| File: | src/gnu/usr.bin/clang/libLLVM/../../../llvm/llvm/lib/Transforms/Utils/SimplifyCFG.cpp |
| Warning: | line 4326, column 24 Called C++ object pointer is null |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
| 1 | //===- SimplifyCFG.cpp - Code to perform CFG simplification ---------------===// | |||
| 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 | // | |||
| 9 | // Peephole optimize the CFG. | |||
| 10 | // | |||
| 11 | //===----------------------------------------------------------------------===// | |||
| 12 | ||||
| 13 | #include "llvm/ADT/APInt.h" | |||
| 14 | #include "llvm/ADT/ArrayRef.h" | |||
| 15 | #include "llvm/ADT/DenseMap.h" | |||
| 16 | #include "llvm/ADT/MapVector.h" | |||
| 17 | #include "llvm/ADT/Optional.h" | |||
| 18 | #include "llvm/ADT/STLExtras.h" | |||
| 19 | #include "llvm/ADT/ScopeExit.h" | |||
| 20 | #include "llvm/ADT/Sequence.h" | |||
| 21 | #include "llvm/ADT/SetOperations.h" | |||
| 22 | #include "llvm/ADT/SetVector.h" | |||
| 23 | #include "llvm/ADT/SmallPtrSet.h" | |||
| 24 | #include "llvm/ADT/SmallVector.h" | |||
| 25 | #include "llvm/ADT/Statistic.h" | |||
| 26 | #include "llvm/ADT/StringRef.h" | |||
| 27 | #include "llvm/Analysis/AssumptionCache.h" | |||
| 28 | #include "llvm/Analysis/ConstantFolding.h" | |||
| 29 | #include "llvm/Analysis/EHPersonalities.h" | |||
| 30 | #include "llvm/Analysis/GuardUtils.h" | |||
| 31 | #include "llvm/Analysis/InstructionSimplify.h" | |||
| 32 | #include "llvm/Analysis/MemorySSA.h" | |||
| 33 | #include "llvm/Analysis/MemorySSAUpdater.h" | |||
| 34 | #include "llvm/Analysis/TargetTransformInfo.h" | |||
| 35 | #include "llvm/Analysis/ValueTracking.h" | |||
| 36 | #include "llvm/IR/Attributes.h" | |||
| 37 | #include "llvm/IR/BasicBlock.h" | |||
| 38 | #include "llvm/IR/CFG.h" | |||
| 39 | #include "llvm/IR/Constant.h" | |||
| 40 | #include "llvm/IR/ConstantRange.h" | |||
| 41 | #include "llvm/IR/Constants.h" | |||
| 42 | #include "llvm/IR/DataLayout.h" | |||
| 43 | #include "llvm/IR/DerivedTypes.h" | |||
| 44 | #include "llvm/IR/Function.h" | |||
| 45 | #include "llvm/IR/GlobalValue.h" | |||
| 46 | #include "llvm/IR/GlobalVariable.h" | |||
| 47 | #include "llvm/IR/IRBuilder.h" | |||
| 48 | #include "llvm/IR/InstrTypes.h" | |||
| 49 | #include "llvm/IR/Instruction.h" | |||
| 50 | #include "llvm/IR/Instructions.h" | |||
| 51 | #include "llvm/IR/IntrinsicInst.h" | |||
| 52 | #include "llvm/IR/Intrinsics.h" | |||
| 53 | #include "llvm/IR/LLVMContext.h" | |||
| 54 | #include "llvm/IR/MDBuilder.h" | |||
| 55 | #include "llvm/IR/Metadata.h" | |||
| 56 | #include "llvm/IR/Module.h" | |||
| 57 | #include "llvm/IR/NoFolder.h" | |||
| 58 | #include "llvm/IR/Operator.h" | |||
| 59 | #include "llvm/IR/PatternMatch.h" | |||
| 60 | #include "llvm/IR/PseudoProbe.h" | |||
| 61 | #include "llvm/IR/Type.h" | |||
| 62 | #include "llvm/IR/Use.h" | |||
| 63 | #include "llvm/IR/User.h" | |||
| 64 | #include "llvm/IR/Value.h" | |||
| 65 | #include "llvm/IR/ValueHandle.h" | |||
| 66 | #include "llvm/Support/BranchProbability.h" | |||
| 67 | #include "llvm/Support/Casting.h" | |||
| 68 | #include "llvm/Support/CommandLine.h" | |||
| 69 | #include "llvm/Support/Debug.h" | |||
| 70 | #include "llvm/Support/ErrorHandling.h" | |||
| 71 | #include "llvm/Support/KnownBits.h" | |||
| 72 | #include "llvm/Support/MathExtras.h" | |||
| 73 | #include "llvm/Support/raw_ostream.h" | |||
| 74 | #include "llvm/Transforms/Utils/BasicBlockUtils.h" | |||
| 75 | #include "llvm/Transforms/Utils/Local.h" | |||
| 76 | #include "llvm/Transforms/Utils/SSAUpdater.h" | |||
| 77 | #include "llvm/Transforms/Utils/ValueMapper.h" | |||
| 78 | #include <algorithm> | |||
| 79 | #include <cassert> | |||
| 80 | #include <climits> | |||
| 81 | #include <cstddef> | |||
| 82 | #include <cstdint> | |||
| 83 | #include <iterator> | |||
| 84 | #include <map> | |||
| 85 | #include <set> | |||
| 86 | #include <tuple> | |||
| 87 | #include <utility> | |||
| 88 | #include <vector> | |||
| 89 | ||||
| 90 | using namespace llvm; | |||
| 91 | using namespace PatternMatch; | |||
| 92 | ||||
| 93 | #define DEBUG_TYPE"simplifycfg" "simplifycfg" | |||
| 94 | ||||
| 95 | cl::opt<bool> llvm::RequireAndPreserveDomTree( | |||
| 96 | "simplifycfg-require-and-preserve-domtree", cl::Hidden, cl::ZeroOrMore, | |||
| 97 | cl::init(false), | |||
| 98 | cl::desc("Temorary development switch used to gradually uplift SimplifyCFG " | |||
| 99 | "into preserving DomTree,")); | |||
| 100 | ||||
| 101 | // Chosen as 2 so as to be cheap, but still to have enough power to fold | |||
| 102 | // a select, so the "clamp" idiom (of a min followed by a max) will be caught. | |||
| 103 | // To catch this, we need to fold a compare and a select, hence '2' being the | |||
| 104 | // minimum reasonable default. | |||
| 105 | static cl::opt<unsigned> PHINodeFoldingThreshold( | |||
| 106 | "phi-node-folding-threshold", cl::Hidden, cl::init(2), | |||
| 107 | cl::desc( | |||
| 108 | "Control the amount of phi node folding to perform (default = 2)")); | |||
| 109 | ||||
| 110 | static cl::opt<unsigned> TwoEntryPHINodeFoldingThreshold( | |||
| 111 | "two-entry-phi-node-folding-threshold", cl::Hidden, cl::init(4), | |||
| 112 | cl::desc("Control the maximal total instruction cost that we are willing " | |||
| 113 | "to speculatively execute to fold a 2-entry PHI node into a " | |||
| 114 | "select (default = 4)")); | |||
| 115 | ||||
| 116 | static cl::opt<bool> | |||
| 117 | HoistCommon("simplifycfg-hoist-common", cl::Hidden, cl::init(true), | |||
| 118 | cl::desc("Hoist common instructions up to the parent block")); | |||
| 119 | ||||
| 120 | static cl::opt<bool> | |||
| 121 | SinkCommon("simplifycfg-sink-common", cl::Hidden, cl::init(true), | |||
| 122 | cl::desc("Sink common instructions down to the end block")); | |||
| 123 | ||||
| 124 | static cl::opt<bool> HoistCondStores( | |||
| 125 | "simplifycfg-hoist-cond-stores", cl::Hidden, cl::init(true), | |||
| 126 | cl::desc("Hoist conditional stores if an unconditional store precedes")); | |||
| 127 | ||||
| 128 | static cl::opt<bool> MergeCondStores( | |||
| 129 | "simplifycfg-merge-cond-stores", cl::Hidden, cl::init(true), | |||
| 130 | cl::desc("Hoist conditional stores even if an unconditional store does not " | |||
| 131 | "precede - hoist multiple conditional stores into a single " | |||
| 132 | "predicated store")); | |||
| 133 | ||||
| 134 | static cl::opt<bool> MergeCondStoresAggressively( | |||
| 135 | "simplifycfg-merge-cond-stores-aggressively", cl::Hidden, cl::init(false), | |||
| 136 | cl::desc("When merging conditional stores, do so even if the resultant " | |||
| 137 | "basic blocks are unlikely to be if-converted as a result")); | |||
| 138 | ||||
| 139 | static cl::opt<bool> SpeculateOneExpensiveInst( | |||
| 140 | "speculate-one-expensive-inst", cl::Hidden, cl::init(true), | |||
| 141 | cl::desc("Allow exactly one expensive instruction to be speculatively " | |||
| 142 | "executed")); | |||
| 143 | ||||
| 144 | static cl::opt<unsigned> MaxSpeculationDepth( | |||
| 145 | "max-speculation-depth", cl::Hidden, cl::init(10), | |||
| 146 | cl::desc("Limit maximum recursion depth when calculating costs of " | |||
| 147 | "speculatively executed instructions")); | |||
| 148 | ||||
| 149 | static cl::opt<int> | |||
| 150 | MaxSmallBlockSize("simplifycfg-max-small-block-size", cl::Hidden, | |||
| 151 | cl::init(10), | |||
| 152 | cl::desc("Max size of a block which is still considered " | |||
| 153 | "small enough to thread through")); | |||
| 154 | ||||
| 155 | // Two is chosen to allow one negation and a logical combine. | |||
| 156 | static cl::opt<unsigned> | |||
| 157 | BranchFoldThreshold("simplifycfg-branch-fold-threshold", cl::Hidden, | |||
| 158 | cl::init(2), | |||
| 159 | cl::desc("Maximum cost of combining conditions when " | |||
| 160 | "folding branches")); | |||
| 161 | ||||
| 162 | STATISTIC(NumBitMaps, "Number of switch instructions turned into bitmaps")static llvm::Statistic NumBitMaps = {"simplifycfg", "NumBitMaps" , "Number of switch instructions turned into bitmaps"}; | |||
| 163 | STATISTIC(NumLinearMaps,static llvm::Statistic NumLinearMaps = {"simplifycfg", "NumLinearMaps" , "Number of switch instructions turned into linear mapping"} | |||
| 164 | "Number of switch instructions turned into linear mapping")static llvm::Statistic NumLinearMaps = {"simplifycfg", "NumLinearMaps" , "Number of switch instructions turned into linear mapping"}; | |||
| 165 | STATISTIC(NumLookupTables,static llvm::Statistic NumLookupTables = {"simplifycfg", "NumLookupTables" , "Number of switch instructions turned into lookup tables"} | |||
| 166 | "Number of switch instructions turned into lookup tables")static llvm::Statistic NumLookupTables = {"simplifycfg", "NumLookupTables" , "Number of switch instructions turned into lookup tables"}; | |||
| 167 | STATISTIC(static llvm::Statistic NumLookupTablesHoles = {"simplifycfg", "NumLookupTablesHoles", "Number of switch instructions turned into lookup tables (holes checked)" } | |||
| 168 | NumLookupTablesHoles,static llvm::Statistic NumLookupTablesHoles = {"simplifycfg", "NumLookupTablesHoles", "Number of switch instructions turned into lookup tables (holes checked)" } | |||
| 169 | "Number of switch instructions turned into lookup tables (holes checked)")static llvm::Statistic NumLookupTablesHoles = {"simplifycfg", "NumLookupTablesHoles", "Number of switch instructions turned into lookup tables (holes checked)" }; | |||
| 170 | STATISTIC(NumTableCmpReuses, "Number of reused switch table lookup compares")static llvm::Statistic NumTableCmpReuses = {"simplifycfg", "NumTableCmpReuses" , "Number of reused switch table lookup compares"}; | |||
| 171 | STATISTIC(NumFoldValueComparisonIntoPredecessors,static llvm::Statistic NumFoldValueComparisonIntoPredecessors = {"simplifycfg", "NumFoldValueComparisonIntoPredecessors", "Number of value comparisons folded into predecessor basic blocks" } | |||
| 172 | "Number of value comparisons folded into predecessor basic blocks")static llvm::Statistic NumFoldValueComparisonIntoPredecessors = {"simplifycfg", "NumFoldValueComparisonIntoPredecessors", "Number of value comparisons folded into predecessor basic blocks" }; | |||
| 173 | STATISTIC(NumFoldBranchToCommonDest,static llvm::Statistic NumFoldBranchToCommonDest = {"simplifycfg" , "NumFoldBranchToCommonDest", "Number of branches folded into predecessor basic block" } | |||
| 174 | "Number of branches folded into predecessor basic block")static llvm::Statistic NumFoldBranchToCommonDest = {"simplifycfg" , "NumFoldBranchToCommonDest", "Number of branches folded into predecessor basic block" }; | |||
| 175 | STATISTIC(static llvm::Statistic NumHoistCommonCode = {"simplifycfg", "NumHoistCommonCode" , "Number of common instruction 'blocks' hoisted up to the begin block" } | |||
| 176 | NumHoistCommonCode,static llvm::Statistic NumHoistCommonCode = {"simplifycfg", "NumHoistCommonCode" , "Number of common instruction 'blocks' hoisted up to the begin block" } | |||
| 177 | "Number of common instruction 'blocks' hoisted up to the begin block")static llvm::Statistic NumHoistCommonCode = {"simplifycfg", "NumHoistCommonCode" , "Number of common instruction 'blocks' hoisted up to the begin block" }; | |||
| 178 | STATISTIC(NumHoistCommonInstrs,static llvm::Statistic NumHoistCommonInstrs = {"simplifycfg", "NumHoistCommonInstrs", "Number of common instructions hoisted up to the begin block" } | |||
| 179 | "Number of common instructions hoisted up to the begin block")static llvm::Statistic NumHoistCommonInstrs = {"simplifycfg", "NumHoistCommonInstrs", "Number of common instructions hoisted up to the begin block" }; | |||
| 180 | STATISTIC(NumSinkCommonCode,static llvm::Statistic NumSinkCommonCode = {"simplifycfg", "NumSinkCommonCode" , "Number of common instruction 'blocks' sunk down to the end block" } | |||
| 181 | "Number of common instruction 'blocks' sunk down to the end block")static llvm::Statistic NumSinkCommonCode = {"simplifycfg", "NumSinkCommonCode" , "Number of common instruction 'blocks' sunk down to the end block" }; | |||
| 182 | STATISTIC(NumSinkCommonInstrs,static llvm::Statistic NumSinkCommonInstrs = {"simplifycfg", "NumSinkCommonInstrs" , "Number of common instructions sunk down to the end block"} | |||
| 183 | "Number of common instructions sunk down to the end block")static llvm::Statistic NumSinkCommonInstrs = {"simplifycfg", "NumSinkCommonInstrs" , "Number of common instructions sunk down to the end block"}; | |||
| 184 | STATISTIC(NumSpeculations, "Number of speculative executed instructions")static llvm::Statistic NumSpeculations = {"simplifycfg", "NumSpeculations" , "Number of speculative executed instructions"}; | |||
| 185 | STATISTIC(NumInvokes,static llvm::Statistic NumInvokes = {"simplifycfg", "NumInvokes" , "Number of invokes with empty resume blocks simplified into calls" } | |||
| 186 | "Number of invokes with empty resume blocks simplified into calls")static llvm::Statistic NumInvokes = {"simplifycfg", "NumInvokes" , "Number of invokes with empty resume blocks simplified into calls" }; | |||
| 187 | ||||
| 188 | namespace { | |||
| 189 | ||||
| 190 | // The first field contains the value that the switch produces when a certain | |||
| 191 | // case group is selected, and the second field is a vector containing the | |||
| 192 | // cases composing the case group. | |||
| 193 | using SwitchCaseResultVectorTy = | |||
| 194 | SmallVector<std::pair<Constant *, SmallVector<ConstantInt *, 4>>, 2>; | |||
| 195 | ||||
| 196 | // The first field contains the phi node that generates a result of the switch | |||
| 197 | // and the second field contains the value generated for a certain case in the | |||
| 198 | // switch for that PHI. | |||
| 199 | using SwitchCaseResultsTy = SmallVector<std::pair<PHINode *, Constant *>, 4>; | |||
| 200 | ||||
| 201 | /// ValueEqualityComparisonCase - Represents a case of a switch. | |||
| 202 | struct ValueEqualityComparisonCase { | |||
| 203 | ConstantInt *Value; | |||
| 204 | BasicBlock *Dest; | |||
| 205 | ||||
| 206 | ValueEqualityComparisonCase(ConstantInt *Value, BasicBlock *Dest) | |||
| 207 | : Value(Value), Dest(Dest) {} | |||
| 208 | ||||
| 209 | bool operator<(ValueEqualityComparisonCase RHS) const { | |||
| 210 | // Comparing pointers is ok as we only rely on the order for uniquing. | |||
| 211 | return Value < RHS.Value; | |||
| 212 | } | |||
| 213 | ||||
| 214 | bool operator==(BasicBlock *RHSDest) const { return Dest == RHSDest; } | |||
| 215 | }; | |||
| 216 | ||||
| 217 | class SimplifyCFGOpt { | |||
| 218 | const TargetTransformInfo &TTI; | |||
| 219 | DomTreeUpdater *DTU; | |||
| 220 | const DataLayout &DL; | |||
| 221 | ArrayRef<WeakVH> LoopHeaders; | |||
| 222 | const SimplifyCFGOptions &Options; | |||
| 223 | bool Resimplify; | |||
| 224 | ||||
| 225 | Value *isValueEqualityComparison(Instruction *TI); | |||
| 226 | BasicBlock *GetValueEqualityComparisonCases( | |||
| 227 | Instruction *TI, std::vector<ValueEqualityComparisonCase> &Cases); | |||
| 228 | bool SimplifyEqualityComparisonWithOnlyPredecessor(Instruction *TI, | |||
| 229 | BasicBlock *Pred, | |||
| 230 | IRBuilder<> &Builder); | |||
| 231 | bool PerformValueComparisonIntoPredecessorFolding(Instruction *TI, Value *&CV, | |||
| 232 | Instruction *PTI, | |||
| 233 | IRBuilder<> &Builder); | |||
| 234 | bool FoldValueComparisonIntoPredecessors(Instruction *TI, | |||
| 235 | IRBuilder<> &Builder); | |||
| 236 | ||||
| 237 | bool simplifyResume(ResumeInst *RI, IRBuilder<> &Builder); | |||
| 238 | bool simplifySingleResume(ResumeInst *RI); | |||
| 239 | bool simplifyCommonResume(ResumeInst *RI); | |||
| 240 | bool simplifyCleanupReturn(CleanupReturnInst *RI); | |||
| 241 | bool simplifyUnreachable(UnreachableInst *UI); | |||
| 242 | bool simplifySwitch(SwitchInst *SI, IRBuilder<> &Builder); | |||
| 243 | bool simplifyIndirectBr(IndirectBrInst *IBI); | |||
| 244 | bool simplifyBranch(BranchInst *Branch, IRBuilder<> &Builder); | |||
| 245 | bool simplifyUncondBranch(BranchInst *BI, IRBuilder<> &Builder); | |||
| 246 | bool simplifyCondBranch(BranchInst *BI, IRBuilder<> &Builder); | |||
| 247 | ||||
| 248 | bool tryToSimplifyUncondBranchWithICmpInIt(ICmpInst *ICI, | |||
| 249 | IRBuilder<> &Builder); | |||
| 250 | ||||
| 251 | bool HoistThenElseCodeToIf(BranchInst *BI, const TargetTransformInfo &TTI, | |||
| 252 | bool EqTermsOnly); | |||
| 253 | bool SpeculativelyExecuteBB(BranchInst *BI, BasicBlock *ThenBB, | |||
| 254 | const TargetTransformInfo &TTI); | |||
| 255 | bool SimplifyTerminatorOnSelect(Instruction *OldTerm, Value *Cond, | |||
| 256 | BasicBlock *TrueBB, BasicBlock *FalseBB, | |||
| 257 | uint32_t TrueWeight, uint32_t FalseWeight); | |||
| 258 | bool SimplifyBranchOnICmpChain(BranchInst *BI, IRBuilder<> &Builder, | |||
| 259 | const DataLayout &DL); | |||
| 260 | bool SimplifySwitchOnSelect(SwitchInst *SI, SelectInst *Select); | |||
| 261 | bool SimplifyIndirectBrOnSelect(IndirectBrInst *IBI, SelectInst *SI); | |||
| 262 | bool TurnSwitchRangeIntoICmp(SwitchInst *SI, IRBuilder<> &Builder); | |||
| 263 | ||||
| 264 | public: | |||
| 265 | SimplifyCFGOpt(const TargetTransformInfo &TTI, DomTreeUpdater *DTU, | |||
| 266 | const DataLayout &DL, ArrayRef<WeakVH> LoopHeaders, | |||
| 267 | const SimplifyCFGOptions &Opts) | |||
| 268 | : TTI(TTI), DTU(DTU), DL(DL), LoopHeaders(LoopHeaders), Options(Opts) { | |||
| 269 | assert((!DTU || !DTU->hasPostDomTree()) &&((void)0) | |||
| 270 | "SimplifyCFG is not yet capable of maintaining validity of a "((void)0) | |||
| 271 | "PostDomTree, so don't ask for it.")((void)0); | |||
| 272 | } | |||
| 273 | ||||
| 274 | bool simplifyOnce(BasicBlock *BB); | |||
| 275 | bool simplifyOnceImpl(BasicBlock *BB); | |||
| 276 | bool run(BasicBlock *BB); | |||
| 277 | ||||
| 278 | // Helper to set Resimplify and return change indication. | |||
| 279 | bool requestResimplify() { | |||
| 280 | Resimplify = true; | |||
| 281 | return true; | |||
| 282 | } | |||
| 283 | }; | |||
| 284 | ||||
| 285 | } // end anonymous namespace | |||
| 286 | ||||
| 287 | /// Return true if it is safe to merge these two | |||
| 288 | /// terminator instructions together. | |||
| 289 | static bool | |||
| 290 | SafeToMergeTerminators(Instruction *SI1, Instruction *SI2, | |||
| 291 | SmallSetVector<BasicBlock *, 4> *FailBlocks = nullptr) { | |||
| 292 | if (SI1 == SI2) | |||
| 293 | return false; // Can't merge with self! | |||
| 294 | ||||
| 295 | // It is not safe to merge these two switch instructions if they have a common | |||
| 296 | // successor, and if that successor has a PHI node, and if *that* PHI node has | |||
| 297 | // conflicting incoming values from the two switch blocks. | |||
| 298 | BasicBlock *SI1BB = SI1->getParent(); | |||
| 299 | BasicBlock *SI2BB = SI2->getParent(); | |||
| 300 | ||||
| 301 | SmallPtrSet<BasicBlock *, 16> SI1Succs(succ_begin(SI1BB), succ_end(SI1BB)); | |||
| 302 | bool Fail = false; | |||
| 303 | for (BasicBlock *Succ : successors(SI2BB)) | |||
| 304 | if (SI1Succs.count(Succ)) | |||
| 305 | for (BasicBlock::iterator BBI = Succ->begin(); isa<PHINode>(BBI); ++BBI) { | |||
| 306 | PHINode *PN = cast<PHINode>(BBI); | |||
| 307 | if (PN->getIncomingValueForBlock(SI1BB) != | |||
| 308 | PN->getIncomingValueForBlock(SI2BB)) { | |||
| 309 | if (FailBlocks) | |||
| 310 | FailBlocks->insert(Succ); | |||
| 311 | Fail = true; | |||
| 312 | } | |||
| 313 | } | |||
| 314 | ||||
| 315 | return !Fail; | |||
| 316 | } | |||
| 317 | ||||
| 318 | /// Update PHI nodes in Succ to indicate that there will now be entries in it | |||
| 319 | /// from the 'NewPred' block. The values that will be flowing into the PHI nodes | |||
| 320 | /// will be the same as those coming in from ExistPred, an existing predecessor | |||
| 321 | /// of Succ. | |||
| 322 | static void AddPredecessorToBlock(BasicBlock *Succ, BasicBlock *NewPred, | |||
| 323 | BasicBlock *ExistPred, | |||
| 324 | MemorySSAUpdater *MSSAU = nullptr) { | |||
| 325 | for (PHINode &PN : Succ->phis()) | |||
| 326 | PN.addIncoming(PN.getIncomingValueForBlock(ExistPred), NewPred); | |||
| 327 | if (MSSAU) | |||
| 328 | if (auto *MPhi = MSSAU->getMemorySSA()->getMemoryAccess(Succ)) | |||
| 329 | MPhi->addIncoming(MPhi->getIncomingValueForBlock(ExistPred), NewPred); | |||
| 330 | } | |||
| 331 | ||||
| 332 | /// Compute an abstract "cost" of speculating the given instruction, | |||
| 333 | /// which is assumed to be safe to speculate. TCC_Free means cheap, | |||
| 334 | /// TCC_Basic means less cheap, and TCC_Expensive means prohibitively | |||
| 335 | /// expensive. | |||
| 336 | static InstructionCost computeSpeculationCost(const User *I, | |||
| 337 | const TargetTransformInfo &TTI) { | |||
| 338 | assert(isSafeToSpeculativelyExecute(I) &&((void)0) | |||
| 339 | "Instruction is not safe to speculatively execute!")((void)0); | |||
| 340 | return TTI.getUserCost(I, TargetTransformInfo::TCK_SizeAndLatency); | |||
| 341 | } | |||
| 342 | ||||
| 343 | /// If we have a merge point of an "if condition" as accepted above, | |||
| 344 | /// return true if the specified value dominates the block. We | |||
| 345 | /// don't handle the true generality of domination here, just a special case | |||
| 346 | /// which works well enough for us. | |||
| 347 | /// | |||
| 348 | /// If AggressiveInsts is non-null, and if V does not dominate BB, we check to | |||
| 349 | /// see if V (which must be an instruction) and its recursive operands | |||
| 350 | /// that do not dominate BB have a combined cost lower than Budget and | |||
| 351 | /// are non-trapping. If both are true, the instruction is inserted into the | |||
| 352 | /// set and true is returned. | |||
| 353 | /// | |||
| 354 | /// The cost for most non-trapping instructions is defined as 1 except for | |||
| 355 | /// Select whose cost is 2. | |||
| 356 | /// | |||
| 357 | /// After this function returns, Cost is increased by the cost of | |||
| 358 | /// V plus its non-dominating operands. If that cost is greater than | |||
| 359 | /// Budget, false is returned and Cost is undefined. | |||
| 360 | static bool dominatesMergePoint(Value *V, BasicBlock *BB, | |||
| 361 | SmallPtrSetImpl<Instruction *> &AggressiveInsts, | |||
| 362 | InstructionCost &Cost, | |||
| 363 | InstructionCost Budget, | |||
| 364 | const TargetTransformInfo &TTI, | |||
| 365 | unsigned Depth = 0) { | |||
| 366 | // It is possible to hit a zero-cost cycle (phi/gep instructions for example), | |||
| 367 | // so limit the recursion depth. | |||
| 368 | // TODO: While this recursion limit does prevent pathological behavior, it | |||
| 369 | // would be better to track visited instructions to avoid cycles. | |||
| 370 | if (Depth == MaxSpeculationDepth) | |||
| 371 | return false; | |||
| 372 | ||||
| 373 | Instruction *I = dyn_cast<Instruction>(V); | |||
| 374 | if (!I) { | |||
| 375 | // Non-instructions all dominate instructions, but not all constantexprs | |||
| 376 | // can be executed unconditionally. | |||
| 377 | if (ConstantExpr *C = dyn_cast<ConstantExpr>(V)) | |||
| 378 | if (C->canTrap()) | |||
| 379 | return false; | |||
| 380 | return true; | |||
| 381 | } | |||
| 382 | BasicBlock *PBB = I->getParent(); | |||
| 383 | ||||
| 384 | // We don't want to allow weird loops that might have the "if condition" in | |||
| 385 | // the bottom of this block. | |||
| 386 | if (PBB == BB) | |||
| 387 | return false; | |||
| 388 | ||||
| 389 | // If this instruction is defined in a block that contains an unconditional | |||
| 390 | // branch to BB, then it must be in the 'conditional' part of the "if | |||
| 391 | // statement". If not, it definitely dominates the region. | |||
| 392 | BranchInst *BI = dyn_cast<BranchInst>(PBB->getTerminator()); | |||
| 393 | if (!BI || BI->isConditional() || BI->getSuccessor(0) != BB) | |||
