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/DFAJumpThreading.h" 149 #include "llvm/Transforms/Scalar/DeadStoreElimination.h" 150 #include "llvm/Transforms/Scalar/DivRemPairs.h" 151 #include "llvm/Transforms/Scalar/EarlyCSE.h" 152 #include "llvm/Transforms/Scalar/Float2Int.h" 153 #include "llvm/Transforms/Scalar/GVN.h" 154 #include "llvm/Transforms/Scalar/GuardWidening.h" 155 #include "llvm/Transforms/Scalar/IVUsersPrinter.h" 156 #include "llvm/Transforms/Scalar/IndVarSimplify.h" 157 #include "llvm/Transforms/Scalar/InductiveRangeCheckElimination.h" 158 #include "llvm/Transforms/Scalar/InferAddressSpaces.h" 159 #include "llvm/Transforms/Scalar/InstSimplifyPass.h" 160 #include "llvm/Transforms/Scalar/JumpThreading.h" 161 #include "llvm/Transforms/Scalar/LICM.h" 162 #include "llvm/Transforms/Scalar/LoopAccessAnalysisPrinter.h" 163 #include "llvm/Transforms/Scalar/LoopBoundSplit.h" 164 #include "llvm/Transforms/Scalar/LoopDataPrefetch.h" 165 #include "llvm/Transforms/Scalar/LoopDeletion.h" 166 #include "llvm/Transforms/Scalar/LoopDistribute.h" 167 #include "llvm/Transforms/Scalar/LoopFlatten.h" 168 #include "llvm/Transforms/Scalar/LoopFuse.h" 169 #include "llvm/Transforms/Scalar/LoopIdiomRecognize.h" 170 #include "llvm/Transforms/Scalar/LoopInstSimplify.h" 171 #include "llvm/Transforms/Scalar/LoopInterchange.h" 172 #include "llvm/Transforms/Scalar/LoopLoadElimination.h" 173 #include "llvm/Transforms/Scalar/LoopPassManager.h" 174 #include "llvm/Transforms/Scalar/LoopPredication.h" 175 #include "llvm/Transforms/Scalar/LoopReroll.h" 176 #include "llvm/Transforms/Scalar/LoopRotation.h" 177 #include "llvm/Transforms/Scalar/LoopSimplifyCFG.h" 178 #include "llvm/Transforms/Scalar/LoopSink.h" 179 #include "llvm/Transforms/Scalar/LoopStrengthReduce.h" 180 #include "llvm/Transforms/Scalar/LoopUnrollAndJamPass.h" 181 #include "llvm/Transforms/Scalar/LoopUnrollPass.h" 182 #include "llvm/Transforms/Scalar/LoopVersioningLICM.h" 183 #include "llvm/Transforms/Scalar/LowerAtomic.h" 184 #include "llvm/Transforms/Scalar/LowerConstantIntrinsics.h" 185 #include "llvm/Transforms/Scalar/LowerExpectIntrinsic.h" 186 #include "llvm/Transforms/Scalar/LowerGuardIntrinsic.h" 187 #include "llvm/Transforms/Scalar/LowerMatrixIntrinsics.h" 188 #include "llvm/Transforms/Scalar/LowerWidenableCondition.h" 189 #include "llvm/Transforms/Scalar/MakeGuardsExplicit.h" 190 #include "llvm/Transforms/Scalar/MemCpyOptimizer.h" 191 #include "llvm/Transforms/Scalar/MergeICmps.h" 192 #include "llvm/Transforms/Scalar/MergedLoadStoreMotion.h" 193 #include "llvm/Transforms/Scalar/NaryReassociate.h" 194 #include "llvm/Transforms/Scalar/NewGVN.h" 195 #include "llvm/Transforms/Scalar/PartiallyInlineLibCalls.h" 196 #include "llvm/Transforms/Scalar/Reassociate.h" 197 #include "llvm/Transforms/Scalar/Reg2Mem.h" 198 #include "llvm/Transforms/Scalar/RewriteStatepointsForGC.h" 199 #include "llvm/Transforms/Scalar/SCCP.h" 200 #include "llvm/Transforms/Scalar/SROA.h" 201 #include "llvm/Transforms/Scalar/ScalarizeMaskedMemIntrin.h" 202 #include "llvm/Transforms/Scalar/Scalarizer.h" 203 #include "llvm/Transforms/Scalar/SeparateConstOffsetFromGEP.h" 204 #include "llvm/Transforms/Scalar/SimpleLoopUnswitch.h" 205 #include "llvm/Transforms/Scalar/SimplifyCFG.h" 206 #include "llvm/Transforms/Scalar/Sink.h" 207 #include "llvm/Transforms/Scalar/SpeculativeExecution.h" 208 #include "llvm/Transforms/Scalar/StraightLineStrengthReduce.h" 209 #include "llvm/Transforms/Scalar/StructurizeCFG.h" 210 #include "llvm/Transforms/Scalar/TailRecursionElimination.h" 211 #include "llvm/Transforms/Scalar/WarnMissedTransforms.h" 212 #include "llvm/Transforms/Utils/AddDiscriminators.h" 213 #include "llvm/Transforms/Utils/AssumeBundleBuilder.h" 214 #include "llvm/Transforms/Utils/BreakCriticalEdges.h" 215 #include "llvm/Transforms/Utils/CanonicalizeAliases.h" 216 #include "llvm/Transforms/Utils/CanonicalizeFreezeInLoops.h" 217 #include "llvm/Transforms/Utils/EntryExitInstrumenter.h" 218 #include "llvm/Transforms/Utils/FixIrreducible.h" 219 #include "llvm/Transforms/Utils/HelloWorld.h" 220 #include "llvm/Transforms/Utils/InjectTLIMappings.h" 221 #include "llvm/Transforms/Utils/InstructionNamer.h" 222 #include "llvm/Transforms/Utils/LCSSA.h" 223 #include "llvm/Transforms/Utils/LibCallsShrinkWrap.h" 224 #include "llvm/Transforms/Utils/LoopSimplify.h" 225 #include "llvm/Transforms/Utils/LoopVersioning.h" 226 #include "llvm/Transforms/Utils/LowerInvoke.h" 227 #include "llvm/Transforms/Utils/LowerSwitch.h" 228 #include "llvm/Transforms/Utils/Mem2Reg.h" 229 #include "llvm/Transforms/Utils/MetaRenamer.h" 230 #include "llvm/Transforms/Utils/NameAnonGlobals.h" 231 #include "llvm/Transforms/Utils/RelLookupTableConverter.h" 232 #include "llvm/Transforms/Utils/StripGCRelocates.h" 233 #include "llvm/Transforms/Utils/StripNonLineTableDebugInfo.h" 234 #include "llvm/Transforms/Utils/SymbolRewriter.h" 235 #include "llvm/Transforms/Utils/UnifyFunctionExitNodes.h" 236 #include "llvm/Transforms/Utils/UnifyLoopExits.h" 237 #include "llvm/Transforms/Vectorize/LoadStoreVectorizer.h" 238 #include "llvm/Transforms/Vectorize/LoopVectorize.h" 239 #include "llvm/Transforms/Vectorize/SLPVectorizer.h" 240 #include "llvm/Transforms/Vectorize/VectorCombine.h" 241 242 using namespace llvm; 243 244 static cl::opt<InliningAdvisorMode> UseInlineAdvisor( 245 "enable-ml-inliner", cl::init(InliningAdvisorMode::Default), cl::Hidden, 246 cl::desc("Enable ML policy for inliner. Currently trained for -Oz only"), 247 cl::values(clEnumValN(InliningAdvisorMode::Default, "default", 248 "Heuristics-based inliner version."), 249 clEnumValN(InliningAdvisorMode::Development, "development", 250 "Use development mode (runtime-loadable model)."), 251 clEnumValN(InliningAdvisorMode::Release, "release", 252 "Use release mode (AOT-compiled model)."))); 253 254 static cl::opt<bool> EnableSyntheticCounts( 255 "enable-npm-synthetic-counts", cl::init(false), cl::Hidden, cl::ZeroOrMore, 256 cl::desc("Run synthetic function entry count generation " 257 "pass")); 258 259 static const Regex DefaultAliasRegex( 260 "^(default|thinlto-pre-link|thinlto|lto-pre-link|lto)<(O[0123sz])>$"); 261 262 /// Flag to enable inline deferral during PGO. 263 static cl::opt<bool> 264 EnablePGOInlineDeferral("enable-npm-pgo-inline-deferral", cl::init(true), 265 cl::Hidden, 266 cl::desc("Enable inline deferral during PGO")); 267 268 static cl::opt<bool> EnableMemProfiler("enable-mem-prof", cl::init(false), 269 cl::Hidden, cl::ZeroOrMore, 270 cl::desc("Enable memory profiler")); 271 272 static cl::opt<bool> PerformMandatoryInliningsFirst( 273 "mandatory-inlining-first", cl::init(true), cl::Hidden, cl::ZeroOrMore, 274 cl::desc("Perform mandatory inlinings module-wide, before performing " 275 "inlining.")); 276 277 static cl::opt<bool> EnableO3NonTrivialUnswitching( 278 "enable-npm-O3-nontrivial-unswitch", cl::init(true), cl::Hidden, 279 cl::ZeroOrMore, cl::desc("Enable non-trivial loop unswitching for -O3")); 280 281 PipelineTuningOptions::PipelineTuningOptions() { 282 LoopInterleaving = true; 283 LoopVectorization = true; 284 SLPVectorization = false; 285 LoopUnrolling = true; 286 ForgetAllSCEVInLoopUnroll = ForgetSCEVInLoopUnroll; 287 LicmMssaOptCap = SetLicmMssaOptCap; 288 LicmMssaNoAccForPromotionCap = SetLicmMssaNoAccForPromotionCap; 289 CallGraphProfile = true; 290 MergeFunctions = false; 291 } 292 293 namespace llvm { 294 cl::opt<bool> PrintPipelinePasses( 295 "print-pipeline-passes", 296 cl::desc("Print a '-passes' compatible string describing the pipeline " 297 "(best-effort only).")); 298 299 extern cl::opt<unsigned> MaxDevirtIterations; 300 extern cl::opt<bool> EnableConstraintElimination; 301 extern cl::opt<bool> EnableFunctionSpecialization; 302 extern cl::opt<bool> EnableGVNHoist; 303 extern cl::opt<bool> EnableGVNSink; 304 extern cl::opt<bool> EnableHotColdSplit; 305 extern cl::opt<bool> EnableIROutliner; 306 extern cl::opt<bool> EnableOrderFileInstrumentation; 307 extern cl::opt<bool> EnableCHR; 308 extern cl::opt<bool> EnableLoopInterchange; 309 extern cl::opt<bool> EnableUnrollAndJam; 310 extern cl::opt<bool> EnableLoopFlatten; 311 extern cl::opt<bool> EnableDFAJumpThreading; 312 extern cl::opt<bool> RunNewGVN; 313 extern cl::opt<bool> RunPartialInlining; 314 extern cl::opt<bool> ExtraVectorizerPasses; 315 316 extern cl::opt<bool> FlattenedProfileUsed; 317 318 extern cl::opt<AttributorRunOption> AttributorRun; 319 extern cl::opt<bool> EnableKnowledgeRetention; 320 321 extern cl::opt<bool> EnableMatrix; 322 323 extern cl::opt<bool> DisablePreInliner; 324 extern cl::opt<int> PreInlineThreshold; 325 } // namespace llvm 326 327 const OptimizationLevel OptimizationLevel::O0 = { 328 /*SpeedLevel*/ 0, 329 /*SizeLevel*/ 0}; 330 const OptimizationLevel OptimizationLevel::O1 = { 331 /*SpeedLevel*/ 1, 332 /*SizeLevel*/ 0}; 333 const OptimizationLevel OptimizationLevel::O2 = { 334 /*SpeedLevel*/ 2, 335 /*SizeLevel*/ 0}; 336 const OptimizationLevel OptimizationLevel::O3 = { 337 /*SpeedLevel*/ 3, 338 /*SizeLevel*/ 0}; 339 const OptimizationLevel OptimizationLevel::Os = { 340 /*SpeedLevel*/ 2, 341 /*SizeLevel*/ 1}; 342 const OptimizationLevel OptimizationLevel::Oz = { 343 /*SpeedLevel*/ 2, 344 /*SizeLevel*/ 2}; 345 346 namespace { 347 348 // The following passes/analyses have custom names, otherwise their name will 349 // include `(anonymous namespace)`. These are special since they are only for 350 // testing purposes and don't live in a header file. 351 352 /// No-op module pass which does nothing. 353 struct NoOpModulePass : PassInfoMixin<NoOpModulePass> { 354 PreservedAnalyses run(Module &M, ModuleAnalysisManager &) { 355 return PreservedAnalyses::all(); 356 } 357 358 static StringRef name() { return "NoOpModulePass"; } 359 }; 360 361 /// No-op module analysis. 362 class NoOpModuleAnalysis : public AnalysisInfoMixin<NoOpModuleAnalysis> { 363 friend AnalysisInfoMixin<NoOpModuleAnalysis>; 364 static AnalysisKey Key; 365 366 public: 367 struct Result {}; 368 Result run(Module &, ModuleAnalysisManager &) { return Result(); } 369 static StringRef name() { return "NoOpModuleAnalysis"; } 370 }; 371 372 /// No-op CGSCC pass which does nothing. 373 struct NoOpCGSCCPass : PassInfoMixin<NoOpCGSCCPass> { 374 PreservedAnalyses run(LazyCallGraph::SCC &C, CGSCCAnalysisManager &, 375 LazyCallGraph &, CGSCCUpdateResult &UR) { 376 return PreservedAnalyses::all(); 377 } 378 static StringRef name() { return "NoOpCGSCCPass"; } 379 }; 380 381 /// No-op CGSCC analysis. 382 class NoOpCGSCCAnalysis : public AnalysisInfoMixin<NoOpCGSCCAnalysis> { 383 friend AnalysisInfoMixin<NoOpCGSCCAnalysis>; 384 static AnalysisKey Key; 385 386 public: 387 struct Result {}; 388 Result run(LazyCallGraph::SCC &, CGSCCAnalysisManager &, LazyCallGraph &G) { 389 return Result(); 390 } 391 static StringRef name() { return "NoOpCGSCCAnalysis"; } 392 }; 393 394 /// No-op function pass which does nothing. 395 struct NoOpFunctionPass : PassInfoMixin<NoOpFunctionPass> { 396 PreservedAnalyses run(Function &F, FunctionAnalysisManager &) { 397 return PreservedAnalyses::all(); 398 } 399 static StringRef name() { return "NoOpFunctionPass"; } 400 }; 401 402 /// No-op function analysis. 403 class NoOpFunctionAnalysis : public AnalysisInfoMixin<NoOpFunctionAnalysis> { 404 friend AnalysisInfoMixin<NoOpFunctionAnalysis>; 405 static AnalysisKey Key; 406 407 public: 408 struct Result {}; 409 Result run(Function &, FunctionAnalysisManager &) { return Result(); } 410 static StringRef name() { return "NoOpFunctionAnalysis"; } 411 }; 412 413 /// No-op loop pass which does nothing. 414 struct NoOpLoopPass : PassInfoMixin<NoOpLoopPass> { 415 PreservedAnalyses run(Loop &L, LoopAnalysisManager &, 416 LoopStandardAnalysisResults &, LPMUpdater &) { 417 return PreservedAnalyses::all(); 418 } 419 static StringRef name() { return "NoOpLoopPass"; } 420 }; 421 422 /// No-op loop analysis. 423 class NoOpLoopAnalysis : public AnalysisInfoMixin<NoOpLoopAnalysis> { 424 friend AnalysisInfoMixin<NoOpLoopAnalysis>; 425 static AnalysisKey Key; 426 427 public: 428 struct Result {}; 429 Result run(Loop &, LoopAnalysisManager &, LoopStandardAnalysisResults &) { 430 return Result(); 431 } 432 static StringRef name() { return "NoOpLoopAnalysis"; } 433 }; 434 435 AnalysisKey NoOpModuleAnalysis::Key; 436 AnalysisKey NoOpCGSCCAnalysis::Key; 437 AnalysisKey NoOpFunctionAnalysis::Key; 438 AnalysisKey NoOpLoopAnalysis::Key; 439 440 /// Whether or not we should populate a PassInstrumentationCallbacks's class to 441 /// pass name map. 442 /// 443 /// This is for optimization purposes so we don't populate it if we never use 444 /// it. This should be updated if new pass instrumentation wants to use the map. 445 /// We currently only use this for --print-before/after. 446 bool shouldPopulateClassToPassNames() { 447 return PrintPipelinePasses || !printBeforePasses().empty() || 448 !printAfterPasses().empty(); 449 } 450 451 } // namespace 452 453 PassBuilder::PassBuilder(TargetMachine *TM, PipelineTuningOptions PTO, 454 Optional<PGOOptions> PGOOpt, 455 PassInstrumentationCallbacks *PIC) 456 : TM(TM), PTO(PTO), PGOOpt(PGOOpt), PIC(PIC) { 457 if (TM) 458 TM->registerPassBuilderCallbacks(*this); 459 if (PIC && shouldPopulateClassToPassNames()) { 460 #define MODULE_PASS(NAME, CREATE_PASS) \ 461 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 462 #define MODULE_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 463 PIC->addClassToPassName(CLASS, NAME); 464 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 465 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 466 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 467 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 468 #define FUNCTION_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 469 PIC->addClassToPassName(CLASS, NAME); 470 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 471 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 472 #define LOOP_PASS(NAME, CREATE_PASS) \ 473 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 474 #define LOOP_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 475 PIC->addClassToPassName(CLASS, NAME); 476 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 477 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 478 #define CGSCC_PASS(NAME, CREATE_PASS) \ 479 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 480 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 481 PIC->addClassToPassName(decltype(CREATE_PASS)::name(), NAME); 482 #include "PassRegistry.def" 483 } 484 } 485 486 void PassBuilder::invokePeepholeEPCallbacks(FunctionPassManager &FPM, 487 OptimizationLevel Level) { 488 for (auto &C : PeepholeEPCallbacks) 489 C(FPM, Level); 490 } 491 492 void PassBuilder::registerModuleAnalyses(ModuleAnalysisManager &MAM) { 493 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 494 MAM.registerPass([&] { return CREATE_PASS; }); 495 #include "PassRegistry.def" 496 497 for (auto &C : ModuleAnalysisRegistrationCallbacks) 498 C(MAM); 499 } 500 501 void PassBuilder::registerCGSCCAnalyses(CGSCCAnalysisManager &CGAM) { 502 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 503 CGAM.registerPass([&] { return CREATE_PASS; }); 504 #include "PassRegistry.def" 505 506 for (auto &C : CGSCCAnalysisRegistrationCallbacks) 507 C(CGAM); 508 } 509 510 void PassBuilder::registerFunctionAnalyses(FunctionAnalysisManager &FAM) { 511 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 512 FAM.registerPass([&] { return CREATE_PASS; }); 513 #include "PassRegistry.def" 514 515 for (auto &C : FunctionAnalysisRegistrationCallbacks) 516 C(FAM); 517 } 518 519 void PassBuilder::registerLoopAnalyses(LoopAnalysisManager &LAM) { 520 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 521 LAM.registerPass([&] { return CREATE_PASS; }); 522 #include "PassRegistry.def" 523 524 for (auto &C : LoopAnalysisRegistrationCallbacks) 525 C(LAM); 526 } 527 528 // Helper to add AnnotationRemarksPass. 