \ 3138 return true; 3139 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 3140 if (PassName == NAME) \ 3141 return true; 3142 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 3143 if (PassName == NAME) \ 3144 return true; 3145 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3146 if (PassName == NAME) \ 3147 return true; 3148 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3149 if (PassName == NAME) \ 3150 return true; 3151 #include "PassRegistry.def" 3152 return false; 3153 } 3154 3155 static void printPassName(StringRef PassName, raw_ostream &OS) { 3156 OS << " " << PassName << "\n"; 3157 } 3158 static void printPassName(StringRef PassName, StringRef Params, 3159 raw_ostream &OS) { 3160 OS << " " << PassName << "<" << Params << ">\n"; 3161 } 3162 3163 void PassBuilder::printPassNames(raw_ostream &OS) { 3164 // TODO: print pass descriptions when they are available 3165 3166 OS << "Module passes:\n"; 3167 #define MODULE_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3168 #include "PassRegistry.def" 3169 3170 OS << "Module analyses:\n"; 3171 #define MODULE_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3172 #include "PassRegistry.def" 3173 3174 OS << "Module alias analyses:\n"; 3175 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3176 #include "PassRegistry.def" 3177 3178 OS << "CGSCC passes:\n"; 3179 #define CGSCC_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3180 #include "PassRegistry.def" 3181 3182 OS << "CGSCC analyses:\n"; 3183 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3184 #include "PassRegistry.def" 3185 3186 OS << "Function passes:\n"; 3187 #define FUNCTION_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3188 #include "PassRegistry.def" 3189 3190 OS << "Function passes with params:\n"; 3191 #define FUNCTION_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 3192 printPassName(NAME, PARAMS, OS); 3193 #include "PassRegistry.def" 3194 3195 OS << "Function analyses:\n"; 3196 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3197 #include "PassRegistry.def" 3198 3199 OS << "Function alias analyses:\n"; 3200 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3201 #include "PassRegistry.def" 3202 3203 OS << "Loop passes:\n"; 3204 #define LOOP_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3205 #include "PassRegistry.def" 3206 3207 OS << "Loop passes with params:\n"; 3208 #define LOOP_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 3209 printPassName(NAME, PARAMS, OS); 3210 #include "PassRegistry.def" 3211 3212 OS << "Loop analyses:\n"; 3213 #define LOOP_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3214 #include "PassRegistry.def" 3215 } 3216 3217 void PassBuilder::registerParseTopLevelPipelineCallback( 3218 const std::function<bool(ModulePassManager &, ArrayRef<PipelineElement>)> 3219 &C) { 3220 TopLevelPipelineParsingCallbacks.push_back(C); 3221 } 3222