| 394 | return true; | |||
| 395 | ||||
| 396 | // If we have seen this instruction before, don't count it again. | |||
| 397 | if (AggressiveInsts.count(I)) | |||
| 398 | return true; | |||
| 399 | ||||
| 400 | // Okay, it looks like the instruction IS in the "condition". Check to | |||
| 401 | // see if it's a cheap instruction to unconditionally compute, and if it | |||
| 402 | // only uses stuff defined outside of the condition. If so, hoist it out. | |||
| 403 | if (!isSafeToSpeculativelyExecute(I)) | |||
| 404 | return false; | |||
| 405 | ||||
| 406 | Cost += computeSpeculationCost(I, TTI); | |||
| 407 | ||||
| 408 | // Allow exactly one instruction to be speculated regardless of its cost | |||
| 409 | // (as long as it is safe to do so). | |||
| 410 | // This is intended to flatten the CFG even if the instruction is a division | |||
| 411 | // or other expensive operation. The speculation of an expensive instruction | |||
| 412 | // is expected to be undone in CodeGenPrepare if the speculation has not | |||
| 413 | // enabled further IR optimizations. | |||
| 414 | if (Cost > Budget && | |||
| 415 | (!SpeculateOneExpensiveInst || !AggressiveInsts.empty() || Depth > 0 || | |||
| 416 | !Cost.isValid())) | |||
| 417 | return false; | |||
| 418 | ||||
| 419 | // Okay, we can only really hoist these out if their operands do | |||
| 420 | // not take us over the cost threshold. | |||
| 421 | for (Use &Op : I->operands()) | |||
| 422 | if (!dominatesMergePoint(Op, BB, AggressiveInsts, Cost, Budget, TTI, | |||
| 423 | Depth + 1)) | |||
| 424 | return false; | |||
| 425 | // Okay, it's safe to do this! Remember this instruction. | |||
| 426 | AggressiveInsts.insert(I); | |||
| 427 | return true; | |||
| 428 | } | |||
| 429 | ||||
| 430 | /// Extract ConstantInt from value, looking through IntToPtr | |||
| 431 | /// and PointerNullValue. Return NULL if value is not a constant int. | |||
| 432 | static ConstantInt *GetConstantInt(Value *V, const DataLayout &DL) { | |||
| 433 | // Normal constant int. | |||
| 434 | ConstantInt *CI = dyn_cast<ConstantInt>(V); | |||
| 435 | if (CI || !isa<Constant>(V) || !V->getType()->isPointerTy()) | |||
| 436 | return CI; | |||
| 437 | ||||
| 438 | // This is some kind of pointer constant. Turn it into a pointer-sized | |||
| 439 | // ConstantInt if possible. | |||
| 440 | IntegerType *PtrTy = cast<IntegerType>(DL.getIntPtrType(V->getType())); | |||
| 441 | ||||
| 442 | // Null pointer means 0, see SelectionDAGBuilder::getValue(const Value*). | |||
| 443 | if (isa<ConstantPointerNull>(V)) | |||
| 444 | return ConstantInt::get(PtrTy, 0); | |||
| 445 | ||||
| 446 | // IntToPtr const int. | |||
| 447 | if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V)) | |||
| 448 | if (CE->getOpcode() == Instruction::IntToPtr) | |||
| 449 | if (ConstantInt *CI = dyn_cast<ConstantInt>(CE->getOperand(0))) { | |||
| 450 | // The constant is very likely to have the right type already. | |||
| 451 | if (CI->getType() == PtrTy) | |||
| 452 | return CI; | |||
| 453 | else | |||
| 454 | return cast<ConstantInt>( | |||
| 455 | ConstantExpr::getIntegerCast(CI, PtrTy, /*isSigned=*/false)); | |||
| 456 | } | |||
| 457 | return nullptr; | |||
| 458 | } | |||
| 459 | ||||
| 460 | namespace { | |||
| 461 | ||||
| 462 | /// Given a chain of or (||) or and (&&) comparison of a value against a | |||
| 463 | /// constant, this will try to recover the information required for a switch | |||
| 464 | /// structure. | |||
| 465 | /// It will depth-first traverse the chain of comparison, seeking for patterns | |||
| 466 | /// like %a == 12 or %a < 4 and combine them to produce a set of integer | |||
| 467 | /// representing the different cases for the switch. | |||
| 468 | /// Note that if the chain is composed of '||' it will build the set of elements | |||
| 469 | /// that matches the comparisons (i.e. any of this value validate the chain) | |||
| 470 | /// while for a chain of '&&' it will build the set elements that make the test | |||
| 471 | /// fail. | |||
| 472 | struct ConstantComparesGatherer { | |||
| 473 | const DataLayout &DL; | |||
| 474 | ||||
| 475 | /// Value found for the switch comparison | |||
| 476 | Value *CompValue = nullptr; | |||
| 477 | ||||
| 478 | /// Extra clause to be checked before the switch | |||
| 479 | Value *Extra = nullptr; | |||
| 480 | ||||
| 481 | /// Set of integers to match in switch | |||
| 482 | SmallVector<ConstantInt *, 8> Vals; | |||
| 483 | ||||
| 484 | /// Number of comparisons matched in the and/or chain | |||
| 485 | unsigned UsedICmps = 0; | |||
| 486 | ||||
| 487 | /// Construct and compute the result for the comparison instruction Cond | |||
| 488 | ConstantComparesGatherer(Instruction *Cond, const DataLayout &DL) : DL(DL) { | |||
| 489 | gather(Cond); | |||
| 490 | } | |||
| 491 | ||||
| 492 | ConstantComparesGatherer(const ConstantComparesGatherer &) = delete; | |||
| 493 | ConstantComparesGatherer & | |||
| 494 | operator=(const ConstantComparesGatherer &) = delete; | |||
| 495 | ||||
| 496 | private: | |||
| 497 | /// Try to set the current value used for the comparison, it succeeds only if | |||
| 498 | /// it wasn't set before or if the new value is the same as the old one | |||
| 499 | bool setValueOnce(Value *NewVal) { | |||
| 500 | if (CompValue && CompValue != NewVal) | |||
| 501 | return false; | |||
| 502 | CompValue = NewVal; | |||
| 503 | return (CompValue != nullptr); | |||
| 504 | } | |||
| 505 | ||||
| 506 | /// Try to match Instruction "I" as a comparison against a constant and | |||
| 507 | /// populates the array Vals with the set of values that match (or do not | |||
| 508 | /// match depending on isEQ). | |||
| 509 | /// Return false on failure. On success, the Value the comparison matched | |||
| 510 | /// against is placed in CompValue. | |||
| 511 | /// If CompValue is already set, the function is expected to fail if a match | |||
| 512 | /// is found but the value compared to is different. | |||
| 513 | bool matchInstruction(Instruction *I, bool isEQ) { | |||
| 514 | // If this is an icmp against a constant, handle this as one of the cases. | |||
| 515 | ICmpInst *ICI; | |||
| 516 | ConstantInt *C; | |||
| 517 | if (!((ICI = dyn_cast<ICmpInst>(I)) && | |||
| 518 | (C = GetConstantInt(I->getOperand(1), DL)))) { | |||
| 519 | return false; | |||
| 520 | } | |||
| 521 | ||||
| 522 | Value *RHSVal; | |||
| 523 | const APInt *RHSC; | |||
| 524 | ||||
| 525 | // Pattern match a special case | |||
| 526 | // (x & ~2^z) == y --> x == y || x == y|2^z | |||
| 527 | // This undoes a transformation done by instcombine to fuse 2 compares. | |||
| 528 | if (ICI->getPredicate() == (isEQ ? ICmpInst::ICMP_EQ : ICmpInst::ICMP_NE)) { | |||
| 529 | // It's a little bit hard to see why the following transformations are | |||
| 530 | // correct. Here is a CVC3 program to verify them for 64-bit values: | |||
| 531 | ||||
| 532 | /* | |||
| 533 | ONE : BITVECTOR(64) = BVZEROEXTEND(0bin1, 63); | |||
| 534 | x : BITVECTOR(64); | |||
| 535 | y : BITVECTOR(64); | |||
| 536 | z : BITVECTOR(64); | |||
| 537 | mask : BITVECTOR(64) = BVSHL(ONE, z); | |||
| 538 | QUERY( (y & ~mask = y) => | |||
| 539 | ((x & ~mask = y) <=> (x = y OR x = (y | mask))) | |||
| 540 | ); | |||
| 541 | QUERY( (y | mask = y) => | |||
| 542 | ((x | mask = y) <=> (x = y OR x = (y & ~mask))) | |||
| 543 | ); | |||
| 544 | */ | |||
| 545 | ||||
| 546 | // Please note that each pattern must be a dual implication (<--> or | |||
| 547 | // iff). One directional implication can create spurious matches. If the | |||
| 548 | // implication is only one-way, an unsatisfiable condition on the left | |||
| 549 | // side can imply a satisfiable condition on the right side. Dual | |||
| 550 | // implication ensures that satisfiable conditions are transformed to | |||
| 551 | // other satisfiable conditions and unsatisfiable conditions are | |||
| 552 | // transformed to other unsatisfiable conditions. | |||
| 553 | ||||
| 554 | // Here is a concrete example of a unsatisfiable condition on the left | |||
| 555 | // implying a satisfiable condition on the right: | |||
| 556 | // | |||
| 557 | // mask = (1 << z) | |||
| 558 | // (x & ~mask) == y --> (x == y || x == (y | mask)) | |||
| 559 | // | |||
| 560 | // Substituting y = 3, z = 0 yields: | |||
| 561 | // (x & -2) == 3 --> (x == 3 || x == 2) | |||
| 562 | ||||
| 563 | // Pattern match a special case: | |||
| 564 | /* | |||
| 565 | QUERY( (y & ~mask = y) => | |||
| 566 | ((x & ~mask = y) <=> (x = y OR x = (y | mask))) | |||
| 567 | ); | |||
| 568 | */ | |||
| 569 | if (match(ICI->getOperand(0), | |||
| 570 | m_And(m_Value(RHSVal), m_APInt(RHSC)))) { | |||
| 571 | APInt Mask = ~*RHSC; | |||
| 572 | if (Mask.isPowerOf2() && (C->getValue() & ~Mask) == C->getValue()) { | |||
| 573 | // If we already have a value for the switch, it has to match! | |||
| 574 | if (!setValueOnce(RHSVal)) | |||
| 575 | return false; | |||
| 576 | ||||
| 577 | Vals.push_back(C); | |||
| 578 | Vals.push_back( | |||
| 579 | ConstantInt::get(C->getContext(), | |||
| 580 | C->getValue() | Mask)); | |||
| 581 | UsedICmps++; | |||
| 582 | return true; | |||
| 583 | } | |||
| 584 | } | |||
| 585 | ||||
| 586 | // Pattern match a special case: | |||
| 587 | /* | |||
| 588 | QUERY( (y | mask = y) => | |||
| 589 | ((x | mask = y) <=> (x = y OR x = (y & ~mask))) | |||
| 590 | ); | |||
| 591 | */ | |||
| 592 | if (match(ICI->getOperand(0), | |||
| 593 | m_Or(m_Value(RHSVal), m_APInt(RHSC)))) { | |||
| 594 | APInt Mask = *RHSC; | |||
| 595 | if (Mask.isPowerOf2() && (C->getValue() | Mask) == C->getValue()) { | |||
| 596 | // If we already have a value for the switch, it has to match! | |||
| 597 | if (!setValueOnce(RHSVal)) | |||
| 598 | return false; | |||
| 599 | ||||
| 600 | Vals.push_back(C); | |||
| 601 | Vals.push_back(ConstantInt::get(C->getContext(), | |||
| 602 | C->getValue() & ~Mask)); | |||
| 603 | UsedICmps++; | |||
| 604 | return true; | |||
| 605 | } | |||
| 606 | } | |||
| 607 | ||||
| 608 | // If we already have a value for the switch, it has to match! | |||
| 609 | if (!setValueOnce(ICI->getOperand(0))) | |||
| 610 | return false; | |||
| 611 | ||||
| 612 | UsedICmps++; | |||
| 613 | Vals.push_back(C); | |||
| 614 | return ICI->getOperand(0); | |||
| 615 | } | |||
| 616 | ||||
| 617 | // If we have "x ult 3", for example, then we can add 0,1,2 to the set. | |||
| 618 | ConstantRange Span = | |||
| 619 | ConstantRange::makeExactICmpRegion(ICI->getPredicate(), C->getValue()); | |||
| 620 | ||||
| 621 | // Shift the range if the compare is fed by an add. This is the range | |||
| 622 | // compare idiom as emitted by instcombine. | |||
| 623 | Value *CandidateVal = I->getOperand(0); | |||
| 624 | if (match(I->getOperand(0), m_Add(m_Value(RHSVal), m_APInt(RHSC)))) { | |||
| 625 | Span = Span.subtract(*RHSC); | |||
| 626 | CandidateVal = RHSVal; | |||
| 627 | } | |||
| 628 | ||||
| 629 | // If this is an and/!= check, then we are looking to build the set of | |||
| 630 | // value that *don't* pass the and chain. I.e. to turn "x ugt 2" into | |||
| 631 | // x != 0 && x != 1. | |||
| 632 | if (!isEQ) | |||
| 633 | Span = Span.inverse(); | |||
| 634 | ||||
| 635 | // If there are a ton of values, we don't want to make a ginormous switch. | |||
| 636 | if (Span.isSizeLargerThan(8) || Span.isEmptySet()) { | |||
| 637 | return false; | |||
| 638 | } | |||
| 639 | ||||
| 640 | // If we already have a value for the switch, it has to match! | |||
| 641 | if (!setValueOnce(CandidateVal)) | |||
| 642 | return false; | |||
| 643 | ||||
| 644 | // Add all values from the range to the set | |||
| 645 | for (APInt Tmp = Span.getLower(); Tmp != Span.getUpper(); ++Tmp) | |||
| 646 | Vals.push_back(ConstantInt::get(I->getContext(), Tmp)); | |||
| 647 | ||||
| 648 | UsedICmps++; | |||
| 649 | return true; | |||
| 650 | } | |||
| 651 | ||||
| 652 | /// Given a potentially 'or'd or 'and'd together collection of icmp | |||
| 653 | /// eq/ne/lt/gt instructions that compare a value against a constant, extract | |||
| 654 | /// the value being compared, and stick the list constants into the Vals | |||
| 655 | /// vector. | |||
| 656 | /// One "Extra" case is allowed to differ from the other. | |||
| 657 | void gather(Value *V) { | |||
| 658 | bool isEQ = match(V, m_LogicalOr(m_Value(), m_Value())); | |||
| 659 | ||||
| 660 | // Keep a stack (SmallVector for efficiency) for depth-first traversal | |||
| 661 | SmallVector<Value *, 8> DFT; | |||
| 662 | SmallPtrSet<Value *, 8> Visited; | |||
| 663 | ||||
| 664 | // Initialize | |||
| 665 | Visited.insert(V); | |||
| 666 | DFT.push_back(V); | |||
| 667 | ||||
| 668 | while (!DFT.empty()) { | |||
| 669 | V = DFT.pop_back_val(); | |||
| 670 | ||||
| 671 | if (Instruction *I = dyn_cast<Instruction>(V)) { | |||
| 672 | // If it is a || (or && depending on isEQ), process the operands. | |||
| 673 | Value *Op0, *Op1; | |||
| 674 | if (isEQ ? match(I, m_LogicalOr(m_Value(Op0), m_Value(Op1))) | |||
| 675 | : match(I, m_LogicalAnd(m_Value(Op0), m_Value(Op1)))) { | |||
| 676 | if (Visited.insert(Op1).second) | |||
| 677 | DFT.push_back(Op1); | |||
| 678 | if (Visited.insert(Op0).second) | |||
| 679 | DFT.push_back(Op0); | |||
| 680 | ||||
| 681 | continue; | |||
| 682 | } | |||
| 683 | ||||
| 684 | // Try to match the current instruction | |||
| 685 | if (matchInstruction(I, isEQ)) | |||
| 686 | // Match succeed, continue the loop | |||
| 687 | continue; | |||
| 688 | } | |||
| 689 | ||||
| 690 | // One element of the sequence of || (or &&) could not be match as a | |||
| 691 | // comparison against the same value as the others. | |||
| 692 | // We allow only one "Extra" case to be checked before the switch | |||
| 693 | if (!Extra) { | |||
| 694 | Extra = V; | |||
| 695 | continue; | |||
| 696 | } | |||
| 697 | // Failed to parse a proper sequence, abort now | |||
| 698 | CompValue = nullptr; | |||
| 699 | break; | |||
| 700 | } | |||
| 701 | } | |||
| 702 | }; | |||
| 703 | ||||
| 704 | } // end anonymous namespace | |||
| 705 | ||||
| 706 | static void EraseTerminatorAndDCECond(Instruction *TI, | |||
| 707 | MemorySSAUpdater *MSSAU = nullptr) { | |||
| 708 | Instruction *Cond = nullptr; | |||
| 709 | if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) { | |||
| 710 | Cond = dyn_cast<Instruction>(SI->getCondition()); | |||
| 711 | } else if (BranchInst *BI = dyn_cast<BranchInst>(TI)) { | |||
| 712 | if (BI->isConditional()) | |||
| 713 | Cond = dyn_cast<Instruction>(BI->getCondition()); | |||
| 714 | } else if (IndirectBrInst *IBI = dyn_cast<IndirectBrInst>(TI)) { | |||
| 715 | Cond = dyn_cast<Instruction>(IBI->getAddress()); | |||
| 716 | } | |||
| 717 | ||||
| 718 | TI->eraseFromParent(); | |||
| 719 | if (Cond) | |||
| 720 | RecursivelyDeleteTriviallyDeadInstructions(Cond, nullptr, MSSAU); | |||
| 721 | } | |||
| 722 | ||||
| 723 | /// Return true if the specified terminator checks | |||
| 724 | /// to see if a value is equal to constant integer value. | |||
| 725 | Value *SimplifyCFGOpt::isValueEqualityComparison(Instruction *TI) { | |||
| 726 | Value *CV = nullptr; | |||
| 727 | if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) { | |||
| 728 | // Do not permit merging of large switch instructions into their | |||
| 729 | // predecessors unless there is only one predecessor. | |||
| 730 | if (!SI->getParent()->hasNPredecessorsOrMore(128 / SI->getNumSuccessors())) | |||
| 731 | CV = SI->getCondition(); | |||
| 732 | } else if (BranchInst *BI = dyn_cast<BranchInst>(TI)) | |||
| 733 | if (BI->isConditional() && BI->getCondition()->hasOneUse()) | |||
| 734 | if (ICmpInst *ICI = dyn_cast<ICmpInst>(BI->getCondition())) { | |||
| 735 | if (ICI->isEquality() && GetConstantInt(ICI->getOperand(1), DL)) | |||
| 736 | CV = ICI->getOperand(0); | |||
| 737 | } | |||
| 738 | ||||
| 739 | // Unwrap any lossless ptrtoint cast. | |||
| 740 | if (CV) { | |||
| 741 | if (PtrToIntInst *PTII = dyn_cast<PtrToIntInst>(CV)) { | |||
| 742 | Value *Ptr = PTII->getPointerOperand(); | |||
| 743 | if (PTII->getType() == DL.getIntPtrType(Ptr->getType())) | |||
| 744 | CV = Ptr; | |||
| 745 | } | |||
| 746 | } | |||
| 747 | return CV; | |||
| 748 | } | |||
| 749 | ||||
| 750 | /// Given a value comparison instruction, | |||
| 751 | /// decode all of the 'cases' that it represents and return the 'default' block. | |||
| 752 | BasicBlock *SimplifyCFGOpt::GetValueEqualityComparisonCases( | |||
| 753 | Instruction *TI, std::vector<ValueEqualityComparisonCase> &Cases) { | |||
| 754 | if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) { | |||
| 755 | Cases.reserve(SI->getNumCases()); | |||
| 756 | for (auto Case : SI->cases()) | |||
| 757 | Cases.push_back(ValueEqualityComparisonCase(Case.getCaseValue(), | |||
| 758 | Case.getCaseSuccessor())); | |||
| 759 | return SI->getDefaultDest(); | |||
| 760 | } | |||
| 761 | ||||
| 762 | BranchInst *BI = cast<BranchInst>(TI); | |||
| 763 | ICmpInst *ICI = cast<ICmpInst>(BI->getCondition()); | |||
| 764 | BasicBlock *Succ = BI->getSuccessor(ICI->getPredicate() == ICmpInst::ICMP_NE); | |||
| 765 | Cases.push_back(ValueEqualityComparisonCase( | |||
| 766 | GetConstantInt(ICI->getOperand(1), DL), Succ)); | |||
| 767 | return BI->getSuccessor(ICI->getPredicate() == ICmpInst::ICMP_EQ); | |||
| 768 | } | |||
| 769 | ||||
| 770 | /// Given a vector of bb/value pairs, remove any entries | |||
| 771 | /// in the list that match the specified block. | |||
| 772 | static void | |||
| 773 | EliminateBlockCases(BasicBlock *BB, | |||
| 774 | std::vector<ValueEqualityComparisonCase> &Cases) { | |||
| 775 | llvm::erase_value(Cases, BB); | |||
| 776 | } | |||
| 777 | ||||
| 778 | /// Return true if there are any keys in C1 that exist in C2 as well. | |||
| 779 | static bool ValuesOverlap(std::vector<ValueEqualityComparisonCase> &C1, | |||
| 780 | std::vector<ValueEqualityComparisonCase> &C2) { | |||
| 781 | std::vector<ValueEqualityComparisonCase> *V1 = &C1, *V2 = &C2; | |||
| 782 | ||||
| 783 | // Make V1 be smaller than V2. | |||
| 784 | if (V1->size() > V2->size()) | |||
| 785 | std::swap(V1, V2); | |||
| 786 | ||||
| 787 | if (V1->empty()) | |||
| 788 | return false; | |||
| 789 | if (V1->size() == 1) { | |||
| 790 | // Just scan V2. | |||
| 791 | ConstantInt *TheVal = (*V1)[0].Value; | |||
| 792 | for (unsigned i = 0, e = V2->size(); i != e; ++i) | |||
| 793 | if (TheVal == (*V2)[i].Value) | |||
| 794 | return true; | |||
| 795 | } | |||
| 796 | ||||
| 797 | // Otherwise, just sort both lists and compare element by element. | |||
| 798 | array_pod_sort(V1->begin(), V1->end()); | |||
| 799 | array_pod_sort(V2->begin(), V2->end()); | |||
| 800 | unsigned i1 = 0, i2 = 0, e1 = V1->size(), e2 = V2->size(); | |||
| 801 | while (i1 != e1 && i2 != e2) { | |||
| 802 | if ((*V1)[i1].Value == (*V2)[i2].Value) | |||
| 803 | return true; | |||
| 804 | if ((*V1)[i1].Value < (*V2)[i2].Value) | |||
| 805 | ++i1; | |||
| 806 | else | |||
| 807 | ++i2; | |||
| 808 | } | |||
| 809 | return false; | |||
| 810 | } | |||
| 811 | ||||
| 812 | // Set branch weights on SwitchInst. This sets the metadata if there is at | |||
| 813 | // least one non-zero weight. | |||
| 814 | static void setBranchWeights(SwitchInst *SI, ArrayRef<uint32_t> Weights) { | |||
| 815 | // Check that there is at least one non-zero weight. Otherwise, pass | |||
| 816 | // nullptr to setMetadata which will erase the existing metadata. | |||
| 817 | MDNode *N = nullptr; | |||
| 818 | if (llvm::any_of(Weights, [](uint32_t W) { return W != 0; })) | |||
| 819 | N = MDBuilder(SI->getParent()->getContext()).createBranchWeights(Weights); | |||
| 820 | SI->setMetadata(LLVMContext::MD_prof, N); | |||
| 821 | } | |||
| 822 | ||||
| 823 | // Similar to the above, but for branch and select instructions that take | |||
| 824 | // exactly 2 weights. | |||
| 825 | static void setBranchWeights(Instruction *I, uint32_t TrueWeight, | |||
| 826 | uint32_t FalseWeight) { | |||
| 827 | assert(isa<BranchInst>(I) || isa<SelectInst>(I))((void)0); | |||
| 828 | // Check that there is at least one non-zero weight. Otherwise, pass | |||
| 829 | // nullptr to setMetadata which will erase the existing metadata. | |||
| 830 | MDNode *N = nullptr; | |||
| 831 | if (TrueWeight || FalseWeight) | |||
| 832 | N = MDBuilder(I->getParent()->getContext()) | |||
| 833 | .createBranchWeights(TrueWeight, FalseWeight); | |||
| 834 | I->setMetadata(LLVMContext::MD_prof, N); | |||
| 835 | } | |||
| 836 | ||||
| 837 | /// If TI is known to be a terminator instruction and its block is known to | |||
| 838 | /// only have a single predecessor block, check to see if that predecessor is | |||
| 839 | /// also a value comparison with the same value, and if that comparison | |||
| 840 | /// determines the outcome of this comparison. If so, simplify TI. This does a | |||
| 841 | /// very limited form of jump threading. | |||
| 842 | bool SimplifyCFGOpt::SimplifyEqualityComparisonWithOnlyPredecessor( | |||
| 843 | Instruction *TI, BasicBlock *Pred, IRBuilder<> &Builder) { | |||
| 844 | Value *PredVal = isValueEqualityComparison(Pred->getTerminator()); | |||
| 845 | if (!PredVal) | |||
| 846 | return false; // Not a value comparison in predecessor. | |||
| 847 | ||||
| 848 | Value *ThisVal = isValueEqualityComparison(TI); | |||
| 849 | assert(ThisVal && "This isn't a value comparison!!")((void)0); | |||
| 850 | if (ThisVal != PredVal) | |||
| 851 | return false; // Different predicates. | |||
| 852 | ||||
| 853 | // TODO: Preserve branch weight metadata, similarly to how | |||
| 854 | // FoldValueComparisonIntoPredecessors preserves it. | |||
| 855 | ||||
| 856 | // Find out information about when control will move from Pred to TI's block. | |||
| 857 | std::vector<ValueEqualityComparisonCase> PredCases; | |||
| 858 | BasicBlock *PredDef = | |||
| 859 | GetValueEqualityComparisonCases(Pred->getTerminator(), PredCases); | |||
| 860 | EliminateBlockCases(PredDef, PredCases); // Remove default from cases. | |||
| 861 | ||||
| 862 | // Find information about how control leaves this block. | |||
| 863 | std::vector<ValueEqualityComparisonCase> ThisCases; | |||
| 864 | BasicBlock *ThisDef = GetValueEqualityComparisonCases(TI, ThisCases); | |||
| 865 | EliminateBlockCases(ThisDef, ThisCases); // Remove default from cases. | |||
| 866 | ||||
| 867 | // If TI's block is the default block from Pred's comparison, potentially | |||
| 868 | // simplify TI based on this knowledge. | |||
| 869 | if (PredDef == TI->getParent()) { | |||
| 870 | // If we are here, we know that the value is none of those cases listed in | |||
| 871 | // PredCases. If there are any cases in ThisCases that are in PredCases, we | |||
| 872 | // can simplify TI. | |||
| 873 | if (!ValuesOverlap(PredCases, ThisCases)) | |||
| 874 | return false; | |||
| 875 | ||||
| 876 | if (isa<BranchInst>(TI)) { | |||
| 877 | // Okay, one of the successors of this condbr is dead. Convert it to a | |||
| 878 | // uncond br. | |||
| 879 | assert(ThisCases.size() == 1 && "Branch can only have one case!")((void)0); | |||
| 880 | // Insert the new branch. | |||
| 881 | Instruction *NI = Builder.CreateBr(ThisDef); | |||
| 882 | (void)NI; | |||
| 883 | ||||
| 884 | // Remove PHI node entries for the dead edge. | |||
| 885 | ThisCases[0].Dest->removePredecessor(PredDef); | |||
| 886 | ||||
| 887 | LLVM_DEBUG(dbgs() << "Threading pred instr: " << *Pred->getTerminator()do { } while (false) | |||
| 888 | << "Through successor TI: " << *TI << "Leaving: " << *NIdo { } while (false) | |||
| 889 | << "\n")do { } while (false); | |||
| 890 | ||||
| 891 | EraseTerminatorAndDCECond(TI); | |||
| 892 | ||||
| 893 | if (DTU) | |||
| 894 | DTU->applyUpdates( | |||
| 895 | {{DominatorTree::Delete, PredDef, ThisCases[0].Dest}}); | |||
| 896 | ||||
| 897 | return true; | |||
| 898 | } | |||
| 899 | ||||