529 static void addAnnotationRemarksPass(ModulePassManager &MPM) { 530 FunctionPassManager FPM; 531 FPM.addPass(AnnotationRemarksPass()); 532 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 533 } 534 535 // Helper to check if the current compilation phase is preparing for LTO 536 static bool isLTOPreLink(ThinOrFullLTOPhase Phase) { 537 return Phase == ThinOrFullLTOPhase::ThinLTOPreLink || 538 Phase == ThinOrFullLTOPhase::FullLTOPreLink; 539 } 540 541 // TODO: Investigate the cost/benefit of tail call elimination on debugging. 542 FunctionPassManager 543 PassBuilder::buildO1FunctionSimplificationPipeline(OptimizationLevel Level, 544 ThinOrFullLTOPhase Phase) { 545 546 FunctionPassManager FPM; 547 548 // Form SSA out of local memory accesses after breaking apart aggregates into 549 // scalars. 550 FPM.addPass(SROA()); 551 552 // Catch trivial redundancies 553 FPM.addPass(EarlyCSEPass(true /* Enable mem-ssa. */)); 554 555 // Hoisting of scalars and load expressions. 556 FPM.addPass(SimplifyCFGPass()); 557 FPM.addPass(InstCombinePass()); 558 559 FPM.addPass(LibCallsShrinkWrapPass()); 560 561 invokePeepholeEPCallbacks(FPM, Level); 562 563 FPM.addPass(SimplifyCFGPass()); 564 565 // Form canonically associated expression trees, and simplify the trees using 566 // basic mathematical properties. For example, this will form (nearly) 567 // minimal multiplication trees. 568 FPM.addPass(ReassociatePass()); 569 570 // Add the primary loop simplification pipeline. 571 // FIXME: Currently this is split into two loop pass pipelines because we run 572 // some function passes in between them. These can and should be removed 573 // and/or replaced by scheduling the loop pass equivalents in the correct 574 // positions. But those equivalent passes aren't powerful enough yet. 575 // Specifically, `SimplifyCFGPass` and `InstCombinePass` are currently still 576 // used. We have `LoopSimplifyCFGPass` which isn't yet powerful enough yet to 577 // fully replace `SimplifyCFGPass`, and the closest to the other we have is 578 // `LoopInstSimplify`. 579 LoopPassManager LPM1, LPM2; 580 581 // Simplify the loop body. We do this initially to clean up after other loop 582 // passes run, either when iterating on a loop or on inner loops with 583 // implications on the outer loop. 584 LPM1.addPass(LoopInstSimplifyPass()); 585 LPM1.addPass(LoopSimplifyCFGPass()); 586 587 // Try to remove as much code from the loop header as possible, 588 // to reduce amount of IR that will have to be duplicated. 589 // TODO: Investigate promotion cap for O1. 590 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 591 592 LPM1.addPass(LoopRotatePass(/* Disable header duplication */ true, 593 isLTOPreLink(Phase))); 594 // TODO: Investigate promotion cap for O1. 595 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 596 LPM1.addPass(SimpleLoopUnswitchPass()); 597 598 LPM2.addPass(LoopIdiomRecognizePass()); 599 LPM2.addPass(IndVarSimplifyPass()); 600 601 for (auto &C : LateLoopOptimizationsEPCallbacks) 602 C(LPM2, Level); 603 604 LPM2.addPass(LoopDeletionPass()); 605 606 if (EnableLoopInterchange) 607 LPM2.addPass(LoopInterchangePass()); 608 609 // Do not enable unrolling in PreLinkThinLTO phase during sample PGO 610 // because it changes IR to makes profile annotation in back compile 611 // inaccurate. The normal unroller doesn't pay attention to forced full unroll 612 // attributes so we need to make sure and allow the full unroll pass to pay 613 // attention to it. 614 if (Phase != ThinOrFullLTOPhase::ThinLTOPreLink || !PGOOpt || 615 PGOOpt->Action != PGOOptions::SampleUse) 616 LPM2.addPass(LoopFullUnrollPass(Level.getSpeedupLevel(), 617 /* OnlyWhenForced= */ !PTO.LoopUnrolling, 618 PTO.ForgetAllSCEVInLoopUnroll)); 619 620 for (auto &C : LoopOptimizerEndEPCallbacks) 621 C(LPM2, Level); 622 623 // We provide the opt remark emitter pass for LICM to use. We only need to do 624 // this once as it is immutable. 625 FPM.addPass( 626 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 627 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM1), 628 /*UseMemorySSA=*/true, 629 /*UseBlockFrequencyInfo=*/true)); 630 FPM.addPass(SimplifyCFGPass()); 631 FPM.addPass(InstCombinePass()); 632 if (EnableLoopFlatten) 633 FPM.addPass(createFunctionToLoopPassAdaptor(LoopFlattenPass())); 634 // The loop passes in LPM2 (LoopFullUnrollPass) do not preserve MemorySSA. 635 // *All* loop passes must preserve it, in order to be able to use it. 636 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM2), 637 /*UseMemorySSA=*/false, 638 /*UseBlockFrequencyInfo=*/false)); 639 640 // Delete small array after loop unroll. 641 FPM.addPass(SROA()); 642 643 // Specially optimize memory movement as it doesn't look like dataflow in SSA. 644 FPM.addPass(MemCpyOptPass()); 645 646 // Sparse conditional constant propagation. 647 // FIXME: It isn't clear why we do this *after* loop passes rather than 648 // before... 649 FPM.addPass(SCCPPass()); 650 651 // Delete dead bit computations (instcombine runs after to fold away the dead 652 // computations, and then ADCE will run later to exploit any new DCE 653 // opportunities that creates). 654 FPM.addPass(BDCEPass()); 655 656 // Run instcombine after redundancy and dead bit elimination to exploit 657 // opportunities opened up by them. 658 FPM.addPass(InstCombinePass()); 659 invokePeepholeEPCallbacks(FPM, Level); 660 661 FPM.addPass(CoroElidePass()); 662 663 for (auto &C : ScalarOptimizerLateEPCallbacks) 664 C(FPM, Level); 665 666 // Finally, do an expensive DCE pass to catch all the dead code exposed by 667 // the simplifications and basic cleanup after all the simplifications. 668 // TODO: Investigate if this is too expensive. 669 FPM.addPass(ADCEPass()); 670 FPM.addPass(SimplifyCFGPass()); 671 FPM.addPass(InstCombinePass()); 672 invokePeepholeEPCallbacks(FPM, Level); 673 674 return FPM; 675 } 676 677 FunctionPassManager 678 PassBuilder::buildFunctionSimplificationPipeline(OptimizationLevel Level, 679 ThinOrFullLTOPhase Phase) { 680 assert(Level != OptimizationLevel::O0 && "Must request optimizations!"); 681 682 // The O1 pipeline has a separate pipeline creation function to simplify 683 // construction readability. 684 if (Level.getSpeedupLevel() == 1) 685 return buildO1FunctionSimplificationPipeline(Level, Phase); 686 687 FunctionPassManager FPM; 688 689 // Form SSA out of local memory accesses after breaking apart aggregates into 690 // scalars. 691 FPM.addPass(SROA()); 692 693 // Catch trivial redundancies 694 FPM.addPass(EarlyCSEPass(true /* Enable mem-ssa. */)); 695 if (EnableKnowledgeRetention) 696 FPM.addPass(AssumeSimplifyPass()); 697 698 // Hoisting of scalars and load expressions. 699 if (EnableGVNHoist) 700 FPM.addPass(GVNHoistPass()); 701 702 // Global value numbering based sinking. 703 if (EnableGVNSink) { 704 FPM.addPass(GVNSinkPass()); 705 FPM.addPass(SimplifyCFGPass()); 706 } 707 708 if (EnableConstraintElimination) 709 FPM.addPass(ConstraintEliminationPass()); 710 711 // Speculative execution if the target has divergent branches; otherwise nop. 712 FPM.addPass(SpeculativeExecutionPass(/* OnlyIfDivergentTarget =*/true)); 713 714 // Optimize based on known information about branches, and cleanup afterward. 715 FPM.addPass(JumpThreadingPass()); 716 FPM.addPass(CorrelatedValuePropagationPass()); 717 718 FPM.addPass(SimplifyCFGPass()); 719 if (Level == OptimizationLevel::O3) 720 FPM.addPass(AggressiveInstCombinePass()); 721 FPM.addPass(InstCombinePass()); 722 723 if (!Level.isOptimizingForSize()) 724 FPM.addPass(LibCallsShrinkWrapPass()); 725 726 invokePeepholeEPCallbacks(FPM, Level); 727 728 // For PGO use pipeline, try to optimize memory intrinsics such as memcpy 729 // using the size value profile. Don't perform this when optimizing for size. 730 if (PGOOpt && PGOOpt->Action == PGOOptions::IRUse && 731 !Level.isOptimizingForSize()) 732 FPM.addPass(PGOMemOPSizeOpt()); 733 734 FPM.addPass(TailCallElimPass()); 735 FPM.addPass(SimplifyCFGPass()); 736 737 // Form canonically associated expression trees, and simplify the trees using 738 // basic mathematical properties. For example, this will form (nearly) 739 // minimal multiplication trees. 740 FPM.addPass(ReassociatePass()); 741 742 // Add the primary loop simplification pipeline. 743 // FIXME: Currently this is split into two loop pass pipelines because we run 744 // some function passes in between them. These can and should be removed 745 // and/or replaced by scheduling the loop pass equivalents in the correct 746 // positions. But those equivalent passes aren't powerful enough yet. 747 // Specifically, `SimplifyCFGPass` and `InstCombinePass` are currently still 748 // used. We have `LoopSimplifyCFGPass` which isn't yet powerful enough yet to 749 // fully replace `SimplifyCFGPass`, and the closest to the other we have is 750 // `LoopInstSimplify`. 751 LoopPassManager LPM1, LPM2; 752 753 // Simplify the loop body. We do this initially to clean up after other loop 754 // passes run, either when iterating on a loop or on inner loops with 755 // implications on the outer loop. 756 LPM1.addPass(LoopInstSimplifyPass()); 757 LPM1.addPass(LoopSimplifyCFGPass()); 758 759 // Try to remove as much code from the loop header as possible, 760 // to reduce amount of IR that will have to be duplicated. 761 // TODO: Investigate promotion cap for O1. 762 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 763 764 // Disable header duplication in loop rotation at -Oz. 765 LPM1.addPass( 766 LoopRotatePass(Level != OptimizationLevel::Oz, isLTOPreLink(Phase))); 767 // TODO: Investigate promotion cap for O1. 768 LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 769 LPM1.addPass( 770 SimpleLoopUnswitchPass(/* NonTrivial */ Level == OptimizationLevel::O3 && 771 EnableO3NonTrivialUnswitching)); 772 LPM2.addPass(LoopIdiomRecognizePass()); 773 LPM2.addPass(IndVarSimplifyPass()); 774 775 for (auto &C : LateLoopOptimizationsEPCallbacks) 776 C(LPM2, Level); 777 778 LPM2.addPass(LoopDeletionPass()); 779 780 if (EnableLoopInterchange) 781 LPM2.addPass(LoopInterchangePass()); 782 783 // Do not enable unrolling in PreLinkThinLTO phase during sample PGO 784 // because it changes IR to makes profile annotation in back compile 785 // inaccurate. The normal unroller doesn't pay attention to forced full unroll 786 // attributes so we need to make sure and allow the full unroll pass to pay 787 // attention to it. 788 if (Phase != ThinOrFullLTOPhase::ThinLTOPreLink || !PGOOpt || 789 PGOOpt->Action != PGOOptions::SampleUse) 790 LPM2.addPass(LoopFullUnrollPass(Level.getSpeedupLevel(), 791 /* OnlyWhenForced= */ !PTO.LoopUnrolling, 792 PTO.ForgetAllSCEVInLoopUnroll)); 793 794 for (auto &C : LoopOptimizerEndEPCallbacks) 795 C(LPM2, Level); 796 797 // We provide the opt remark emitter pass for LICM to use. We only need to do 798 // this once as it is immutable. 799 FPM.addPass( 800 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 801 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM1), 802 /*UseMemorySSA=*/true, 803 /*UseBlockFrequencyInfo=*/true)); 804 FPM.addPass(SimplifyCFGPass()); 805 FPM.addPass(InstCombinePass()); 806 if (EnableLoopFlatten) 807 FPM.addPass(createFunctionToLoopPassAdaptor(LoopFlattenPass())); 808 // The loop passes in LPM2 (LoopIdiomRecognizePass, IndVarSimplifyPass, 809 // LoopDeletionPass and LoopFullUnrollPass) do not preserve MemorySSA. 810 // *All* loop passes must preserve it, in order to be able to use it. 811 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM2), 812 /*UseMemorySSA=*/false, 813 /*UseBlockFrequencyInfo=*/false)); 814 815 // Delete small array after loop unroll. 816 FPM.addPass(SROA()); 817 818 // Eliminate redundancies. 819 FPM.addPass(MergedLoadStoreMotionPass()); 820 if (RunNewGVN) 821 FPM.addPass(NewGVNPass()); 822 else 823 FPM.addPass(GVN()); 824 825 // Sparse conditional constant propagation. 826 // FIXME: It isn't clear why we do this *after* loop passes rather than 827 // before... 828 FPM.addPass(SCCPPass()); 829 830 // Delete dead bit computations (instcombine runs after to fold away the dead 831 // computations, and then ADCE will run later to exploit any new DCE 832 // opportunities that creates). 833 FPM.addPass(BDCEPass()); 834 835 // Run instcombine after redundancy and dead bit elimination to exploit 836 // opportunities opened up by them. 837 FPM.addPass(InstCombinePass()); 838 invokePeepholeEPCallbacks(FPM, Level); 839 840 // Re-consider control flow based optimizations after redundancy elimination, 841 // redo DCE, etc. 842 if (EnableDFAJumpThreading && Level.getSizeLevel() == 0) 843 FPM.addPass(DFAJumpThreadingPass()); 844 845 FPM.addPass(JumpThreadingPass()); 846 FPM.addPass(CorrelatedValuePropagationPass()); 847 848 // Finally, do an expensive DCE pass to catch all the dead code exposed by 849 // the simplifications and basic cleanup after all the simplifications. 850 // TODO: Investigate if this is too expensive. 851 FPM.addPass(ADCEPass()); 852 853 // Specially optimize memory movement as it doesn't look like dataflow in SSA. 854 FPM.addPass(MemCpyOptPass()); 855 856 FPM.addPass(DSEPass()); 857 FPM.addPass(createFunctionToLoopPassAdaptor( 858 LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap), 859 /*UseMemorySSA=*/true, /*UseBlockFrequencyInfo=*/true)); 860 861 FPM.addPass(CoroElidePass()); 862 863 for (auto &C : ScalarOptimizerLateEPCallbacks) 864 C(FPM, Level); 865 866 FPM.addPass(SimplifyCFGPass( 867 SimplifyCFGOptions().hoistCommonInsts(true).sinkCommonInsts(true))); 868 FPM.addPass(InstCombinePass()); 869 invokePeepholeEPCallbacks(FPM, Level); 870 871 if (EnableCHR && Level == OptimizationLevel::O3 && PGOOpt && 872 (PGOOpt->Action == PGOOptions::IRUse || 873 PGOOpt->Action == PGOOptions::SampleUse)) 874 FPM.addPass(ControlHeightReductionPass()); 875 876 return FPM; 877 } 878 879 void PassBuilder::addRequiredLTOPreLinkPasses(ModulePassManager &MPM) { 880 MPM.addPass(CanonicalizeAliasesPass()); 881 MPM.addPass(NameAnonGlobalPass()); 882 } 883 884 void PassBuilder::addPGOInstrPasses(ModulePassManager &MPM, 885 OptimizationLevel Level, bool RunProfileGen, 886 bool IsCS, std::string ProfileFile, 887 std::string ProfileRemappingFile) { 888 assert(Level != OptimizationLevel::O0 && "Not expecting O0 here!"); 889 if (!IsCS && !DisablePreInliner) { 890 InlineParams IP; 891 892 IP.DefaultThreshold = PreInlineThreshold; 893 894 // FIXME: The hint threshold has the same value used by the regular inliner 895 // when not optimzing for size. This should probably be lowered after 896 // performance testing. 897 // FIXME: this comment is cargo culted from the old pass manager, revisit). 898 IP.HintThreshold = Level.isOptimizingForSize() ? PreInlineThreshold : 325; 899 ModuleInlinerWrapperPass MIWP(IP); 900 CGSCCPassManager &CGPipeline = MIWP.getPM(); 901 902 FunctionPassManager FPM; 903 FPM.addPass(SROA()); 904 FPM.addPass(EarlyCSEPass()); // Catch trivial redundancies. 905 FPM.addPass(SimplifyCFGPass()); // Merge & remove basic blocks. 906 FPM.addPass(InstCombinePass()); // Combine silly sequences. 