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