| 900 | SwitchInstProfUpdateWrapper SI = *cast<SwitchInst>(TI); | |||
| 901 | // Okay, TI has cases that are statically dead, prune them away. | |||
| 902 | SmallPtrSet<Constant *, 16> DeadCases; | |||
| 903 | for (unsigned i = 0, e = PredCases.size(); i != e; ++i) | |||
| 904 | DeadCases.insert(PredCases[i].Value); | |||
| 905 | ||||
| 906 | LLVM_DEBUG(dbgs() << "Threading pred instr: " << *Pred->getTerminator()do { } while (false) | |||
| 907 | << "Through successor TI: " << *TI)do { } while (false); | |||
| 908 | ||||
| 909 | SmallDenseMap<BasicBlock *, int, 8> NumPerSuccessorCases; | |||
| 910 | for (SwitchInst::CaseIt i = SI->case_end(), e = SI->case_begin(); i != e;) { | |||
| 911 | --i; | |||
| 912 | auto *Successor = i->getCaseSuccessor(); | |||
| 913 | if (DTU) | |||
| 914 | ++NumPerSuccessorCases[Successor]; | |||
| 915 | if (DeadCases.count(i->getCaseValue())) { | |||
| 916 | Successor->removePredecessor(PredDef); | |||
| 917 | SI.removeCase(i); | |||
| 918 | if (DTU) | |||
| 919 | --NumPerSuccessorCases[Successor]; | |||
| 920 | } | |||
| 921 | } | |||
| 922 | ||||
| 923 | if (DTU) { | |||
| 924 | std::vector<DominatorTree::UpdateType> Updates; | |||
| 925 | for (const std::pair<BasicBlock *, int> &I : NumPerSuccessorCases) | |||
| 926 | if (I.second == 0) | |||
| 927 | Updates.push_back({DominatorTree::Delete, PredDef, I.first}); | |||
| 928 | DTU->applyUpdates(Updates); | |||
| 929 | } | |||
| 930 | ||||
| 931 | LLVM_DEBUG(dbgs() << "Leaving: " << *TI << "\n")do { } while (false); | |||
| 932 | return true; | |||
| 933 | } | |||
| 934 | ||||
| 935 | // Otherwise, TI's block must correspond to some matched value. Find out | |||
| 936 | // which value (or set of values) this is. | |||
| 937 | ConstantInt *TIV = nullptr; | |||
| 938 | BasicBlock *TIBB = TI->getParent(); | |||
| 939 | for (unsigned i = 0, e = PredCases.size(); i != e; ++i) | |||
| 940 | if (PredCases[i].Dest == TIBB) { | |||
| 941 | if (TIV) | |||
| 942 | return false; // Cannot handle multiple values coming to this block. | |||
| 943 | TIV = PredCases[i].Value; | |||
| 944 | } | |||
| 945 | assert(TIV && "No edge from pred to succ?")((void)0); | |||
| 946 | ||||
| 947 | // Okay, we found the one constant that our value can be if we get into TI's | |||
| 948 | // BB. Find out which successor will unconditionally be branched to. | |||
| 949 | BasicBlock *TheRealDest = nullptr; | |||
| 950 | for (unsigned i = 0, e = ThisCases.size(); i != e; ++i) | |||
| 951 | if (ThisCases[i].Value == TIV) { | |||
| 952 | TheRealDest = ThisCases[i].Dest; | |||
| 953 | break; | |||
| 954 | } | |||
| 955 | ||||
| 956 | // If not handled by any explicit cases, it is handled by the default case. | |||
| 957 | if (!TheRealDest) | |||
| 958 | TheRealDest = ThisDef; | |||
| 959 | ||||
| 960 | SmallPtrSet<BasicBlock *, 2> RemovedSuccs; | |||
| 961 | ||||
| 962 | // Remove PHI node entries for dead edges. | |||
| 963 | BasicBlock *CheckEdge = TheRealDest; | |||
| 964 | for (BasicBlock *Succ : successors(TIBB)) | |||
| 965 | if (Succ != CheckEdge) { | |||
| 966 | if (Succ != TheRealDest) | |||
| 967 | RemovedSuccs.insert(Succ); | |||
| 968 | Succ->removePredecessor(TIBB); | |||
| 969 | } else | |||
| 970 | CheckEdge = nullptr; | |||
| 971 | ||||
| 972 | // Insert the new branch. | |||
| 973 | Instruction *NI = Builder.CreateBr(TheRealDest); | |||
| 974 | (void)NI; | |||
| 975 | ||||
| 976 | LLVM_DEBUG(dbgs() << "Threading pred instr: " << *Pred->getTerminator()do { } while (false) | |||
| 977 | << "Through successor TI: " << *TI << "Leaving: " << *NIdo { } while (false) | |||
| 978 | << "\n")do { } while (false); | |||
| 979 | ||||
| 980 | EraseTerminatorAndDCECond(TI); | |||
| 981 | if (DTU) { | |||
| 982 | SmallVector<DominatorTree::UpdateType, 2> Updates; | |||
| 983 | Updates.reserve(RemovedSuccs.size()); | |||
| 984 | for (auto *RemovedSucc : RemovedSuccs) | |||
| 985 | Updates.push_back({DominatorTree::Delete, TIBB, RemovedSucc}); | |||
| 986 | DTU->applyUpdates(Updates); | |||
| 987 | } | |||
| 988 | return true; | |||
| 989 | } | |||
| 990 | ||||
| 991 | namespace { | |||
| 992 | ||||
| 993 | /// This class implements a stable ordering of constant | |||
| 994 | /// integers that does not depend on their address. This is important for | |||
| 995 | /// applications that sort ConstantInt's to ensure uniqueness. | |||
| 996 | struct ConstantIntOrdering { | |||
| 997 | bool operator()(const ConstantInt *LHS, const ConstantInt *RHS) const { | |||
| 998 | return LHS->getValue().ult(RHS->getValue()); | |||
| 999 | } | |||
| 1000 | }; | |||
| 1001 | ||||
| 1002 | } // end anonymous namespace | |||
| 1003 | ||||
| 1004 | static int ConstantIntSortPredicate(ConstantInt *const *P1, | |||
| 1005 | ConstantInt *const *P2) { | |||
| 1006 | const ConstantInt *LHS = *P1; | |||
| 1007 | const ConstantInt *RHS = *P2; | |||
| 1008 | if (LHS == RHS) | |||
| 1009 | return 0; | |||
| 1010 | return LHS->getValue().ult(RHS->getValue()) ? 1 : -1; | |||
| 1011 | } | |||
| 1012 | ||||
| 1013 | static inline bool HasBranchWeights(const Instruction *I) { | |||
| 1014 | MDNode *ProfMD = I->getMetadata(LLVMContext::MD_prof); | |||
| 1015 | if (ProfMD && ProfMD->getOperand(0)) | |||
| 1016 | if (MDString *MDS = dyn_cast<MDString>(ProfMD->getOperand(0))) | |||
| 1017 | return MDS->getString().equals("branch_weights"); | |||
| 1018 | ||||
| 1019 | return false; | |||
| 1020 | } | |||
| 1021 | ||||
| 1022 | /// Get Weights of a given terminator, the default weight is at the front | |||
| 1023 | /// of the vector. If TI is a conditional eq, we need to swap the branch-weight | |||
| 1024 | /// metadata. | |||
| 1025 | static void GetBranchWeights(Instruction *TI, | |||
| 1026 | SmallVectorImpl<uint64_t> &Weights) { | |||
| 1027 | MDNode *MD = TI->getMetadata(LLVMContext::MD_prof); | |||
| 1028 | assert(MD)((void)0); | |||
| 1029 | for (unsigned i = 1, e = MD->getNumOperands(); i < e; ++i) { | |||
| 1030 | ConstantInt *CI = mdconst::extract<ConstantInt>(MD->getOperand(i)); | |||
| 1031 | Weights.push_back(CI->getValue().getZExtValue()); | |||
| 1032 | } | |||
| 1033 | ||||
| 1034 | // If TI is a conditional eq, the default case is the false case, | |||
| 1035 | // and the corresponding branch-weight data is at index 2. We swap the | |||
| 1036 | // default weight to be the first entry. | |||
| 1037 | if (BranchInst *BI = dyn_cast<BranchInst>(TI)) { | |||
| 1038 | assert(Weights.size() == 2)((void)0); | |||
| 1039 | ICmpInst *ICI = cast<ICmpInst>(BI->getCondition()); | |||
| 1040 | if (ICI->getPredicate() == ICmpInst::ICMP_EQ) | |||
| 1041 | std::swap(Weights.front(), Weights.back()); | |||
| 1042 | } | |||
| 1043 | } | |||
| 1044 | ||||
| 1045 | /// Keep halving the weights until all can fit in uint32_t. | |||
| 1046 | static void FitWeights(MutableArrayRef<uint64_t> Weights) { | |||
| 1047 | uint64_t Max = *std::max_element(Weights.begin(), Weights.end()); | |||
| 1048 | if (Max > UINT_MAX(2147483647 *2U +1U)) { | |||
| 1049 | unsigned Offset = 32 - countLeadingZeros(Max); | |||
| 1050 | for (uint64_t &I : Weights) | |||
| 1051 | I >>= Offset; | |||
| 1052 | } | |||
| 1053 | } | |||
| 1054 | ||||
| 1055 | static void CloneInstructionsIntoPredecessorBlockAndUpdateSSAUses( | |||
| 1056 | BasicBlock *BB, BasicBlock *PredBlock, ValueToValueMapTy &VMap) { | |||
| 1057 | Instruction *PTI = PredBlock->getTerminator(); | |||
| 1058 | ||||
| 1059 | // If we have bonus instructions, clone them into the predecessor block. | |||
| 1060 | // Note that there may be multiple predecessor blocks, so we cannot move | |||
| 1061 | // bonus instructions to a predecessor block. | |||
| 1062 | for (Instruction &BonusInst : *BB) { | |||
| 1063 | if (isa<DbgInfoIntrinsic>(BonusInst) || BonusInst.isTerminator()) | |||
| 1064 | continue; | |||
| 1065 | ||||
| 1066 | Instruction *NewBonusInst = BonusInst.clone(); | |||
| 1067 | ||||
| 1068 | if (PTI->getDebugLoc() != NewBonusInst->getDebugLoc()) { | |||
| 1069 | // Unless the instruction has the same !dbg location as the original | |||
| 1070 | // branch, drop it. When we fold the bonus instructions we want to make | |||
| 1071 | // sure we reset their debug locations in order to avoid stepping on | |||
| 1072 | // dead code caused by folding dead branches. | |||
| 1073 | NewBonusInst->setDebugLoc(DebugLoc()); | |||
| 1074 | } | |||
| 1075 | ||||
| 1076 | RemapInstruction(NewBonusInst, VMap, | |||
| 1077 | RF_NoModuleLevelChanges | RF_IgnoreMissingLocals); | |||
| 1078 | VMap[&BonusInst] = NewBonusInst; | |||
| 1079 | ||||
| 1080 | // If we moved a load, we cannot any longer claim any knowledge about | |||
| 1081 | // its potential value. The previous information might have been valid | |||
| 1082 | // only given the branch precondition. | |||
| 1083 | // For an analogous reason, we must also drop all the metadata whose | |||
| 1084 | // semantics we don't understand. We *can* preserve !annotation, because | |||
| 1085 | // it is tied to the instruction itself, not the value or position. | |||
| 1086 | // Similarly strip attributes on call parameters that may cause UB in | |||
| 1087 | // location the call is moved to. | |||
| 1088 | NewBonusInst->dropUndefImplyingAttrsAndUnknownMetadata( | |||
| 1089 | LLVMContext::MD_annotation); | |||
| 1090 | ||||
| 1091 | PredBlock->getInstList().insert(PTI->getIterator(), NewBonusInst); | |||
| 1092 | NewBonusInst->takeName(&BonusInst); | |||
| 1093 | BonusInst.setName(NewBonusInst->getName() + ".old"); | |||
| 1094 | ||||
| 1095 | // Update (liveout) uses of bonus instructions, | |||
| 1096 | // now that the bonus instruction has been cloned into predecessor. | |||
| 1097 | // Note that we expect to be in a block-closed SSA form for this to work! | |||
| 1098 | for (Use &U : make_early_inc_range(BonusInst.uses())) { | |||
| 1099 | auto *UI = cast<Instruction>(U.getUser()); | |||
| 1100 | auto *PN = dyn_cast<PHINode>(UI); | |||
| 1101 | if (!PN) { | |||
| 1102 | assert(UI->getParent() == BB && BonusInst.comesBefore(UI) &&((void)0) | |||
| 1103 | "If the user is not a PHI node, then it should be in the same "((void)0) | |||
| 1104 | "block as, and come after, the original bonus instruction.")((void)0); | |||
| 1105 | continue; // Keep using the original bonus instruction. | |||
| 1106 | } | |||
| 1107 | // Is this the block-closed SSA form PHI node? | |||
| 1108 | if (PN->getIncomingBlock(U) == BB) | |||
| 1109 | continue; // Great, keep using the original bonus instruction. | |||
| 1110 | // The only other alternative is an "use" when coming from | |||
| 1111 | // the predecessor block - here we should refer to the cloned bonus instr. | |||
| 1112 | assert(PN->getIncomingBlock(U) == PredBlock &&((void)0) | |||
| 1113 | "Not in block-closed SSA form?")((void)0); | |||
| 1114 | U.set(NewBonusInst); | |||
| 1115 | } | |||
| 1116 | } | |||
| 1117 | } | |||
| 1118 | ||||
| 1119 | bool SimplifyCFGOpt::PerformValueComparisonIntoPredecessorFolding( | |||
| 1120 | Instruction *TI, Value *&CV, Instruction *PTI, IRBuilder<> &Builder) { | |||
| 1121 | BasicBlock *BB = TI->getParent(); | |||
| 1122 | BasicBlock *Pred = PTI->getParent(); | |||
| 1123 | ||||
| 1124 | SmallVector<DominatorTree::UpdateType, 32> Updates; | |||
| 1125 | ||||
| 1126 | // Figure out which 'cases' to copy from SI to PSI. | |||
| 1127 | std::vector<ValueEqualityComparisonCase> BBCases; | |||
| 1128 | BasicBlock *BBDefault = GetValueEqualityComparisonCases(TI, BBCases); | |||
| 1129 | ||||
| 1130 | std::vector<ValueEqualityComparisonCase> PredCases; | |||
| 1131 | BasicBlock *PredDefault = GetValueEqualityComparisonCases(PTI, PredCases); | |||
| 1132 | ||||
| 1133 | // Based on whether the default edge from PTI goes to BB or not, fill in | |||
| 1134 | // PredCases and PredDefault with the new switch cases we would like to | |||
| 1135 | // build. | |||
| 1136 | SmallMapVector<BasicBlock *, int, 8> NewSuccessors; | |||
| 1137 | ||||
| 1138 | // Update the branch weight metadata along the way | |||
| 1139 | SmallVector<uint64_t, 8> Weights; | |||
| 1140 | bool PredHasWeights = HasBranchWeights(PTI); | |||
| 1141 | bool SuccHasWeights = HasBranchWeights(TI); | |||
| 1142 | ||||
| 1143 | if (PredHasWeights) { | |||
| 1144 | GetBranchWeights(PTI, Weights); | |||
| 1145 | // branch-weight metadata is inconsistent here. | |||
| 1146 | if (Weights.size() != 1 + PredCases.size()) | |||
| 1147 | PredHasWeights = SuccHasWeights = false; | |||
| 1148 | } else if (SuccHasWeights) | |||
| 1149 | // If there are no predecessor weights but there are successor weights, | |||
| 1150 | // populate Weights with 1, which will later be scaled to the sum of | |||
| 1151 | // successor's weights | |||
| 1152 | Weights.assign(1 + PredCases.size(), 1); | |||
| 1153 | ||||
| 1154 | SmallVector<uint64_t, 8> SuccWeights; | |||
| 1155 | if (SuccHasWeights) { | |||
| 1156 | GetBranchWeights(TI, SuccWeights); | |||
| 1157 | // branch-weight metadata is inconsistent here. | |||
| 1158 | if (SuccWeights.size() != 1 + BBCases.size()) | |||
| 1159 | PredHasWeights = SuccHasWeights = false; | |||
| 1160 | } else if (PredHasWeights) | |||
| 1161 | SuccWeights.assign(1 + BBCases.size(), 1); | |||
| 1162 | ||||
| 1163 | if (PredDefault == BB) { | |||
| 1164 | // If this is the default destination from PTI, only the edges in TI | |||
| 1165 | // that don't occur in PTI, or that branch to BB will be activated. | |||
| 1166 | std::set<ConstantInt *, ConstantIntOrdering> PTIHandled; | |||
| 1167 | for (unsigned i = 0, e = PredCases.size(); i != e; ++i) | |||
| 1168 | if (PredCases[i].Dest != BB) | |||
| 1169 | PTIHandled.insert(PredCases[i].Value); | |||
| 1170 | else { | |||
| 1171 | // The default destination is BB, we don't need explicit targets. | |||
| 1172 | std::swap(PredCases[i], PredCases.back()); | |||
| 1173 | ||||
| 1174 | if (PredHasWeights || SuccHasWeights) { | |||
| 1175 | // Increase weight for the default case. | |||
| 1176 | Weights[0] += Weights[i + 1]; | |||
| 1177 | std::swap(Weights[i + 1], Weights.back()); | |||
| 1178 | Weights.pop_back(); | |||
| 1179 | } | |||
| 1180 | ||||
| 1181 | PredCases.pop_back(); | |||
| 1182 | --i; | |||
| 1183 | --e; | |||
| 1184 | } | |||
| 1185 | ||||
| 1186 | // Reconstruct the new switch statement we will be building. | |||
| 1187 | if (PredDefault != BBDefault) { | |||
| 1188 | PredDefault->removePredecessor(Pred); | |||
| 1189 | if (DTU && PredDefault != BB) | |||
| 1190 | Updates.push_back({DominatorTree::Delete, Pred, PredDefault}); | |||
| 1191 | PredDefault = BBDefault; | |||
| 1192 | ++NewSuccessors[BBDefault]; | |||
| 1193 | } | |||
| 1194 | ||||
| 1195 | unsigned CasesFromPred = Weights.size(); | |||
| 1196 | uint64_t ValidTotalSuccWeight = 0; | |||
| 1197 | for (unsigned i = 0, e = BBCases.size(); i != e; ++i) | |||
| 1198 | if (!PTIHandled.count(BBCases[i].Value) && BBCases[i].Dest != BBDefault) { | |||
| 1199 | PredCases.push_back(BBCases[i]); | |||
| 1200 | ++NewSuccessors[BBCases[i].Dest]; | |||
| 1201 | if (SuccHasWeights || PredHasWeights) { | |||
| 1202 | // The default weight is at index 0, so weight for the ith case | |||
| 1203 | // should be at index i+1. Scale the cases from successor by | |||
| 1204 | // PredDefaultWeight (Weights[0]). | |||
| 1205 | Weights.push_back(Weights[0] * SuccWeights[i + 1]); | |||
| 1206 | ValidTotalSuccWeight += SuccWeights[i + 1]; | |||
| 1207 | } | |||
| 1208 | } | |||
| 1209 | ||||
| 1210 | if (SuccHasWeights || PredHasWeights) { | |||
| 1211 | ValidTotalSuccWeight += SuccWeights[0]; | |||
| 1212 | // Scale the cases from predecessor by ValidTotalSuccWeight. | |||
| 1213 | for (unsigned i = 1; i < CasesFromPred; ++i) | |||
| 1214 | Weights[i] *= ValidTotalSuccWeight; | |||
| 1215 | // Scale the default weight by SuccDefaultWeight (SuccWeights[0]). | |||
| 1216 | Weights[0] *= SuccWeights[0]; | |||
| 1217 | } | |||
| 1218 | } else { | |||
| 1219 | // If this is not the default destination from PSI, only the edges | |||
| 1220 | // in SI that occur in PSI with a destination of BB will be | |||
| 1221 | // activated. | |||
| 1222 | std::set<ConstantInt *, ConstantIntOrdering> PTIHandled; | |||
| 1223 | std::map<ConstantInt *, uint64_t> WeightsForHandled; | |||
| 1224 | for (unsigned i = 0, e = PredCases.size(); i != e; ++i) | |||
| 1225 | if (PredCases[i].Dest == BB) { | |||
| 1226 | PTIHandled.insert(PredCases[i].Value); | |||
| 1227 | ||||
| 1228 | if (PredHasWeights || SuccHasWeights) { | |||
| 1229 | WeightsForHandled[PredCases[i].Value] = Weights[i + 1]; | |||
| 1230 | std::swap(Weights[i + 1], Weights.back()); | |||
| 1231 | Weights.pop_back(); | |||
| 1232 | } | |||
| 1233 | ||||
| 1234 | std::swap(PredCases[i], PredCases.back()); | |||
| 1235 | PredCases.pop_back(); | |||
| 1236 | --i; | |||
| 1237 | --e; | |||
| 1238 | } | |||
| 1239 | ||||
| 1240 | // Okay, now we know which constants were sent to BB from the | |||
| 1241 | // predecessor. Figure out where they will all go now. | |||
| 1242 | for (unsigned i = 0, e = BBCases.size(); i != e; ++i) | |||
| 1243 | if (PTIHandled.count(BBCases[i].Value)) { | |||
| 1244 | // If this is one we are capable of getting... | |||
| 1245 | if (PredHasWeights || SuccHasWeights) | |||
| 1246 | Weights.push_back(WeightsForHandled[BBCases[i].Value]); | |||
| 1247 | PredCases.push_back(BBCases[i]); | |||
| 1248 | ++NewSuccessors[BBCases[i].Dest]; | |||
| 1249 | PTIHandled.erase(BBCases[i].Value); // This constant is taken care of | |||
| 1250 | } | |||
| 1251 | ||||
| 1252 | // If there are any constants vectored to BB that TI doesn't handle, | |||
| 1253 | // they must go to the default destination of TI. | |||
| 1254 | for (ConstantInt *I : PTIHandled) { | |||
| 1255 | if (PredHasWeights || SuccHasWeights) | |||
| 1256 | Weights.push_back(WeightsForHandled[I]); | |||
| 1257 | PredCases.push_back(ValueEqualityComparisonCase(I, BBDefault)); | |||
| 1258 | ++NewSuccessors[BBDefault]; | |||
| 1259 | } | |||
| 1260 | } | |||
| 1261 | ||||
| 1262 | // Okay, at this point, we know which new successor Pred will get. Make | |||
| 1263 | // sure we update the number of entries in the PHI nodes for these | |||
| 1264 | // successors. | |||
| 1265 | SmallPtrSet<BasicBlock *, 2> SuccsOfPred; | |||
| 1266 | if (DTU) { | |||
| 1267 | SuccsOfPred = {succ_begin(Pred), succ_end(Pred)}; | |||
| 1268 | Updates.reserve(Updates.size() + NewSuccessors.size()); | |||
| 1269 | } | |||
| 1270 | for (const std::pair<BasicBlock *, int /*Num*/> &NewSuccessor : | |||
| 1271 | NewSuccessors) { | |||
| 1272 | for (auto I : seq(0, NewSuccessor.second)) { | |||
| 1273 | (void)I; | |||
| 1274 | AddPredecessorToBlock(NewSuccessor.first, Pred, BB); | |||
| 1275 | } | |||
| 1276 | if (DTU && !SuccsOfPred.contains(NewSuccessor.first)) | |||
| 1277 | Updates.push_back({DominatorTree::Insert, Pred, NewSuccessor.first}); | |||
| 1278 | } | |||
| 1279 | ||||
| 1280 | Builder.SetInsertPoint(PTI); | |||
| 1281 | // Convert pointer to int before we switch. | |||
| 1282 | if (CV->getType()->isPointerTy()) { | |||
| 1283 | CV = | |||
| 1284 | Builder.CreatePtrToInt(CV, DL.getIntPtrType(CV->getType()), "magicptr"); | |||
| 1285 | } | |||
| 1286 | ||||
| 1287 | // Now that the successors are updated, create the new Switch instruction. | |||
| 1288 | SwitchInst *NewSI = Builder.CreateSwitch(CV, PredDefault, PredCases.size()); | |||
| 1289 | NewSI->setDebugLoc(PTI->getDebugLoc()); | |||
| 1290 | for (ValueEqualityComparisonCase &V : PredCases) | |||
| 1291 | NewSI->addCase(V.Value, V.Dest); | |||
| 1292 | ||||
| 1293 | if (PredHasWeights || SuccHasWeights) { | |||
| 1294 | // Halve the weights if any of them cannot fit in an uint32_t | |||
| 1295 | FitWeights(Weights); | |||
| 1296 | ||||
| 1297 | SmallVector<uint32_t, 8> MDWeights(Weights.begin(), Weights.end()); | |||
| 1298 | ||||
| 1299 | setBranchWeights(NewSI, MDWeights); | |||
| 1300 | } | |||
| 1301 | ||||
| 1302 | EraseTerminatorAndDCECond(PTI); | |||
| 1303 | ||||
| 1304 | // Okay, last check. If BB is still a successor of PSI, then we must | |||
| 1305 | // have an infinite loop case. If so, add an infinitely looping block | |||
| 1306 | // to handle the case to preserve the behavior of the code. | |||
| 1307 | BasicBlock *InfLoopBlock = nullptr; | |||
| 1308 | for (unsigned i = 0, e = NewSI->getNumSuccessors(); i != e; ++i) | |||
| 1309 | if (NewSI->getSuccessor(i) == BB) { | |||
| 1310 | if (!InfLoopBlock) { | |||
| 1311 | // Insert it at the end of the function, because it's either code, | |||
| 1312 | // or it won't matter if it's hot. :) | |||
| 1313 | InfLoopBlock = | |||
| 1314 | BasicBlock::Create(BB->getContext(), "infloop", BB->getParent()); | |||
| 1315 | BranchInst::Create(InfLoopBlock, InfLoopBlock); | |||
| 1316 | if (DTU) | |||
| 1317 | Updates.push_back( | |||
| 1318 | {DominatorTree::Insert, InfLoopBlock, InfLoopBlock}); | |||
| 1319 | } | |||
| 1320 | NewSI->setSuccessor(i, InfLoopBlock); | |||
| 1321 | } | |||
| 1322 | ||||
| 1323 | if (DTU) { | |||
| 1324 | if (InfLoopBlock) | |||
| 1325 | Updates.push_back({DominatorTree::Insert, Pred, InfLoopBlock}); | |||
| 1326 | ||||
| 1327 | Updates.push_back({DominatorTree::Delete, Pred, BB}); | |||
| 1328 | ||||
| 1329 | DTU->applyUpdates(Updates); | |||
| 1330 | } | |||
| 1331 | ||||
| 1332 | ++NumFoldValueComparisonIntoPredecessors; | |||
| 1333 | return true; | |||
| 1334 | } | |||
| 1335 | ||||
| 1336 | /// The specified terminator is a value equality comparison instruction | |||
| 1337 | /// (either a switch or a branch on "X == c"). | |||
| 1338 | /// See if any of the predecessors of the terminator block are value comparisons | |||
| 1339 | /// on the same value. If so, and if safe to do so, fold them together. | |||
| 1340 | bool SimplifyCFGOpt::FoldValueComparisonIntoPredecessors(Instruction *TI, | |||
| 1341 | IRBuilder<> &Builder) { | |||
| 1342 | BasicBlock *BB = TI->getParent(); | |||
| 1343 | Value *CV = isValueEqualityComparison(TI); // CondVal | |||
| 1344 | assert(CV && "Not a comparison?")((void)0); | |||
| 1345 | ||||
| 1346 | bool Changed = false; | |||
| 1347 | ||||
| 1348 | SmallSetVector<BasicBlock *, 16> Preds(pred_begin(BB), pred_end(BB)); | |||
| 1349 | while (!Preds.empty()) { | |||
| 1350 | BasicBlock *Pred = Preds.pop_back_val(); | |||
| 1351 | Instruction *PTI = Pred->getTerminator(); | |||
| 1352 | ||||
| 1353 | // Don't try to fold into itself. | |||
| 1354 | if (Pred == BB) | |||
| 1355 | continue; | |||
| 1356 | ||||
| 1357 | // See if the predecessor is a comparison with the same value. | |||
| 1358 | Value *PCV = isValueEqualityComparison(PTI); // PredCondVal | |||
| 1359 | if (PCV != CV) | |||
| 1360 | continue; | |||
| 1361 | ||||
| 1362 | SmallSetVector<BasicBlock *, 4> FailBlocks; | |||
| 1363 | if (!SafeToMergeTerminators(TI, PTI, &FailBlocks)) { | |||
| 1364 | for (auto *Succ : FailBlocks) { | |||
| 1365 | if (!SplitBlockPredecessors(Succ, TI->getParent(), ".fold.split", DTU)) | |||
| 1366 | return false; | |||
| 1367 | } | |||
| 1368 | } | |||
| 1369 | ||||
| 1370 | PerformValueComparisonIntoPredecessorFolding(TI, CV, PTI, Builder); | |||
| 1371 | Changed = true; | |||
| 1372 | } | |||
| 1373 | return Changed; | |||
| 1374 | } | |||
| 1375 | ||||
| 1376 | // If we would need to insert a select that uses the value of this invoke | |||
| 1377 | // (comments in HoistThenElseCodeToIf explain why we would need to do this), we | |||
| 1378 | // can't hoist the invoke, as there is nowhere to put the select in this case. | |||
| 1379 | static bool isSafeToHoistInvoke(BasicBlock *BB1, BasicBlock *BB2, | |||
| 1380 | Instruction *I1, Instruction *I2) { | |||
| 1381 | for (BasicBlock *Succ : successors(BB1)) { | |||
| 1382 | for (const PHINode &PN : Succ->phis()) { | |||
| 1383 | Value *BB1V = PN.getIncomingValueForBlock(BB1); | |||
| 1384 | Value *BB2V = PN.getIncomingValueForBlock(BB2); | |||
| 1385 | if (BB1V != BB2V && (BB1V == I1 || BB2V == I2)) { | |||