907 invokePeepholeEPCallbacks(FPM, Level); 908 909 CGPipeline.addPass(createCGSCCToFunctionPassAdaptor(std::move(FPM))); 910 911 MPM.addPass(std::move(MIWP)); 912 913 // Delete anything that is now dead to make sure that we don't instrument 914 // dead code. Instrumentation can end up keeping dead code around and 915 // dramatically increase code size. 916 MPM.addPass(GlobalDCEPass()); 917 } 918 919 if (!RunProfileGen) { 920 assert(!ProfileFile.empty() && "Profile use expecting a profile file!"); 921 MPM.addPass(PGOInstrumentationUse(ProfileFile, ProfileRemappingFile, IsCS)); 922 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 923 // RequireAnalysisPass for PSI before subsequent non-module passes. 924 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 925 return; 926 } 927 928 // Perform PGO instrumentation. 929 MPM.addPass(PGOInstrumentationGen(IsCS)); 930 931 FunctionPassManager FPM; 932 // Disable header duplication in loop rotation at -Oz. 933 FPM.addPass(createFunctionToLoopPassAdaptor( 934 LoopRotatePass(Level != OptimizationLevel::Oz), /*UseMemorySSA=*/false, 935 /*UseBlockFrequencyInfo=*/false)); 936 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 937 938 // Add the profile lowering pass. 939 InstrProfOptions Options; 940 if (!ProfileFile.empty()) 941 Options.InstrProfileOutput = ProfileFile; 942 // Do counter promotion at Level greater than O0. 943 Options.DoCounterPromotion = true; 944 Options.UseBFIInPromotion = IsCS; 945 MPM.addPass(InstrProfiling(Options, IsCS)); 946 } 947 948 void PassBuilder::addPGOInstrPassesForO0(ModulePassManager &MPM, 949 bool RunProfileGen, bool IsCS, 950 std::string ProfileFile, 951 std::string ProfileRemappingFile) { 952 if (!RunProfileGen) { 953 assert(!ProfileFile.empty() && "Profile use expecting a profile file!"); 954 MPM.addPass(PGOInstrumentationUse(ProfileFile, ProfileRemappingFile, IsCS)); 955 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 956 // RequireAnalysisPass for PSI before subsequent non-module passes. 957 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 958 return; 959 } 960 961 // Perform PGO instrumentation. 962 MPM.addPass(PGOInstrumentationGen(IsCS)); 963 // Add the profile lowering pass. 964 InstrProfOptions Options; 965 if (!ProfileFile.empty()) 966 Options.InstrProfileOutput = ProfileFile; 967 // Do not do counter promotion at O0. 968 Options.DoCounterPromotion = false; 969 Options.UseBFIInPromotion = IsCS; 970 MPM.addPass(InstrProfiling(Options, IsCS)); 971 } 972 973 static InlineParams getInlineParamsFromOptLevel(OptimizationLevel Level) { 974 return getInlineParams(Level.getSpeedupLevel(), Level.getSizeLevel()); 975 } 976 977 ModuleInlinerWrapperPass 978 PassBuilder::buildInlinerPipeline(OptimizationLevel Level, 979 ThinOrFullLTOPhase Phase) { 980 InlineParams IP = getInlineParamsFromOptLevel(Level); 981 if (Phase == ThinOrFullLTOPhase::ThinLTOPreLink && PGOOpt && 982 PGOOpt->Action == PGOOptions::SampleUse) 983 IP.HotCallSiteThreshold = 0; 984 985 if (PGOOpt) 986 IP.EnableDeferral = EnablePGOInlineDeferral; 987 988 ModuleInlinerWrapperPass MIWP(IP, PerformMandatoryInliningsFirst, 989 UseInlineAdvisor, MaxDevirtIterations); 990 991 // Require the GlobalsAA analysis for the module so we can query it within 992 // the CGSCC pipeline. 993 MIWP.addModulePass(RequireAnalysisPass<GlobalsAA, Module>()); 994 // Invalidate AAManager so it can be recreated and pick up the newly available 995 // GlobalsAA. 996 MIWP.addModulePass( 997 createModuleToFunctionPassAdaptor(InvalidateAnalysisPass<AAManager>())); 998 999 // Require the ProfileSummaryAnalysis for the module so we can query it within 1000 // the inliner pass. 1001 MIWP.addModulePass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 1002 1003 // Now begin the main postorder CGSCC pipeline. 1004 // FIXME: The current CGSCC pipeline has its origins in the legacy pass 1005 // manager and trying to emulate its precise behavior. Much of this doesn't 1006 // make a lot of sense and we should revisit the core CGSCC structure. 1007 CGSCCPassManager &MainCGPipeline = MIWP.getPM(); 1008 1009 // Note: historically, the PruneEH pass was run first to deduce nounwind and 1010 // generally clean up exception handling overhead. It isn't clear this is 1011 // valuable as the inliner doesn't currently care whether it is inlining an 1012 // invoke or a call. 1013 1014 if (AttributorRun & AttributorRunOption::CGSCC) 1015 MainCGPipeline.addPass(AttributorCGSCCPass()); 1016 1017 // Now deduce any function attributes based in the current code. 1018 MainCGPipeline.addPass(PostOrderFunctionAttrsPass()); 1019 1020 // When at O3 add argument promotion to the pass pipeline. 1021 // FIXME: It isn't at all clear why this should be limited to O3. 1022 if (Level == OptimizationLevel::O3) 1023 MainCGPipeline.addPass(ArgumentPromotionPass()); 1024 1025 // Try to perform OpenMP specific optimizations. This is a (quick!) no-op if 1026 // there are no OpenMP runtime calls present in the module. 1027 if (Level == OptimizationLevel::O2 || Level == OptimizationLevel::O3) 1028 MainCGPipeline.addPass(OpenMPOptCGSCCPass()); 1029 1030 for (auto &C : CGSCCOptimizerLateEPCallbacks) 1031 C(MainCGPipeline, Level); 1032 1033 // Lastly, add the core function simplification pipeline nested inside the 1034 // CGSCC walk. 1035 MainCGPipeline.addPass(createCGSCCToFunctionPassAdaptor( 1036 buildFunctionSimplificationPipeline(Level, Phase))); 1037 1038 MainCGPipeline.addPass(CoroSplitPass(Level != OptimizationLevel::O0)); 1039 1040 return MIWP; 1041 } 1042 1043 ModulePassManager 1044 PassBuilder::buildModuleSimplificationPipeline(OptimizationLevel Level, 1045 ThinOrFullLTOPhase Phase) { 1046 ModulePassManager MPM; 1047 1048 // Place pseudo probe instrumentation as the first pass of the pipeline to 1049 // minimize the impact of optimization changes. 1050 if (PGOOpt && PGOOpt->PseudoProbeForProfiling && 1051 Phase != ThinOrFullLTOPhase::ThinLTOPostLink) 1052 MPM.addPass(SampleProfileProbePass(TM)); 1053 1054 bool HasSampleProfile = PGOOpt && (PGOOpt->Action == PGOOptions::SampleUse); 1055 1056 // In ThinLTO mode, when flattened profile is used, all the available 1057 // profile information will be annotated in PreLink phase so there is 1058 // no need to load the profile again in PostLink. 1059 bool LoadSampleProfile = 1060 HasSampleProfile && 1061 !(FlattenedProfileUsed && Phase == ThinOrFullLTOPhase::ThinLTOPostLink); 1062 1063 // During the ThinLTO backend phase we perform early indirect call promotion 1064 // here, before globalopt. Otherwise imported available_externally functions 1065 // look unreferenced and are removed. If we are going to load the sample 1066 // profile then defer until later. 1067 // TODO: See if we can move later and consolidate with the location where 1068 // we perform ICP when we are loading a sample profile. 1069 // TODO: We pass HasSampleProfile (whether there was a sample profile file 1070 // passed to the compile) to the SamplePGO flag of ICP. This is used to 1071 // determine whether the new direct calls are annotated with prof metadata. 1072 // Ideally this should be determined from whether the IR is annotated with 1073 // sample profile, and not whether the a sample profile was provided on the 1074 // command line. E.g. for flattened profiles where we will not be reloading 1075 // the sample profile in the ThinLTO backend, we ideally shouldn't have to 1076 // provide the sample profile file. 1077 if (Phase == ThinOrFullLTOPhase::ThinLTOPostLink && !LoadSampleProfile) 1078 MPM.addPass(PGOIndirectCallPromotion(true /* InLTO */, HasSampleProfile)); 1079 1080 // Do basic inference of function attributes from known properties of system 1081 // libraries and other oracles. 1082 MPM.addPass(InferFunctionAttrsPass()); 1083 1084 // Create an early function pass manager to cleanup the output of the 1085 // frontend. 1086 FunctionPassManager EarlyFPM; 1087 // Lower llvm.expect to metadata before attempting transforms. 1088 // Compare/branch metadata may alter the behavior of passes like SimplifyCFG. 1089 EarlyFPM.addPass(LowerExpectIntrinsicPass()); 1090 EarlyFPM.addPass(SimplifyCFGPass()); 1091 EarlyFPM.addPass(SROA()); 1092 EarlyFPM.addPass(EarlyCSEPass()); 1093 EarlyFPM.addPass(CoroEarlyPass()); 1094 if (Level == OptimizationLevel::O3) 1095 EarlyFPM.addPass(CallSiteSplittingPass()); 1096 1097 // In SamplePGO ThinLTO backend, we need instcombine before profile annotation 1098 // to convert bitcast to direct calls so that they can be inlined during the 1099 // profile annotation prepration step. 1100 // More details about SamplePGO design can be found in: 1101 // https://research.google.com/pubs/pub45290.html 1102 // FIXME: revisit how SampleProfileLoad/Inliner/ICP is structured. 1103 if (LoadSampleProfile) 1104 EarlyFPM.addPass(InstCombinePass()); 1105 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(EarlyFPM))); 1106 1107 if (LoadSampleProfile) { 1108 // Annotate sample profile right after early FPM to ensure freshness of 1109 // the debug info. 1110 MPM.addPass(SampleProfileLoaderPass(PGOOpt->ProfileFile, 1111 PGOOpt->ProfileRemappingFile, Phase)); 1112 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 1113 // RequireAnalysisPass for PSI before subsequent non-module passes. 1114 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 1115 // Do not invoke ICP in the LTOPrelink phase as it makes it hard 1116 // for the profile annotation to be accurate in the LTO backend. 1117 if (Phase != ThinOrFullLTOPhase::ThinLTOPreLink && 1118 Phase != ThinOrFullLTOPhase::FullLTOPreLink) 1119 // We perform early indirect call promotion here, before globalopt. 1120 // This is important for the ThinLTO backend phase because otherwise 1121 // imported available_externally functions look unreferenced and are 1122 // removed. 1123 MPM.addPass( 1124 PGOIndirectCallPromotion(true /* IsInLTO */, true /* SamplePGO */)); 1125 } 1126 1127 // Try to perform OpenMP specific optimizations on the module. This is a 1128 // (quick!) no-op if there are no OpenMP runtime calls present in the module. 1129 if (Level != OptimizationLevel::O0) 1130 MPM.addPass(OpenMPOptPass()); 1131 1132 if (AttributorRun & AttributorRunOption::MODULE) 1133 MPM.addPass(AttributorPass()); 1134 1135 // Lower type metadata and the type.test intrinsic in the ThinLTO 1136 // post link pipeline after ICP. This is to enable usage of the type 1137 // tests in ICP sequences. 1138 if (Phase == ThinOrFullLTOPhase::ThinLTOPostLink) 1139 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1140 1141 for (auto &C : PipelineEarlySimplificationEPCallbacks) 1142 C(MPM, Level); 1143 1144 // Specialize functions with IPSCCP. 1145 if (EnableFunctionSpecialization) 1146 MPM.addPass(FunctionSpecializationPass()); 1147 1148 // Interprocedural constant propagation now that basic cleanup has occurred 1149 // and prior to optimizing globals. 1150 // FIXME: This position in the pipeline hasn't been carefully considered in 1151 // years, it should be re-analyzed. 1152 MPM.addPass(IPSCCPPass()); 1153 1154 // Attach metadata to indirect call sites indicating the set of functions 1155 // they may target at run-time. This should follow IPSCCP. 1156 MPM.addPass(CalledValuePropagationPass()); 1157 1158 // Optimize globals to try and fold them into constants. 1159 MPM.addPass(GlobalOptPass()); 1160 1161 // Promote any localized globals to SSA registers. 1162 // FIXME: Should this instead by a run of SROA? 1163 // FIXME: We should probably run instcombine and simplifycfg afterward to 1164 // delete control flows that are dead once globals have been folded to 1165 // constants. 1166 MPM.addPass(createModuleToFunctionPassAdaptor(PromotePass())); 1167 1168 // Remove any dead arguments exposed by cleanups and constant folding 1169 // globals. 1170 MPM.addPass(DeadArgumentEliminationPass()); 1171 1172 // Create a small function pass pipeline to cleanup after all the global 1173 // optimizations. 1174 FunctionPassManager GlobalCleanupPM; 1175 GlobalCleanupPM.addPass(InstCombinePass()); 1176 invokePeepholeEPCallbacks(GlobalCleanupPM, Level); 1177 1178 GlobalCleanupPM.addPass(SimplifyCFGPass()); 1179 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(GlobalCleanupPM))); 1180 1181 // Add all the requested passes for instrumentation PGO, if requested. 1182 if (PGOOpt && Phase != ThinOrFullLTOPhase::ThinLTOPostLink && 1183 (PGOOpt->Action == PGOOptions::IRInstr || 1184 PGOOpt->Action == PGOOptions::IRUse)) { 1185 addPGOInstrPasses(MPM, Level, 1186 /* RunProfileGen */ PGOOpt->Action == PGOOptions::IRInstr, 1187 /* IsCS */ false, PGOOpt->ProfileFile, 1188 PGOOpt->ProfileRemappingFile); 1189 MPM.addPass(PGOIndirectCallPromotion(false, false)); 1190 } 1191 if (PGOOpt && Phase != ThinOrFullLTOPhase::ThinLTOPostLink && 1192 PGOOpt->CSAction == PGOOptions::CSIRInstr) 1193 MPM.addPass(PGOInstrumentationGenCreateVar(PGOOpt->CSProfileGenFile)); 1194 1195 // Synthesize function entry counts for non-PGO compilation. 1196 if (EnableSyntheticCounts && !PGOOpt) 1197 MPM.addPass(SyntheticCountsPropagation()); 1198 1199 MPM.addPass(buildInlinerPipeline(Level, Phase)); 1200 1201 if (EnableMemProfiler && Phase != ThinOrFullLTOPhase::ThinLTOPreLink) { 1202 MPM.addPass(createModuleToFunctionPassAdaptor(MemProfilerPass())); 1203 MPM.addPass(ModuleMemProfilerPass()); 1204 } 1205 1206 return MPM; 1207 } 1208 1209 /// TODO: Should LTO cause any differences to this set of passes? 1210 void PassBuilder::addVectorPasses(OptimizationLevel Level, 1211 FunctionPassManager &FPM, bool IsFullLTO) { 1212 FPM.addPass(LoopVectorizePass( 1213 LoopVectorizeOptions(!PTO.LoopInterleaving, !PTO.LoopVectorization))); 1214 1215 if (IsFullLTO) { 1216 // The vectorizer may have significantly shortened a loop body; unroll 1217 // again. Unroll small loops to hide loop backedge latency and saturate any 1218 // parallel execution resources of an out-of-order processor. We also then 1219 // need to clean up redundancies and loop invariant code. 1220 // FIXME: It would be really good to use a loop-integrated instruction 1221 // combiner for cleanup here so that the unrolling and LICM can be pipelined 1222 // across the loop nests. 1223 // We do UnrollAndJam in a separate LPM to ensure it happens before unroll 1224 if (EnableUnrollAndJam && PTO.LoopUnrolling) 1225 FPM.addPass(createFunctionToLoopPassAdaptor( 1226 LoopUnrollAndJamPass(Level.getSpeedupLevel()))); 1227 FPM.addPass(LoopUnrollPass(LoopUnrollOptions( 1228 Level.getSpeedupLevel(), /*OnlyWhenForced=*/!PTO.LoopUnrolling, 1229 PTO.ForgetAllSCEVInLoopUnroll))); 1230 FPM.addPass(WarnMissedTransformationsPass()); 1231 } 1232 1233 if (!IsFullLTO) { 1234 // Eliminate loads by forwarding stores from the previous iteration to loads 1235 // of the current iteration. 1236 FPM.addPass(LoopLoadEliminationPass()); 1237 } 1238 // Cleanup after the loop optimization passes. 1239 FPM.addPass(InstCombinePass()); 1240 1241 if (Level.getSpeedupLevel() > 1 && ExtraVectorizerPasses) { 1242 // At higher optimization levels, try to clean up any runtime overlap and 1243 // alignment checks inserted by the vectorizer. We want to track correlated 1244 // runtime checks for two inner loops in the same outer loop, fold any 1245 // common computations, hoist loop-invariant aspects out of any outer loop, 1246 // and unswitch the runtime checks if possible. Once hoisted, we may have 1247 // dead (or speculatable) control flows or more combining opportunities. 1248 FPM.addPass(EarlyCSEPass()); 1249 FPM.addPass(CorrelatedValuePropagationPass()); 1250 FPM.addPass(InstCombinePass()); 1251 LoopPassManager LPM; 1252 LPM.