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