| 1386 | return false; | |||
| 1387 | } | |||
| 1388 | } | |||
| 1389 | } | |||
| 1390 | return true; | |||
| 1391 | } | |||
| 1392 | ||||
| 1393 | static bool passingValueIsAlwaysUndefined(Value *V, Instruction *I, bool PtrValueMayBeModified = false); | |||
| 1394 | ||||
| 1395 | /// Given a conditional branch that goes to BB1 and BB2, hoist any common code | |||
| 1396 | /// in the two blocks up into the branch block. The caller of this function | |||
| 1397 | /// guarantees that BI's block dominates BB1 and BB2. If EqTermsOnly is given, | |||
| 1398 | /// only perform hoisting in case both blocks only contain a terminator. In that | |||
| 1399 | /// case, only the original BI will be replaced and selects for PHIs are added. | |||
| 1400 | bool SimplifyCFGOpt::HoistThenElseCodeToIf(BranchInst *BI, | |||
| 1401 | const TargetTransformInfo &TTI, | |||
| 1402 | bool EqTermsOnly) { | |||
| 1403 | // This does very trivial matching, with limited scanning, to find identical | |||
| 1404 | // instructions in the two blocks. In particular, we don't want to get into | |||
| 1405 | // O(M*N) situations here where M and N are the sizes of BB1 and BB2. As | |||
| 1406 | // such, we currently just scan for obviously identical instructions in an | |||
| 1407 | // identical order. | |||
| 1408 | BasicBlock *BB1 = BI->getSuccessor(0); // The true destination. | |||
| 1409 | BasicBlock *BB2 = BI->getSuccessor(1); // The false destination | |||
| 1410 | ||||
| 1411 | // If either of the blocks has it's address taken, then we can't do this fold, | |||
| 1412 | // because the code we'd hoist would no longer run when we jump into the block | |||
| 1413 | // by it's address. | |||
| 1414 | if (BB1->hasAddressTaken() || BB2->hasAddressTaken()) | |||
| 1415 | return false; | |||
| 1416 | ||||
| 1417 | BasicBlock::iterator BB1_Itr = BB1->begin(); | |||
| 1418 | BasicBlock::iterator BB2_Itr = BB2->begin(); | |||
| 1419 | ||||
| 1420 | Instruction *I1 = &*BB1_Itr++, *I2 = &*BB2_Itr++; | |||
| 1421 | // Skip debug info if it is not identical. | |||
| 1422 | DbgInfoIntrinsic *DBI1 = dyn_cast<DbgInfoIntrinsic>(I1); | |||
| 1423 | DbgInfoIntrinsic *DBI2 = dyn_cast<DbgInfoIntrinsic>(I2); | |||
| 1424 | if (!DBI1 || !DBI2 || !DBI1->isIdenticalToWhenDefined(DBI2)) { | |||
| 1425 | while (isa<DbgInfoIntrinsic>(I1)) | |||
| 1426 | I1 = &*BB1_Itr++; | |||
| 1427 | while (isa<DbgInfoIntrinsic>(I2)) | |||
| 1428 | I2 = &*BB2_Itr++; | |||
| 1429 | } | |||
| 1430 | // FIXME: Can we define a safety predicate for CallBr? | |||
| 1431 | if (isa<PHINode>(I1) || !I1->isIdenticalToWhenDefined(I2) || | |||
| 1432 | (isa<InvokeInst>(I1) && !isSafeToHoistInvoke(BB1, BB2, I1, I2)) || | |||
| 1433 | isa<CallBrInst>(I1)) | |||
| 1434 | return false; | |||
| 1435 | ||||
| 1436 | BasicBlock *BIParent = BI->getParent(); | |||
| 1437 | ||||
| 1438 | bool Changed = false; | |||
| 1439 | ||||
| 1440 | auto _ = make_scope_exit([&]() { | |||
| 1441 | if (Changed) | |||
| 1442 | ++NumHoistCommonCode; | |||
| 1443 | }); | |||
| 1444 | ||||
| 1445 | // Check if only hoisting terminators is allowed. This does not add new | |||
| 1446 | // instructions to the hoist location. | |||
| 1447 | if (EqTermsOnly) { | |||
| 1448 | // Skip any debug intrinsics, as they are free to hoist. | |||
| 1449 | auto *I1NonDbg = &*skipDebugIntrinsics(I1->getIterator()); | |||
| 1450 | auto *I2NonDbg = &*skipDebugIntrinsics(I2->getIterator()); | |||
| 1451 | if (!I1NonDbg->isIdenticalToWhenDefined(I2NonDbg)) | |||
| 1452 | return false; | |||
| 1453 | if (!I1NonDbg->isTerminator()) | |||
| 1454 | return false; | |||
| 1455 | // Now we know that we only need to hoist debug instrinsics and the | |||
| 1456 | // terminator. Let the loop below handle those 2 cases. | |||
| 1457 | } | |||
| 1458 | ||||
| 1459 | do { | |||
| 1460 | // If we are hoisting the terminator instruction, don't move one (making a | |||
| 1461 | // broken BB), instead clone it, and remove BI. | |||
| 1462 | if (I1->isTerminator()) | |||
| 1463 | goto HoistTerminator; | |||
| 1464 | ||||
| 1465 | // If we're going to hoist a call, make sure that the two instructions we're | |||
| 1466 | // commoning/hoisting are both marked with musttail, or neither of them is | |||
| 1467 | // marked as such. Otherwise, we might end up in a situation where we hoist | |||
| 1468 | // from a block where the terminator is a `ret` to a block where the terminator | |||
| 1469 | // is a `br`, and `musttail` calls expect to be followed by a return. | |||
| 1470 | auto *C1 = dyn_cast<CallInst>(I1); | |||
| 1471 | auto *C2 = dyn_cast<CallInst>(I2); | |||
| 1472 | if (C1 && C2) | |||
| 1473 | if (C1->isMustTailCall() != C2->isMustTailCall()) | |||
| 1474 | return Changed; | |||
| 1475 | ||||
| 1476 | if (!TTI.isProfitableToHoist(I1) || !TTI.isProfitableToHoist(I2)) | |||
| 1477 | return Changed; | |||
| 1478 | ||||
| 1479 | // If any of the two call sites has nomerge attribute, stop hoisting. | |||
| 1480 | if (const auto *CB1 = dyn_cast<CallBase>(I1)) | |||
| 1481 | if (CB1->cannotMerge()) | |||
| 1482 | return Changed; | |||
| 1483 | if (const auto *CB2 = dyn_cast<CallBase>(I2)) | |||
| 1484 | if (CB2->cannotMerge()) | |||
| 1485 | return Changed; | |||
| 1486 | ||||
| 1487 | if (isa<DbgInfoIntrinsic>(I1) || isa<DbgInfoIntrinsic>(I2)) { | |||
| 1488 | assert (isa<DbgInfoIntrinsic>(I1) && isa<DbgInfoIntrinsic>(I2))((void)0); | |||
| 1489 | // The debug location is an integral part of a debug info intrinsic | |||
| 1490 | // and can't be separated from it or replaced. Instead of attempting | |||
| 1491 | // to merge locations, simply hoist both copies of the intrinsic. | |||
| 1492 | BIParent->getInstList().splice(BI->getIterator(), | |||
| 1493 | BB1->getInstList(), I1); | |||
| 1494 | BIParent->getInstList().splice(BI->getIterator(), | |||
| 1495 | BB2->getInstList(), I2); | |||
| 1496 | Changed = true; | |||
| 1497 | } else { | |||
| 1498 | // For a normal instruction, we just move one to right before the branch, | |||
| 1499 | // then replace all uses of the other with the first. Finally, we remove | |||
| 1500 | // the now redundant second instruction. | |||
| 1501 | BIParent->getInstList().splice(BI->getIterator(), | |||
| 1502 | BB1->getInstList(), I1); | |||
| 1503 | if (!I2->use_empty()) | |||
| 1504 | I2->replaceAllUsesWith(I1); | |||
| 1505 | I1->andIRFlags(I2); | |||
| 1506 | unsigned KnownIDs[] = {LLVMContext::MD_tbaa, | |||
| 1507 | LLVMContext::MD_range, | |||
| 1508 | LLVMContext::MD_fpmath, | |||
| 1509 | LLVMContext::MD_invariant_load, | |||
| 1510 | LLVMContext::MD_nonnull, | |||
| 1511 | LLVMContext::MD_invariant_group, | |||
| 1512 | LLVMContext::MD_align, | |||
| 1513 | LLVMContext::MD_dereferenceable, | |||
| 1514 | LLVMContext::MD_dereferenceable_or_null, | |||
| 1515 | LLVMContext::MD_mem_parallel_loop_access, | |||
| 1516 | LLVMContext::MD_access_group, | |||
| 1517 | LLVMContext::MD_preserve_access_index}; | |||
| 1518 | combineMetadata(I1, I2, KnownIDs, true); | |||
| 1519 | ||||
| 1520 | // I1 and I2 are being combined into a single instruction. Its debug | |||
| 1521 | // location is the merged locations of the original instructions. | |||
| 1522 | I1->applyMergedLocation(I1->getDebugLoc(), I2->getDebugLoc()); | |||
| 1523 | ||||
| 1524 | I2->eraseFromParent(); | |||
| 1525 | Changed = true; | |||
| 1526 | } | |||
| 1527 | ++NumHoistCommonInstrs; | |||
| 1528 | ||||
| 1529 | I1 = &*BB1_Itr++; | |||
| 1530 | I2 = &*BB2_Itr++; | |||
| 1531 | // Skip debug info if it is not identical. | |||
| 1532 | DbgInfoIntrinsic *DBI1 = dyn_cast<DbgInfoIntrinsic>(I1); | |||
| 1533 | DbgInfoIntrinsic *DBI2 = dyn_cast<DbgInfoIntrinsic>(I2); | |||
| 1534 | if (!DBI1 || !DBI2 || !DBI1->isIdenticalToWhenDefined(DBI2)) { | |||
| 1535 | while (isa<DbgInfoIntrinsic>(I1)) | |||
| 1536 | I1 = &*BB1_Itr++; | |||
| 1537 | while (isa<DbgInfoIntrinsic>(I2)) | |||
| 1538 | I2 = &*BB2_Itr++; | |||
| 1539 | } | |||
| 1540 | } while (I1->isIdenticalToWhenDefined(I2)); | |||
| 1541 | ||||
| 1542 | return true; | |||
| 1543 | ||||
| 1544 | HoistTerminator: | |||
| 1545 | // It may not be possible to hoist an invoke. | |||
| 1546 | // FIXME: Can we define a safety predicate for CallBr? | |||
| 1547 | if (isa<InvokeInst>(I1) && !isSafeToHoistInvoke(BB1, BB2, I1, I2)) | |||
| 1548 | return Changed; | |||
| 1549 | ||||
| 1550 | // TODO: callbr hoisting currently disabled pending further study. | |||
| 1551 | if (isa<CallBrInst>(I1)) | |||
| 1552 | return Changed; | |||
| 1553 | ||||
| 1554 | for (BasicBlock *Succ : successors(BB1)) { | |||
| 1555 | for (PHINode &PN : Succ->phis()) { | |||
| 1556 | Value *BB1V = PN.getIncomingValueForBlock(BB1); | |||
| 1557 | Value *BB2V = PN.getIncomingValueForBlock(BB2); | |||
| 1558 | if (BB1V == BB2V) | |||
| 1559 | continue; | |||
| 1560 | ||||
| 1561 | // Check for passingValueIsAlwaysUndefined here because we would rather | |||
| 1562 | // eliminate undefined control flow then converting it to a select. | |||
| 1563 | if (passingValueIsAlwaysUndefined(BB1V, &PN) || | |||
| 1564 | passingValueIsAlwaysUndefined(BB2V, &PN)) | |||
| 1565 | return Changed; | |||
| 1566 | ||||
| 1567 | if (isa<ConstantExpr>(BB1V) && !isSafeToSpeculativelyExecute(BB1V)) | |||
| 1568 | return Changed; | |||
| 1569 | if (isa<ConstantExpr>(BB2V) && !isSafeToSpeculativelyExecute(BB2V)) | |||
| 1570 | return Changed; | |||
| 1571 | } | |||
| 1572 | } | |||
| 1573 | ||||
| 1574 | // Okay, it is safe to hoist the terminator. | |||
| 1575 | Instruction *NT = I1->clone(); | |||
| 1576 | BIParent->getInstList().insert(BI->getIterator(), NT); | |||
| 1577 | if (!NT->getType()->isVoidTy()) { | |||
| 1578 | I1->replaceAllUsesWith(NT); | |||
| 1579 | I2->replaceAllUsesWith(NT); | |||
| 1580 | NT->takeName(I1); | |||
| 1581 | } | |||
| 1582 | Changed = true; | |||
| 1583 | ++NumHoistCommonInstrs; | |||
| 1584 | ||||
| 1585 | // Ensure terminator gets a debug location, even an unknown one, in case | |||
| 1586 | // it involves inlinable calls. | |||
| 1587 | NT->applyMergedLocation(I1->getDebugLoc(), I2->getDebugLoc()); | |||
| 1588 | ||||
| 1589 | // PHIs created below will adopt NT's merged DebugLoc. | |||
| 1590 | IRBuilder<NoFolder> Builder(NT); | |||
| 1591 | ||||
| 1592 | // Hoisting one of the terminators from our successor is a great thing. | |||
| 1593 | // Unfortunately, the successors of the if/else blocks may have PHI nodes in | |||
| 1594 | // them. If they do, all PHI entries for BB1/BB2 must agree for all PHI | |||
| 1595 | // nodes, so we insert select instruction to compute the final result. | |||
| 1596 | std::map<std::pair<Value *, Value *>, SelectInst *> InsertedSelects; | |||
| 1597 | for (BasicBlock *Succ : successors(BB1)) { | |||
| 1598 | for (PHINode &PN : Succ->phis()) { | |||
| 1599 | Value *BB1V = PN.getIncomingValueForBlock(BB1); | |||
| 1600 | Value *BB2V = PN.getIncomingValueForBlock(BB2); | |||
| 1601 | if (BB1V == BB2V) | |||
| 1602 | continue; | |||
| 1603 | ||||
| 1604 | // These values do not agree. Insert a select instruction before NT | |||
| 1605 | // that determines the right value. | |||
| 1606 | SelectInst *&SI = InsertedSelects[std::make_pair(BB1V, BB2V)]; | |||
| 1607 | if (!SI) { | |||
| 1608 | // Propagate fast-math-flags from phi node to its replacement select. | |||
| 1609 | IRBuilder<>::FastMathFlagGuard FMFGuard(Builder); | |||
| 1610 | if (isa<FPMathOperator>(PN)) | |||
| 1611 | Builder.setFastMathFlags(PN.getFastMathFlags()); | |||
| 1612 | ||||
| 1613 | SI = cast<SelectInst>( | |||
| 1614 | Builder.CreateSelect(BI->getCondition(), BB1V, BB2V, | |||
| 1615 | BB1V->getName() + "." + BB2V->getName(), BI)); | |||
| 1616 | } | |||
| 1617 | ||||
| 1618 | // Make the PHI node use the select for all incoming values for BB1/BB2 | |||
| 1619 | for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i) | |||
| 1620 | if (PN.getIncomingBlock(i) == BB1 || PN.getIncomingBlock(i) == BB2) | |||
| 1621 | PN.setIncomingValue(i, SI); | |||
| 1622 | } | |||
| 1623 | } | |||
| 1624 | ||||
| 1625 | SmallVector<DominatorTree::UpdateType, 4> Updates; | |||
| 1626 | ||||
| 1627 | // Update any PHI nodes in our new successors. | |||
| 1628 | for (BasicBlock *Succ : successors(BB1)) { | |||
| 1629 | AddPredecessorToBlock(Succ, BIParent, BB1); | |||
| 1630 | if (DTU) | |||
| 1631 | Updates.push_back({DominatorTree::Insert, BIParent, Succ}); | |||
| 1632 | } | |||
| 1633 | ||||
| 1634 | if (DTU) | |||
| 1635 | for (BasicBlock *Succ : successors(BI)) | |||
| 1636 | Updates.push_back({DominatorTree::Delete, BIParent, Succ}); | |||
| 1637 | ||||
| 1638 | EraseTerminatorAndDCECond(BI); | |||
| 1639 | if (DTU) | |||
| 1640 | DTU->applyUpdates(Updates); | |||
| 1641 | return Changed; | |||
| 1642 | } | |||
| 1643 | ||||
| 1644 | // Check lifetime markers. | |||
| 1645 | static bool isLifeTimeMarker(const Instruction *I) { | |||
| 1646 | if (auto II = dyn_cast<IntrinsicInst>(I)) { | |||
| 1647 | switch (II->getIntrinsicID()) { | |||
| 1648 | default: | |||
| 1649 | break; | |||
| 1650 | case Intrinsic::lifetime_start: | |||
| 1651 | case Intrinsic::lifetime_end: | |||
| 1652 | return true; | |||
| 1653 | } | |||
| 1654 | } | |||
| 1655 | return false; | |||
| 1656 | } | |||
| 1657 | ||||
| 1658 | // TODO: Refine this. This should avoid cases like turning constant memcpy sizes | |||
| 1659 | // into variables. | |||
| 1660 | static bool replacingOperandWithVariableIsCheap(const Instruction *I, | |||
| 1661 | int OpIdx) { | |||
| 1662 | return !isa<IntrinsicInst>(I); | |||
| 1663 | } | |||
| 1664 | ||||
| 1665 | // All instructions in Insts belong to different blocks that all unconditionally | |||
| 1666 | // branch to a common successor. Analyze each instruction and return true if it | |||
| 1667 | // would be possible to sink them into their successor, creating one common | |||
| 1668 | // instruction instead. For every value that would be required to be provided by | |||
| 1669 | // PHI node (because an operand varies in each input block), add to PHIOperands. | |||
| 1670 | static bool canSinkInstructions( | |||
| 1671 | ArrayRef<Instruction *> Insts, | |||
| 1672 | DenseMap<Instruction *, SmallVector<Value *, 4>> &PHIOperands) { | |||
| 1673 | // Prune out obviously bad instructions to move. Each instruction must have | |||
| 1674 | // exactly zero or one use, and we check later that use is by a single, common | |||
| 1675 | // PHI instruction in the successor. | |||
| 1676 | bool HasUse = !Insts.front()->user_empty(); | |||
| 1677 | for (auto *I : Insts) { | |||
| 1678 | // These instructions may change or break semantics if moved. | |||
| 1679 | if (isa<PHINode>(I) || I->isEHPad() || isa<AllocaInst>(I) || | |||
| 1680 | I->getType()->isTokenTy()) | |||
| 1681 | return false; | |||
| 1682 | ||||
| 1683 | // Do not try to sink an instruction in an infinite loop - it can cause | |||
| 1684 | // this algorithm to infinite loop. | |||
| 1685 | if (I->getParent()->getSingleSuccessor() == I->getParent()) | |||
| 1686 | return false; | |||
| 1687 | ||||
| 1688 | // Conservatively return false if I is an inline-asm instruction. Sinking | |||
| 1689 | // and merging inline-asm instructions can potentially create arguments | |||
| 1690 | // that cannot satisfy the inline-asm constraints. | |||
| 1691 | // If the instruction has nomerge attribute, return false. | |||
| 1692 | if (const auto *C = dyn_cast<CallBase>(I)) | |||
| 1693 | if (C->isInlineAsm() || C->cannotMerge()) | |||
| 1694 | return false; | |||
| 1695 | ||||
| 1696 | // Each instruction must have zero or one use. | |||
| 1697 | if (HasUse && !I->hasOneUse()) | |||
| 1698 | return false; | |||
| 1699 | if (!HasUse && !I->user_empty()) | |||
| 1700 | return false; | |||
| 1701 | } | |||
| 1702 | ||||
| 1703 | const Instruction *I0 = Insts.front(); | |||
| 1704 | for (auto *I : Insts) | |||
| 1705 | if (!I->isSameOperationAs(I0)) | |||
| 1706 | return false; | |||
| 1707 | ||||
| 1708 | // All instructions in Insts are known to be the same opcode. If they have a | |||
| 1709 | // use, check that the only user is a PHI or in the same block as the | |||
| 1710 | // instruction, because if a user is in the same block as an instruction we're | |||
| 1711 | // contemplating sinking, it must already be determined to be sinkable. | |||
| 1712 | if (HasUse) { | |||
| 1713 | auto *PNUse = dyn_cast<PHINode>(*I0->user_begin()); | |||
| 1714 | auto *Succ = I0->getParent()->getTerminator()->getSuccessor(0); | |||
| 1715 | if (!all_of(Insts, [&PNUse,&Succ](const Instruction *I) -> bool { | |||
| 1716 | auto *U = cast<Instruction>(*I->user_begin()); | |||
| 1717 | return (PNUse && | |||
| 1718 | PNUse->getParent() == Succ && | |||
| 1719 | PNUse->getIncomingValueForBlock(I->getParent()) == I) || | |||
| 1720 | U->getParent() == I->getParent(); | |||
| 1721 | })) | |||
| 1722 | return false; | |||
| 1723 | } | |||
| 1724 | ||||
| 1725 | // Because SROA can't handle speculating stores of selects, try not to sink | |||
| 1726 | // loads, stores or lifetime markers of allocas when we'd have to create a | |||
| 1727 | // PHI for the address operand. Also, because it is likely that loads or | |||
| 1728 | // stores of allocas will disappear when Mem2Reg/SROA is run, don't sink | |||
| 1729 | // them. | |||
| 1730 | // This can cause code churn which can have unintended consequences down | |||
| 1731 | // the line - see https://llvm.org/bugs/show_bug.cgi?id=30244. | |||
| 1732 | // FIXME: This is a workaround for a deficiency in SROA - see | |||
| 1733 | // https://llvm.org/bugs/show_bug.cgi?id=30188 | |||
| 1734 | if (isa<StoreInst>(I0) && any_of(Insts, [](const Instruction *I) { | |||
| 1735 | return isa<AllocaInst>(I->getOperand(1)->stripPointerCasts()); | |||
| 1736 | })) | |||
| 1737 | return false; | |||
| 1738 | if (isa<LoadInst>(I0) && any_of(Insts, [](const Instruction *I) { | |||
| 1739 | return isa<AllocaInst>(I->getOperand(0)->stripPointerCasts()); | |||
| 1740 | })) | |||
| 1741 | return false; | |||
| 1742 | if (isLifeTimeMarker(I0) && any_of(Insts, [](const Instruction *I) { | |||
| 1743 | return isa<AllocaInst>(I->getOperand(1)->stripPointerCasts()); | |||
| 1744 | })) | |||
| 1745 | return false; | |||
| 1746 | ||||
| 1747 | // For calls to be sinkable, they must all be indirect, or have same callee. | |||
| 1748 | // I.e. if we have two direct calls to different callees, we don't want to | |||
| 1749 | // turn that into an indirect call. Likewise, if we have an indirect call, | |||
| 1750 | // and a direct call, we don't actually want to have a single indirect call. | |||
| 1751 | if (isa<CallBase>(I0)) { | |||
| 1752 | auto IsIndirectCall = [](const Instruction *I) { | |||
| 1753 | return cast<CallBase>(I)->isIndirectCall(); | |||
| 1754 | }; | |||
| 1755 | bool HaveIndirectCalls = any_of(Insts, IsIndirectCall); | |||
| 1756 | bool AllCallsAreIndirect = all_of(Insts, IsIndirectCall); | |||
| 1757 | if (HaveIndirectCalls) { | |||
| 1758 | if (!AllCallsAreIndirect) | |||
| 1759 | return false; | |||
| 1760 | } else { | |||
| 1761 | // All callees must be identical. | |||
| 1762 | Value *Callee = nullptr; | |||
| 1763 | for (const Instruction *I : Insts) { | |||
| 1764 | Value *CurrCallee = cast<CallBase>(I)->getCalledOperand(); | |||
| 1765 | if (!Callee) | |||
| 1766 | Callee = CurrCallee; | |||
| 1767 | else if (Callee != CurrCallee) | |||
| 1768 | return false; | |||
| 1769 | } | |||
| 1770 | } | |||
| 1771 | } | |||
| 1772 | ||||
| 1773 | for (unsigned OI = 0, OE = I0->getNumOperands(); OI != OE; ++OI) { | |||
| 1774 | Value *Op = I0->getOperand(OI); | |||
| 1775 | if (Op->getType()->isTokenTy()) | |||
| 1776 | // Don't touch any operand of token type. | |||
| 1777 | return false; | |||
| 1778 | ||||
| 1779 | auto SameAsI0 = [&I0, OI](const Instruction *I) { | |||
| 1780 | assert(I->getNumOperands() == I0->getNumOperands())((void)0); | |||
| 1781 | return I->getOperand(OI) == I0->getOperand(OI); | |||
| 1782 | }; | |||
| 1783 | if (!all_of(Insts, SameAsI0)) { | |||
| 1784 | if ((isa<Constant>(Op) && !replacingOperandWithVariableIsCheap(I0, OI)) || | |||
| 1785 | !canReplaceOperandWithVariable(I0, OI)) | |||
| 1786 | // We can't create a PHI from this GEP. | |||
| 1787 | return false; | |||
| 1788 | for (auto *I : Insts) | |||
| 1789 | PHIOperands[I].push_back(I->getOperand(OI)); | |||
| 1790 | } | |||
| 1791 | } | |||
| 1792 | return true; | |||
| 1793 | } | |||
| 1794 | ||||
| 1795 | // Assuming canSinkInstructions(Blocks) has returned true, sink the last | |||
| 1796 | // instruction of every block in Blocks to their common successor, commoning | |||
| 1797 | // into one instruction. | |||
| 1798 | static bool sinkLastInstruction(ArrayRef<BasicBlock*> Blocks) { | |||
| 1799 | auto *BBEnd = Blocks[0]->getTerminator()->getSuccessor(0); | |||
| 1800 | ||||
| 1801 | // canSinkInstructions returning true guarantees that every block has at | |||
| 1802 | // least one non-terminator instruction. | |||
| 1803 | SmallVector<Instruction*,4> Insts; | |||
| 1804 | for (auto *BB : Blocks) { | |||
| 1805 | Instruction *I = BB->getTerminator(); | |||
| 1806 | do { | |||
| 1807 | I = I->getPrevNode(); | |||
| 1808 | } while (isa<DbgInfoIntrinsic>(I) && I != &BB->front()); | |||
| 1809 | if (!isa<DbgInfoIntrinsic>(I)) | |||
| 1810 | Insts.push_back(I); | |||
| 1811 | } | |||
| 1812 | ||||
| 1813 | // The only checking we need to do now is that all users of all instructions | |||
| 1814 | // are the same PHI node. canSinkInstructions should have checked this but | |||
| 1815 | // it is slightly over-aggressive - it gets confused by commutative | |||
| 1816 | // instructions so double-check it here. | |||
| 1817 | Instruction *I0 = Insts.front(); | |||
| 1818 | if (!I0->user_empty()) { | |||
| 1819 | auto *PNUse = dyn_cast<PHINode>(*I0->user_begin()); | |||
| 1820 | if (!all_of(Insts, [&PNUse](const Instruction *I) -> bool { | |||
| 1821 | auto *U = cast<Instruction>(*I->user_begin()); | |||
| 1822 | return U == PNUse; | |||
| 1823 | })) | |||
| 1824 | return false; | |||
| 1825 | } | |||
| 1826 | ||||
| 1827 | // We don't need to do any more checking here; canSinkInstructions should | |||
| 1828 | // have done it all for us. | |||
| 1829 | SmallVector<Value*, 4> NewOperands; | |||
| 1830 | for (unsigned O = 0, E = I0->getNumOperands(); O != E; ++O) { | |||
| 1831 | // This check is different to that in canSinkInstructions. There, we | |||
| 1832 | // cared about the global view once simplifycfg (and instcombine) have | |||
| 1833 | // completed - it takes into account PHIs that become trivially | |||
| 1834 | // simplifiable. However here we need a more local view; if an operand | |||
| 1835 | // differs we create a PHI and rely on instcombine to clean up the very | |||
| 1836 | // small mess we may make. | |||
| 1837 | bool NeedPHI = any_of(Insts, [&I0, O](const Instruction *I) { | |||
| 1838 | return I->getOperand(O) != I0->getOperand(O); | |||
| 1839 | }); | |||
| 1840 | if (!NeedPHI) { | |||
| 1841 | NewOperands.push_back(I0->getOperand(O)); | |||
| 1842 | continue; | |||
| 1843 | } | |||
| 1844 | ||||
| 1845 | // Create a new PHI in the successor block and populate it. | |||
| 1846 | auto *Op = I0->getOperand(O); | |||
| 1847 | assert(!Op->getType()->isTokenTy() && "Can't PHI tokens!")((void)0); | |||
| 1848 | auto *PN = PHINode::Create(Op->getType(), Insts.size(), | |||
| 1849 | Op->getName() + ".sink", &BBEnd->front()); | |||
| 1850 | for (auto *I : Insts) | |||
| 1851 | PN->addIncoming(I->getOperand(O), I->getParent()); | |||
| 1852 | NewOperands.push_back(PN); | |||
| 1853 | } | |||
| 1854 | ||||
| 1855 | // Arbitrarily use I0 as the new "common" instruction; remap its operands | |||
| 1856 | // and move it to the start of the successor block. | |||
| 1857 | for (unsigned O = 0, E = I0->getNumOperands(); O != E; ++O) | |||
| 1858 | I0->getOperandUse(O).set(NewOperands[O]); | |||