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap)); 1253 LPM.addPass(SimpleLoopUnswitchPass(/* NonTrivial */ Level == 1254 OptimizationLevel::O3)); 1255 FPM.addPass( 1256 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 1257 FPM.addPass( 1258 createFunctionToLoopPassAdaptor(std::move(LPM), /*UseMemorySSA=*/true, 1259 /*UseBlockFrequencyInfo=*/true)); 1260 FPM.addPass(SimplifyCFGPass()); 1261 FPM.addPass(InstCombinePass()); 1262 } 1263 1264 // Now that we've formed fast to execute loop structures, we do further 1265 // optimizations. These are run afterward as they might block doing complex 1266 // analyses and transforms such as what are needed for loop vectorization. 1267 1268 // Cleanup after loop vectorization, etc. Simplification passes like CVP and 1269 // GVN, loop transforms, and others have already run, so it's now better to 1270 // convert to more optimized IR using more aggressive simplify CFG options. 1271 // The extra sinking transform can create larger basic blocks, so do this 1272 // before SLP vectorization. 1273 FPM.addPass(SimplifyCFGPass(SimplifyCFGOptions() 1274 .forwardSwitchCondToPhi(true) 1275 .convertSwitchToLookupTable(true) 1276 .needCanonicalLoops(false) 1277 .hoistCommonInsts(true) 1278 .sinkCommonInsts(true))); 1279 1280 if (IsFullLTO) { 1281 FPM.addPass(SCCPPass()); 1282 FPM.addPass(InstCombinePass()); 1283 FPM.addPass(BDCEPass()); 1284 } 1285 1286 // Optimize parallel scalar instruction chains into SIMD instructions. 1287 if (PTO.SLPVectorization) { 1288 FPM.addPass(SLPVectorizerPass()); 1289 if (Level.getSpeedupLevel() > 1 && ExtraVectorizerPasses) { 1290 FPM.addPass(EarlyCSEPass()); 1291 } 1292 } 1293 // Enhance/cleanup vector code. 1294 FPM.addPass(VectorCombinePass()); 1295 1296 if (!IsFullLTO) { 1297 FPM.addPass(InstCombinePass()); 1298 // Unroll small loops to hide loop backedge latency and saturate any 1299 // parallel execution resources of an out-of-order processor. We also then 1300 // need to clean up redundancies and loop invariant code. 1301 // FIXME: It would be really good to use a loop-integrated instruction 1302 // combiner for cleanup here so that the unrolling and LICM can be pipelined 1303 // across the loop nests. 1304 // We do UnrollAndJam in a separate LPM to ensure it happens before unroll 1305 if (EnableUnrollAndJam && PTO.LoopUnrolling) { 1306 FPM.addPass(createFunctionToLoopPassAdaptor( 1307 LoopUnrollAndJamPass(Level.getSpeedupLevel()))); 1308 } 1309 FPM.addPass(LoopUnrollPass(LoopUnrollOptions( 1310 Level.getSpeedupLevel(), /*OnlyWhenForced=*/!PTO.LoopUnrolling, 1311 PTO.ForgetAllSCEVInLoopUnroll))); 1312 FPM.addPass(WarnMissedTransformationsPass()); 1313 FPM.addPass(InstCombinePass()); 1314 FPM.addPass( 1315 RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>()); 1316 FPM.addPass(createFunctionToLoopPassAdaptor( 1317 LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap), 1318 /*UseMemorySSA=*/true, /*UseBlockFrequencyInfo=*/true)); 1319 } 1320 1321 // Now that we've vectorized and unrolled loops, we may have more refined 1322 // alignment information, try to re-derive it here. 1323 FPM.addPass(AlignmentFromAssumptionsPass()); 1324 1325 if (IsFullLTO) 1326 FPM.addPass(InstCombinePass()); 1327 } 1328 1329 ModulePassManager 1330 PassBuilder::buildModuleOptimizationPipeline(OptimizationLevel Level, 1331 bool LTOPreLink) { 1332 ModulePassManager MPM; 1333 1334 // Optimize globals now that the module is fully simplified. 1335 MPM.addPass(GlobalOptPass()); 1336 MPM.addPass(GlobalDCEPass()); 1337 1338 // Run partial inlining pass to partially inline functions that have 1339 // large bodies. 1340 if (RunPartialInlining) 1341 MPM.addPass(PartialInlinerPass()); 1342 1343 // Remove avail extern fns and globals definitions since we aren't compiling 1344 // an object file for later LTO. For LTO we want to preserve these so they 1345 // are eligible for inlining at link-time. Note if they are unreferenced they 1346 // will be removed by GlobalDCE later, so this only impacts referenced 1347 // available externally globals. Eventually they will be suppressed during 1348 // codegen, but eliminating here enables more opportunity for GlobalDCE as it 1349 // may make globals referenced by available external functions dead and saves 1350 // running remaining passes on the eliminated functions. These should be 1351 // preserved during prelinking for link-time inlining decisions. 1352 if (!LTOPreLink) 1353 MPM.addPass(EliminateAvailableExternallyPass()); 1354 1355 if (EnableOrderFileInstrumentation) 1356 MPM.addPass(InstrOrderFilePass()); 1357 1358 // Do RPO function attribute inference across the module to forward-propagate 1359 // attributes where applicable. 1360 // FIXME: Is this really an optimization rather than a canonicalization? 1361 MPM.addPass(ReversePostOrderFunctionAttrsPass()); 1362 1363 // Do a post inline PGO instrumentation and use pass. This is a context 1364 // sensitive PGO pass. We don't want to do this in LTOPreLink phrase as 1365 // cross-module inline has not been done yet. The context sensitive 1366 // instrumentation is after all the inlines are done. 1367 if (!LTOPreLink && PGOOpt) { 1368 if (PGOOpt->CSAction == PGOOptions::CSIRInstr) 1369 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ true, 1370 /* IsCS */ true, PGOOpt->CSProfileGenFile, 1371 PGOOpt->ProfileRemappingFile); 1372 else if (PGOOpt->CSAction == PGOOptions::CSIRUse) 1373 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ false, 1374 /* IsCS */ true, PGOOpt->ProfileFile, 1375 PGOOpt->ProfileRemappingFile); 1376 } 1377 1378 // Re-require GloblasAA here prior to function passes. This is particularly 1379 // useful as the above will have inlined, DCE'ed, and function-attr 1380 // propagated everything. We should at this point have a reasonably minimal 1381 // and richly annotated call graph. By computing aliasing and mod/ref 1382 // information for all local globals here, the late loop passes and notably 1383 // the vectorizer will be able to use them to help recognize vectorizable 1384 // memory operations. 1385 MPM.addPass(RequireAnalysisPass<GlobalsAA, Module>()); 1386 1387 FunctionPassManager OptimizePM; 1388 OptimizePM.addPass(Float2IntPass()); 1389 OptimizePM.addPass(LowerConstantIntrinsicsPass()); 1390 1391 if (EnableMatrix) { 1392 OptimizePM.addPass(LowerMatrixIntrinsicsPass()); 1393 OptimizePM.addPass(EarlyCSEPass()); 1394 } 1395 1396 // FIXME: We need to run some loop optimizations to re-rotate loops after 1397 // simplifycfg and others undo their rotation. 1398 1399 // Optimize the loop execution. These passes operate on entire loop nests 1400 // rather than on each loop in an inside-out manner, and so they are actually 1401 // function passes. 1402 1403 for (auto &C : VectorizerStartEPCallbacks) 1404 C(OptimizePM, Level); 1405 1406 // First rotate loops that may have been un-rotated by prior passes. 1407 // Disable header duplication at -Oz. 1408 OptimizePM.addPass(createFunctionToLoopPassAdaptor( 1409 LoopRotatePass(Level != OptimizationLevel::Oz, LTOPreLink), 1410 /*UseMemorySSA=*/false, /*UseBlockFrequencyInfo=*/false)); 1411 1412 // Distribute loops to allow partial vectorization. I.e. isolate dependences 1413 // into separate loop that would otherwise inhibit vectorization. This is 1414 // currently only performed for loops marked with the metadata 1415 // llvm.loop.distribute=true or when -enable-loop-distribute is specified. 1416 OptimizePM.addPass(LoopDistributePass()); 1417 1418 // Populates the VFABI attribute with the scalar-to-vector mappings 1419 // from the TargetLibraryInfo. 1420 OptimizePM.addPass(InjectTLIMappings()); 1421 1422 addVectorPasses(Level, OptimizePM, /* IsFullLTO */ false); 1423 1424 // Split out cold code. Splitting is done late to avoid hiding context from 1425 // other optimizations and inadvertently regressing performance. The tradeoff 1426 // is that this has a higher code size cost than splitting early. 1427 if (EnableHotColdSplit && !LTOPreLink) 1428 MPM.addPass(HotColdSplittingPass()); 1429 1430 // Search the code for similar regions of code. If enough similar regions can 1431 // be found where extracting the regions into their own function will decrease 1432 // the size of the program, we extract the regions, a deduplicate the 1433 // structurally similar regions. 1434 if (EnableIROutliner) 1435 MPM.addPass(IROutlinerPass()); 1436 1437 // Merge functions if requested. 1438 if (PTO.MergeFunctions) 1439 MPM.addPass(MergeFunctionsPass()); 1440 1441 // LoopSink pass sinks instructions hoisted by LICM, which serves as a 1442 // canonicalization pass that enables other optimizations. As a result, 1443 // LoopSink pass needs to be a very late IR pass to avoid undoing LICM 1444 // result too early. 1445 OptimizePM.addPass(LoopSinkPass()); 1446 1447 // And finally clean up LCSSA form before generating code. 1448 OptimizePM.addPass(InstSimplifyPass()); 1449 1450 // This hoists/decomposes div/rem ops. It should run after other sink/hoist 1451 // passes to avoid re-sinking, but before SimplifyCFG because it can allow 1452 // flattening of blocks. 1453 OptimizePM.addPass(DivRemPairsPass()); 1454 1455 // LoopSink (and other loop passes since the last simplifyCFG) might have 1456 // resulted in single-entry-single-exit or empty blocks. Clean up the CFG. 1457 OptimizePM.addPass(SimplifyCFGPass()); 1458 1459 OptimizePM.addPass(CoroCleanupPass()); 1460 1461 // Add the core optimizing pipeline. 1462 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(OptimizePM))); 1463 1464 for (auto &C : OptimizerLastEPCallbacks) 1465 C(MPM, Level); 1466 1467 if (PTO.CallGraphProfile) 1468 MPM.addPass(CGProfilePass()); 1469 1470 // Now we need to do some global optimization transforms. 1471 // FIXME: It would seem like these should come first in the optimization 1472 // pipeline and maybe be the bottom of the canonicalization pipeline? Weird 1473 // ordering here. 1474 MPM.addPass(GlobalDCEPass()); 1475 MPM.addPass(ConstantMergePass()); 1476 1477 // TODO: Relative look table converter pass caused an issue when full lto is 1478 // enabled. See https://reviews.llvm.org/D94355 for more details. 1479 // Until the issue fixed, disable this pass during pre-linking phase. 1480 if (!LTOPreLink) 1481 MPM.addPass(RelLookupTableConverterPass()); 1482 1483 return MPM; 1484 } 1485 1486 ModulePassManager 1487 PassBuilder::buildPerModuleDefaultPipeline(OptimizationLevel Level, 1488 bool LTOPreLink) { 1489 assert(Level != OptimizationLevel::O0 && 1490 "Must request optimizations for the default pipeline!"); 1491 1492 ModulePassManager MPM; 1493 1494 // Convert @llvm.global.annotations to !annotation metadata. 1495 MPM.addPass(Annotation2MetadataPass()); 1496 1497 // Force any function attributes we want the rest of the pipeline to observe. 1498 MPM.addPass(ForceFunctionAttrsPass()); 1499 1500 // Apply module pipeline start EP callback. 1501 for (auto &C : PipelineStartEPCallbacks) 1502 C(MPM, Level); 1503 1504 if (PGOOpt && PGOOpt->DebugInfoForProfiling) 1505 MPM.addPass(createModuleToFunctionPassAdaptor(AddDiscriminatorsPass())); 1506 1507 // Add the core simplification pipeline. 1508 MPM.addPass(buildModuleSimplificationPipeline( 1509 Level, LTOPreLink ? ThinOrFullLTOPhase::FullLTOPreLink 1510 : ThinOrFullLTOPhase::None)); 1511 1512 // Now add the optimization pipeline. 1513 MPM.addPass(buildModuleOptimizationPipeline(Level, LTOPreLink)); 1514 1515 if (PGOOpt && PGOOpt->PseudoProbeForProfiling) 1516 MPM.addPass(PseudoProbeUpdatePass()); 1517 1518 // Emit annotation remarks. 1519 addAnnotationRemarksPass(MPM); 1520 1521 if (LTOPreLink) 1522 addRequiredLTOPreLinkPasses(MPM); 1523 1524 return MPM; 1525 } 1526 1527 ModulePassManager 1528 PassBuilder::buildThinLTOPreLinkDefaultPipeline(OptimizationLevel Level) { 1529 assert(Level != OptimizationLevel::O0 && 1530 "Must request optimizations for the default pipeline!"); 1531 1532 ModulePassManager MPM; 1533 1534 // Convert @llvm.global.annotations to !annotation metadata. 1535 MPM.addPass(Annotation2MetadataPass()); 1536 1537 // Force any function attributes we want the rest of the pipeline to observe. 1538 MPM.addPass(ForceFunctionAttrsPass()); 1539 1540 if (PGOOpt && PGOOpt->DebugInfoForProfiling) 1541 MPM.addPass(createModuleToFunctionPassAdaptor(AddDiscriminatorsPass())); 1542 1543 // Apply module pipeline start EP callback. 1544 for (auto &C : PipelineStartEPCallbacks) 1545 C(MPM, Level); 1546 1547 // If we are planning to perform ThinLTO later, we don't bloat the code with 1548 // unrolling/vectorization/... now. Just simplify the module as much as we 1549 // can. 1550 MPM.addPass(buildModuleSimplificationPipeline( 1551 Level, ThinOrFullLTOPhase::ThinLTOPreLink)); 1552 1553 // Run partial inlining pass to partially inline functions that have 1554 // large bodies. 1555 // FIXME: It isn't clear whether this is really the right place to run this 1556 // in ThinLTO. Because there is another canonicalization and simplification 1557 // phase that will run after the thin link, running this here ends up with 1558 // less information than will be available later and it may grow functions in 1559 // ways that aren't beneficial. 1560 if (RunPartialInlining) 1561 MPM.addPass(PartialInlinerPass()); 1562 1563 // Reduce the size of the IR as much as possible. 1564 MPM.addPass(GlobalOptPass()); 1565 1566 // Module simplification splits coroutines, but does not fully clean up 1567 // coroutine intrinsics. To ensure ThinLTO optimization passes don't trip up 1568 // on these, we schedule the cleanup here. 1569 MPM.addPass(createModuleToFunctionPassAdaptor(CoroCleanupPass())); 1570 1571 if (PGOOpt && PGOOpt->PseudoProbeForProfiling) 1572 MPM.addPass(PseudoProbeUpdatePass()); 1573 1574 // Handle OptimizerLastEPCallbacks added by clang on PreLink. Actual 1575 // optimization is going to be done in PostLink stage, but clang can't 1576 // add callbacks there in case of in-process ThinLTO called by linker. 1577 for (auto &C : OptimizerLastEPCallbacks) 1578 C(MPM, Level); 1579 1580 // Emit annotation remarks. 1581 addAnnotationRemarksPass(MPM); 1582 1583 addRequiredLTOPreLinkPasses(MPM); 1584 1585 return MPM; 1586 } 1587 1588 ModulePassManager PassBuilder::buildThinLTODefaultPipeline( 1589 OptimizationLevel Level, const ModuleSummaryIndex *ImportSummary) { 1590 ModulePassManager MPM; 1591 1592 // Convert @llvm.global.annotations to !annotation metadata. 1593 MPM.addPass(Annotation2MetadataPass()); 1594 1595 if (ImportSummary) { 1596 // These passes import type identifier resolutions for whole-program 1597 // devirtualization and CFI. They must run early because other passes may 1598 // disturb the specific instruction patterns that these passes look for, 1599 // creating dependencies on resolutions that may not appear in the summary. 1600 // 1601 // For example, GVN may transform the pattern assume(type.test) appearing in 1602 // two basic blocks into assume(phi(type.test, type.test)), which would 1603 // transform a dependency on a WPD resolution into a dependency on a type 1604 // identifier resolution for CFI. 1605 // 1606 // Also, WPD has access to more precise information than ICP and can 1607 // devirtualize more effectively, so it should operate on the IR first. 1608 // 1609 // The WPD and LowerTypeTest passes need to run at -O0 to lower type 1610 // metadata and intrinsics. 1611 MPM.addPass(WholeProgramDevirtPass(nullptr, ImportSummary)); 1612 MPM.addPass(LowerTypeTestsPass(nullptr, ImportSummary)); 1613 } 1614 1615 if (Level == OptimizationLevel::O0) { 1616 // Run a second time to clean up any type tests left behind by WPD for use 1617 // in ICP. 1618 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1619 // Drop available_externally and unreferenced globals. This is necessary 1620 // with ThinLTO in order to avoid leaving undefined references to dead 1621 // globals in the object file. 1622 MPM.addPass(EliminateAvailableExternallyPass()); 1623 MPM.addPass(GlobalDCEPass()); 1624 return MPM; 1625 } 1626 1627 // Force any function attributes we want the rest of the pipeline to observe. 