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