| 1859 | I0->moveBefore(&*BBEnd->getFirstInsertionPt()); | |||
| 1860 | ||||
| 1861 | // Update metadata and IR flags, and merge debug locations. | |||
| 1862 | for (auto *I : Insts) | |||
| 1863 | if (I != I0) { | |||
| 1864 | // The debug location for the "common" instruction is the merged locations | |||
| 1865 | // of all the commoned instructions. We start with the original location | |||
| 1866 | // of the "common" instruction and iteratively merge each location in the | |||
| 1867 | // loop below. | |||
| 1868 | // This is an N-way merge, which will be inefficient if I0 is a CallInst. | |||
| 1869 | // However, as N-way merge for CallInst is rare, so we use simplified API | |||
| 1870 | // instead of using complex API for N-way merge. | |||
| 1871 | I0->applyMergedLocation(I0->getDebugLoc(), I->getDebugLoc()); | |||
| 1872 | combineMetadataForCSE(I0, I, true); | |||
| 1873 | I0->andIRFlags(I); | |||
| 1874 | } | |||
| 1875 | ||||
| 1876 | if (!I0->user_empty()) { | |||
| 1877 | // canSinkLastInstruction checked that all instructions were used by | |||
| 1878 | // one and only one PHI node. Find that now, RAUW it to our common | |||
| 1879 | // instruction and nuke it. | |||
| 1880 | auto *PN = cast<PHINode>(*I0->user_begin()); | |||
| 1881 | PN->replaceAllUsesWith(I0); | |||
| 1882 | PN->eraseFromParent(); | |||
| 1883 | } | |||
| 1884 | ||||
| 1885 | // Finally nuke all instructions apart from the common instruction. | |||
| 1886 | for (auto *I : Insts) | |||
| 1887 | if (I != I0) | |||
| 1888 | I->eraseFromParent(); | |||
| 1889 | ||||
| 1890 | return true; | |||
| 1891 | } | |||
| 1892 | ||||
| 1893 | namespace { | |||
| 1894 | ||||
| 1895 | // LockstepReverseIterator - Iterates through instructions | |||
| 1896 | // in a set of blocks in reverse order from the first non-terminator. | |||
| 1897 | // For example (assume all blocks have size n): | |||
| 1898 | // LockstepReverseIterator I([B1, B2, B3]); | |||
| 1899 | // *I-- = [B1[n], B2[n], B3[n]]; | |||
| 1900 | // *I-- = [B1[n-1], B2[n-1], B3[n-1]]; | |||
| 1901 | // *I-- = [B1[n-2], B2[n-2], B3[n-2]]; | |||
| 1902 | // ... | |||
| 1903 | class LockstepReverseIterator { | |||
| 1904 | ArrayRef<BasicBlock*> Blocks; | |||
| 1905 | SmallVector<Instruction*,4> Insts; | |||
| 1906 | bool Fail; | |||
| 1907 | ||||
| 1908 | public: | |||
| 1909 | LockstepReverseIterator(ArrayRef<BasicBlock*> Blocks) : Blocks(Blocks) { | |||
| 1910 | reset(); | |||
| 1911 | } | |||
| 1912 | ||||
| 1913 | void reset() { | |||
| 1914 | Fail = false; | |||
| 1915 | Insts.clear(); | |||
| 1916 | for (auto *BB : Blocks) { | |||
| 1917 | Instruction *Inst = BB->getTerminator(); | |||
| 1918 | for (Inst = Inst->getPrevNode(); Inst && isa<DbgInfoIntrinsic>(Inst);) | |||
| 1919 | Inst = Inst->getPrevNode(); | |||
| 1920 | if (!Inst) { | |||
| 1921 | // Block wasn't big enough. | |||
| 1922 | Fail = true; | |||
| 1923 | return; | |||
| 1924 | } | |||
| 1925 | Insts.push_back(Inst); | |||
| 1926 | } | |||
| 1927 | } | |||
| 1928 | ||||
| 1929 | bool isValid() const { | |||
| 1930 | return !Fail; | |||
| 1931 | } | |||
| 1932 | ||||
| 1933 | void operator--() { | |||
| 1934 | if (Fail) | |||
| 1935 | return; | |||
| 1936 | for (auto *&Inst : Insts) { | |||
| 1937 | for (Inst = Inst->getPrevNode(); Inst && isa<DbgInfoIntrinsic>(Inst);) | |||
| 1938 | Inst = Inst->getPrevNode(); | |||
| 1939 | // Already at beginning of block. | |||
| 1940 | if (!Inst) { | |||
| 1941 | Fail = true; | |||
| 1942 | return; | |||
| 1943 | } | |||
| 1944 | } | |||
| 1945 | } | |||
| 1946 | ||||
| 1947 | void operator++() { | |||
| 1948 | if (Fail) | |||
| 1949 | return; | |||
| 1950 | for (auto *&Inst : Insts) { | |||
| 1951 | for (Inst = Inst->getNextNode(); Inst && isa<DbgInfoIntrinsic>(Inst);) | |||
| 1952 | Inst = Inst->getNextNode(); | |||
| 1953 | // Already at end of block. | |||
| 1954 | if (!Inst) { | |||
| 1955 | Fail = true; | |||
| 1956 | return; | |||
| 1957 | } | |||
| 1958 | } | |||
| 1959 | } | |||
| 1960 | ||||
| 1961 | ArrayRef<Instruction*> operator * () const { | |||
| 1962 | return Insts; | |||
| 1963 | } | |||
| 1964 | }; | |||
| 1965 | ||||
| 1966 | } // end anonymous namespace | |||
| 1967 | ||||
| 1968 | /// Check whether BB's predecessors end with unconditional branches. If it is | |||
| 1969 | /// true, sink any common code from the predecessors to BB. | |||
| 1970 | static bool SinkCommonCodeFromPredecessors(BasicBlock *BB, | |||
| 1971 | DomTreeUpdater *DTU) { | |||
| 1972 | // We support two situations: | |||
| 1973 | // (1) all incoming arcs are unconditional | |||
| 1974 | // (2) there are non-unconditional incoming arcs | |||
| 1975 | // | |||
| 1976 | // (2) is very common in switch defaults and | |||
| 1977 | // else-if patterns; | |||
| 1978 | // | |||
| 1979 | // if (a) f(1); | |||
| 1980 | // else if (b) f(2); | |||
| 1981 | // | |||
| 1982 | // produces: | |||
| 1983 | // | |||
| 1984 | // [if] | |||
| 1985 | // / \ | |||
| 1986 | // [f(1)] [if] | |||
| 1987 | // | | \ | |||
| 1988 | // | | | | |||
| 1989 | // | [f(2)]| | |||
| 1990 | // \ | / | |||
| 1991 | // [ end ] | |||
| 1992 | // | |||
| 1993 | // [end] has two unconditional predecessor arcs and one conditional. The | |||
| 1994 | // conditional refers to the implicit empty 'else' arc. This conditional | |||
| 1995 | // arc can also be caused by an empty default block in a switch. | |||
| 1996 | // | |||
| 1997 | // In this case, we attempt to sink code from all *unconditional* arcs. | |||
| 1998 | // If we can sink instructions from these arcs (determined during the scan | |||
| 1999 | // phase below) we insert a common successor for all unconditional arcs and | |||
| 2000 | // connect that to [end], to enable sinking: | |||
| 2001 | // | |||
| 2002 | // [if] | |||
| 2003 | // / \ | |||
| 2004 | // [x(1)] [if] | |||
| 2005 | // | | \ | |||
| 2006 | // | | \ | |||
| 2007 | // | [x(2)] | | |||
| 2008 | // \ / | | |||
| 2009 | // [sink.split] | | |||
| 2010 | // \ / | |||
| 2011 | // [ end ] | |||
| 2012 | // | |||
| 2013 | SmallVector<BasicBlock*,4> UnconditionalPreds; | |||
| 2014 | bool HaveNonUnconditionalPredecessors = false; | |||
| 2015 | for (auto *PredBB : predecessors(BB)) { | |||
| 2016 | auto *PredBr = dyn_cast<BranchInst>(PredBB->getTerminator()); | |||
| 2017 | if (PredBr && PredBr->isUnconditional()) | |||
| 2018 | UnconditionalPreds.push_back(PredBB); | |||
| 2019 | else | |||
| 2020 | HaveNonUnconditionalPredecessors = true; | |||
| 2021 | } | |||
| 2022 | if (UnconditionalPreds.size() < 2) | |||
| 2023 | return false; | |||
| 2024 | ||||
| 2025 | // We take a two-step approach to tail sinking. First we scan from the end of | |||
| 2026 | // each block upwards in lockstep. If the n'th instruction from the end of each | |||
| 2027 | // block can be sunk, those instructions are added to ValuesToSink and we | |||
| 2028 | // carry on. If we can sink an instruction but need to PHI-merge some operands | |||
| 2029 | // (because they're not identical in each instruction) we add these to | |||
| 2030 | // PHIOperands. | |||
| 2031 | int ScanIdx = 0; | |||
| 2032 | SmallPtrSet<Value*,4> InstructionsToSink; | |||
| 2033 | DenseMap<Instruction*, SmallVector<Value*,4>> PHIOperands; | |||
| 2034 | LockstepReverseIterator LRI(UnconditionalPreds); | |||
| 2035 | while (LRI.isValid() && | |||
| 2036 | canSinkInstructions(*LRI, PHIOperands)) { | |||
| 2037 | LLVM_DEBUG(dbgs() << "SINK: instruction can be sunk: " << *(*LRI)[0]do { } while (false) | |||
| 2038 | << "\n")do { } while (false); | |||
| 2039 | InstructionsToSink.insert((*LRI).begin(), (*LRI).end()); | |||
| 2040 | ++ScanIdx; | |||
| 2041 | --LRI; | |||
| 2042 | } | |||
| 2043 | ||||
| 2044 | // If no instructions can be sunk, early-return. | |||
| 2045 | if (ScanIdx == 0) | |||
| 2046 | return false; | |||
| 2047 | ||||
| 2048 | // Okay, we *could* sink last ScanIdx instructions. But how many can we | |||
| 2049 | // actually sink before encountering instruction that is unprofitable to sink? | |||
| 2050 | auto ProfitableToSinkInstruction = [&](LockstepReverseIterator &LRI) { | |||
| 2051 | unsigned NumPHIdValues = 0; | |||
| 2052 | for (auto *I : *LRI) | |||
| 2053 | for (auto *V : PHIOperands[I]) { | |||
| 2054 | if (InstructionsToSink.count(V) == 0) | |||
| 2055 | ++NumPHIdValues; | |||
| 2056 | // FIXME: this check is overly optimistic. We may end up not sinking | |||
| 2057 | // said instruction, due to the very same profitability check. | |||
| 2058 | // See @creating_too_many_phis in sink-common-code.ll. | |||
| 2059 | } | |||
| 2060 | LLVM_DEBUG(dbgs() << "SINK: #phid values: " << NumPHIdValues << "\n")do { } while (false); | |||
| 2061 | unsigned NumPHIInsts = NumPHIdValues / UnconditionalPreds.size(); | |||
| 2062 | if ((NumPHIdValues % UnconditionalPreds.size()) != 0) | |||
| 2063 | NumPHIInsts++; | |||
| 2064 | ||||
| 2065 | return NumPHIInsts <= 1; | |||
| 2066 | }; | |||
| 2067 | ||||
| 2068 | // We've determined that we are going to sink last ScanIdx instructions, | |||
| 2069 | // and recorded them in InstructionsToSink. Now, some instructions may be | |||
| 2070 | // unprofitable to sink. But that determination depends on the instructions | |||
| 2071 | // that we are going to sink. | |||
| 2072 | ||||
| 2073 | // First, forward scan: find the first instruction unprofitable to sink, | |||
| 2074 | // recording all the ones that are profitable to sink. | |||
| 2075 | // FIXME: would it be better, after we detect that not all are profitable. | |||
| 2076 | // to either record the profitable ones, or erase the unprofitable ones? | |||
| 2077 | // Maybe we need to choose (at runtime) the one that will touch least instrs? | |||
| 2078 | LRI.reset(); | |||
| 2079 | int Idx = 0; | |||
| 2080 | SmallPtrSet<Value *, 4> InstructionsProfitableToSink; | |||
| 2081 | while (Idx < ScanIdx) { | |||
| 2082 | if (!ProfitableToSinkInstruction(LRI)) { | |||
| 2083 | // Too many PHIs would be created. | |||
| 2084 | LLVM_DEBUG(do { } while (false) | |||
| 2085 | dbgs() << "SINK: stopping here, too many PHIs would be created!\n")do { } while (false); | |||
| 2086 | break; | |||
| 2087 | } | |||
| 2088 | InstructionsProfitableToSink.insert((*LRI).begin(), (*LRI).end()); | |||
| 2089 | --LRI; | |||
| 2090 | ++Idx; | |||
| 2091 | } | |||
| 2092 | ||||
| 2093 | // If no instructions can be sunk, early-return. | |||
| 2094 | if (Idx == 0) | |||
| 2095 | return false; | |||
| 2096 | ||||
| 2097 | // Did we determine that (only) some instructions are unprofitable to sink? | |||
| 2098 | if (Idx < ScanIdx) { | |||
| 2099 | // Okay, some instructions are unprofitable. | |||
| 2100 | ScanIdx = Idx; | |||
| 2101 | InstructionsToSink = InstructionsProfitableToSink; | |||
| 2102 | ||||
| 2103 | // But, that may make other instructions unprofitable, too. | |||
| 2104 | // So, do a backward scan, do any earlier instructions become unprofitable? | |||
| 2105 | assert(!ProfitableToSinkInstruction(LRI) &&((void)0) | |||
| 2106 | "We already know that the last instruction is unprofitable to sink")((void)0); | |||
| 2107 | ++LRI; | |||
| 2108 | --Idx; | |||
| 2109 | while (Idx >= 0) { | |||
| 2110 | // If we detect that an instruction becomes unprofitable to sink, | |||
| 2111 | // all earlier instructions won't be sunk either, | |||
| 2112 | // so preemptively keep InstructionsProfitableToSink in sync. | |||
| 2113 | // FIXME: is this the most performant approach? | |||
| 2114 | for (auto *I : *LRI) | |||
| 2115 | InstructionsProfitableToSink.erase(I); | |||
| 2116 | if (!ProfitableToSinkInstruction(LRI)) { | |||
| 2117 | // Everything starting with this instruction won't be sunk. | |||
| 2118 | ScanIdx = Idx; | |||
| 2119 | InstructionsToSink = InstructionsProfitableToSink; | |||
| 2120 | } | |||
| 2121 | ++LRI; | |||
| 2122 | --Idx; | |||
| 2123 | } | |||
| 2124 | } | |||
| 2125 | ||||
| 2126 | // If no instructions can be sunk, early-return. | |||
| 2127 | if (ScanIdx == 0) | |||
| 2128 | return false; | |||
| 2129 | ||||
| 2130 | bool Changed = false; | |||
| 2131 | ||||
| 2132 | if (HaveNonUnconditionalPredecessors) { | |||
| 2133 | // It is always legal to sink common instructions from unconditional | |||
| 2134 | // predecessors. However, if not all predecessors are unconditional, | |||
| 2135 | // this transformation might be pessimizing. So as a rule of thumb, | |||
| 2136 | // don't do it unless we'd sink at least one non-speculatable instruction. | |||
| 2137 | // See https://bugs.llvm.org/show_bug.cgi?id=30244 | |||
| 2138 | LRI.reset(); | |||
| 2139 | int Idx = 0; | |||
| 2140 | bool Profitable = false; | |||
| 2141 | while (Idx < ScanIdx) { | |||
| 2142 | if (!isSafeToSpeculativelyExecute((*LRI)[0])) { | |||
| 2143 | Profitable = true; | |||
| 2144 | break; | |||
| 2145 | } | |||
| 2146 | --LRI; | |||
| 2147 | ++Idx; | |||
| 2148 | } | |||
| 2149 | if (!Profitable) | |||
| 2150 | return false; | |||
| 2151 | ||||
| 2152 | LLVM_DEBUG(dbgs() << "SINK: Splitting edge\n")do { } while (false); | |||
| 2153 | // We have a conditional edge and we're going to sink some instructions. | |||
| 2154 | // Insert a new block postdominating all blocks we're going to sink from. | |||
| 2155 | if (!SplitBlockPredecessors(BB, UnconditionalPreds, ".sink.split", DTU)) | |||
| 2156 | // Edges couldn't be split. | |||
| 2157 | return false; | |||
| 2158 | Changed = true; | |||
| 2159 | } | |||
| 2160 | ||||
| 2161 | // Now that we've analyzed all potential sinking candidates, perform the | |||
| 2162 | // actual sink. We iteratively sink the last non-terminator of the source | |||
| 2163 | // blocks into their common successor unless doing so would require too | |||
| 2164 | // many PHI instructions to be generated (currently only one PHI is allowed | |||
| 2165 | // per sunk instruction). | |||
| 2166 | // | |||
| 2167 | // We can use InstructionsToSink to discount values needing PHI-merging that will | |||
| 2168 | // actually be sunk in a later iteration. This allows us to be more | |||
| 2169 | // aggressive in what we sink. This does allow a false positive where we | |||
| 2170 | // sink presuming a later value will also be sunk, but stop half way through | |||
| 2171 | // and never actually sink it which means we produce more PHIs than intended. | |||
| 2172 | // This is unlikely in practice though. | |||
| 2173 | int SinkIdx = 0; | |||
| 2174 | for (; SinkIdx != ScanIdx; ++SinkIdx) { | |||
| 2175 | LLVM_DEBUG(dbgs() << "SINK: Sink: "do { } while (false) | |||
| 2176 | << *UnconditionalPreds[0]->getTerminator()->getPrevNode()do { } while (false) | |||
| 2177 | << "\n")do { } while (false); | |||
| 2178 | ||||
| 2179 | // Because we've sunk every instruction in turn, the current instruction to | |||
| 2180 | // sink is always at index 0. | |||
| 2181 | LRI.reset(); | |||
| 2182 | ||||
| 2183 | if (!sinkLastInstruction(UnconditionalPreds)) { | |||
| 2184 | LLVM_DEBUG(do { } while (false) | |||
| 2185 | dbgs()do { } while (false) | |||
| 2186 | << "SINK: stopping here, failed to actually sink instruction!\n")do { } while (false); | |||
| 2187 | break; | |||
| 2188 | } | |||
| 2189 | ||||
| 2190 | NumSinkCommonInstrs++; | |||
| 2191 | Changed = true; | |||
| 2192 | } | |||
| 2193 | if (SinkIdx != 0) | |||
| 2194 | ++NumSinkCommonCode; | |||
| 2195 | return Changed; | |||
| 2196 | } | |||
| 2197 | ||||
| 2198 | /// Determine if we can hoist sink a sole store instruction out of a | |||
| 2199 | /// conditional block. | |||
| 2200 | /// | |||
| 2201 | /// We are looking for code like the following: | |||
| 2202 | /// BrBB: | |||
| 2203 | /// store i32 %add, i32* %arrayidx2 | |||
| 2204 | /// ... // No other stores or function calls (we could be calling a memory | |||
| 2205 | /// ... // function). | |||
| 2206 | /// %cmp = icmp ult %x, %y | |||
| 2207 | /// br i1 %cmp, label %EndBB, label %ThenBB | |||
| 2208 | /// ThenBB: | |||
| 2209 | /// store i32 %add5, i32* %arrayidx2 | |||
| 2210 | /// br label EndBB | |||
| 2211 | /// EndBB: | |||
| 2212 | /// ... | |||
| 2213 | /// We are going to transform this into: | |||
| 2214 | /// BrBB: | |||
| 2215 | /// store i32 %add, i32* %arrayidx2 | |||
| 2216 | /// ... // | |||
| 2217 | /// %cmp = icmp ult %x, %y | |||
| 2218 | /// %add.add5 = select i1 %cmp, i32 %add, %add5 | |||
| 2219 | /// store i32 %add.add5, i32* %arrayidx2 | |||
| 2220 | /// ... | |||
| 2221 | /// | |||
| 2222 | /// \return The pointer to the value of the previous store if the store can be | |||
| 2223 | /// hoisted into the predecessor block. 0 otherwise. | |||
| 2224 | static Value *isSafeToSpeculateStore(Instruction *I, BasicBlock *BrBB, | |||
| 2225 | BasicBlock *StoreBB, BasicBlock *EndBB) { | |||
| 2226 | StoreInst *StoreToHoist = dyn_cast<StoreInst>(I); | |||
| 2227 | if (!StoreToHoist) | |||
| 2228 | return nullptr; | |||
| 2229 | ||||
| 2230 | // Volatile or atomic. | |||
| 2231 | if (!StoreToHoist->isSimple()) | |||
| 2232 | return nullptr; | |||
| 2233 | ||||
| 2234 | Value *StorePtr = StoreToHoist->getPointerOperand(); | |||
| 2235 | Type *StoreTy = StoreToHoist->getValueOperand()->getType(); | |||
| 2236 | ||||
| 2237 | // Look for a store to the same pointer in BrBB. | |||
| 2238 | unsigned MaxNumInstToLookAt = 9; | |||
| 2239 | // Skip pseudo probe intrinsic calls which are not really killing any memory | |||
| 2240 | // accesses. | |||
| 2241 | for (Instruction &CurI : reverse(BrBB->instructionsWithoutDebug(true))) { | |||
| 2242 | if (!MaxNumInstToLookAt) | |||
| 2243 | break; | |||
| 2244 | --MaxNumInstToLookAt; | |||
| 2245 | ||||
| 2246 | // Could be calling an instruction that affects memory like free(). | |||
| 2247 | if (CurI.mayWriteToMemory() && !isa<StoreInst>(CurI)) | |||
| 2248 | return nullptr; | |||
| 2249 | ||||
| 2250 | if (auto *SI = dyn_cast<StoreInst>(&CurI)) { | |||
| 2251 | // Found the previous store to same location and type. Make sure it is | |||
| 2252 | // simple, to avoid introducing a spurious non-atomic write after an | |||
| 2253 | // atomic write. | |||
| 2254 | if (SI->getPointerOperand() == StorePtr && | |||
| 2255 | SI->getValueOperand()->getType() == StoreTy && SI->isSimple()) | |||
| 2256 | // Found the previous store, return its value operand. | |||
| 2257 | return SI->getValueOperand(); | |||
| 2258 | return nullptr; // Unknown store. | |||
| 2259 | } | |||
| 2260 | } | |||
| 2261 | ||||
| 2262 | return nullptr; | |||
| 2263 | } | |||
| 2264 | ||||
| 2265 | /// Estimate the cost of the insertion(s) and check that the PHI nodes can be | |||
| 2266 | /// converted to selects. | |||
| 2267 | static bool validateAndCostRequiredSelects(BasicBlock *BB, BasicBlock *ThenBB, | |||
| 2268 | BasicBlock *EndBB, | |||
| 2269 | unsigned &SpeculatedInstructions, | |||
| 2270 | InstructionCost &Cost, | |||
| 2271 | const TargetTransformInfo &TTI) { | |||
| 2272 | TargetTransformInfo::TargetCostKind CostKind = | |||
| 2273 | BB->getParent()->hasMinSize() | |||
| 2274 | ? TargetTransformInfo::TCK_CodeSize | |||
| 2275 | : TargetTransformInfo::TCK_SizeAndLatency; | |||
| 2276 | ||||
| 2277 | bool HaveRewritablePHIs = false; | |||
| 2278 | for (PHINode &PN : EndBB->phis()) { | |||
| 2279 | Value *OrigV = PN.getIncomingValueForBlock(BB); | |||
| 2280 | Value *ThenV = PN.getIncomingValueForBlock(ThenBB); | |||
| 2281 | ||||
| 2282 | // FIXME: Try to remove some of the duplication with HoistThenElseCodeToIf. | |||
| 2283 | // Skip PHIs which are trivial. | |||
| 2284 | if (ThenV == OrigV) | |||
| 2285 | continue; | |||
| 2286 | ||||
| 2287 | Cost += TTI.getCmpSelInstrCost(Instruction::Select, PN.getType(), nullptr, | |||
| 2288 | CmpInst::BAD_ICMP_PREDICATE, CostKind); | |||
| 2289 | ||||
| 2290 | // Don't convert to selects if we could remove undefined behavior instead. | |||
| 2291 | if (passingValueIsAlwaysUndefined(OrigV, &PN) || | |||
| 2292 | passingValueIsAlwaysUndefined(ThenV, &PN)) | |||
| 2293 | return false; | |||
| 2294 | ||||
| 2295 | HaveRewritablePHIs = true; | |||
| 2296 | ConstantExpr *OrigCE = dyn_cast<ConstantExpr>(OrigV); | |||
| 2297 | ConstantExpr *ThenCE = dyn_cast<ConstantExpr>(ThenV); | |||
| 2298 | if (!OrigCE && !ThenCE) | |||
| 2299 | continue; // Known safe and cheap. | |||
| 2300 | ||||
| 2301 | if ((ThenCE && !isSafeToSpeculativelyExecute(ThenCE)) || | |||
| 2302 | (OrigCE && !isSafeToSpeculativelyExecute(OrigCE))) | |||
| 2303 | return false; | |||
| 2304 | InstructionCost OrigCost = OrigCE ? computeSpeculationCost(OrigCE, TTI) : 0; | |||
| 2305 | InstructionCost ThenCost = ThenCE ? computeSpeculationCost(ThenCE, TTI) : 0; | |||
| 2306 | InstructionCost MaxCost = | |||
| 2307 | 2 * PHINodeFoldingThreshold * TargetTransformInfo::TCC_Basic; | |||
| 2308 | if (OrigCost + ThenCost > MaxCost) | |||
| 2309 | return false; | |||
| 2310 | ||||
| 2311 | // Account for the cost of an unfolded ConstantExpr which could end up | |||
| 2312 | // getting expanded into Instructions. | |||
| 2313 | // FIXME: This doesn't account for how many operations are combined in the | |||
| 2314 | // constant expression. | |||
| 2315 | ++SpeculatedInstructions; | |||
| 2316 | if (SpeculatedInstructions > 1) | |||
| 2317 | return false; | |||
| 2318 | } | |||
| 2319 | ||||
| 2320 | return HaveRewritablePHIs; | |||
| 2321 | } | |||
| 2322 | ||||
| 2323 | /// Speculate a conditional basic block flattening the CFG. | |||
| 2324 | /// | |||
| 2325 | /// Note that this is a very risky transform currently. Speculating | |||
| 2326 | /// instructions like this is most often not desirable. Instead, there is an MI | |||
| 2327 | /// pass which can do it with full awareness of the resource constraints. | |||
| 2328 | /// However, some cases are "obvious" and we should do directly. An example of | |||
| 2329 | /// this is speculating a single, reasonably cheap instruction. | |||
| 2330 | /// | |||
| 2331 | /// There is only one distinct advantage to flattening the CFG at the IR level: | |||
| 2332 | /// it makes very common but simplistic optimizations such as are common in | |||
| 2333 | /// instcombine and the DAG combiner more powerful by removing CFG edges and | |||
| 2334 | /// modeling their effects with easier to reason about SSA value graphs. | |||
| 2335 | /// | |||
| 2336 | /// | |||
| 2337 | /// An illustration of this transform is turning this IR: | |||
| 2338 | /// \code | |||
| 2339 | /// BB: | |||
| 2340 | /// %cmp = icmp ult %x, %y | |||
| 2341 | /// br i1 %cmp, label %EndBB, label %ThenBB | |||
| 2342 | /// ThenBB: | |||
| 2343 | /// %sub = sub %x, %y | |||
| 2344 | /// br label BB2 | |||
| 2345 | /// EndBB: | |||
| 2346 | /// %phi = phi [ %sub, %ThenBB ], [ 0, %EndBB ] | |||
| 2347 | /// ... | |||
| 2348 | /// \endcode | |||
| 2349 | /// | |||
| 2350 | /// Into this IR: | |||
| 2351 | /// \code | |||
| 2352 | /// BB: | |||
| 2353 | /// %cmp = icmp ult %x, %y | |||
| 2354 | /// %sub = sub %x, %y | |||
| 2355 | /// %cond = select i1 %cmp, 0, %sub | |||
| 2356 | /// ... | |||
| 2357 | /// \endcode | |||
| 2358 | /// | |||
| 2359 | /// \returns true if the conditional block is removed. | |||
| 2360 | bool SimplifyCFGOpt::SpeculativelyExecuteBB(BranchInst *BI, BasicBlock *ThenBB, | |||
| 2361 | const TargetTransformInfo &TTI) { | |||
| 2362 | // Be conservative for now. FP select instruction can often be expensive. | |||
| 2363 | Value *BrCond = BI->getCondition(); | |||
| 2364 | if (isa<FCmpInst>(BrCond)) | |||
| 2365 | return false; | |||
| 2366 | ||||
| 2367 | BasicBlock *BB = BI->getParent(); | |||
| 2368 | BasicBlock *EndBB = ThenBB->getTerminator()->getSuccessor(0); | |||
| 2369 | InstructionCost Budget = | |||
| 2370 | PHINodeFoldingThreshold * TargetTransformInfo::TCC_Basic; | |||
| 2371 | ||||
| 2372 | // If ThenBB is actually on the false edge of the conditional branch, remember | |||
| 2373 | // to swap the select operands later. | |||
| 2374 | bool Invert = false; | |||
| 2375 | if (ThenBB != BI->getSuccessor(0)) { | |||