1628 MPM.addPass(ForceFunctionAttrsPass()); 1629 1630 // Add the core simplification pipeline. 1631 MPM.addPass(buildModuleSimplificationPipeline( 1632 Level, ThinOrFullLTOPhase::ThinLTOPostLink)); 1633 1634 // Now add the optimization pipeline. 1635 MPM.addPass(buildModuleOptimizationPipeline(Level)); 1636 1637 // Emit annotation remarks. 1638 addAnnotationRemarksPass(MPM); 1639 1640 return MPM; 1641 } 1642 1643 ModulePassManager 1644 PassBuilder::buildLTOPreLinkDefaultPipeline(OptimizationLevel Level) { 1645 assert(Level != OptimizationLevel::O0 && 1646 "Must request optimizations for the default pipeline!"); 1647 // FIXME: We should use a customized pre-link pipeline! 1648 return buildPerModuleDefaultPipeline(Level, 1649 /* LTOPreLink */ true); 1650 } 1651 1652 ModulePassManager 1653 PassBuilder::buildLTODefaultPipeline(OptimizationLevel Level, 1654 ModuleSummaryIndex *ExportSummary) { 1655 ModulePassManager MPM; 1656 1657 // Convert @llvm.global.annotations to !annotation metadata. 1658 MPM.addPass(Annotation2MetadataPass()); 1659 1660 // Create a function that performs CFI checks for cross-DSO calls with targets 1661 // in the current module. 1662 MPM.addPass(CrossDSOCFIPass()); 1663 1664 if (Level == OptimizationLevel::O0) { 1665 // The WPD and LowerTypeTest passes need to run at -O0 to lower type 1666 // metadata and intrinsics. 1667 MPM.addPass(WholeProgramDevirtPass(ExportSummary, nullptr)); 1668 MPM.addPass(LowerTypeTestsPass(ExportSummary, nullptr)); 1669 // Run a second time to clean up any type tests left behind by WPD for use 1670 // in ICP. 1671 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1672 1673 // Emit annotation remarks. 1674 addAnnotationRemarksPass(MPM); 1675 1676 return MPM; 1677 } 1678 1679 if (PGOOpt && PGOOpt->Action == PGOOptions::SampleUse) { 1680 // Load sample profile before running the LTO optimization pipeline. 1681 MPM.addPass(SampleProfileLoaderPass(PGOOpt->ProfileFile, 1682 PGOOpt->ProfileRemappingFile, 1683 ThinOrFullLTOPhase::FullLTOPostLink)); 1684 // Cache ProfileSummaryAnalysis once to avoid the potential need to insert 1685 // RequireAnalysisPass for PSI before subsequent non-module passes. 1686 MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>()); 1687 } 1688 1689 // Remove unused virtual tables to improve the quality of code generated by 1690 // whole-program devirtualization and bitset lowering. 1691 MPM.addPass(GlobalDCEPass()); 1692 1693 // Force any function attributes we want the rest of the pipeline to observe. 1694 MPM.addPass(ForceFunctionAttrsPass()); 1695 1696 // Do basic inference of function attributes from known properties of system 1697 // libraries and other oracles. 1698 MPM.addPass(InferFunctionAttrsPass()); 1699 1700 if (Level.getSpeedupLevel() > 1) { 1701 FunctionPassManager EarlyFPM; 1702 EarlyFPM.addPass(CallSiteSplittingPass()); 1703 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(EarlyFPM))); 1704 1705 // Indirect call promotion. This should promote all the targets that are 1706 // left by the earlier promotion pass that promotes intra-module targets. 1707 // This two-step promotion is to save the compile time. For LTO, it should 1708 // produce the same result as if we only do promotion here. 1709 MPM.addPass(PGOIndirectCallPromotion( 1710 true /* InLTO */, PGOOpt && PGOOpt->Action == PGOOptions::SampleUse)); 1711 1712 if (EnableFunctionSpecialization) 1713 MPM.addPass(FunctionSpecializationPass()); 1714 // Propagate constants at call sites into the functions they call. This 1715 // opens opportunities for globalopt (and inlining) by substituting function 1716 // pointers passed as arguments to direct uses of functions. 1717 MPM.addPass(IPSCCPPass()); 1718 1719 // Attach metadata to indirect call sites indicating the set of functions 1720 // they may target at run-time. This should follow IPSCCP. 1721 MPM.addPass(CalledValuePropagationPass()); 1722 } 1723 1724 // Now deduce any function attributes based in the current code. 1725 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor( 1726 PostOrderFunctionAttrsPass())); 1727 1728 // Do RPO function attribute inference across the module to forward-propagate 1729 // attributes where applicable. 1730 // FIXME: Is this really an optimization rather than a canonicalization? 1731 MPM.addPass(ReversePostOrderFunctionAttrsPass()); 1732 1733 // Use in-range annotations on GEP indices to split globals where beneficial. 1734 MPM.addPass(GlobalSplitPass()); 1735 1736 // Run whole program optimization of virtual call when the list of callees 1737 // is fixed. 1738 MPM.addPass(WholeProgramDevirtPass(ExportSummary, nullptr)); 1739 1740 // Stop here at -O1. 1741 if (Level == OptimizationLevel::O1) { 1742 // The LowerTypeTestsPass needs to run to lower type metadata and the 1743 // type.test intrinsics. The pass does nothing if CFI is disabled. 1744 MPM.addPass(LowerTypeTestsPass(ExportSummary, nullptr)); 1745 // Run a second time to clean up any type tests left behind by WPD for use 1746 // in ICP (which is performed earlier than this in the regular LTO 1747 // pipeline). 1748 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1749 1750 // Emit annotation remarks. 1751 addAnnotationRemarksPass(MPM); 1752 1753 return MPM; 1754 } 1755 1756 // Optimize globals to try and fold them into constants. 1757 MPM.addPass(GlobalOptPass()); 1758 1759 // Promote any localized globals to SSA registers. 1760 MPM.addPass(createModuleToFunctionPassAdaptor(PromotePass())); 1761 1762 // Linking modules together can lead to duplicate global constant, only 1763 // keep one copy of each constant. 1764 MPM.addPass(ConstantMergePass()); 1765 1766 // Remove unused arguments from functions. 1767 MPM.addPass(DeadArgumentEliminationPass()); 1768 1769 // Reduce the code after globalopt and ipsccp. Both can open up significant 1770 // simplification opportunities, and both can propagate functions through 1771 // function pointers. When this happens, we often have to resolve varargs 1772 // calls, etc, so let instcombine do this. 1773 FunctionPassManager PeepholeFPM; 1774 if (Level == OptimizationLevel::O3) 1775 PeepholeFPM.addPass(AggressiveInstCombinePass()); 1776 PeepholeFPM.addPass(InstCombinePass()); 1777 invokePeepholeEPCallbacks(PeepholeFPM, Level); 1778 1779 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(PeepholeFPM))); 1780 1781 // Note: historically, the PruneEH pass was run first to deduce nounwind and 1782 // generally clean up exception handling overhead. It isn't clear this is 1783 // valuable as the inliner doesn't currently care whether it is inlining an 1784 // invoke or a call. 1785 // Run the inliner now. 1786 MPM.addPass(ModuleInlinerWrapperPass(getInlineParamsFromOptLevel(Level))); 1787 1788 // Optimize globals again after we ran the inliner. 1789 MPM.addPass(GlobalOptPass()); 1790 1791 // Garbage collect dead functions. 1792 MPM.addPass(GlobalDCEPass()); 1793 1794 // If we didn't decide to inline a function, check to see if we can 1795 // transform it to pass arguments by value instead of by reference. 1796 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(ArgumentPromotionPass())); 1797 1798 FunctionPassManager FPM; 1799 // The IPO Passes may leave cruft around. Clean up after them. 1800 FPM.addPass(InstCombinePass()); 1801 invokePeepholeEPCallbacks(FPM, Level); 1802 1803 FPM.addPass(JumpThreadingPass(/*InsertFreezeWhenUnfoldingSelect*/ true)); 1804 1805 // Do a post inline PGO instrumentation and use pass. This is a context 1806 // sensitive PGO pass. 1807 if (PGOOpt) { 1808 if (PGOOpt->CSAction == PGOOptions::CSIRInstr) 1809 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ true, 1810 /* IsCS */ true, PGOOpt->CSProfileGenFile, 1811 PGOOpt->ProfileRemappingFile); 1812 else if (PGOOpt->CSAction == PGOOptions::CSIRUse) 1813 addPGOInstrPasses(MPM, Level, /* RunProfileGen */ false, 1814 /* IsCS */ true, PGOOpt->ProfileFile, 1815 PGOOpt->ProfileRemappingFile); 1816 } 1817 1818 // Break up allocas 1819 FPM.addPass(SROA()); 1820 1821 // LTO provides additional opportunities for tailcall elimination due to 1822 // link-time inlining, and visibility of nocapture attribute. 1823 FPM.addPass(TailCallElimPass()); 1824 1825 // Run a few AA driver optimizations here and now to cleanup the code. 1826 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 1827 1828 MPM.addPass( 1829 createModuleToPostOrderCGSCCPassAdaptor(PostOrderFunctionAttrsPass())); 1830 1831 // Require the GlobalsAA analysis for the module so we can query it within 1832 // MainFPM. 1833 MPM.addPass(RequireAnalysisPass<GlobalsAA, Module>()); 1834 // Invalidate AAManager so it can be recreated and pick up the newly available 1835 // GlobalsAA. 1836 MPM.addPass( 1837 createModuleToFunctionPassAdaptor(InvalidateAnalysisPass<AAManager>())); 1838 1839 FunctionPassManager MainFPM; 1840 MainFPM.addPass(createFunctionToLoopPassAdaptor( 1841 LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap), 1842 /*USeMemorySSA=*/true, /*UseBlockFrequencyInfo=*/true)); 1843 1844 if (RunNewGVN) 1845 MainFPM.addPass(NewGVNPass()); 1846 else 1847 MainFPM.addPass(GVN()); 1848 1849 // Remove dead memcpy()'s. 1850 MainFPM.addPass(MemCpyOptPass()); 1851 1852 // Nuke dead stores. 1853 MainFPM.addPass(DSEPass()); 1854 MainFPM.addPass(MergedLoadStoreMotionPass()); 1855 1856 // More loops are countable; try to optimize them. 1857 if (EnableLoopFlatten && Level.getSpeedupLevel() > 1) 1858 MainFPM.addPass(createFunctionToLoopPassAdaptor(LoopFlattenPass())); 1859 1860 if (EnableConstraintElimination) 1861 MainFPM.addPass(ConstraintEliminationPass()); 1862 1863 LoopPassManager LPM; 1864 LPM.addPass(IndVarSimplifyPass()); 1865 LPM.addPass(LoopDeletionPass()); 1866 // FIXME: Add loop interchange. 1867 1868 // Unroll small loops and perform peeling. 1869 LPM.addPass(LoopFullUnrollPass(Level.getSpeedupLevel(), 1870 /* OnlyWhenForced= */ !PTO.LoopUnrolling, 1871 PTO.ForgetAllSCEVInLoopUnroll)); 1872 // The loop passes in LPM (LoopFullUnrollPass) do not preserve MemorySSA. 1873 // *All* loop passes must preserve it, in order to be able to use it. 1874 MainFPM.addPass(createFunctionToLoopPassAdaptor( 1875 std::move(LPM), /*UseMemorySSA=*/false, /*UseBlockFrequencyInfo=*/true)); 1876 1877 MainFPM.addPass(LoopDistributePass()); 1878 1879 addVectorPasses(Level, MainFPM, /* IsFullLTO */ true); 1880 1881 invokePeepholeEPCallbacks(MainFPM, Level); 1882 MainFPM.addPass(JumpThreadingPass(/*InsertFreezeWhenUnfoldingSelect*/ true)); 1883 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(MainFPM))); 1884 1885 // Lower type metadata and the type.test intrinsic. This pass supports 1886 // clang's control flow integrity mechanisms (-fsanitize=cfi*) and needs 1887 // to be run at link time if CFI is enabled. This pass does nothing if 1888 // CFI is disabled. 1889 MPM.addPass(LowerTypeTestsPass(ExportSummary, nullptr)); 1890 // Run a second time to clean up any type tests left behind by WPD for use 1891 // in ICP (which is performed earlier than this in the regular LTO pipeline). 1892 MPM.addPass(LowerTypeTestsPass(nullptr, nullptr, true)); 1893 1894 // Enable splitting late in the FullLTO post-link pipeline. This is done in 1895 // the same stage in the old pass manager (\ref addLateLTOOptimizationPasses). 1896 if (EnableHotColdSplit) 1897 MPM.addPass(HotColdSplittingPass()); 1898 1899 // Add late LTO optimization passes. 1900 // Delete basic blocks, which optimization passes may have killed. 1901 MPM.addPass(createModuleToFunctionPassAdaptor( 1902 SimplifyCFGPass(SimplifyCFGOptions().hoistCommonInsts(true)))); 1903 1904 // Drop bodies of available eternally objects to improve GlobalDCE. 1905 MPM.addPass(EliminateAvailableExternallyPass()); 1906 1907 // Now that we have optimized the program, discard unreachable functions. 1908 MPM.addPass(GlobalDCEPass()); 1909 1910 if (PTO.MergeFunctions) 1911 MPM.addPass(MergeFunctionsPass()); 1912 1913 // Emit annotation remarks. 1914 addAnnotationRemarksPass(MPM); 1915 1916 return MPM; 1917 } 1918 1919 ModulePassManager PassBuilder::buildO0DefaultPipeline(OptimizationLevel Level, 1920 bool LTOPreLink) { 1921 assert(Level == OptimizationLevel::O0 && 1922 "buildO0DefaultPipeline should only be used with O0"); 1923 1924 ModulePassManager MPM; 1925 1926 if (PGOOpt && (PGOOpt->Action == PGOOptions::IRInstr || 1927 PGOOpt->Action == PGOOptions::IRUse)) 1928 addPGOInstrPassesForO0( 1929 MPM, 1930 /* RunProfileGen */ (PGOOpt->Action == PGOOptions::IRInstr), 1931 /* IsCS */ false, PGOOpt->ProfileFile, PGOOpt->ProfileRemappingFile); 1932 1933 for (auto &C : PipelineStartEPCallbacks) 1934 C(MPM, Level); 1935 1936 if (PGOOpt && PGOOpt->DebugInfoForProfiling) 1937 MPM.addPass(createModuleToFunctionPassAdaptor(AddDiscriminatorsPass())); 1938 1939 for (auto &C : PipelineEarlySimplificationEPCallbacks) 1940 C(MPM, Level); 1941 1942 // Build a minimal pipeline based on the semantics required by LLVM, 1943 // which is just that always inlining occurs. Further, disable generating 1944 // lifetime intrinsics to avoid enabling further optimizations during 1945 // code generation. 1946 MPM.addPass(AlwaysInlinerPass( 1947 /*InsertLifetimeIntrinsics=*/false)); 1948 1949 if (PTO.MergeFunctions) 1950 MPM.addPass(MergeFunctionsPass()); 1951 1952 if (EnableMatrix) 1953 MPM.addPass( 1954 createModuleToFunctionPassAdaptor(LowerMatrixIntrinsicsPass(true))); 1955 1956 if (!CGSCCOptimizerLateEPCallbacks.empty()) { 1957 CGSCCPassManager CGPM; 1958 for (auto &C : CGSCCOptimizerLateEPCallbacks) 1959 C(CGPM, Level); 1960 if (!CGPM.isEmpty()) 1961 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPM))); 1962 } 1963 if (!LateLoopOptimizationsEPCallbacks.empty()) { 1964 LoopPassManager LPM; 1965 for (auto &C : LateLoopOptimizationsEPCallbacks) 1966 C(LPM, Level); 1967 if (!LPM.isEmpty()) { 1968 MPM.addPass(createModuleToFunctionPassAdaptor( 1969 createFunctionToLoopPassAdaptor(std::move(LPM)))); 1970 } 1971 } 1972 if (!LoopOptimizerEndEPCallbacks.empty()) { 1973 LoopPassManager LPM; 1974 for (auto &C : LoopOptimizerEndEPCallbacks) 1975 C(LPM, Level); 1976 if (!LPM.isEmpty()) { 1977 MPM.addPass(createModuleToFunctionPassAdaptor( 1978 createFunctionToLoopPassAdaptor(std::move(LPM)))); 1979 } 1980 } 1981 if (!ScalarOptimizerLateEPCallbacks.empty()) { 1982 FunctionPassManager FPM; 1983 for (auto &C : ScalarOptimizerLateEPCallbacks) 1984 C(FPM, Level); 1985 if (!FPM.isEmpty()) 1986 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 1987 } 1988 if (!VectorizerStartEPCallbacks.empty()) { 1989 FunctionPassManager FPM; 1990 for (auto &C : VectorizerStartEPCallbacks) 1991 C(FPM, Level); 1992 if (!FPM.isEmpty()) 1993 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 1994 } 1995 1996 MPM.addPass(createModuleToFunctionPassAdaptor(CoroEarlyPass())); 1997 CGSCCPassManager CGPM; 1998 CGPM.addPass(CoroSplitPass()); 1999 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPM))); 2000 MPM.addPass(createModuleToFunctionPassAdaptor(CoroCleanupPass())); 2001 2002 for (auto &C : OptimizerLastEPCallbacks) 2003 C(MPM, Level); 2004 2005 if (LTOPreLink) 2006 addRequiredLTOPreLinkPasses(MPM); 2007 2008 return MPM; 2009 } 2010 2011 AAManager PassBuilder::buildDefaultAAPipeline() { 2012 AAManager AA; 2013 2014 // The order in which these are registered determines their priority when 2015 // being queried. 2016 2017 // First we register the basic alias analysis that provides the majority of 2018 // per-function local AA logic. This is a stateless, on-demand local set of 2019 // AA techniques. 2020 AA.registerFunctionAnalysis<BasicAA>(); 2021 2022 // Next we query fast, specialized alias analyses that wrap IR-embedded 2023 // information about aliasing. 2024 AA.registerFunctionAnalysis<ScopedNoAliasAA>(); 2025 AA.registerFunctionAnalysis<TypeBasedAA>(); 2026 2027 // Add support for querying global aliasing information when available. 2028 // Because the `AAManager` is a function analysis and `GlobalsAA` is a module 2029 // analysis, all that the `AAManager` can do is query for any *cached* 2030 // results from `GlobalsAA` through a readonly proxy. 2031 AA.registerModuleAnalysis<GlobalsAA>(); 2032 2033 // Add target-specific alias analyses. 2034 if (TM) 2035 TM->registerDefaultAliasAnalyses(AA); 2036 2037 return AA; 2038 } 2039 2040 static Optional<int> parseRepeatPassName(StringRef Name) { 2041 if (!Name.consume_front("repeat<") || !Name.consume_back(">")) 2042 return None; 2043 int Count; 2044 if (Name.getAsInteger(0, Count) || Count <= 0) 2045 return None; 2046 return Count; 2047 } 2048 2049 static Optional<int> parseDevirtPassName(StringRef Name) { 2050 if (!Name.consume_front("devirt<") || !Name.consume_back(">")) 2051 return None; 2052 int Count; 2053 if (Name.getAsInteger(0, Count) || Count < 0) 2054 return None; 2055 return Count; 2056 } 2057 2058 static bool checkParametrizedPassName(StringRef Name, StringRef PassName) { 2059 if (!Name.consume_front(PassName)) 2060 return false; 2061 // normal pass name w/o parameters == default parameters 2062 if (Name.empty()) 2063 return true; 2064 return Name.startswith("<") && Name.endswith(">"); 2065 } 2066 2067 namespace { 2068 2069 /// This performs customized parsing of pass name with parameters. 2070 /// 2071 /// We do not need parametrization of passes in textual pipeline very often, 2072 /// yet on a rare occasion ability to specify parameters right there can be 2073 /// useful. 2074 /// 2075 /// \p Name - parameterized specification of a pass from a textual pipeline 2076 /// is a string in a form of : 2077 /// PassName '<' parameter-list '>' 2078 /// 2079 /// Parameter list is being parsed by the parser callable argument, \p Parser, 2080 /// It takes a string-ref of parameters and returns either StringError or a 2081 /// parameter list in a form of a custom parameters type, all wrapped into 2082 /// Expected<> template class. 2083 /// 2084 template <typename ParametersParseCallableT> 2085 auto parsePassParameters(ParametersParseCallableT &&Parser, StringRef Name, 2086 StringRef PassName) -> decltype(Parser(StringRef{})) { 2087 using ParametersT = typename decltype(Parser(StringRef{}))::value_type; 2088 2089 StringRef Params = Name; 2090 if (!Params.consume_front(PassName)) { 2091 assert(false && 2092 "unable to strip pass name from parametrized pass specification"); 2093 } 2094 if (!Params.empty() && 2095 (!Params.consume_front("<") || !Params.consume_back(">"))) { 2096 assert(false && "invalid format for parametrized pass name"); 2097 } 2098 2099 Expected<ParametersT> Result = Parser(Params); 2100 assert((Result || Result.template errorIsA<StringError>()) && 2101 "Pass parameter parser can only return StringErrors."); 2102 return Result; 2103 } 2104 2105 /// Parser of parameters for LoopUnroll pass. 2106 Expected<LoopUnrollOptions> parseLoopUnrollOptions(StringRef Params) { 2107 LoopUnrollOptions UnrollOpts; 2108 while (!Params.empty()) { 2109 StringRef ParamName; 2110 std::tie(ParamName, Params) = Params.split(';'); 2111 int OptLevel = StringSwitch<int>(ParamName) 2112 .Case("O0", 0) 2113 .Case("O1", 1) 2114 .Case("O2", 2) 2115 .Case("O3", 3) 2116 .Default(-1); 2117 if (OptLevel >= 0) { 2118 UnrollOpts.setOptLevel(OptLevel); 2119 continue; 2120 } 2121 if (ParamName.consume_front("full-unroll-max=")) { 2122 int Count; 2123 if (ParamName.getAsInteger(0, Count)) 2124 return make_error<StringError>( 2125 formatv("invalid LoopUnrollPass parameter '{0}' ", ParamName).str(), 2126 inconvertibleErrorCode()); 2127 UnrollOpts.setFullUnrollMaxCount(Count); 2128 continue; 2129 } 2130 2131 bool Enable = !ParamName.consume_front("no-"); 2132 if (ParamName == "partial") { 2133 UnrollOpts.setPartial(Enable); 2134 } else if (ParamName == "peeling") { 2135 UnrollOpts.setPeeling(Enable); 2136 } else if (ParamName == "profile-peeling") { 2137 UnrollOpts.setProfileBasedPeeling(Enable); 2138 } else if (ParamName == "runtime") { 2139 UnrollOpts.setRuntime(Enable); 2140 } else if (ParamName == "upperbound") { 2141 UnrollOpts.setUpperBound(Enable); 2142 } else { 2143 return make_error<StringError>( 2144 formatv("invalid LoopUnrollPass parameter '{0}' ", ParamName).str(), 2145 inconvertibleErrorCode()); 2146 } 2147 } 2148 return UnrollOpts; 2149 } 2150 2151 Expected<bool> parseSinglePassOption(StringRef Params, StringRef OptionName, 2152 StringRef PassName) { 2153 bool Result = false; 2154 while (!Params.empty()) { 2155 StringRef ParamName; 2156 std::tie(ParamName, Params) = Params.split(';'); 2157 2158 if (ParamName == OptionName) { 2159 Result = true; 2160 } else { 2161 return make_error<StringError>( 2162 formatv("invalid {1} pass parameter '{0}' ", ParamName, PassName) 2163 .str(), 2164 inconvertibleErrorCode()); 2165 } 2166 } 2167 return Result; 2168 } 2169 2170 Expected<bool> parseEarlyCSEPassOptions(StringRef Params) { 2171 return parseSinglePassOption(Params, "memssa", "EarlyCSE"); 2172 } 2173 2174 Expected<bool> parseEntryExitInstrumenterPassOptions(StringRef Params) { 2175 return parseSinglePassOption(Params, "post-inline", "EntryExitInstrumenter"); 2176 } 2177 2178 Expected<bool> parseLoopExtractorPassOptions(StringRef Params) { 2179 return parseSinglePassOption(Params, "single", "LoopExtractor"); 2180 } 2181 2182 Expected<bool> parseLowerMatrixIntrinsicsPassOptions(StringRef Params) { 2183 return parseSinglePassOption(Params, "minimal", "LowerMatrixIntrinsics"); 2184 } 2185 2186 Expected<AddressSanitizerOptions> parseASanPassOptions(StringRef Params) { 2187 AddressSanitizerOptions Result; 2188 while (!Params.empty()) { 2189 StringRef ParamName; 2190 std::tie(ParamName, Params) = Params.split(';'); 2191 2192 if (ParamName == "kernel") { 2193 Result.CompileKernel = true; 2194 } else { 2195 return make_error<StringError>( 2196 formatv("invalid AddressSanitizer pass parameter '{0}' ", ParamName) 2197 .str(), 2198 inconvertibleErrorCode()); 2199 } 2200 } 2201 return Result; 2202 } 2203 2204 Expected<HWAddressSanitizerOptions> parseHWASanPassOptions(StringRef Params) { 2205 HWAddressSanitizerOptions Result; 2206 while (!Params.empty()) { 2207 StringRef ParamName; 2208 std::tie(ParamName, Params) = Params.split(';'); 2209 2210 if (ParamName == "recover") { 2211 Result.Recover = true; 2212 } else if (ParamName == "kernel") { 2213 Result.CompileKernel = true; 2214 } else { 2215 return make_error<StringError>( 2216 formatv("invalid HWAddressSanitizer pass parameter '{0}' ", ParamName) 2217 .str(), 2218 inconvertibleErrorCode()); 2219 } 2220 } 2221 return Result; 2222 } 2223 2224 Expected<MemorySanitizerOptions> parseMSanPassOptions(StringRef Params) { 2225 MemorySanitizerOptions Result; 2226 while (!Params.empty()) { 2227 StringRef ParamName; 2228 std::tie(ParamName, Params) = Params.split(';'); 2229 2230 if (ParamName == "recover") { 2231 Result.Recover = true; 2232 } else if (ParamName == "kernel") { 2233 Result.Kernel = true; 2234 } else if (ParamName.consume_front("track-origins=")) { 2235 if (ParamName.getAsInteger(0, Result.TrackOrigins)) 2236 return make_error<StringError>( 2237 formatv("invalid argument to MemorySanitizer pass track-origins " 2238 "parameter: '{0}' ", 2239 ParamName) 2240 .str(), 2241 inconvertibleErrorCode()); 2242 } else { 2243 return make_error<StringError>( 2244 formatv("invalid MemorySanitizer pass parameter '{0}' ", ParamName) 2245 .str(), 2246 inconvertibleErrorCode()); 2247 } 2248 } 2249 return Result; 2250 } 2251 2252 /// Parser of parameters for SimplifyCFG pass. 2253 Expected<SimplifyCFGOptions> parseSimplifyCFGOptions(StringRef Params) { 2254 SimplifyCFGOptions Result; 2255 while (!Params.empty()) { 2256 StringRef ParamName; 2257 std::tie(ParamName, Params) = Params.split(';'); 2258 2259 bool Enable = !ParamName.consume_front("no-"); 2260 if (ParamName == "forward-switch-cond") { 2261 Result.forwardSwitchCondToPhi(Enable); 2262 } else if (ParamName == "switch-to-lookup") { 2263 Result.convertSwitchToLookupTable(Enable); 2264 } else if (ParamName == "keep-loops") { 2265 Result.needCanonicalLoops(Enable); 2266 } else if (ParamName == "hoist-common-insts") { 2267 Result.hoistCommonInsts(Enable); 2268 } else if (ParamName == "sink-common-insts") { 2269 Result.sinkCommonInsts(Enable); 2270 } else if (Enable && ParamName.consume_front("bonus-inst-threshold=")) { 2271 APInt BonusInstThreshold; 2272 if (ParamName.getAsInteger(0, BonusInstThreshold)) 2273 return make_error<StringError>( 2274 formatv("invalid argument to SimplifyCFG pass bonus-threshold " 2275 "parameter: '{0}' ", 2276 ParamName).str(), 2277 inconvertibleErrorCode()); 2278 Result.bonusInstThreshold(BonusInstThreshold.getSExtValue()); 2279 } else { 2280 return make_error<StringError>( 2281 formatv("invalid SimplifyCFG pass parameter '{0}' ", ParamName).str(), 2282 inconvertibleErrorCode()); 2283 } 2284 } 2285 return Result; 2286 } 2287 2288 /// Parser of parameters for LoopVectorize pass. 2289 Expected<LoopVectorizeOptions> parseLoopVectorizeOptions(StringRef Params) { 2290 LoopVectorizeOptions Opts; 2291 while (!Params.empty()) { 2292 StringRef ParamName; 2293 std::tie(ParamName, Params) = Params.split(';'); 2294 2295 bool Enable = !ParamName.consume_front("no-"); 2296 if (ParamName == "interleave-forced-only") { 2297 Opts.setInterleaveOnlyWhenForced(Enable); 2298 } else if (ParamName == "vectorize-forced-only") { 2299 Opts.setVectorizeOnlyWhenForced(Enable); 2300 } else { 2301 return make_error<StringError>( 2302 formatv("invalid LoopVectorize parameter '{0}' ", ParamName).str(), 2303 inconvertibleErrorCode()); 2304 } 2305 } 2306 return Opts; 2307 } 2308 2309 Expected<std::pair<bool, bool>> parseLoopUnswitchOptions(StringRef Params) { 2310 std::pair<bool, bool> Result = {false, true}; 2311 while (!Params.empty()) { 2312 StringRef ParamName; 2313 std::tie(ParamName, Params) = Params.split(';'); 2314 2315 bool Enable = !ParamName.consume_front("no-"); 2316 if (ParamName == "nontrivial") { 2317 Result.first = Enable; 2318 } else if (ParamName == "trivial") { 2319 Result.second = Enable; 2320 } else { 2321 return make_error<StringError>( 2322 formatv("invalid LoopUnswitch pass parameter '{0}' ", ParamName) 2323 .str(), 2324 inconvertibleErrorCode()); 2325 } 2326 } 2327 return Result; 2328 } 2329 2330 Expected<bool> parseMergedLoadStoreMotionOptions(StringRef Params) { 2331 bool Result = false; 2332 while (!Params.empty()) { 2333 StringRef ParamName; 2334 std::tie(ParamName, Params) = Params.split(';'); 2335 2336 bool Enable = !ParamName.consume_front("no-"); 2337 if (ParamName == "split-footer-bb") { 2338 Result = Enable; 2339 } else { 2340 return make_error<StringError>( 2341 formatv("invalid MergedLoadStoreMotion pass parameter '{0}' ", 2342 ParamName) 2343 .str(), 2344 inconvertibleErrorCode()); 2345 } 2346 } 2347 return Result; 2348 } 2349 2350 Expected<GVNOptions> parseGVNOptions(StringRef Params) { 2351 GVNOptions Result; 2352 while (!Params.empty()) { 2353 StringRef ParamName; 2354 std::tie(ParamName, Params) = Params.split(';'); 2355 2356 bool Enable = !ParamName.consume_front("no-"); 2357 if (ParamName == "pre") { 2358 Result.setPRE(Enable); 2359 } else if (ParamName == "load-pre") { 2360 Result.setLoadPRE(Enable); 2361 } else if (ParamName == "split-backedge-load-pre") { 2362 Result.setLoadPRESplitBackedge(Enable); 2363 } else if (ParamName == "memdep") { 2364 Result.setMemDep(Enable); 2365 } else { 2366 return make_error<StringError>( 2367 formatv("invalid GVN pass parameter '{0}' ", ParamName).str(), 2368 inconvertibleErrorCode()); 2369 } 2370 } 2371 return Result; 2372 } 2373 2374 Expected<StackLifetime::LivenessType> 2375 parseStackLifetimeOptions(StringRef Params) { 2376 StackLifetime::LivenessType Result = StackLifetime::LivenessType::May; 2377 while (!Params.empty()) { 2378 StringRef ParamName; 2379 std::tie(ParamName, Params) = Params.split(';'); 2380 2381 if (ParamName == "may") { 2382 Result = StackLifetime::LivenessType::May; 2383 } else if (ParamName == "must") { 2384 Result = StackLifetime::LivenessType::Must; 2385 } else { 2386 return make_error<StringError>( 2387 formatv("invalid StackLifetime parameter '{0}' ", ParamName).str(), 2388 inconvertibleErrorCode()); 2389 } 2390 } 2391 return Result; 2392 } 2393 2394 } // namespace 2395 2396 /// Tests whether a pass name starts with a valid prefix for a default pipeline 2397 /// alias. 2398 static bool startsWithDefaultPipelineAliasPrefix(StringRef Name) { 2399 return Name.startswith("default") || Name.startswith("thinlto") || 2400 Name.startswith("lto"); 2401 } 2402 2403 /// Tests whether registered callbacks will accept a given pass name. 2404 /// 2405 /// When parsing a pipeline text, the type of the outermost pipeline may be 2406 /// omitted, in which case the type is automatically determined from the first 2407 /// pass name in the text. This may be a name that is handled through one of the 2408 /// callbacks. We check this through the oridinary parsing callbacks by setting 2409 /// up a dummy PassManager in order to not force the client to also handle this 2410 /// type of query. 2411 template <typename PassManagerT, typename CallbacksT> 2412 static bool callbacksAcceptPassName(StringRef Name, CallbacksT &Callbacks) { 2413 if (!Callbacks.empty()) { 2414 PassManagerT DummyPM; 2415 for (auto &CB : Callbacks) 2416 if (CB(Name, DummyPM, {})) 2417 return true; 2418 } 2419 return false; 2420 } 2421 2422 template <typename CallbacksT> 2423 static bool isModulePassName(StringRef Name, CallbacksT &Callbacks) { 2424 // Manually handle aliases for pre-configured pipeline fragments. 2425 if (startsWithDefaultPipelineAliasPrefix(Name)) 2426 return DefaultAliasRegex.match(Name); 2427 2428 // Explicitly handle pass manager names. 2429 if (Name == "module") 2430 return true; 2431 if (Name == "cgscc") 2432 return true; 2433 if (Name == "function") 2434 return true; 2435 2436 // Explicitly handle custom-parsed pass names. 2437 if (parseRepeatPassName(Name)) 2438 return true; 2439 2440 #define MODULE_PASS(NAME, CREATE_PASS) \ 2441 if (Name == NAME) \ 2442 return true; 2443 #define MODULE_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 2444 if (checkParametrizedPassName(Name, NAME)) \ 2445 return true; 2446 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 2447 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2448 return true; 2449 #include "PassRegistry.def" 2450 2451 return callbacksAcceptPassName<ModulePassManager>(Name, Callbacks); 2452 } 2453 2454 template <typename CallbacksT> 2455 static bool isCGSCCPassName(StringRef Name, CallbacksT &Callbacks) { 2456 // Explicitly handle pass manager names. 2457 if (Name == "cgscc") 2458 return true; 2459 if (Name == "function") 2460 return true; 2461 2462 // Explicitly handle custom-parsed pass names. 