| 2376 | assert(ThenBB == BI->getSuccessor(1) && "No edge from 'if' block?")((void)0); | |||
| 2377 | Invert = true; | |||
| 2378 | } | |||
| 2379 | assert(EndBB == BI->getSuccessor(!Invert) && "No edge from to end block")((void)0); | |||
| 2380 | ||||
| 2381 | // If the branch is non-unpredictable, and is predicted to *not* branch to | |||
| 2382 | // the `then` block, then avoid speculating it. | |||
| 2383 | if (!BI->getMetadata(LLVMContext::MD_unpredictable)) { | |||
| 2384 | uint64_t TWeight, FWeight; | |||
| 2385 | if (BI->extractProfMetadata(TWeight, FWeight) && (TWeight + FWeight) != 0) { | |||
| 2386 | uint64_t EndWeight = Invert ? TWeight : FWeight; | |||
| 2387 | BranchProbability BIEndProb = | |||
| 2388 | BranchProbability::getBranchProbability(EndWeight, TWeight + FWeight); | |||
| 2389 | BranchProbability Likely = TTI.getPredictableBranchThreshold(); | |||
| 2390 | if (BIEndProb >= Likely) | |||
| 2391 | return false; | |||
| 2392 | } | |||
| 2393 | } | |||
| 2394 | ||||
| 2395 | // Keep a count of how many times instructions are used within ThenBB when | |||
| 2396 | // they are candidates for sinking into ThenBB. Specifically: | |||
| 2397 | // - They are defined in BB, and | |||
| 2398 | // - They have no side effects, and | |||
| 2399 | // - All of their uses are in ThenBB. | |||
| 2400 | SmallDenseMap<Instruction *, unsigned, 4> SinkCandidateUseCounts; | |||
| 2401 | ||||
| 2402 | SmallVector<Instruction *, 4> SpeculatedDbgIntrinsics; | |||
| 2403 | ||||
| 2404 | unsigned SpeculatedInstructions = 0; | |||
| 2405 | Value *SpeculatedStoreValue = nullptr; | |||
| 2406 | StoreInst *SpeculatedStore = nullptr; | |||
| 2407 | for (BasicBlock::iterator BBI = ThenBB->begin(), | |||
| 2408 | BBE = std::prev(ThenBB->end()); | |||
| 2409 | BBI != BBE; ++BBI) { | |||
| 2410 | Instruction *I = &*BBI; | |||
| 2411 | // Skip debug info. | |||
| 2412 | if (isa<DbgInfoIntrinsic>(I)) { | |||
| 2413 | SpeculatedDbgIntrinsics.push_back(I); | |||
| 2414 | continue; | |||
| 2415 | } | |||
| 2416 | ||||
| 2417 | // Skip pseudo probes. The consequence is we lose track of the branch | |||
| 2418 | // probability for ThenBB, which is fine since the optimization here takes | |||
| 2419 | // place regardless of the branch probability. | |||
| 2420 | if (isa<PseudoProbeInst>(I)) { | |||
| 2421 | // The probe should be deleted so that it will not be over-counted when | |||
| 2422 | // the samples collected on the non-conditional path are counted towards | |||
| 2423 | // the conditional path. We leave it for the counts inference algorithm to | |||
| 2424 | // figure out a proper count for an unknown probe. | |||
| 2425 | SpeculatedDbgIntrinsics.push_back(I); | |||
| 2426 | continue; | |||
| 2427 | } | |||
| 2428 | ||||
| 2429 | // Only speculatively execute a single instruction (not counting the | |||
| 2430 | // terminator) for now. | |||
| 2431 | ++SpeculatedInstructions; | |||
| 2432 | if (SpeculatedInstructions > 1) | |||
| 2433 | return false; | |||
| 2434 | ||||
| 2435 | // Don't hoist the instruction if it's unsafe or expensive. | |||
| 2436 | if (!isSafeToSpeculativelyExecute(I) && | |||
| 2437 | !(HoistCondStores && (SpeculatedStoreValue = isSafeToSpeculateStore( | |||
| 2438 | I, BB, ThenBB, EndBB)))) | |||
| 2439 | return false; | |||
| 2440 | if (!SpeculatedStoreValue && | |||
| 2441 | computeSpeculationCost(I, TTI) > | |||
| 2442 | PHINodeFoldingThreshold * TargetTransformInfo::TCC_Basic) | |||
| 2443 | return false; | |||
| 2444 | ||||
| 2445 | // Store the store speculation candidate. | |||
| 2446 | if (SpeculatedStoreValue) | |||
| 2447 | SpeculatedStore = cast<StoreInst>(I); | |||
| 2448 | ||||
| 2449 | // Do not hoist the instruction if any of its operands are defined but not | |||
| 2450 | // used in BB. The transformation will prevent the operand from | |||
| 2451 | // being sunk into the use block. | |||
| 2452 | for (Use &Op : I->operands()) { | |||
| 2453 | Instruction *OpI = dyn_cast<Instruction>(Op); | |||
| 2454 | if (!OpI || OpI->getParent() != BB || OpI->mayHaveSideEffects()) | |||
| 2455 | continue; // Not a candidate for sinking. | |||
| 2456 | ||||
| 2457 | ++SinkCandidateUseCounts[OpI]; | |||
| 2458 | } | |||
| 2459 | } | |||
| 2460 | ||||
| 2461 | // Consider any sink candidates which are only used in ThenBB as costs for | |||
| 2462 | // speculation. Note, while we iterate over a DenseMap here, we are summing | |||
| 2463 | // and so iteration order isn't significant. | |||
| 2464 | for (SmallDenseMap<Instruction *, unsigned, 4>::iterator | |||
| 2465 | I = SinkCandidateUseCounts.begin(), | |||
| 2466 | E = SinkCandidateUseCounts.end(); | |||
| 2467 | I != E; ++I) | |||
| 2468 | if (I->first->hasNUses(I->second)) { | |||
| 2469 | ++SpeculatedInstructions; | |||
| 2470 | if (SpeculatedInstructions > 1) | |||
| 2471 | return false; | |||
| 2472 | } | |||
| 2473 | ||||
| 2474 | // Check that we can insert the selects and that it's not too expensive to do | |||
| 2475 | // so. | |||
| 2476 | bool Convert = SpeculatedStore != nullptr; | |||
| 2477 | InstructionCost Cost = 0; | |||
| 2478 | Convert |= validateAndCostRequiredSelects(BB, ThenBB, EndBB, | |||
| 2479 | SpeculatedInstructions, | |||
| 2480 | Cost, TTI); | |||
| 2481 | if (!Convert || Cost > Budget) | |||
| 2482 | return false; | |||
| 2483 | ||||
| 2484 | // If we get here, we can hoist the instruction and if-convert. | |||
| 2485 | LLVM_DEBUG(dbgs() << "SPECULATIVELY EXECUTING BB" << *ThenBB << "\n";)do { } while (false); | |||
| 2486 | ||||
| 2487 | // Insert a select of the value of the speculated store. | |||
| 2488 | if (SpeculatedStoreValue) { | |||
| 2489 | IRBuilder<NoFolder> Builder(BI); | |||
| 2490 | Value *TrueV = SpeculatedStore->getValueOperand(); | |||
| 2491 | Value *FalseV = SpeculatedStoreValue; | |||
| 2492 | if (Invert) | |||
| 2493 | std::swap(TrueV, FalseV); | |||
| 2494 | Value *S = Builder.CreateSelect( | |||
| 2495 | BrCond, TrueV, FalseV, "spec.store.select", BI); | |||
| 2496 | SpeculatedStore->setOperand(0, S); | |||
| 2497 | SpeculatedStore->applyMergedLocation(BI->getDebugLoc(), | |||
| 2498 | SpeculatedStore->getDebugLoc()); | |||
| 2499 | } | |||
| 2500 | ||||
| 2501 | // Metadata can be dependent on the condition we are hoisting above. | |||
| 2502 | // Conservatively strip all metadata on the instruction. Drop the debug loc | |||
| 2503 | // to avoid making it appear as if the condition is a constant, which would | |||
| 2504 | // be misleading while debugging. | |||
| 2505 | // Similarly strip attributes that maybe dependent on condition we are | |||
| 2506 | // hoisting above. | |||
| 2507 | for (auto &I : *ThenBB) { | |||
| 2508 | if (!SpeculatedStoreValue || &I != SpeculatedStore) | |||
| 2509 | I.setDebugLoc(DebugLoc()); | |||
| 2510 | I.dropUndefImplyingAttrsAndUnknownMetadata(); | |||
| 2511 | } | |||
| 2512 | ||||
| 2513 | // Hoist the instructions. | |||
| 2514 | BB->getInstList().splice(BI->getIterator(), ThenBB->getInstList(), | |||
| 2515 | ThenBB->begin(), std::prev(ThenBB->end())); | |||
| 2516 | ||||
| 2517 | // Insert selects and rewrite the PHI operands. | |||
| 2518 | IRBuilder<NoFolder> Builder(BI); | |||
| 2519 | for (PHINode &PN : EndBB->phis()) { | |||
| 2520 | unsigned OrigI = PN.getBasicBlockIndex(BB); | |||
| 2521 | unsigned ThenI = PN.getBasicBlockIndex(ThenBB); | |||
| 2522 | Value *OrigV = PN.getIncomingValue(OrigI); | |||
| 2523 | Value *ThenV = PN.getIncomingValue(ThenI); | |||
| 2524 | ||||
| 2525 | // Skip PHIs which are trivial. | |||
| 2526 | if (OrigV == ThenV) | |||
| 2527 | continue; | |||
| 2528 | ||||
| 2529 | // Create a select whose true value is the speculatively executed value and | |||
| 2530 | // false value is the pre-existing value. Swap them if the branch | |||
| 2531 | // destinations were inverted. | |||
| 2532 | Value *TrueV = ThenV, *FalseV = OrigV; | |||
| 2533 | if (Invert) | |||
| 2534 | std::swap(TrueV, FalseV); | |||
| 2535 | Value *V = Builder.CreateSelect(BrCond, TrueV, FalseV, "spec.select", BI); | |||
| 2536 | PN.setIncomingValue(OrigI, V); | |||
| 2537 | PN.setIncomingValue(ThenI, V); | |||
| 2538 | } | |||
| 2539 | ||||
| 2540 | // Remove speculated dbg intrinsics. | |||
| 2541 | // FIXME: Is it possible to do this in a more elegant way? Moving/merging the | |||
| 2542 | // dbg value for the different flows and inserting it after the select. | |||
| 2543 | for (Instruction *I : SpeculatedDbgIntrinsics) | |||
| 2544 | I->eraseFromParent(); | |||
| 2545 | ||||
| 2546 | ++NumSpeculations; | |||
| 2547 | return true; | |||
| 2548 | } | |||
| 2549 | ||||
| 2550 | /// Return true if we can thread a branch across this block. | |||
| 2551 | static bool BlockIsSimpleEnoughToThreadThrough(BasicBlock *BB) { | |||
| 2552 | int Size = 0; | |||
| 2553 | ||||
| 2554 | SmallPtrSet<const Value *, 32> EphValues; | |||
| 2555 | auto IsEphemeral = [&](const Value *V) { | |||
| 2556 | if (isa<AssumeInst>(V)) | |||
| 2557 | return true; | |||
| 2558 | return isSafeToSpeculativelyExecute(V) && | |||
| 2559 | all_of(V->users(), | |||
| 2560 | [&](const User *U) { return EphValues.count(U); }); | |||
| 2561 | }; | |||
| 2562 | ||||
| 2563 | // Walk the loop in reverse so that we can identify ephemeral values properly | |||
| 2564 | // (values only feeding assumes). | |||
| 2565 | for (Instruction &I : reverse(BB->instructionsWithoutDebug())) { | |||
| 2566 | // Can't fold blocks that contain noduplicate or convergent calls. | |||
| 2567 | if (CallInst *CI = dyn_cast<CallInst>(&I)) | |||
| 2568 | if (CI->cannotDuplicate() || CI->isConvergent()) | |||
| 2569 | return false; | |||
| 2570 | ||||
| 2571 | // Ignore ephemeral values which are deleted during codegen. | |||
| 2572 | if (IsEphemeral(&I)) | |||
| 2573 | EphValues.insert(&I); | |||
| 2574 | // We will delete Phis while threading, so Phis should not be accounted in | |||
| 2575 | // block's size. | |||
| 2576 | else if (!isa<PHINode>(I)) { | |||
| 2577 | if (Size++ > MaxSmallBlockSize) | |||
| 2578 | return false; // Don't clone large BB's. | |||
| 2579 | } | |||
| 2580 | ||||
| 2581 | // We can only support instructions that do not define values that are | |||
| 2582 | // live outside of the current basic block. | |||
| 2583 | for (User *U : I.users()) { | |||
| 2584 | Instruction *UI = cast<Instruction>(U); | |||
| 2585 | if (UI->getParent() != BB || isa<PHINode>(UI)) | |||
| 2586 | return false; | |||
| 2587 | } | |||
| 2588 | ||||
| 2589 | // Looks ok, continue checking. | |||
| 2590 | } | |||
| 2591 | ||||
| 2592 | return true; | |||
| 2593 | } | |||
| 2594 | ||||
| 2595 | /// If we have a conditional branch on a PHI node value that is defined in the | |||
| 2596 | /// same block as the branch and if any PHI entries are constants, thread edges | |||
| 2597 | /// corresponding to that entry to be branches to their ultimate destination. | |||
| 2598 | static bool FoldCondBranchOnPHI(BranchInst *BI, DomTreeUpdater *DTU, | |||
| 2599 | const DataLayout &DL, AssumptionCache *AC) { | |||
| 2600 | BasicBlock *BB = BI->getParent(); | |||
| 2601 | PHINode *PN = dyn_cast<PHINode>(BI->getCondition()); | |||
| 2602 | // NOTE: we currently cannot transform this case if the PHI node is used | |||
| 2603 | // outside of the block. | |||
| 2604 | if (!PN || PN->getParent() != BB || !PN->hasOneUse()) | |||
| 2605 | return false; | |||
| 2606 | ||||
| 2607 | // Degenerate case of a single entry PHI. | |||
| 2608 | if (PN->getNumIncomingValues() == 1) { | |||
| 2609 | FoldSingleEntryPHINodes(PN->getParent()); | |||
| 2610 | return true; | |||
| 2611 | } | |||
| 2612 | ||||
| 2613 | // Now we know that this block has multiple preds and two succs. | |||
| 2614 | if (!BlockIsSimpleEnoughToThreadThrough(BB)) | |||
| 2615 | return false; | |||
| 2616 | ||||
| 2617 | // Okay, this is a simple enough basic block. See if any phi values are | |||
| 2618 | // constants. | |||
| 2619 | for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) { | |||
| 2620 | ConstantInt *CB = dyn_cast<ConstantInt>(PN->getIncomingValue(i)); | |||
| 2621 | if (!CB || !CB->getType()->isIntegerTy(1)) | |||
| 2622 | continue; | |||
| 2623 | ||||
| 2624 | // Okay, we now know that all edges from PredBB should be revectored to | |||
| 2625 | // branch to RealDest. | |||
| 2626 | BasicBlock *PredBB = PN->getIncomingBlock(i); | |||
| 2627 | BasicBlock *RealDest = BI->getSuccessor(!CB->getZExtValue()); | |||
| 2628 | ||||
| 2629 | if (RealDest == BB) | |||
| 2630 | continue; // Skip self loops. | |||
| 2631 | // Skip if the predecessor's terminator is an indirect branch. | |||
| 2632 | if (isa<IndirectBrInst>(PredBB->getTerminator())) | |||
| 2633 | continue; | |||
| 2634 | ||||
| 2635 | SmallVector<DominatorTree::UpdateType, 3> Updates; | |||
| 2636 | ||||
| 2637 | // The dest block might have PHI nodes, other predecessors and other | |||
| 2638 | // difficult cases. Instead of being smart about this, just insert a new | |||
| 2639 | // block that jumps to the destination block, effectively splitting | |||
| 2640 | // the edge we are about to create. | |||
| 2641 | BasicBlock *EdgeBB = | |||
| 2642 | BasicBlock::Create(BB->getContext(), RealDest->getName() + ".critedge", | |||
| 2643 | RealDest->getParent(), RealDest); | |||
| 2644 | BranchInst *CritEdgeBranch = BranchInst::Create(RealDest, EdgeBB); | |||
| 2645 | if (DTU) | |||
| 2646 | Updates.push_back({DominatorTree::Insert, EdgeBB, RealDest}); | |||
| 2647 | CritEdgeBranch->setDebugLoc(BI->getDebugLoc()); | |||
| 2648 | ||||
| 2649 | // Update PHI nodes. | |||
| 2650 | AddPredecessorToBlock(RealDest, EdgeBB, BB); | |||
| 2651 | ||||
| 2652 | // BB may have instructions that are being threaded over. Clone these | |||
| 2653 | // instructions into EdgeBB. We know that there will be no uses of the | |||
| 2654 | // cloned instructions outside of EdgeBB. | |||
| 2655 | BasicBlock::iterator InsertPt = EdgeBB->begin(); | |||
| 2656 | DenseMap<Value *, Value *> TranslateMap; // Track translated values. | |||
| 2657 | for (BasicBlock::iterator BBI = BB->begin(); &*BBI != BI; ++BBI) { | |||
| 2658 | if (PHINode *PN = dyn_cast<PHINode>(BBI)) { | |||
| 2659 | TranslateMap[PN] = PN->getIncomingValueForBlock(PredBB); | |||
| 2660 | continue; | |||
| 2661 | } | |||
| 2662 | // Clone the instruction. | |||
| 2663 | Instruction *N = BBI->clone(); | |||
| 2664 | if (BBI->hasName()) | |||
| 2665 | N->setName(BBI->getName() + ".c"); | |||
| 2666 | ||||
| 2667 | // Update operands due to translation. | |||
| 2668 | for (Use &Op : N->operands()) { | |||
| 2669 | DenseMap<Value *, Value *>::iterator PI = TranslateMap.find(Op); | |||
| 2670 | if (PI != TranslateMap.end()) | |||
| 2671 | Op = PI->second; | |||
| 2672 | } | |||
| 2673 | ||||
| 2674 | // Check for trivial simplification. | |||
| 2675 | if (Value *V = SimplifyInstruction(N, {DL, nullptr, nullptr, AC})) { | |||
| 2676 | if (!BBI->use_empty()) | |||
| 2677 | TranslateMap[&*BBI] = V; | |||
| 2678 | if (!N->mayHaveSideEffects()) { | |||
| 2679 | N->deleteValue(); // Instruction folded away, don't need actual inst | |||
| 2680 | N = nullptr; | |||
| 2681 | } | |||
| 2682 | } else { | |||
| 2683 | if (!BBI->use_empty()) | |||
| 2684 | TranslateMap[&*BBI] = N; | |||
| 2685 | } | |||
| 2686 | if (N) { | |||
| 2687 | // Insert the new instruction into its new home. | |||
| 2688 | EdgeBB->getInstList().insert(InsertPt, N); | |||
| 2689 | ||||
| 2690 | // Register the new instruction with the assumption cache if necessary. | |||
| 2691 | if (auto *Assume = dyn_cast<AssumeInst>(N)) | |||
| 2692 | if (AC) | |||
| 2693 | AC->registerAssumption(Assume); | |||
| 2694 | } | |||
| 2695 | } | |||
| 2696 | ||||
| 2697 | // Loop over all of the edges from PredBB to BB, changing them to branch | |||
| 2698 | // to EdgeBB instead. | |||
| 2699 | Instruction *PredBBTI = PredBB->getTerminator(); | |||
| 2700 | for (unsigned i = 0, e = PredBBTI->getNumSuccessors(); i != e; ++i) | |||
| 2701 | if (PredBBTI->getSuccessor(i) == BB) { | |||
| 2702 | BB->removePredecessor(PredBB); | |||
| 2703 | PredBBTI->setSuccessor(i, EdgeBB); | |||
| 2704 | } | |||
| 2705 | ||||
| 2706 | if (DTU) { | |||
| 2707 | Updates.push_back({DominatorTree::Insert, PredBB, EdgeBB}); | |||
| 2708 | Updates.push_back({DominatorTree::Delete, PredBB, BB}); | |||
| 2709 | ||||
| 2710 | DTU->applyUpdates(Updates); | |||
| 2711 | } | |||
| 2712 | ||||
| 2713 | // Recurse, simplifying any other constants. | |||
| 2714 | return FoldCondBranchOnPHI(BI, DTU, DL, AC) || true; | |||
| 2715 | } | |||
| 2716 | ||||
| 2717 | return false; | |||
| 2718 | } | |||
| 2719 | ||||
| 2720 | /// Given a BB that starts with the specified two-entry PHI node, | |||
| 2721 | /// see if we can eliminate it. | |||
| 2722 | static bool FoldTwoEntryPHINode(PHINode *PN, const TargetTransformInfo &TTI, | |||
| 2723 | DomTreeUpdater *DTU, const DataLayout &DL) { | |||
| 2724 | // Ok, this is a two entry PHI node. Check to see if this is a simple "if | |||
| 2725 | // statement", which has a very simple dominance structure. Basically, we | |||
| 2726 | // are trying to find the condition that is being branched on, which | |||
| 2727 | // subsequently causes this merge to happen. We really want control | |||
| 2728 | // dependence information for this check, but simplifycfg can't keep it up | |||
| 2729 | // to date, and this catches most of the cases we care about anyway. | |||
| 2730 | BasicBlock *BB = PN->getParent(); | |||
| 2731 | ||||
| 2732 | BasicBlock *IfTrue, *IfFalse; | |||
| 2733 | BranchInst *DomBI = GetIfCondition(BB, IfTrue, IfFalse); | |||
| 2734 | if (!DomBI) | |||
| 2735 | return false; | |||
| 2736 | Value *IfCond = DomBI->getCondition(); | |||
| 2737 | // Don't bother if the branch will be constant folded trivially. | |||
| 2738 | if (isa<ConstantInt>(IfCond)) | |||
| 2739 | return false; | |||
| 2740 | ||||
| 2741 | BasicBlock *DomBlock = DomBI->getParent(); | |||
| 2742 | SmallVector<BasicBlock *, 2> IfBlocks; | |||
| 2743 | llvm::copy_if( | |||
| 2744 | PN->blocks(), std::back_inserter(IfBlocks), [](BasicBlock *IfBlock) { | |||
| 2745 | return cast<BranchInst>(IfBlock->getTerminator())->isUnconditional(); | |||
| 2746 | }); | |||
| 2747 | assert((IfBlocks.size() == 1 || IfBlocks.size() == 2) &&((void)0) | |||
| 2748 | "Will have either one or two blocks to speculate.")((void)0); | |||
| 2749 | ||||
| 2750 | // If the branch is non-unpredictable, see if we either predictably jump to | |||
| 2751 | // the merge bb (if we have only a single 'then' block), or if we predictably | |||
| 2752 | // jump to one specific 'then' block (if we have two of them). | |||
| 2753 | // It isn't beneficial to speculatively execute the code | |||
| 2754 | // from the block that we know is predictably not entered. | |||
| 2755 | if (!DomBI->getMetadata(LLVMContext::MD_unpredictable)) { | |||
| 2756 | uint64_t TWeight, FWeight; | |||
| 2757 | if (DomBI->extractProfMetadata(TWeight, FWeight) && | |||
| 2758 | (TWeight + FWeight) != 0) { | |||
| 2759 | BranchProbability BITrueProb = | |||
| 2760 | BranchProbability::getBranchProbability(TWeight, TWeight + FWeight); | |||
| 2761 | BranchProbability Likely = TTI.getPredictableBranchThreshold(); | |||
| 2762 | BranchProbability BIFalseProb = BITrueProb.getCompl(); | |||
| 2763 | if (IfBlocks.size() == 1) { | |||
| 2764 | BranchProbability BIBBProb = | |||
| 2765 | DomBI->getSuccessor(0) == BB ? BITrueProb : BIFalseProb; | |||
| 2766 | if (BIBBProb >= Likely) | |||
| 2767 | return false; | |||
| 2768 | } else { | |||
| 2769 | if (BITrueProb >= Likely || BIFalseProb >= Likely) | |||
| 2770 | return false; | |||
| 2771 | } | |||
| 2772 | } | |||
| 2773 | } | |||
| 2774 | ||||
| 2775 | // Don't try to fold an unreachable block. For example, the phi node itself | |||
| 2776 | // can't be the candidate if-condition for a select that we want to form. | |||
| 2777 | if (auto *IfCondPhiInst = dyn_cast<PHINode>(IfCond)) | |||
| 2778 | if (IfCondPhiInst->getParent() == BB) | |||
| 2779 | return false; | |||
| 2780 | ||||
| 2781 | // Okay, we found that we can merge this two-entry phi node into a select. | |||
| 2782 | // Doing so would require us to fold *all* two entry phi nodes in this block. | |||
| 2783 | // At some point this becomes non-profitable (particularly if the target | |||
| 2784 | // doesn't support cmov's). Only do this transformation if there are two or | |||
| 2785 | // fewer PHI nodes in this block. | |||
| 2786 | unsigned NumPhis = 0; | |||
| 2787 | for (BasicBlock::iterator I = BB->begin(); isa<PHINode>(I); ++NumPhis, ++I) | |||
| 2788 | if (NumPhis > 2) | |||
| 2789 | return false; | |||
| 2790 | ||||
| 2791 | // Loop over the PHI's seeing if we can promote them all to select | |||
| 2792 | // instructions. While we are at it, keep track of the instructions | |||
| 2793 | // that need to be moved to the dominating block. | |||
| 2794 | SmallPtrSet<Instruction *, 4> AggressiveInsts; | |||
| 2795 | InstructionCost Cost = 0; | |||
| 2796 | InstructionCost Budget = | |||
| 2797 | TwoEntryPHINodeFoldingThreshold * TargetTransformInfo::TCC_Basic; | |||
| 2798 | ||||
| 2799 | bool Changed = false; | |||
| 2800 | for (BasicBlock::iterator II = BB->begin(); isa<PHINode>(II);) { | |||
| 2801 | PHINode *PN = cast<PHINode>(II++); | |||
| 2802 | if (Value *V = SimplifyInstruction(PN, {DL, PN})) { | |||
| 2803 | PN->replaceAllUsesWith(V); | |||
| 2804 | PN->eraseFromParent(); | |||
| 2805 | Changed = true; | |||
| 2806 | continue; | |||
| 2807 | } | |||
| 2808 | ||||
| 2809 | if (!dominatesMergePoint(PN->getIncomingValue(0), BB, AggressiveInsts, | |||
| 2810 | Cost, Budget, TTI) || | |||
| 2811 | !dominatesMergePoint(PN->getIncomingValue(1), BB, AggressiveInsts, | |||
| 2812 | Cost, Budget, TTI)) | |||
| 2813 | return Changed; | |||
| 2814 | } | |||
| 2815 | ||||
| 2816 | // If we folded the first phi, PN dangles at this point. Refresh it. If | |||
| 2817 | // we ran out of PHIs then we simplified them all. | |||
| 2818 | PN = dyn_cast<PHINode>(BB->begin()); | |||
| 2819 | if (!PN) | |||
| 2820 | return true; | |||
| 2821 | ||||
| 2822 | // Return true if at least one of these is a 'not', and another is either | |||
| 2823 | // a 'not' too, or a constant. | |||
| 2824 | auto CanHoistNotFromBothValues = [](Value *V0, Value *V1) { | |||
| 2825 | if (!match(V0, m_Not(m_Value()))) | |||
| 2826 | std::swap(V0, V1); | |||
| 2827 | auto Invertible = m_CombineOr(m_Not(m_Value()), m_AnyIntegralConstant()); | |||
| 2828 | return match(V0, m_Not(m_Value())) && match(V1, Invertible); | |||
| 2829 | }; | |||
| 2830 | ||||
| 2831 | // Don't fold i1 branches on PHIs which contain binary operators or | |||
| 2832 | // (possibly inverted) select form of or/ands, unless one of | |||
| 2833 | // the incoming values is an 'not' and another one is freely invertible. | |||
| 2834 | // These can often be turned into switches and other things. | |||
| 2835 | auto IsBinOpOrAnd = [](Value *V) { | |||
| 2836 | return match( | |||
| 2837 | V, m_CombineOr( | |||
| 2838 | m_BinOp(), | |||
| 2839 | m_CombineOr(m_Select(m_Value(), m_ImmConstant(), m_Value()), | |||
| 2840 | m_Select(m_Value(), m_Value(), m_ImmConstant())))); | |||
| 2841 | }; | |||
| 2842 | if (PN->getType()->isIntegerTy(1) && | |||
| 2843 | (IsBinOpOrAnd(PN->getIncomingValue(0)) || | |||
| 2844 | IsBinOpOrAnd(PN->getIncomingValue(1)) || IsBinOpOrAnd(IfCond)) && | |||
| 2845 | !CanHoistNotFromBothValues(PN->getIncomingValue(0), | |||
| 2846 | PN->getIncomingValue(1))) | |||
| 2847 | return Changed; | |||
| 2848 | ||||
| 2849 | // If all PHI nodes are promotable, check to make sure that all instructions | |||
| 2850 | // in the predecessor blocks can be promoted as well. If not, we won't be able | |||
| 2851 | // to get rid of the control flow, so it's not worth promoting to select | |||
| 2852 | // instructions. | |||