2463 if (parseRepeatPassName(Name)) 2464 return true; 2465 if (parseDevirtPassName(Name)) 2466 return true; 2467 2468 #define CGSCC_PASS(NAME, CREATE_PASS) \ 2469 if (Name == NAME) \ 2470 return true; 2471 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 2472 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2473 return true; 2474 #include "PassRegistry.def" 2475 2476 return callbacksAcceptPassName<CGSCCPassManager>(Name, Callbacks); 2477 } 2478 2479 template <typename CallbacksT> 2480 static bool isFunctionPassName(StringRef Name, CallbacksT &Callbacks) { 2481 // Explicitly handle pass manager names. 2482 if (Name == "function") 2483 return true; 2484 if (Name == "loop" || Name == "loop-mssa") 2485 return true; 2486 2487 // Explicitly handle custom-parsed pass names. 2488 if (parseRepeatPassName(Name)) 2489 return true; 2490 2491 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2492 if (Name == NAME) \ 2493 return true; 2494 #define FUNCTION_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 2495 if (checkParametrizedPassName(Name, NAME)) \ 2496 return true; 2497 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 2498 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2499 return true; 2500 #include "PassRegistry.def" 2501 2502 return callbacksAcceptPassName<FunctionPassManager>(Name, Callbacks); 2503 } 2504 2505 template <typename CallbacksT> 2506 static bool isLoopPassName(StringRef Name, CallbacksT &Callbacks, 2507 bool &UseMemorySSA) { 2508 UseMemorySSA = false; 2509 2510 // Explicitly handle custom-parsed pass names. 2511 if (parseRepeatPassName(Name)) 2512 return true; 2513 2514 if (Name == "licm") { 2515 UseMemorySSA = true; 2516 return true; 2517 } 2518 2519 #define LOOP_PASS(NAME, CREATE_PASS) \ 2520 if (Name == NAME) \ 2521 return true; 2522 #define LOOP_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 2523 if (checkParametrizedPassName(Name, NAME)) \ 2524 return true; 2525 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 2526 if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \ 2527 return true; 2528 #include "PassRegistry.def" 2529 2530 return callbacksAcceptPassName<LoopPassManager>(Name, Callbacks); 2531 } 2532 2533 Optional<std::vector<PassBuilder::PipelineElement>> 2534 PassBuilder::parsePipelineText(StringRef Text) { 2535 std::vector<PipelineElement> ResultPipeline; 2536 2537 SmallVector<std::vector<PipelineElement> *, 4> PipelineStack = { 2538 &ResultPipeline}; 2539 for (;;) { 2540 std::vector<PipelineElement> &Pipeline = *PipelineStack.back(); 2541 size_t Pos = Text.find_first_of(",()"); 2542 Pipeline.push_back({Text.substr(0, Pos), {}}); 2543 2544 // If we have a single terminating name, we're done. 2545 if (Pos == Text.npos) 2546 break; 2547 2548 char Sep = Text[Pos]; 2549 Text = Text.substr(Pos + 1); 2550 if (Sep == ',') 2551 // Just a name ending in a comma, continue. 2552 continue; 2553 2554 if (Sep == '(') { 2555 // Push the inner pipeline onto the stack to continue processing. 2556 PipelineStack.push_back(&Pipeline.back().InnerPipeline); 2557 continue; 2558 } 2559 2560 assert(Sep == ')' && "Bogus separator!"); 2561 // When handling the close parenthesis, we greedily consume them to avoid 2562 // empty strings in the pipeline. 2563 do { 2564 // If we try to pop the outer pipeline we have unbalanced parentheses. 2565 if (PipelineStack.size() == 1) 2566 return None; 2567 2568 PipelineStack.pop_back(); 2569 } while (Text.consume_front(")")); 2570 2571 // Check if we've finished parsing. 2572 if (Text.empty()) 2573 break; 2574 2575 // Otherwise, the end of an inner pipeline always has to be followed by 2576 // a comma, and then we can continue. 2577 if (!Text.consume_front(",")) 2578 return None; 2579 } 2580 2581 if (PipelineStack.size() > 1) 2582 // Unbalanced paretheses. 2583 return None; 2584 2585 assert(PipelineStack.back() == &ResultPipeline && 2586 "Wrong pipeline at the bottom of the stack!"); 2587 return {std::move(ResultPipeline)}; 2588 } 2589 2590 Error PassBuilder::parseModulePass(ModulePassManager &MPM, 2591 const PipelineElement &E) { 2592 auto &Name = E.Name; 2593 auto &InnerPipeline = E.InnerPipeline; 2594 2595 // First handle complex passes like the pass managers which carry pipelines. 2596 if (!InnerPipeline.empty()) { 2597 if (Name == "module") { 2598 ModulePassManager NestedMPM; 2599 if (auto Err = parseModulePassPipeline(NestedMPM, InnerPipeline)) 2600 return Err; 2601 MPM.addPass(std::move(NestedMPM)); 2602 return Error::success(); 2603 } 2604 if (Name == "cgscc") { 2605 CGSCCPassManager CGPM; 2606 if (auto Err = parseCGSCCPassPipeline(CGPM, InnerPipeline)) 2607 return Err; 2608 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPM))); 2609 return Error::success(); 2610 } 2611 if (Name == "function") { 2612 FunctionPassManager FPM; 2613 if (auto Err = parseFunctionPassPipeline(FPM, InnerPipeline)) 2614 return Err; 2615 MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM))); 2616 return Error::success(); 2617 } 2618 if (auto Count = parseRepeatPassName(Name)) { 2619 ModulePassManager NestedMPM; 2620 if (auto Err = parseModulePassPipeline(NestedMPM, InnerPipeline)) 2621 return Err; 2622 MPM.addPass(createRepeatedPass(*Count, std::move(NestedMPM))); 2623 return Error::success(); 2624 } 2625 2626 for (auto &C : ModulePipelineParsingCallbacks) 2627 if (C(Name, MPM, InnerPipeline)) 2628 return Error::success(); 2629 2630 // Normal passes can't have pipelines. 2631 return make_error<StringError>( 2632 formatv("invalid use of '{0}' pass as module pipeline", Name).str(), 2633 inconvertibleErrorCode()); 2634 ; 2635 } 2636 2637 // Manually handle aliases for pre-configured pipeline fragments. 2638 if (startsWithDefaultPipelineAliasPrefix(Name)) { 2639 SmallVector<StringRef, 3> Matches; 2640 if (!DefaultAliasRegex.match(Name, &Matches)) 2641 return make_error<StringError>( 2642 formatv("unknown default pipeline alias '{0}'", Name).str(), 2643 inconvertibleErrorCode()); 2644 2645 assert(Matches.size() == 3 && "Must capture two matched strings!"); 2646 2647 OptimizationLevel L = StringSwitch<OptimizationLevel>(Matches[2]) 2648 .Case("O0", OptimizationLevel::O0) 2649 .Case("O1", OptimizationLevel::O1) 2650 .Case("O2", OptimizationLevel::O2) 2651 .Case("O3", OptimizationLevel::O3) 2652 .Case("Os", OptimizationLevel::Os) 2653 .Case("Oz", OptimizationLevel::Oz); 2654 if (L == OptimizationLevel::O0 && Matches[1] != "thinlto" && 2655 Matches[1] != "lto") { 2656 MPM.addPass(buildO0DefaultPipeline(L, Matches[1] == "thinlto-pre-link" || 2657 Matches[1] == "lto-pre-link")); 2658 return Error::success(); 2659 } 2660 2661 // This is consistent with old pass manager invoked via opt, but 2662 // inconsistent with clang. Clang doesn't enable loop vectorization 2663 // but does enable slp vectorization at Oz. 2664 PTO.LoopVectorization = 2665 L.getSpeedupLevel() > 1 && L != OptimizationLevel::Oz; 2666 PTO.SLPVectorization = 2667 L.getSpeedupLevel() > 1 && L != OptimizationLevel::Oz; 2668 2669 if (Matches[1] == "default") { 2670 MPM.addPass(buildPerModuleDefaultPipeline(L)); 2671 } else if (Matches[1] == "thinlto-pre-link") { 2672 MPM.addPass(buildThinLTOPreLinkDefaultPipeline(L)); 2673 } else if (Matches[1] == "thinlto") { 2674 MPM.addPass(buildThinLTODefaultPipeline(L, nullptr)); 2675 } else if (Matches[1] == "lto-pre-link") { 2676 MPM.addPass(buildLTOPreLinkDefaultPipeline(L)); 2677 } else { 2678 assert(Matches[1] == "lto" && "Not one of the matched options!"); 2679 MPM.addPass(buildLTODefaultPipeline(L, nullptr)); 2680 } 2681 return Error::success(); 2682 } 2683 2684 // Finally expand the basic registered passes from the .inc file. 2685 #define MODULE_PASS(NAME, CREATE_PASS) \ 2686 if (Name == NAME) { \ 2687 MPM.addPass(CREATE_PASS); \ 2688 return Error::success(); \ 2689 } 2690 #define MODULE_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 2691 if (checkParametrizedPassName(Name, NAME)) { \ 2692 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2693 if (!Params) \ 2694 return Params.takeError(); \ 2695 MPM.addPass(CREATE_PASS(Params.get())); \ 2696 return Error::success(); \ 2697 } 2698 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 2699 if (Name == "require<" NAME ">") { \ 2700 MPM.addPass( \ 2701 RequireAnalysisPass< \ 2702 std::remove_reference<decltype(CREATE_PASS)>::type, Module>()); \ 2703 return Error::success(); \ 2704 } \ 2705 if (Name == "invalidate<" NAME ">") { \ 2706 MPM.addPass(InvalidateAnalysisPass< \ 2707 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2708 return Error::success(); \ 2709 } 2710 #define CGSCC_PASS(NAME, CREATE_PASS) \ 2711 if (Name == NAME) { \ 2712 MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(CREATE_PASS)); \ 2713 return Error::success(); \ 2714 } 2715 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2716 if (Name == NAME) { \ 2717 MPM.addPass(createModuleToFunctionPassAdaptor(CREATE_PASS)); \ 2718 return Error::success(); \ 2719 } 2720 #define FUNCTION_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 2721 if (checkParametrizedPassName(Name, NAME)) { \ 2722 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2723 if (!Params) \ 2724 return Params.takeError(); \ 2725 MPM.addPass(createModuleToFunctionPassAdaptor(CREATE_PASS(Params.get()))); \ 2726 return Error::success(); \ 2727 } 2728 #define LOOP_PASS(NAME, CREATE_PASS) \ 2729 if (Name == NAME) { \ 2730 MPM.addPass(createModuleToFunctionPassAdaptor( \ 2731 createFunctionToLoopPassAdaptor(CREATE_PASS, false, false))); \ 2732 return Error::success(); \ 2733 } 2734 #define LOOP_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 2735 if (checkParametrizedPassName(Name, NAME)) { \ 2736 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2737 if (!Params) \ 2738 return Params.takeError(); \ 2739 MPM.addPass( \ 2740 createModuleToFunctionPassAdaptor(createFunctionToLoopPassAdaptor( \ 2741 CREATE_PASS(Params.get()), false, false))); \ 2742 return Error::success(); \ 2743 } 2744 #include "PassRegistry.def" 2745 2746 for (auto &C : ModulePipelineParsingCallbacks) 2747 if (C(Name, MPM, InnerPipeline)) 2748 return Error::success(); 2749 return make_error<StringError>( 2750 formatv("unknown module pass '{0}'", Name).str(), 2751 inconvertibleErrorCode()); 2752 } 2753 2754 Error PassBuilder::parseCGSCCPass(CGSCCPassManager &CGPM, 2755 const PipelineElement &E) { 2756 auto &Name = E.Name; 2757 auto &InnerPipeline = E.InnerPipeline; 2758 2759 // First handle complex passes like the pass managers which carry pipelines. 2760 if (!InnerPipeline.empty()) { 2761 if (Name == "cgscc") { 2762 CGSCCPassManager NestedCGPM; 2763 if (auto Err = parseCGSCCPassPipeline(NestedCGPM, InnerPipeline)) 2764 return Err; 2765 // Add the nested pass manager with the appropriate adaptor. 2766 CGPM.addPass(std::move(NestedCGPM)); 2767 return Error::success(); 2768 } 2769 if (Name == "function") { 2770 FunctionPassManager FPM; 2771 if (auto Err = parseFunctionPassPipeline(FPM, InnerPipeline)) 2772 return Err; 2773 // Add the nested pass manager with the appropriate adaptor. 2774 CGPM.addPass(createCGSCCToFunctionPassAdaptor(std::move(FPM))); 2775 return Error::success(); 2776 } 2777 if (auto Count = parseRepeatPassName(Name)) { 2778 CGSCCPassManager NestedCGPM; 2779 if (auto Err = parseCGSCCPassPipeline(NestedCGPM, InnerPipeline)) 2780 return Err; 2781 CGPM.addPass(createRepeatedPass(*Count, std::move(NestedCGPM))); 2782 return Error::success(); 2783 } 2784 if (auto MaxRepetitions = parseDevirtPassName(Name)) { 2785 CGSCCPassManager NestedCGPM; 2786 if (auto Err = parseCGSCCPassPipeline(NestedCGPM, InnerPipeline)) 2787 return Err; 2788 CGPM.addPass( 2789 createDevirtSCCRepeatedPass(std::move(NestedCGPM), *MaxRepetitions)); 2790 return Error::success(); 2791 } 2792 2793 for (auto &C : CGSCCPipelineParsingCallbacks) 2794 if (C(Name, CGPM, InnerPipeline)) 2795 return Error::success(); 2796 2797 // Normal passes can't have pipelines. 2798 return make_error<StringError>( 2799 formatv("invalid use of '{0}' pass as cgscc pipeline", Name).str(), 2800 inconvertibleErrorCode()); 2801 } 2802 2803 // Now expand the basic registered passes from the .inc file. 2804 #define CGSCC_PASS(NAME, CREATE_PASS) \ 2805 if (Name == NAME) { \ 2806 CGPM.addPass(CREATE_PASS); \ 2807 return Error::success(); \ 2808 } 2809 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 2810 if (Name == "require<" NAME ">") { \ 2811 CGPM.addPass(RequireAnalysisPass< \ 2812 std::remove_reference<decltype(CREATE_PASS)>::type, \ 2813 LazyCallGraph::SCC, CGSCCAnalysisManager, LazyCallGraph &, \ 2814 CGSCCUpdateResult &>()); \ 2815 return Error::success(); \ 2816 } \ 2817 if (Name == "invalidate<" NAME ">") { \ 2818 CGPM.addPass(InvalidateAnalysisPass< \ 2819 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2820 return Error::success(); \ 2821 } 2822 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2823 if (Name == NAME) { \ 2824 CGPM.addPass(createCGSCCToFunctionPassAdaptor(CREATE_PASS)); \ 2825 return Error::success(); \ 2826 } 2827 #define FUNCTION_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 2828 if (checkParametrizedPassName(Name, NAME)) { \ 2829 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2830 if (!Params) \ 2831 return Params.takeError(); \ 2832 CGPM.addPass(createCGSCCToFunctionPassAdaptor(CREATE_PASS(Params.get()))); \ 2833 return Error::success(); \ 2834 } 2835 #define LOOP_PASS(NAME, CREATE_PASS) \ 2836 if (Name == NAME) { \ 2837 CGPM.addPass(createCGSCCToFunctionPassAdaptor( \ 2838 createFunctionToLoopPassAdaptor(CREATE_PASS, false, false))); \ 2839 return Error::success(); \ 2840 } 2841 #define LOOP_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 2842 if (checkParametrizedPassName(Name, NAME)) { \ 2843 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2844 if (!Params) \ 2845 return Params.takeError(); \ 2846 CGPM.addPass( \ 2847 createCGSCCToFunctionPassAdaptor(createFunctionToLoopPassAdaptor( \ 2848 CREATE_PASS(Params.get()), false, false))); \ 2849 return Error::success(); \ 2850 } 2851 #include "PassRegistry.def" 2852 2853 for (auto &C : CGSCCPipelineParsingCallbacks) 2854 if (C(Name, CGPM, InnerPipeline)) 2855 return Error::success(); 2856 return make_error<StringError>( 2857 formatv("unknown cgscc pass '{0}'", Name).str(), 2858 inconvertibleErrorCode()); 2859 } 2860 2861 Error PassBuilder::parseFunctionPass(FunctionPassManager &FPM, 2862 const PipelineElement &E) { 2863 auto &Name = E.Name; 2864 auto &InnerPipeline = E.InnerPipeline; 2865 2866 // First handle complex passes like the pass managers which carry pipelines. 2867 if (!InnerPipeline.empty()) { 2868 if (Name == "function") { 2869 FunctionPassManager NestedFPM; 2870 if (auto Err = parseFunctionPassPipeline(NestedFPM, InnerPipeline)) 2871 return Err; 2872 // Add the nested pass manager with the appropriate adaptor. 2873 FPM.addPass(std::move(NestedFPM)); 2874 return Error::success(); 2875 } 2876 if (Name == "loop" || Name == "loop-mssa") { 2877 LoopPassManager LPM; 2878 if (auto Err = parseLoopPassPipeline(LPM, InnerPipeline)) 2879 return Err; 2880 // Add the nested pass manager with the appropriate adaptor. 2881 bool UseMemorySSA = (Name == "loop-mssa"); 2882 bool UseBFI = llvm::any_of( 2883 InnerPipeline, [](auto Pipeline) { return Pipeline.Name == "licm"; }); 2884 FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM), UseMemorySSA, 2885 UseBFI)); 2886 return Error::success(); 2887 } 2888 if (auto Count = parseRepeatPassName(Name)) { 2889 FunctionPassManager NestedFPM; 2890 if (auto Err = parseFunctionPassPipeline(NestedFPM, InnerPipeline)) 2891 return Err; 2892 FPM.addPass(createRepeatedPass(*Count, std::move(NestedFPM))); 2893 return Error::success(); 2894 } 2895 2896 for (auto &C : FunctionPipelineParsingCallbacks) 2897 if (C(Name, FPM, InnerPipeline)) 2898 return Error::success(); 2899 2900 // Normal passes can't have pipelines. 2901 return make_error<StringError>( 2902 formatv("invalid use of '{0}' pass as function pipeline", Name).str(), 2903 inconvertibleErrorCode()); 2904 } 2905 2906 // Now expand the basic registered passes from the .inc file. 2907 #define FUNCTION_PASS(NAME, CREATE_PASS) \ 2908 if (Name == NAME) { \ 2909 FPM.addPass(CREATE_PASS); \ 2910 return Error::success(); \ 2911 } 2912 #define FUNCTION_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 2913 if (checkParametrizedPassName(Name, NAME)) { \ 2914 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2915 if (!Params) \ 2916 return Params.takeError(); \ 2917 FPM.addPass(CREATE_PASS(Params.get())); \ 2918 return Error::success(); \ 2919 } 2920 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 2921 if (Name == "require<" NAME ">") { \ 2922 FPM.addPass( \ 2923 RequireAnalysisPass< \ 2924 std::remove_reference<decltype(CREATE_PASS)>::type, Function>()); \ 2925 return Error::success(); \ 2926 } \ 2927 if (Name == "invalidate<" NAME ">") { \ 2928 FPM.addPass(InvalidateAnalysisPass< \ 2929 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 2930 return Error::success(); \ 2931 } 2932 // FIXME: UseMemorySSA is set to false. Maybe we could do things like: 2933 // bool UseMemorySSA = !("canon-freeze" || "loop-predication" || 2934 // "guard-widening"); 2935 // The risk is that it may become obsolete if we're not careful. 2936 #define LOOP_PASS(NAME, CREATE_PASS) \ 2937 if (Name == NAME) { \ 2938 FPM.addPass(createFunctionToLoopPassAdaptor(CREATE_PASS, false, false)); \ 2939 return Error::success(); \ 2940 } 2941 #define LOOP_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 2942 if (checkParametrizedPassName(Name, NAME)) { \ 2943 auto Params = parsePassParameters(PARSER, Name, NAME); \ 2944 if (!Params) \ 2945 return Params.takeError(); \ 2946 FPM.addPass(createFunctionToLoopPassAdaptor(CREATE_PASS(Params.get()), \ 2947 false, false)); \ 2948 return Error::success(); \ 2949 } 2950 #include "PassRegistry.def" 2951 2952 for (auto &C : FunctionPipelineParsingCallbacks) 2953 if (C(Name, FPM, InnerPipeline)) 2954 return Error::success(); 2955 return make_error<StringError>( 2956 formatv("unknown function pass '{0}'", Name).str(), 2957 inconvertibleErrorCode()); 2958 } 2959 2960 Error PassBuilder::parseLoopPass(LoopPassManager &LPM, 2961 const PipelineElement &E) { 2962 StringRef Name = E.Name; 2963 auto &InnerPipeline = E.InnerPipeline; 2964 2965 // First handle complex passes like the pass managers which carry pipelines. 2966 if (!InnerPipeline.empty()) { 2967 if (Name == "loop") { 2968 LoopPassManager NestedLPM; 2969 if (auto Err = parseLoopPassPipeline(NestedLPM, InnerPipeline)) 2970 return Err; 2971 // Add the nested pass manager with the appropriate adaptor. 2972 LPM.addPass(std::move(NestedLPM)); 2973 return Error::success(); 2974 } 2975 if (auto Count = parseRepeatPassName(Name)) { 2976 LoopPassManager NestedLPM; 2977 if (auto Err = parseLoopPassPipeline(NestedLPM, InnerPipeline)) 2978 return Err; 2979 LPM.addPass(createRepeatedPass(*Count, std::move(NestedLPM))); 2980 return Error::success(); 2981 } 2982 2983 for (auto &C : LoopPipelineParsingCallbacks) 2984 if (C(Name, LPM, InnerPipeline)) 2985 return Error::success(); 2986 2987 // Normal passes can't have pipelines. 2988 return make_error<StringError>( 2989 formatv("invalid use of '{0}' pass as loop pipeline", Name).str(), 2990 inconvertibleErrorCode()); 2991 } 2992 2993 // Now expand the basic registered passes from the .inc file. 2994 #define LOOP_PASS(NAME, CREATE_PASS) \ 2995 if (Name == NAME) { \ 2996 LPM.addPass(CREATE_PASS); \ 2997 return Error::success(); \ 2998 } 2999 #define LOOP_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 3000 if (checkParametrizedPassName(Name, NAME)) { \ 3001 auto Params = parsePassParameters(PARSER, Name, NAME); \ 3002 if (!Params) \ 3003 return Params.takeError(); \ 3004 LPM.addPass(CREATE_PASS(Params.get())); \ 3005 return Error::success(); \ 3006 } 3007 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 3008 if (Name == "require<" NAME ">") { \ 3009 LPM.addPass(RequireAnalysisPass< \ 3010 std::remove_reference<decltype(CREATE_PASS)>::type, Loop, \ 3011 LoopAnalysisManager, LoopStandardAnalysisResults &, \ 3012 LPMUpdater &>()); \ 3013 return Error::success(); \ 3014 } \ 3015 if (Name == "invalidate<" NAME ">") { \ 3016 LPM.addPass(InvalidateAnalysisPass< \ 3017 std::remove_reference<decltype(CREATE_PASS)>::type>()); \ 3018 return Error::success(); \ 3019 } 3020 #include "PassRegistry.def" 3021 3022 for (auto &C : LoopPipelineParsingCallbacks) 3023 if (C(Name, LPM, InnerPipeline)) 3024 return Error::success(); 3025 return make_error<StringError>(formatv("unknown loop pass '{0}'", Name).str(), 3026 inconvertibleErrorCode()); 3027 } 3028 3029 bool PassBuilder::parseAAPassName(AAManager &AA, StringRef Name) { 3030 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3031 if (Name == NAME) { \ 3032 AA.registerModuleAnalysis< \ 3033 std::remove_reference<decltype(CREATE_PASS)>::type>(); \ 3034 return true; \ 3035 } 3036 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3037 if (Name == NAME) { \ 3038 AA.registerFunctionAnalysis< \ 3039 std::remove_reference<decltype(CREATE_PASS)>::type>(); \ 3040 return true; \ 3041 } 3042 #include "PassRegistry.def" 3043 3044 for (auto &C : AAParsingCallbacks) 3045 if (C(Name, AA)) 3046 return true; 3047 return false; 3048 } 3049 3050 Error PassBuilder::parseLoopPassPipeline(LoopPassManager &LPM, 3051 ArrayRef<PipelineElement> Pipeline) { 3052 for (const auto &Element : Pipeline) { 3053 if (auto Err = parseLoopPass(LPM, Element)) 3054 return Err; 3055 } 3056 return Error::success(); 3057 } 3058 3059 Error PassBuilder::parseFunctionPassPipeline( 3060 FunctionPassManager &FPM, ArrayRef<PipelineElement> Pipeline) { 3061 for (const auto &Element : Pipeline) { 3062 if (auto Err = parseFunctionPass(FPM, Element)) 3063 return Err; 3064 } 3065 return Error::success(); 3066 } 3067 3068 Error PassBuilder::parseCGSCCPassPipeline(CGSCCPassManager &CGPM, 3069 ArrayRef<PipelineElement> Pipeline) { 3070 for (const auto &Element : Pipeline) { 3071 if (auto Err = parseCGSCCPass(CGPM, Element)) 3072 return Err; 3073 } 3074 return Error::success(); 3075 } 3076 3077 void PassBuilder::crossRegisterProxies(LoopAnalysisManager &LAM, 3078 FunctionAnalysisManager &FAM, 3079 CGSCCAnalysisManager &CGAM, 3080 ModuleAnalysisManager &MAM) { 3081 MAM.registerPass([&] { return FunctionAnalysisManagerModuleProxy(FAM); }); 3082 MAM.registerPass([&] { return CGSCCAnalysisManagerModuleProxy(CGAM); }); 3083 CGAM.registerPass([&] { return ModuleAnalysisManagerCGSCCProxy(MAM); }); 3084 FAM.registerPass([&] { return CGSCCAnalysisManagerFunctionProxy(CGAM); }); 3085 FAM.registerPass([&] { return ModuleAnalysisManagerFunctionProxy(MAM); }); 3086 FAM.registerPass([&] { return LoopAnalysisManagerFunctionProxy(LAM); }); 3087 LAM.registerPass([&] { return FunctionAnalysisManagerLoopProxy(FAM); }); 3088 } 3089 3090 Error PassBuilder::parseModulePassPipeline(ModulePassManager &MPM, 3091 ArrayRef<PipelineElement> Pipeline) { 3092 for (const auto &Element : Pipeline) { 3093 if (auto Err = parseModulePass(MPM, Element)) 3094 return Err; 3095 } 3096 return Error::success(); 3097 } 3098 3099 // Primary pass pipeline description parsing routine for a \c ModulePassManager 3100 // FIXME: Should this routine accept a TargetMachine or require the caller to 3101 // pre-populate the analysis managers with target-specific stuff? 3102 Error PassBuilder::parsePassPipeline(ModulePassManager &MPM, 3103 StringRef PipelineText) { 3104 auto Pipeline = parsePipelineText(PipelineText); 3105 if (!Pipeline || Pipeline->empty()) 3106 return make_error<StringError>( 3107 formatv("invalid pipeline '{0}'", PipelineText).str(), 3108 inconvertibleErrorCode()); 3109 3110 // If the first name isn't at the module layer, wrap the pipeline up 3111 // automatically. 3112 StringRef FirstName = Pipeline->front().Name; 3113 3114 if (!isModulePassName(FirstName, ModulePipelineParsingCallbacks)) { 3115 bool UseMemorySSA; 3116 if (isCGSCCPassName(FirstName, CGSCCPipelineParsingCallbacks)) { 3117 Pipeline = {{"cgscc", std::move(*Pipeline)}}; 3118 } else if (isFunctionPassName(FirstName, 3119 FunctionPipelineParsingCallbacks)) { 3120 Pipeline = {{"function", std::move(*Pipeline)}}; 3121 } else if (isLoopPassName(FirstName, LoopPipelineParsingCallbacks, 3122 UseMemorySSA)) { 3123 Pipeline = {{"function", {{UseMemorySSA ? "loop-mssa" : "loop", 3124 std::move(*Pipeline)}}}}; 3125 } else { 3126 for (auto &C : TopLevelPipelineParsingCallbacks) 3127 if (C(MPM, *Pipeline)) 3128 return Error::success(); 3129 3130 // Unknown pass or pipeline name! 3131 auto &InnerPipeline = Pipeline->front().InnerPipeline; 3132 return make_error<StringError>( 3133 formatv("unknown {0} name '{1}'", 3134 (InnerPipeline.empty() ? "pass" : "pipeline"), FirstName) 3135 .str(), 3136 inconvertibleErrorCode()); 3137 } 3138 } 3139 3140 if (auto Err = parseModulePassPipeline(MPM, *Pipeline)) 3141 return Err; 3142 return Error::success(); 3143 } 3144 3145 // Primary pass pipeline description parsing routine for a \c CGSCCPassManager 3146 Error PassBuilder::parsePassPipeline(CGSCCPassManager &CGPM, 3147 StringRef PipelineText) { 3148 auto Pipeline = parsePipelineText(PipelineText); 3149 if (!Pipeline || Pipeline->empty()) 3150 return make_error<StringError>( 3151 formatv("invalid pipeline '{0}'", PipelineText).str(), 3152 inconvertibleErrorCode()); 3153 3154 StringRef FirstName = Pipeline->front().Name; 3155 if (!isCGSCCPassName(FirstName, CGSCCPipelineParsingCallbacks)) 3156 return make_error<StringError>( 3157 formatv("unknown cgscc pass '{0}' in pipeline '{1}'", FirstName, 3158 PipelineText) 3159 .str(), 3160 inconvertibleErrorCode()); 3161 3162 if (auto Err = parseCGSCCPassPipeline(CGPM, *Pipeline)) 3163 return Err; 3164 return Error::success(); 3165 } 3166 3167 // Primary pass pipeline description parsing routine for a \c 3168 // FunctionPassManager 3169 Error PassBuilder::parsePassPipeline(FunctionPassManager &FPM, 3170 StringRef PipelineText) { 3171 auto Pipeline = parsePipelineText(PipelineText); 3172 if (!Pipeline || Pipeline->empty()) 3173 return make_error<StringError>( 3174 formatv("invalid pipeline '{0}'", PipelineText).str(), 3175 inconvertibleErrorCode()); 3176 3177 StringRef FirstName = Pipeline->front().Name; 3178 if (!isFunctionPassName(FirstName, FunctionPipelineParsingCallbacks)) 3179 return make_error<StringError>( 3180 formatv("unknown function pass '{0}' in pipeline '{1}'", FirstName, 3181 PipelineText) 3182 .str(), 3183 inconvertibleErrorCode()); 3184 3185 if (auto Err = parseFunctionPassPipeline(FPM, *Pipeline)) 3186 return Err; 3187 return Error::success(); 3188 } 3189 3190 // Primary pass pipeline description parsing routine for a \c LoopPassManager 3191 Error PassBuilder::parsePassPipeline(LoopPassManager &CGPM, 3192 StringRef PipelineText) { 3193 auto Pipeline = parsePipelineText(PipelineText); 3194 if (!Pipeline || Pipeline->empty()) 3195 return make_error<StringError>( 3196 formatv("invalid pipeline '{0}'", PipelineText).str(), 3197 inconvertibleErrorCode()); 3198 3199 if (auto Err = parseLoopPassPipeline(CGPM, *Pipeline)) 3200 return Err; 3201 3202 return Error::success(); 3203 } 3204 3205 Error PassBuilder::parseAAPipeline(AAManager &AA, StringRef PipelineText) { 3206 // If the pipeline just consists of the word 'default' just replace the AA 3207 // manager with our default one. 3208 if (PipelineText == "default") { 3209 AA = buildDefaultAAPipeline(); 3210 return Error::success(); 3211 } 3212 3213 while (!PipelineText.empty()) { 3214 StringRef Name; 3215 std::tie(Name, PipelineText) = PipelineText.split(','); 3216 if (!parseAAPassName(AA, Name)) 3217 return make_error<StringError>( 3218 formatv("unknown alias analysis name '{0}'", Name).str(), 3219 inconvertibleErrorCode()); 3220 } 3221 3222 return Error::success(); 3223 } 3224 3225 bool PassBuilder::isAAPassName(StringRef PassName) { 3226 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3227 if (PassName == NAME) \ 3228 return true; 3229 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3230 if (PassName == NAME) \ 3231 return true; 3232 #include "PassRegistry.def" 3233 return false; 3234 } 3235 3236 bool PassBuilder::isAnalysisPassName(StringRef PassName) { 3237 #define MODULE_ANALYSIS(NAME, CREATE_PASS) \ 3238 if (PassName == NAME) \ 3239 return true; 3240 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \ 3241 if (PassName == NAME) \ 3242 return true; 3243 #define LOOP_ANALYSIS(NAME, CREATE_PASS) \ 3244 if (PassName == NAME) \ 3245 return true; 3246 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) \ 3247 if (PassName == NAME) \ 3248 return true; 3249 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3250 if (PassName == NAME) \ 3251 return true; 3252 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) \ 3253 if (PassName == NAME) \ 3254 return true; 3255 #include "PassRegistry.def" 3256 return false; 3257 } 3258 3259 static void printPassName(StringRef PassName, raw_ostream &OS) { 3260 OS << " " << PassName << "\n"; 3261 } 3262 static void printPassName(StringRef PassName, StringRef Params, 3263 raw_ostream &OS) { 3264 OS << " " << PassName << "<" << Params << ">\n"; 3265 } 3266 3267 void PassBuilder::printPassNames(raw_ostream &OS) { 3268 // TODO: print pass descriptions when they are available 3269 3270 OS << "Module passes:\n"; 3271 #define MODULE_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3272 #include "PassRegistry.def" 3273 3274 OS << "Module passes with params:\n"; 3275 #define MODULE_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 3276 printPassName(NAME, PARAMS, OS); 3277 #include "PassRegistry.def" 3278 3279 OS << "Module analyses:\n"; 3280 #define MODULE_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3281 #include "PassRegistry.def" 3282 3283 OS << "Module alias analyses:\n"; 3284 #define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3285 #include "PassRegistry.def" 3286 3287 OS << "CGSCC passes:\n"; 3288 #define CGSCC_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3289 #include "PassRegistry.def" 3290 3291 OS << "CGSCC analyses:\n"; 3292 #define CGSCC_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3293 #include "PassRegistry.def" 3294 3295 OS << "Function passes:\n"; 3296 #define FUNCTION_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3297 #include "PassRegistry.def" 3298 3299 OS << "Function passes with params:\n"; 3300 #define FUNCTION_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 3301 printPassName(NAME, PARAMS, OS); 3302 #include "PassRegistry.def" 3303 3304 OS << "Function analyses:\n"; 3305 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3306 #include "PassRegistry.def" 3307 3308 OS << "Function alias analyses:\n"; 3309 #define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3310 #include "PassRegistry.def" 3311 3312 OS << "Loop passes:\n"; 3313 #define LOOP_PASS(NAME, CREATE_PASS) printPassName(NAME, OS); 3314 #include "PassRegistry.def" 3315 3316 OS << "Loop passes with params:\n"; 3317 #define LOOP_PASS_WITH_PARAMS(NAME, CLASS, CREATE_PASS, PARSER, PARAMS) \ 3318 printPassName(NAME, PARAMS, OS); 3319 #include "PassRegistry.def" 3320 3321 OS << "Loop analyses:\n"; 3322 #define LOOP_ANALYSIS(NAME, CREATE_PASS) printPassName(NAME, OS); 3323 #include "PassRegistry.def" 3324 } 3325 3326 void PassBuilder::registerParseTopLevelPipelineCallback( 3327 const std::function<bool(ModulePassManager &, ArrayRef<PipelineElement>)> 3328 &C) { 3329 TopLevelPipelineParsingCallbacks.push_back(C); 3330 } 3331