| 2853 | for (BasicBlock *IfBlock : IfBlocks) | |||
| 2854 | for (BasicBlock::iterator I = IfBlock->begin(); !I->isTerminator(); ++I) | |||
| 2855 | if (!AggressiveInsts.count(&*I) && !isa<DbgInfoIntrinsic>(I) && | |||
| 2856 | !isa<PseudoProbeInst>(I)) { | |||
| 2857 | // This is not an aggressive instruction that we can promote. | |||
| 2858 | // Because of this, we won't be able to get rid of the control flow, so | |||
| 2859 | // the xform is not worth it. | |||
| 2860 | return Changed; | |||
| 2861 | } | |||
| 2862 | ||||
| 2863 | // If either of the blocks has it's address taken, we can't do this fold. | |||
| 2864 | if (any_of(IfBlocks, | |||
| 2865 | [](BasicBlock *IfBlock) { return IfBlock->hasAddressTaken(); })) | |||
| 2866 | return Changed; | |||
| 2867 | ||||
| 2868 | LLVM_DEBUG(dbgs() << "FOUND IF CONDITION! " << *IfConddo { } while (false) | |||
| 2869 | << " T: " << IfTrue->getName()do { } while (false) | |||
| 2870 | << " F: " << IfFalse->getName() << "\n")do { } while (false); | |||
| 2871 | ||||
| 2872 | // If we can still promote the PHI nodes after this gauntlet of tests, | |||
| 2873 | // do all of the PHI's now. | |||
| 2874 | ||||
| 2875 | // Move all 'aggressive' instructions, which are defined in the | |||
| 2876 | // conditional parts of the if's up to the dominating block. | |||
| 2877 | for (BasicBlock *IfBlock : IfBlocks) | |||
| 2878 | hoistAllInstructionsInto(DomBlock, DomBI, IfBlock); | |||
| 2879 | ||||
| 2880 | IRBuilder<NoFolder> Builder(DomBI); | |||
| 2881 | // Propagate fast-math-flags from phi nodes to replacement selects. | |||
| 2882 | IRBuilder<>::FastMathFlagGuard FMFGuard(Builder); | |||
| 2883 | while (PHINode *PN = dyn_cast<PHINode>(BB->begin())) { | |||
| 2884 | if (isa<FPMathOperator>(PN)) | |||
| 2885 | Builder.setFastMathFlags(PN->getFastMathFlags()); | |||
| 2886 | ||||
| 2887 | // Change the PHI node into a select instruction. | |||
| 2888 | Value *TrueVal = PN->getIncomingValueForBlock(IfTrue); | |||
| 2889 | Value *FalseVal = PN->getIncomingValueForBlock(IfFalse); | |||
| 2890 | ||||
| 2891 | Value *Sel = Builder.CreateSelect(IfCond, TrueVal, FalseVal, "", DomBI); | |||
| 2892 | PN->replaceAllUsesWith(Sel); | |||
| 2893 | Sel->takeName(PN); | |||
| 2894 | PN->eraseFromParent(); | |||
| 2895 | } | |||
| 2896 | ||||
| 2897 | // At this point, all IfBlocks are empty, so our if statement | |||
| 2898 | // has been flattened. Change DomBlock to jump directly to our new block to | |||
| 2899 | // avoid other simplifycfg's kicking in on the diamond. | |||
| 2900 | Builder.CreateBr(BB); | |||
| 2901 | ||||
| 2902 | SmallVector<DominatorTree::UpdateType, 3> Updates; | |||
| 2903 | if (DTU) { | |||
| 2904 | Updates.push_back({DominatorTree::Insert, DomBlock, BB}); | |||
| 2905 | for (auto *Successor : successors(DomBlock)) | |||
| 2906 | Updates.push_back({DominatorTree::Delete, DomBlock, Successor}); | |||
| 2907 | } | |||
| 2908 | ||||
| 2909 | DomBI->eraseFromParent(); | |||
| 2910 | if (DTU) | |||
| 2911 | DTU->applyUpdates(Updates); | |||
| 2912 | ||||
| 2913 | return true; | |||
| 2914 | } | |||
| 2915 | ||||
| 2916 | static Value *createLogicalOp(IRBuilderBase &Builder, | |||
| 2917 | Instruction::BinaryOps Opc, Value *LHS, | |||
| 2918 | Value *RHS, const Twine &Name = "") { | |||
| 2919 | // Try to relax logical op to binary op. | |||
| 2920 | if (impliesPoison(RHS, LHS)) | |||
| 2921 | return Builder.CreateBinOp(Opc, LHS, RHS, Name); | |||
| 2922 | if (Opc == Instruction::And) | |||
| 2923 | return Builder.CreateLogicalAnd(LHS, RHS, Name); | |||
| 2924 | if (Opc == Instruction::Or) | |||
| 2925 | return Builder.CreateLogicalOr(LHS, RHS, Name); | |||
| 2926 | llvm_unreachable("Invalid logical opcode")__builtin_unreachable(); | |||
| 2927 | } | |||
| 2928 | ||||
| 2929 | /// Return true if either PBI or BI has branch weight available, and store | |||
| 2930 | /// the weights in {Pred|Succ}{True|False}Weight. If one of PBI and BI does | |||
| 2931 | /// not have branch weight, use 1:1 as its weight. | |||
| 2932 | static bool extractPredSuccWeights(BranchInst *PBI, BranchInst *BI, | |||
| 2933 | uint64_t &PredTrueWeight, | |||
| 2934 | uint64_t &PredFalseWeight, | |||
| 2935 | uint64_t &SuccTrueWeight, | |||
| 2936 | uint64_t &SuccFalseWeight) { | |||
| 2937 | bool PredHasWeights = | |||
| 2938 | PBI->extractProfMetadata(PredTrueWeight, PredFalseWeight); | |||
| 2939 | bool SuccHasWeights = | |||
| 2940 | BI->extractProfMetadata(SuccTrueWeight, SuccFalseWeight); | |||
| 2941 | if (PredHasWeights || SuccHasWeights) { | |||
| 2942 | if (!PredHasWeights) | |||
| 2943 | PredTrueWeight = PredFalseWeight = 1; | |||
| 2944 | if (!SuccHasWeights) | |||
| 2945 | SuccTrueWeight = SuccFalseWeight = 1; | |||
| 2946 | return true; | |||
| 2947 | } else { | |||
| 2948 | return false; | |||
| 2949 | } | |||
| 2950 | } | |||
| 2951 | ||||
| 2952 | /// Determine if the two branches share a common destination and deduce a glue | |||
| 2953 | /// that joins the branches' conditions to arrive at the common destination if | |||
| 2954 | /// that would be profitable. | |||
| 2955 | static Optional<std::pair<Instruction::BinaryOps, bool>> | |||
| 2956 | shouldFoldCondBranchesToCommonDestination(BranchInst *BI, BranchInst *PBI, | |||
| 2957 | const TargetTransformInfo *TTI) { | |||
| 2958 | assert(BI && PBI && BI->isConditional() && PBI->isConditional() &&((void)0) | |||
| 2959 | "Both blocks must end with a conditional branches.")((void)0); | |||
| 2960 | assert(is_contained(predecessors(BI->getParent()), PBI->getParent()) &&((void)0) | |||
| 2961 | "PredBB must be a predecessor of BB.")((void)0); | |||
| 2962 | ||||
| 2963 | // We have the potential to fold the conditions together, but if the | |||
| 2964 | // predecessor branch is predictable, we may not want to merge them. | |||
| 2965 | uint64_t PTWeight, PFWeight; | |||
| 2966 | BranchProbability PBITrueProb, Likely; | |||
| 2967 | if (TTI && !PBI->getMetadata(LLVMContext::MD_unpredictable) && | |||
| 2968 | PBI->extractProfMetadata(PTWeight, PFWeight) && | |||
| 2969 | (PTWeight + PFWeight) != 0) { | |||
| 2970 | PBITrueProb = | |||
| 2971 | BranchProbability::getBranchProbability(PTWeight, PTWeight + PFWeight); | |||
| 2972 | Likely = TTI->getPredictableBranchThreshold(); | |||
| 2973 | } | |||
| 2974 | ||||
| 2975 | if (PBI->getSuccessor(0) == BI->getSuccessor(0)) { | |||
| 2976 | // Speculate the 2nd condition unless the 1st is probably true. | |||
| 2977 | if (PBITrueProb.isUnknown() || PBITrueProb < Likely) | |||
| 2978 | return {{Instruction::Or, false}}; | |||
| 2979 | } else if (PBI->getSuccessor(1) == BI->getSuccessor(1)) { | |||
| 2980 | // Speculate the 2nd condition unless the 1st is probably false. | |||
| 2981 | if (PBITrueProb.isUnknown() || PBITrueProb.getCompl() < Likely) | |||
| 2982 | return {{Instruction::And, false}}; | |||
| 2983 | } else if (PBI->getSuccessor(0) == BI->getSuccessor(1)) { | |||
| 2984 | // Speculate the 2nd condition unless the 1st is probably true. | |||
| 2985 | if (PBITrueProb.isUnknown() || PBITrueProb < Likely) | |||
| 2986 | return {{Instruction::And, true}}; | |||
| 2987 | } else if (PBI->getSuccessor(1) == BI->getSuccessor(0)) { | |||
| 2988 | // Speculate the 2nd condition unless the 1st is probably false. | |||
| 2989 | if (PBITrueProb.isUnknown() || PBITrueProb.getCompl() < Likely) | |||
| 2990 | return {{Instruction::Or, true}}; | |||
| 2991 | } | |||
| 2992 | return None; | |||
| 2993 | } | |||
| 2994 | ||||
| 2995 | static bool performBranchToCommonDestFolding(BranchInst *BI, BranchInst *PBI, | |||
| 2996 | DomTreeUpdater *DTU, | |||
| 2997 | MemorySSAUpdater *MSSAU, | |||
| 2998 | const TargetTransformInfo *TTI) { | |||
| 2999 | BasicBlock *BB = BI->getParent(); | |||
| 3000 | BasicBlock *PredBlock = PBI->getParent(); | |||
| 3001 | ||||
| 3002 | // Determine if the two branches share a common destination. | |||
| 3003 | Instruction::BinaryOps Opc; | |||
| 3004 | bool InvertPredCond; | |||
| 3005 | std::tie(Opc, InvertPredCond) = | |||
| 3006 | *shouldFoldCondBranchesToCommonDestination(BI, PBI, TTI); | |||
| 3007 | ||||
| 3008 | LLVM_DEBUG(dbgs() << "FOLDING BRANCH TO COMMON DEST:\n" << *PBI << *BB)do { } while (false); | |||
| 3009 | ||||
| 3010 | IRBuilder<> Builder(PBI); | |||
| 3011 | // The builder is used to create instructions to eliminate the branch in BB. | |||
| 3012 | // If BB's terminator has !annotation metadata, add it to the new | |||
| 3013 | // instructions. | |||
| 3014 | Builder.CollectMetadataToCopy(BB->getTerminator(), | |||
| 3015 | {LLVMContext::MD_annotation}); | |||
| 3016 | ||||
| 3017 | // If we need to invert the condition in the pred block to match, do so now. | |||
| 3018 | if (InvertPredCond) { | |||
| 3019 | Value *NewCond = PBI->getCondition(); | |||
| 3020 | if (NewCond->hasOneUse() && isa<CmpInst>(NewCond)) { | |||
| 3021 | CmpInst *CI = cast<CmpInst>(NewCond); | |||
| 3022 | CI->setPredicate(CI->getInversePredicate()); | |||
| 3023 | } else { | |||
| 3024 | NewCond = | |||
| 3025 | Builder.CreateNot(NewCond, PBI->getCondition()->getName() + ".not"); | |||
| 3026 | } | |||
| 3027 | ||||
| 3028 | PBI->setCondition(NewCond); | |||
| 3029 | PBI->swapSuccessors(); | |||
| 3030 | } | |||
| 3031 | ||||
| 3032 | BasicBlock *UniqueSucc = | |||
| 3033 | PBI->getSuccessor(0) == BB ? BI->getSuccessor(0) : BI->getSuccessor(1); | |||
| 3034 | ||||
| 3035 | // Before cloning instructions, notify the successor basic block that it | |||
| 3036 | // is about to have a new predecessor. This will update PHI nodes, | |||
| 3037 | // which will allow us to update live-out uses of bonus instructions. | |||
| 3038 | AddPredecessorToBlock(UniqueSucc, PredBlock, BB, MSSAU); | |||
| 3039 | ||||
| 3040 | // Try to update branch weights. | |||
| 3041 | uint64_t PredTrueWeight, PredFalseWeight, SuccTrueWeight, SuccFalseWeight; | |||
| 3042 | if (extractPredSuccWeights(PBI, BI, PredTrueWeight, PredFalseWeight, | |||
| 3043 | SuccTrueWeight, SuccFalseWeight)) { | |||
| 3044 | SmallVector<uint64_t, 8> NewWeights; | |||
| 3045 | ||||
| 3046 | if (PBI->getSuccessor(0) == BB) { | |||
| 3047 | // PBI: br i1 %x, BB, FalseDest | |||
| 3048 | // BI: br i1 %y, UniqueSucc, FalseDest | |||
| 3049 | // TrueWeight is TrueWeight for PBI * TrueWeight for BI. | |||
| 3050 | NewWeights.push_back(PredTrueWeight * SuccTrueWeight); | |||
| 3051 | // FalseWeight is FalseWeight for PBI * TotalWeight for BI + | |||
| 3052 | // TrueWeight for PBI * FalseWeight for BI. | |||
| 3053 | // We assume that total weights of a BranchInst can fit into 32 bits. | |||
| 3054 | // Therefore, we will not have overflow using 64-bit arithmetic. | |||
| 3055 | NewWeights.push_back(PredFalseWeight * | |||
| 3056 | (SuccFalseWeight + SuccTrueWeight) + | |||
| 3057 | PredTrueWeight * SuccFalseWeight); | |||
| 3058 | } else { | |||
| 3059 | // PBI: br i1 %x, TrueDest, BB | |||
| 3060 | // BI: br i1 %y, TrueDest, UniqueSucc | |||
| 3061 | // TrueWeight is TrueWeight for PBI * TotalWeight for BI + | |||
| 3062 | // FalseWeight for PBI * TrueWeight for BI. | |||
| 3063 | NewWeights.push_back(PredTrueWeight * (SuccFalseWeight + SuccTrueWeight) + | |||
| 3064 | PredFalseWeight * SuccTrueWeight); | |||
| 3065 | // FalseWeight is FalseWeight for PBI * FalseWeight for BI. | |||
| 3066 | NewWeights.push_back(PredFalseWeight * SuccFalseWeight); | |||
| 3067 | } | |||
| 3068 | ||||
| 3069 | // Halve the weights if any of them cannot fit in an uint32_t | |||
| 3070 | FitWeights(NewWeights); | |||
| 3071 | ||||
| 3072 | SmallVector<uint32_t, 8> MDWeights(NewWeights.begin(), NewWeights.end()); | |||
| 3073 | setBranchWeights(PBI, MDWeights[0], MDWeights[1]); | |||
| 3074 | ||||
| 3075 | // TODO: If BB is reachable from all paths through PredBlock, then we | |||
| 3076 | // could replace PBI's branch probabilities with BI's. | |||
| 3077 | } else | |||
| 3078 | PBI->setMetadata(LLVMContext::MD_prof, nullptr); | |||
| 3079 | ||||
| 3080 | // Now, update the CFG. | |||
| 3081 | PBI->setSuccessor(PBI->getSuccessor(0) != BB, UniqueSucc); | |||
| 3082 | ||||
| 3083 | if (DTU) | |||
| 3084 | DTU->applyUpdates({{DominatorTree::Insert, PredBlock, UniqueSucc}, | |||
| 3085 | {DominatorTree::Delete, PredBlock, BB}}); | |||
| 3086 | ||||
| 3087 | // If BI was a loop latch, it may have had associated loop metadata. | |||
| 3088 | // We need to copy it to the new latch, that is, PBI. | |||
| 3089 | if (MDNode *LoopMD = BI->getMetadata(LLVMContext::MD_loop)) | |||
| 3090 | PBI->setMetadata(LLVMContext::MD_loop, LoopMD); | |||
| 3091 | ||||
| 3092 | ValueToValueMapTy VMap; // maps original values to cloned values | |||
| 3093 | CloneInstructionsIntoPredecessorBlockAndUpdateSSAUses(BB, PredBlock, VMap); | |||
| 3094 | ||||
| 3095 | // Now that the Cond was cloned into the predecessor basic block, | |||
| 3096 | // or/and the two conditions together. | |||
| 3097 | Value *BICond = VMap[BI->getCondition()]; | |||
| 3098 | PBI->setCondition( | |||
| 3099 | createLogicalOp(Builder, Opc, PBI->getCondition(), BICond, "or.cond")); | |||
| 3100 | ||||
| 3101 | // Copy any debug value intrinsics into the end of PredBlock. | |||
| 3102 | for (Instruction &I : *BB) { | |||
| 3103 | if (isa<DbgInfoIntrinsic>(I)) { | |||
| 3104 | Instruction *NewI = I.clone(); | |||
| 3105 | RemapInstruction(NewI, VMap, | |||
| 3106 | RF_NoModuleLevelChanges | RF_IgnoreMissingLocals); | |||
| 3107 | NewI->insertBefore(PBI); | |||
| 3108 | } | |||
| 3109 | } | |||
| 3110 | ||||
| 3111 | ++NumFoldBranchToCommonDest; | |||
| 3112 | return true; | |||
| 3113 | } | |||
| 3114 | ||||
| 3115 | /// If this basic block is simple enough, and if a predecessor branches to us | |||
| 3116 | /// and one of our successors, fold the block into the predecessor and use | |||
| 3117 | /// logical operations to pick the right destination. | |||
| 3118 | bool llvm::FoldBranchToCommonDest(BranchInst *BI, DomTreeUpdater *DTU, | |||
| 3119 | MemorySSAUpdater *MSSAU, | |||
| 3120 | const TargetTransformInfo *TTI, | |||
| 3121 | unsigned BonusInstThreshold) { | |||
| 3122 | // If this block ends with an unconditional branch, | |||
| 3123 | // let SpeculativelyExecuteBB() deal with it. | |||
| 3124 | if (!BI->isConditional()) | |||
| 3125 | return false; | |||
| 3126 | ||||
| 3127 | BasicBlock *BB = BI->getParent(); | |||
| 3128 | TargetTransformInfo::TargetCostKind CostKind = | |||
| 3129 | BB->getParent()->hasMinSize() ? TargetTransformInfo::TCK_CodeSize | |||
| 3130 | : TargetTransformInfo::TCK_SizeAndLatency; | |||
| 3131 | ||||
| 3132 | Instruction *Cond = dyn_cast<Instruction>(BI->getCondition()); | |||
| 3133 | ||||
| 3134 | if (!Cond || (!isa<CmpInst>(Cond) && !isa<BinaryOperator>(Cond)) || | |||
| 3135 | Cond->getParent() != BB || !Cond->hasOneUse()) | |||
| 3136 | return false; | |||
| 3137 | ||||
| 3138 | // Cond is known to be a compare or binary operator. Check to make sure that | |||
| 3139 | // neither operand is a potentially-trapping constant expression. | |||
| 3140 | if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Cond->getOperand(0))) | |||
| 3141 | if (CE->canTrap()) | |||
| 3142 | return false; | |||
| 3143 | if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Cond->getOperand(1))) | |||
| 3144 | if (CE->canTrap()) | |||
| 3145 | return false; | |||
| 3146 | ||||
| 3147 | // Finally, don't infinitely unroll conditional loops. | |||
| 3148 | if (is_contained(successors(BB), BB)) | |||
| 3149 | return false; | |||
| 3150 | ||||
| 3151 | // With which predecessors will we want to deal with? | |||
| 3152 | SmallVector<BasicBlock *, 8> Preds; | |||
| 3153 | for (BasicBlock *PredBlock : predecessors(BB)) { | |||
| 3154 | BranchInst *PBI = dyn_cast<BranchInst>(PredBlock->getTerminator()); | |||
| 3155 | ||||
| 3156 | // Check that we have two conditional branches. If there is a PHI node in | |||
| 3157 | // the common successor, verify that the same value flows in from both | |||
| 3158 | // blocks. | |||
| 3159 | if (!PBI || PBI->isUnconditional() || !SafeToMergeTerminators(BI, PBI)) | |||
| 3160 | continue; | |||
| 3161 | ||||
| 3162 | // Determine if the two branches share a common destination. | |||
| 3163 | Instruction::BinaryOps Opc; | |||
| 3164 | bool InvertPredCond; | |||
| 3165 | if (auto Recipe = shouldFoldCondBranchesToCommonDestination(BI, PBI, TTI)) | |||
| 3166 | std::tie(Opc, InvertPredCond) = *Recipe; | |||
| 3167 | else | |||
| 3168 | continue; | |||
| 3169 | ||||
| 3170 | // Check the cost of inserting the necessary logic before performing the | |||
| 3171 | // transformation. | |||
| 3172 | if (TTI) { | |||
| 3173 | Type *Ty = BI->getCondition()->getType(); | |||
| 3174 | InstructionCost Cost = TTI->getArithmeticInstrCost(Opc, Ty, CostKind); | |||
| 3175 | if (InvertPredCond && (!PBI->getCondition()->hasOneUse() || | |||
| 3176 | !isa<CmpInst>(PBI->getCondition()))) | |||
| 3177 | Cost += TTI->getArithmeticInstrCost(Instruction::Xor, Ty, CostKind); | |||
| 3178 | ||||
| 3179 | if (Cost > BranchFoldThreshold) | |||
| 3180 | continue; | |||
| 3181 | } | |||
| 3182 | ||||
| 3183 | // Ok, we do want to deal with this predecessor. Record it. | |||
| 3184 | Preds.emplace_back(PredBlock); | |||
| 3185 | } | |||
| 3186 | ||||
| 3187 | // If there aren't any predecessors into which we can fold, | |||
| 3188 | // don't bother checking the cost. | |||
| 3189 | if (Preds.empty()) | |||
| 3190 | return false; | |||
| 3191 | ||||
| 3192 | // Only allow this transformation if computing the condition doesn't involve | |||
| 3193 | // too many instructions and these involved instructions can be executed | |||
| 3194 | // unconditionally. We denote all involved instructions except the condition | |||
| 3195 | // as "bonus instructions", and only allow this transformation when the | |||
| 3196 | // number of the bonus instructions we'll need to create when cloning into | |||
| 3197 | // each predecessor does not exceed a certain threshold. | |||
| 3198 | unsigned NumBonusInsts = 0; | |||
| 3199 | const unsigned PredCount = Preds.size(); | |||
| 3200 | for (Instruction &I : *BB) { | |||
| 3201 | // Don't check the branch condition comparison itself. | |||
| 3202 | if (&I == Cond) | |||
| 3203 | continue; | |||
| 3204 | // Ignore dbg intrinsics, and the terminator. | |||
| 3205 | if (isa<DbgInfoIntrinsic>(I) || isa<BranchInst>(I)) | |||
| 3206 | continue; | |||
| 3207 | // I must be safe to execute unconditionally. | |||
| 3208 | if (!isSafeToSpeculativelyExecute(&I)) | |||
| 3209 | return false; | |||
| 3210 | ||||
| 3211 | // Account for the cost of duplicating this instruction into each | |||
| 3212 | // predecessor. | |||
| 3213 | NumBonusInsts += PredCount; | |||
| 3214 | // Early exits once we reach the limit. | |||
| 3215 | if (NumBonusInsts > BonusInstThreshold) | |||
| 3216 | return false; | |||
| 3217 | ||||
| 3218 | auto IsBCSSAUse = [BB, &I](Use &U) { | |||
| 3219 | auto *UI = cast<Instruction>(U.getUser()); | |||
| 3220 | if (auto *PN = dyn_cast<PHINode>(UI)) | |||
| 3221 | return PN->getIncomingBlock(U) == BB; | |||
| 3222 | return UI->getParent() == BB && I.comesBefore(UI); | |||
| 3223 | }; | |||
| 3224 | ||||
| 3225 | // Does this instruction require rewriting of uses? | |||
| 3226 | if (!all_of(I.uses(), IsBCSSAUse)) | |||
| 3227 | return false; | |||
| 3228 | } | |||
| 3229 | ||||
| 3230 | // Ok, we have the budget. Perform the transformation. | |||
| 3231 | for (BasicBlock *PredBlock : Preds) { | |||
| 3232 | auto *PBI = cast<BranchInst>(PredBlock->getTerminator()); | |||
| 3233 | return performBranchToCommonDestFolding(BI, PBI, DTU, MSSAU, TTI); | |||
| 3234 | } | |||
| 3235 | return false; | |||
| 3236 | } | |||
| 3237 | ||||
| 3238 | // If there is only one store in BB1 and BB2, return it, otherwise return | |||
| 3239 | // nullptr. | |||
| 3240 | static StoreInst *findUniqueStoreInBlocks(BasicBlock *BB1, BasicBlock *BB2) { | |||
| 3241 | StoreInst *S = nullptr; | |||
| 3242 | for (auto *BB : {BB1, BB2}) { | |||
| 3243 | if (!BB) | |||
| 3244 | continue; | |||
| 3245 | for (auto &I : *BB) | |||
| 3246 | if (auto *SI = dyn_cast<StoreInst>(&I)) { | |||
| 3247 | if (S) | |||
| 3248 | // Multiple stores seen. | |||
| 3249 | return nullptr; | |||
| 3250 | else | |||
| 3251 | S = SI; | |||
| 3252 | } | |||
| 3253 | } | |||
| 3254 | return S; | |||
| 3255 | } | |||
| 3256 | ||||
| 3257 | static Value *ensureValueAvailableInSuccessor(Value *V, BasicBlock *BB, | |||
| 3258 | Value *AlternativeV = nullptr) { | |||
| 3259 | // PHI is going to be a PHI node that allows the value V that is defined in | |||
| 3260 | // BB to be referenced in BB's only successor. | |||
| 3261 | // | |||
| 3262 | // If AlternativeV is nullptr, the only value we care about in PHI is V. It | |||
| 3263 | // doesn't matter to us what the other operand is (it'll never get used). We | |||
| 3264 | // could just create a new PHI with an undef incoming value, but that could | |||
| 3265 | // increase register pressure if EarlyCSE/InstCombine can't fold it with some | |||
| 3266 | // other PHI. So here we directly look for some PHI in BB's successor with V | |||
| 3267 | // as an incoming operand. If we find one, we use it, else we create a new | |||
| 3268 | // one. | |||
| 3269 | // | |||
| 3270 | // If AlternativeV is not nullptr, we care about both incoming values in PHI. | |||
| 3271 | // PHI must be exactly: phi <ty> [ %BB, %V ], [ %OtherBB, %AlternativeV] | |||
| 3272 | // where OtherBB is the single other predecessor of BB's only successor. | |||
| 3273 | PHINode *PHI = nullptr; | |||
| 3274 | BasicBlock *Succ = BB->getSingleSuccessor(); | |||
| 3275 | ||||
| 3276 | for (auto I = Succ->begin(); isa<PHINode>(I); ++I) | |||
| 3277 | if (cast<PHINode>(I)->getIncomingValueForBlock(BB) == V) { | |||
| 3278 | PHI = cast<PHINode>(I); | |||
| 3279 | if (!AlternativeV) | |||
| 3280 | break; | |||
| 3281 | ||||
| 3282 | assert(Succ->hasNPredecessors(2))((void)0); | |||
| 3283 | auto PredI = pred_begin(Succ); | |||
| 3284 | BasicBlock *OtherPredBB = *PredI == BB ? *++PredI : *PredI; | |||
| 3285 | if (PHI->getIncomingValueForBlock(OtherPredBB) == AlternativeV) | |||
| 3286 | break; | |||
| 3287 | PHI = nullptr; | |||
| 3288 | } | |||
| 3289 | if (PHI) | |||
| 3290 | return PHI; | |||
| 3291 | ||||
| 3292 | // If V is not an instruction defined in BB, just return it. | |||
| 3293 | if (!AlternativeV && | |||
| 3294 | (!isa<Instruction>(V) || cast<Instruction>(V)->getParent() != BB)) | |||
| 3295 | return V; | |||
| 3296 | ||||
| 3297 | PHI = PHINode::Create(V->getType(), 2, "simplifycfg.merge", &Succ->front()); | |||
| 3298 | PHI->addIncoming(V, BB); | |||
| 3299 | for (BasicBlock *PredBB : predecessors(Succ)) | |||
| 3300 | if (PredBB != BB) | |||
| 3301 | PHI->addIncoming( | |||
| 3302 | AlternativeV ? AlternativeV : UndefValue::get(V->getType()), PredBB); | |||
| 3303 | return PHI; | |||
| 3304 | } | |||
| 3305 | ||||
| 3306 | static bool mergeConditionalStoreToAddress( | |||
| 3307 | BasicBlock *PTB, BasicBlock *PFB, BasicBlock *QTB, BasicBlock *QFB, | |||
| 3308 | BasicBlock *PostBB, Value *Address, bool InvertPCond, bool InvertQCond, | |||
| 3309 | DomTreeUpdater *DTU, const DataLayout &DL, const TargetTransformInfo &TTI) { | |||
| 3310 | // For every pointer, there must be exactly two stores, one coming from | |||
| 3311 | // PTB or PFB, and the other from QTB or QFB. We don't support more than one | |||
| 3312 | // store (to any address) in PTB,PFB or QTB,QFB. | |||
| 3313 | // FIXME: We could relax this restriction with a bit more work and performance | |||
| 3314 | // testing. | |||
| 3315 | StoreInst *PStore = findUniqueStoreInBlocks(PTB, PFB); | |||
| 3316 | StoreInst *QStore = findUniqueStoreInBlocks(QTB, QFB); | |||
| 3317 | if (!PStore || !QStore) | |||
| 3318 | return false; | |||
| 3319 | ||||
| 3320 | // Now check the stores are compatible. | |||
| 3321 | if (!QStore->isUnordered() || !PStore->isUnordered()) | |||
| 3322 | return false; | |||
| 3323 | ||||
| 3324 | // Check that sinking the store won't cause program behavior changes. Sinking | |||
| 3325 | // the store out of the Q blocks won't change any behavior as we're sinking | |||
| 3326 | // from a block to its unconditional successor. But we're moving a store from | |||
| 3327 | // the P blocks down through the middle block (QBI) and past both QFB and QTB. | |||
| 3328 | // So we need to check that there are no aliasing loads or stores in | |||
| 3329 | // QBI, QTB and QFB. We also need to check there are no conflicting memory | |||
| 3330 | // operations between PStore and the end of its parent block. | |||
| 3331 | // | |||
| 3332 | // The ideal way to do this is to query AliasAnalysis, but we don't | |||
| 3333 | // preserve AA currently so that is dangerous. Be super safe and just | |||
| 3334 | // check there are no other memory operations at all. | |||
| 3335 | for (auto &I : *QFB->getSinglePredecessor()) | |||
| 3336 | if (I.mayReadOrWriteMemory()) | |||
| 3337 | return false; | |||
| 3338 | for (auto &I : *QFB) | |||
| 3339 | if (&I != QStore && I.mayReadOrWriteMemory()) | |||
| 3340 | return false; | |||
| 3341 | if (QTB) | |||
| 3342 | for (auto &I : *QTB) | |||
| 3343 | if (&I != QStore && I.mayReadOrWriteMemory()) | |||
| 3344 | return false; | |||
| 3345 | for (auto I = BasicBlock::iterator(PStore), E = PStore->getParent()->end(); | |||
| 3346 | I != E; ++I) | |||
| 3347 | if (&*I != PStore && I->mayReadOrWriteMemory()) | |||
| 3348 | return false; | |||
| 3349 | ||||
| 3350 | // If we're not in aggressive mode, we only optimize if we have some | |||
| 3351 | // confidence that by optimizing we'll allow P and/or Q to be if-converted. | |||
| 3352 | auto IsWorthwhile = [&](BasicBlock *BB, ArrayRef<StoreInst *> FreeStores) { | |||
| 3353 | if (!BB) | |||
| 3354 | return true; | |||
| 3355 | // Heuristic: if the block can be if-converted/phi-folded and the | |||
| 3356 | // instructions inside are all cheap (arithmetic/GEPs), it's worthwhile to | |||
| 3357 | // thread this store. | |||
| 3358 | InstructionCost Cost = 0; | |||
| 3359 | InstructionCost Budget = | |||
| 3360 | PHINodeFoldingThreshold * TargetTransformInfo::TCC_Basic; | |||
| 3361 | for (auto &I : BB->instructionsWithoutDebug()) { | |||
| 3362 | // Consider terminator instruction to be free. | |||
| 3363 | if (I.isTerminator()) | |||
| 3364 | continue; | |||
| 3365 | // If this is one the stores that we want to speculate out of this BB, | |||
| 3366 | // then don't count it's cost, consider it to be free. | |||
| 3367 | if (auto *S = dyn_cast<StoreInst>(&I)) | |||
| 3368 | if (llvm::find(FreeStores, S)) | |||
| 3369 | continue; | |||
| 3370 | // Else, we have a white-list of instructions that we are ak speculating. | |||
| 3371 | if (!isa<BinaryOperator>(I) && !isa<GetElementPtrInst>(I)) | |||
| 3372 | return false; // Not in white-list - not worthwhile folding. | |||
| 3373 | // And finally, if this is a non-free instruction that we are okay | |||
| 3374 | // speculating, ensure that we consider the speculation budget. | |||
| 3375 | Cost += TTI.getUserCost(&I, TargetTransformInfo::TCK_SizeAndLatency); | |||
| 3376 | if (Cost > Budget) | |||
| 3377 | return false; // Eagerly refuse to fold as soon as we're out of budget. | |||
| 3378 | } | |||
| 3379 | assert(Cost <= Budget &&((void)0) | |||
| 3380 | "When we run out of budget we will eagerly return from within the "((void)0) | |||
| 3381 | "per-instruction loop.")((void)0); | |||
| 3382 | return true; | |||
| 3383 | }; | |||
| 3384 | ||||
| 3385 | const std::array<StoreInst *, 2> FreeStores = {PStore, QStore}; | |||
| 3386 | if (!MergeCondStoresAggressively && | |||
| 3387 | (!IsWorthwhile(PTB, FreeStores) || !IsWorthwhile(PFB, FreeStores) || | |||
| 3388 | !IsWorthwhile(QTB, FreeStores) || !IsWorthwhile(QFB, FreeStores))) | |||
| 3389 | return false; | |||
| 3390 | ||||
| 3391 | // If PostBB has more than two predecessors, we need to split it so we can | |||
| 3392 | // sink the store. | |||
| 3393 | if (std::next(pred_begin(PostBB), 2) != pred_end(PostBB)) { | |||
| 3394 | // We know that QFB's only successor is PostBB. And QFB has a single | |||
| 3395 | // predecessor. If QTB exists, then its only successor is also PostBB. | |||
| 3396 | // If QTB does not exist, then QFB's only predecessor has a conditional | |||
| 3397 | // branch to QFB and PostBB. | |||
| 3398 | BasicBlock *TruePred = QTB ? QTB : QFB->getSinglePredecessor(); | |||
| 3399 | BasicBlock *NewBB = | |||
| 3400 | SplitBlockPredecessors(PostBB, {QFB, TruePred}, "condstore.split", DTU); | |||
| 3401 | if (!NewBB) | |||
| 3402 | return false; | |||
| 3403 | PostBB = NewBB; | |||
| 3404 | } | |||
| 3405 | ||||
| 3406 | // OK, we're going to sink the stores to PostBB. The store has to be | |||
| 3407 | // conditional though, so first create the predicate. | |||
| 3408 | Value *PCond = cast<BranchInst>(PFB->getSinglePredecessor()->getTerminator()) | |||
| 3409 | ->getCondition(); | |||
| 3410 | Value *QCond = cast<BranchInst>(QFB->getSinglePredecessor()->getTerminator()) | |||
| 3411 | ->getCondition(); | |||
| 3412 | ||||
| 3413 | Value *PPHI = ensureValueAvailableInSuccessor(PStore->getValueOperand(), | |||
| 3414 | PStore->getParent()); | |||
| 3415 | Value *QPHI = ensureValueAvailableInSuccessor(QStore->getValueOperand(), | |||
| 3416 | QStore->getParent(), PPHI); | |||
| 3417 | ||||
| 3418 | IRBuilder<> QB(&*PostBB->getFirstInsertionPt()); | |||
| 3419 | ||||
| 3420 | Value *PPred = PStore->getParent() == PTB ? PCond : QB.CreateNot(PCond); | |||
| 3421 | Value *QPred = QStore->getParent() == QTB ? QCond : QB.CreateNot(QCond); | |||
| 3422 | ||||
| 3423 | if (InvertPCond) | |||
| 3424 | PPred = QB.CreateNot(PPred); | |||
| 3425 | if (InvertQCond) | |||
| 3426 | QPred = QB.CreateNot(QPred); | |||
| 3427 | Value *CombinedPred = QB.CreateOr(PPred, QPred); | |||
| 3428 | ||||
| 3429 | auto *T = SplitBlockAndInsertIfThen(CombinedPred, &*QB.GetInsertPoint(), | |||
| 3430 | /*Unreachable=*/false, | |||
| 3431 | /*BranchWeights=*/nullptr, DTU); | |||
| 3432 | QB.SetInsertPoint(T); | |||
| 3433 | StoreInst *SI = cast<StoreInst>(QB.CreateStore(QPHI, Address)); | |||
| 3434 | AAMDNodes AAMD; | |||
| 3435 | PStore->getAAMetadata(AAMD, /*Merge=*/false); | |||
| 3436 | PStore->getAAMetadata(AAMD, /*Merge=*/true); | |||
| 3437 | SI->setAAMetadata(AAMD); | |||
| 3438 | // Choose the minimum alignment. If we could prove both stores execute, we | |||
| 3439 | // could use biggest one. In this case, though, we only know that one of the | |||
| 3440 | // stores executes. And we don't know it's safe to take the alignment from a | |||
| 3441 | // store that doesn't execute. | |||
| 3442 | SI->setAlignment(std::min(PStore->getAlign(), QStore->getAlign())); | |||
| 3443 | ||||
| 3444 | QStore->eraseFromParent(); | |||
| 3445 | PStore->eraseFromParent(); | |||
| 3446 | ||||
| 3447 | return true; | |||
| 3448 | } | |||
| 3449 | ||||
| 3450 | static bool mergeConditionalStores(BranchInst *PBI, BranchInst *QBI, | |||
| 3451 | DomTreeUpdater *DTU, const DataLayout &DL, | |||
| 3452 | const TargetTransformInfo &TTI) { | |||
| 3453 | // The intention here is to find diamonds or triangles (see below) where each | |||
| 3454 | // conditional block contains a store to the same address. Both of these | |||
| 3455 | // stores are conditional, so they can't be unconditionally sunk. But it may | |||
| 3456 | // be profitable to speculatively sink the stores into one merged store at the | |||
| 3457 | // end, and predicate the merged store on the union of the two conditions of | |||
| 3458 | // PBI and QBI. | |||
| 3459 | // | |||
| 3460 | // This can reduce the number of stores executed if both of the conditions are | |||
| 3461 | // true, and can allow the blocks to become small enough to be if-converted. | |||
| 3462 | // This optimization will also chain, so that ladders of test-and-set | |||
| 3463 | // sequences can be if-converted away. | |||
| 3464 | // | |||
| 3465 | // We only deal with simple diamonds or triangles: | |||
| 3466 | // | |||
| 3467 | // PBI or PBI or a combination of the two | |||
| 3468 | // / \ | \ | |||
| 3469 | // PTB PFB | PFB | |||
| 3470 | // \ / | / | |||
| 3471 | // QBI QBI | |||
| 3472 | // / \ | \ | |||
| 3473 | // QTB QFB | QFB | |||
| 3474 | // \ / | / | |||
| 3475 | // PostBB PostBB | |||
| 3476 | // | |||
| 3477 | // We model triangles as a type of diamond with a nullptr "true" block. | |||
| 3478 | // Triangles are canonicalized so that the fallthrough edge is represented by | |||
| 3479 | // a true condition, as in the diagram above. | |||
| 3480 | BasicBlock *PTB = PBI->getSuccessor(0); | |||
| 3481 | BasicBlock *PFB = PBI->getSuccessor(1); | |||
| 3482 | BasicBlock *QTB = QBI->getSuccessor(0); | |||
| 3483 | BasicBlock *QFB = QBI->getSuccessor(1); | |||
| 3484 | BasicBlock *PostBB = QFB->getSingleSuccessor(); | |||
| 3485 | ||||
| 3486 | // Make sure we have a good guess for PostBB. If QTB's only successor is | |||
| 3487 | // QFB, then QFB is a better PostBB. | |||
| 3488 | if (QTB->getSingleSuccessor() == QFB) | |||
| 3489 | PostBB = QFB; | |||
| 3490 | ||||
| 3491 | // If we couldn't find a good PostBB, stop. | |||
| 3492 | if (!PostBB) | |||
| 3493 | return false; | |||
| 3494 | ||||
| 3495 | bool InvertPCond = false, InvertQCond = false; | |||
| 3496 | // Canonicalize fallthroughs to the true branches. | |||
| 3497 | if (PFB == QBI->getParent()) { | |||
| 3498 | std::swap(PFB, PTB); | |||
| 3499 | InvertPCond = true; | |||
| 3500 | } | |||
| 3501 | if (QFB == PostBB) { | |||
| 3502 | std::swap(QFB, QTB); | |||
| 3503 | InvertQCond = true; | |||
| 3504 | } | |||
| 3505 | ||||
| 3506 | // From this point on we can assume PTB or QTB may be fallthroughs but PFB | |||
| 3507 | // and QFB may not. Model fallthroughs as a nullptr block. | |||
| 3508 | if (PTB == QBI->getParent()) | |||
| 3509 | PTB = nullptr; | |||
| 3510 | if (QTB == PostBB) | |||
| 3511 | QTB = nullptr; | |||
| 3512 | ||||
| 3513 | // Legality bailouts. We must have at least the non-fallthrough blocks and | |||
| 3514 | // the post-dominating block, and the non-fallthroughs must only have one | |||
| 3515 | // predecessor. | |||
| 3516 | auto HasOnePredAndOneSucc = [](BasicBlock *BB, BasicBlock *P, BasicBlock *S) { | |||
| 3517 | return BB->getSinglePredecessor() == P && BB->getSingleSuccessor() == S; | |||
| 3518 | }; | |||
| 3519 | if (!HasOnePredAndOneSucc(PFB, PBI->getParent(), QBI->getParent()) || | |||
| 3520 | !HasOnePredAndOneSucc(QFB, QBI->getParent(), PostBB)) | |||
| 3521 | return false; | |||
| 3522 | if ((PTB && !HasOnePredAndOneSucc(PTB, PBI->getParent(), QBI->getParent())) || | |||
| 3523 | (QTB && !HasOnePredAndOneSucc(QTB, QBI->getParent(), PostBB))) | |||
| 3524 | return false; | |||
| 3525 | if (!QBI->getParent()->hasNUses(2)) | |||
| 3526 | return false; | |||
| 3527 | ||||
| 3528 | // OK, this is a sequence of two diamonds or triangles. | |||
| 3529 | // Check if there are stores in PTB or PFB that are repeated in QTB or QFB. | |||
| 3530 | SmallPtrSet<Value *, 4> PStoreAddresses, QStoreAddresses; | |||
| 3531 | for (auto *BB : {PTB, PFB}) { | |||
| 3532 | if (!BB) | |||
| 3533 | continue; | |||
| 3534 | for (auto &I : *BB) | |||
| 3535 | if (StoreInst *SI = dyn_cast<StoreInst>(&I)) | |||
| 3536 | PStoreAddresses.insert(SI->getPointerOperand()); | |||
| 3537 | } | |||
| 3538 | for (auto *BB : {QTB, QFB}) { | |||
| 3539 | if (!BB) | |||
| 3540 | continue; | |||
| 3541 | for (auto &I : *BB) | |||
| 3542 | if (StoreInst *SI = dyn_cast<StoreInst>(&I)) | |||
| 3543 | QStoreAddresses.insert(SI->getPointerOperand()); | |||
| 3544 | } | |||
| 3545 | ||||
| 3546 | set_intersect(PStoreAddresses, QStoreAddresses); | |||
| 3547 | // set_intersect mutates PStoreAddresses in place. Rename it here to make it | |||
| 3548 | // clear what it contains. | |||
| 3549 | auto &CommonAddresses = PStoreAddresses; | |||
| 3550 | ||||
| 3551 | bool Changed = false; | |||
| 3552 | for (auto *Address : CommonAddresses) | |||
| 3553 | Changed |= | |||
| 3554 | mergeConditionalStoreToAddress(PTB, PFB, QTB, QFB, PostBB, Address, | |||
| 3555 | InvertPCond, InvertQCond, DTU, DL, TTI); | |||
| 3556 | return Changed; | |||
| 3557 | } | |||
| 3558 | ||||
| 3559 | /// If the previous block ended with a widenable branch, determine if reusing | |||
| 3560 | /// the target block is profitable and legal. This will have the effect of | |||
| 3561 | /// "widening" PBI, but doesn't require us to reason about hosting safety. | |||
| 3562 | static bool tryWidenCondBranchToCondBranch(BranchInst *PBI, BranchInst *BI, | |||
| 3563 | DomTreeUpdater *DTU) { | |||
| 3564 | // TODO: This can be generalized in two important ways: | |||
| 3565 | // 1) We can allow phi nodes in IfFalseBB and simply reuse all the input | |||
| 3566 | // values from the PBI edge. | |||
| 3567 | // 2) We can sink side effecting instructions into BI's fallthrough | |||
| 3568 | // successor provided they doesn't contribute to computation of | |||
| 3569 | // BI's condition. | |||
| 3570 | Value *CondWB, *WC; | |||
| 3571 | BasicBlock *IfTrueBB, *IfFalseBB; | |||
| 3572 | if (!parseWidenableBranch(PBI, CondWB, WC, IfTrueBB, IfFalseBB) || | |||
| 3573 | IfTrueBB != BI->getParent() || !BI->getParent()->getSinglePredecessor()) | |||
| 3574 | return false; | |||
| 3575 | if (!IfFalseBB->phis().empty()) | |||
| 3576 | return false; // TODO | |||
| 3577 | // Use lambda to lazily compute expensive condition after cheap ones. | |||
| 3578 | auto NoSideEffects = [](BasicBlock &BB) { | |||
| 3579 | return !llvm::any_of(BB, [](const Instruction &I) { | |||
| 3580 | return I.mayWriteToMemory() || I.mayHaveSideEffects(); | |||
| 3581 | }); | |||
| 3582 | }; | |||
| 3583 | if (BI->getSuccessor(1) != IfFalseBB && // no inf looping | |||
| 3584 | BI->getSuccessor(1)->getTerminatingDeoptimizeCall() && // profitability | |||
| 3585 | NoSideEffects(*BI->getParent())) { | |||
| 3586 | auto *OldSuccessor = BI->getSuccessor(1); | |||
| 3587 | OldSuccessor->removePredecessor(BI->getParent()); | |||
| 3588 | BI->setSuccessor(1, IfFalseBB); | |||
| 3589 | if (DTU) | |||
| 3590 | DTU->applyUpdates( | |||
| 3591 | {{DominatorTree::Insert, BI->getParent(), IfFalseBB}, | |||
| 3592 | {DominatorTree::Delete, BI->getParent(), OldSuccessor}}); | |||
| 3593 | return true; | |||
| 3594 | } | |||
| 3595 | if (BI->getSuccessor(0) != IfFalseBB && // no inf looping | |||
| 3596 | BI->getSuccessor(0)->getTerminatingDeoptimizeCall() && // profitability | |||
| 3597 | NoSideEffects(*BI->getParent())) { | |||
| 3598 | auto *OldSuccessor = BI->getSuccessor(0); | |||
| 3599 | OldSuccessor->removePredecessor(BI->getParent()); | |||
| 3600 | BI->setSuccessor(0, IfFalseBB); | |||
| 3601 | if (DTU) | |||
| 3602 | DTU->applyUpdates( | |||
| 3603 | {{DominatorTree::Insert, BI->getParent(), IfFalseBB}, | |||
| 3604 | {DominatorTree::Delete, BI->getParent(), OldSuccessor}}); | |||
| 3605 | return true; | |||
| 3606 | } | |||
| 3607 | return false; | |||
| 3608 | } | |||
| 3609 | ||||
| 3610 | /// If we have a conditional branch as a predecessor of another block, | |||
| 3611 | /// this function tries to simplify it. We know | |||
| 3612 | /// that PBI and BI are both conditional branches, and BI is in one of the | |||
| 3613 | /// successor blocks of PBI - PBI branches to BI. | |||
| 3614 | static bool SimplifyCondBranchToCondBranch(BranchInst *PBI, BranchInst *BI, | |||
| 3615 | DomTreeUpdater *DTU, | |||
| 3616 | const DataLayout &DL, | |||
| 3617 | const TargetTransformInfo &TTI) { | |||
| 3618 | assert(PBI->isConditional() && BI->isConditional())((void)0); | |||
| 3619 | BasicBlock *BB = BI->getParent(); | |||
| 3620 | ||||
| 3621 | // If this block ends with a branch instruction, and if there is a | |||
| 3622 | // predecessor that ends on a branch of the same condition, make | |||
| 3623 | // this conditional branch redundant. | |||
| 3624 | if (PBI->getCondition() == BI->getCondition() && | |||
| 3625 | PBI->getSuccessor(0) != PBI->getSuccessor(1)) { | |||
| 3626 | // Okay, the outcome of this conditional branch is statically | |||
| 3627 | // knowable. If this block had a single pred, handle specially. | |||
| 3628 | if (BB->getSinglePredecessor()) { | |||
| 3629 | // Turn this into a branch on constant. | |||
| 3630 | bool CondIsTrue = PBI->getSuccessor(0) == BB; | |||
| 3631 | BI->setCondition( | |||
| 3632 | ConstantInt::get(Type::getInt1Ty(BB->getContext()), CondIsTrue)); | |||
| 3633 | return true; // Nuke the branch on constant. | |||
| 3634 | } | |||
| 3635 | ||||
| 3636 | // Otherwise, if there are multiple predecessors, insert a PHI that merges | |||
| 3637 | // in the constant and simplify the block result. Subsequent passes of | |||
| 3638 | // simplifycfg will thread the block. | |||
| 3639 | if (BlockIsSimpleEnoughToThreadThrough(BB)) { | |||
| 3640 | pred_iterator PB = pred_begin(BB), PE = pred_end(BB); | |||
| 3641 | PHINode *NewPN = PHINode::Create( | |||
| 3642 | Type::getInt1Ty(BB->getContext()), std::distance(PB, PE), | |||
| 3643 | BI->getCondition()->getName() + ".pr", &BB->front()); | |||
| 3644 | // Okay, we're going to insert the PHI node. Since PBI is not the only | |||
| 3645 | // predecessor, compute the PHI'd conditional value for all of the preds. | |||
| 3646 | // Any predecessor where the condition is not computable we keep symbolic. | |||
| 3647 | for (pred_iterator PI = PB; PI != PE; ++PI) { | |||
| 3648 | BasicBlock *P = *PI; | |||
| 3649 | if ((PBI = dyn_cast<BranchInst>(P->getTerminator())) && PBI != BI && | |||
| 3650 | PBI->isConditional() && PBI->getCondition() == BI->getCondition() && | |||
| 3651 | PBI->getSuccessor(0) != PBI->getSuccessor(1)) { | |||
| 3652 | bool CondIsTrue = PBI->getSuccessor(0) == BB; | |||
| 3653 | NewPN->addIncoming( | |||
| 3654 | ConstantInt::get(Type::getInt1Ty(BB->getContext()), CondIsTrue), | |||
| 3655 | P); | |||
| 3656 | } else { | |||
| 3657 | NewPN->addIncoming(BI->getCondition(), P); | |||
| 3658 | } | |||
| 3659 | } | |||
| 3660 | ||||
| 3661 | BI->setCondition(NewPN); | |||
| 3662 | return true; | |||
| 3663 | } | |||
| 3664 | } | |||
| 3665 | ||||
| 3666 | // If the previous block ended with a widenable branch, determine if reusing | |||
| 3667 | // the target block is profitable and legal. This will have the effect of | |||
| 3668 | // "widening" PBI, but doesn't require us to reason about hosting safety. | |||
| 3669 | if (tryWidenCondBranchToCondBranch(PBI, BI, DTU)) | |||
| 3670 | return true; | |||
| 3671 | ||||
| 3672 | if (auto *CE = dyn_cast<ConstantExpr>(BI->getCondition())) | |||
| 3673 | if (CE->canTrap()) | |||
| 3674 | return false; | |||
| 3675 | ||||
| 3676 | // If both branches are conditional and both contain stores to the same | |||
| 3677 | // address, remove the stores from the conditionals and create a conditional | |||
| 3678 | // merged store at the end. | |||
| 3679 | if (MergeCondStores && mergeConditionalStores(PBI, BI, DTU, DL, TTI)) | |||
| 3680 | return true; | |||
| 3681 | ||||
| 3682 | // If this is a conditional branch in an empty block, and if any | |||
| 3683 | // predecessors are a conditional branch to one of our destinations, | |||
| 3684 | // fold the conditions into logical ops and one cond br. | |||
| 3685 | ||||
| 3686 | // Ignore dbg intrinsics. | |||
| 3687 | if (&*BB->instructionsWithoutDebug().begin() != BI) | |||
| 3688 | return false; | |||
| 3689 | ||||
| 3690 | int PBIOp, BIOp; | |||
| 3691 | if (PBI->getSuccessor(0) == BI->getSuccessor(0)) { | |||
| 3692 | PBIOp = 0; | |||
| 3693 | BIOp = 0; | |||
| 3694 | } else if (PBI->getSuccessor(0) == BI->getSuccessor(1)) { | |||
| 3695 | PBIOp = 0; | |||
| 3696 | BIOp = 1; | |||
| 3697 | } else if (PBI->getSuccessor(1) == BI->getSuccessor(0)) { | |||
| 3698 | PBIOp = 1; | |||
| 3699 | BIOp = 0; | |||
| 3700 | } else if (PBI->getSuccessor(1) == BI->getSuccessor(1)) { | |||
| 3701 | PBIOp = 1; | |||
| 3702 | BIOp = 1; | |||
| 3703 | } else { | |||
| 3704 | return false; | |||
| 3705 | } | |||
| 3706 | ||||
| 3707 | // Check to make sure that the other destination of this branch | |||
| 3708 | // isn't BB itself. If so, this is an infinite loop that will | |||
| 3709 | // keep getting unwound. | |||
| 3710 | if (PBI->getSuccessor(PBIOp) == BB) | |||
| 3711 | return false; | |||
| 3712 | ||||
| 3713 | // Do not perform this transformation if it would require | |||
| 3714 | // insertion of a large number of select instructions. For targets | |||
| 3715 | // without predication/cmovs, this is a big pessimization. | |||
| 3716 | ||||
| 3717 | // Also do not perform this transformation if any phi node in the common | |||
| 3718 | // destination block can trap when reached by BB or PBB (PR17073). In that | |||
| 3719 | // case, it would be unsafe to hoist the operation into a select instruction. | |||
| 3720 | ||||
| 3721 | BasicBlock *CommonDest = PBI->getSuccessor(PBIOp); | |||
| 3722 | BasicBlock *RemovedDest = PBI->getSuccessor(PBIOp ^ 1); | |||
| 3723 | unsigned NumPhis = 0; | |||
| 3724 | for (BasicBlock::iterator II = CommonDest->begin(); isa<PHINode>(II); | |||
| 3725 | ++II, ++NumPhis) { | |||
| 3726 | if (NumPhis > 2) // Disable this xform. | |||
| 3727 | return false; | |||
| 3728 | ||||
| 3729 | PHINode *PN = cast<PHINode>(II); | |||
| 3730 | Value *BIV = PN->getIncomingValueForBlock(BB); | |||
| 3731 | if (ConstantExpr *CE = dyn_cast<ConstantExpr>(BIV)) | |||
| 3732 | if (CE->canTrap()) | |||
| 3733 | return false; | |||
| 3734 | ||||
| 3735 | unsigned PBBIdx = PN->getBasicBlockIndex(PBI->getParent()); | |||
| 3736 | Value *PBIV = PN->getIncomingValue(PBBIdx); | |||
| 3737 | if (ConstantExpr *CE = dyn_cast<ConstantExpr>(PBIV)) | |||
| 3738 | if (CE->canTrap()) | |||
| 3739 | return false; | |||
| 3740 | } | |||
| 3741 | ||||
| 3742 | // Finally, if everything is ok, fold the branches to logical ops. | |||
| 3743 | BasicBlock *OtherDest = BI->getSuccessor(BIOp ^ 1); | |||
| 3744 | ||||
| 3745 | LLVM_DEBUG(dbgs() << "FOLDING BRs:" << *PBI->getParent()do { } while (false) | |||
| 3746 | << "AND: " << *BI->getParent())do { } while (false); | |||
| 3747 | ||||
| 3748 | SmallVector<DominatorTree::UpdateType, 5> Updates; | |||
| 3749 | ||||
| 3750 | // If OtherDest *is* BB, then BB is a basic block with a single conditional | |||
| 3751 | // branch in it, where one edge (OtherDest) goes back to itself but the other | |||
| 3752 | // exits. We don't *know* that the program avoids the infinite loop | |||
| 3753 | // (even though that seems likely). If we do this xform naively, we'll end up | |||
| 3754 | // recursively unpeeling the loop. Since we know that (after the xform is | |||
| 3755 | // done) that the block *is* infinite if reached, we just make it an obviously | |||
| 3756 | // infinite loop with no cond branch. | |||
| 3757 | if (OtherDest == BB) { | |||
| 3758 | // Insert it at the end of the function, because it's either code, | |||
| 3759 | // or it won't matter if it's hot. :) | |||
| 3760 | BasicBlock *InfLoopBlock = | |||
| 3761 | BasicBlock::Create(BB->getContext(), "infloop", BB->getParent()); | |||
| 3762 | BranchInst::Create(InfLoopBlock, InfLoopBlock); | |||
| 3763 | if (DTU) | |||
| 3764 | Updates.push_back({DominatorTree::Insert, InfLoopBlock, InfLoopBlock}); | |||
| 3765 | OtherDest = InfLoopBlock; | |||
| 3766 | } | |||
| 3767 | ||||
| 3768 | LLVM_DEBUG(dbgs() << *PBI->getParent()->getParent())do { } while (false); | |||
| 3769 | ||||
| 3770 | // BI may have other predecessors. Because of this, we leave | |||
| 3771 | // it alone, but modify PBI. | |||
| 3772 | ||||
| 3773 | // Make sure we get to CommonDest on True&True directions. | |||
| 3774 | Value *PBICond = PBI->getCondition(); | |||
| 3775 | IRBuilder<NoFolder> Builder(PBI); | |||
| 3776 | if (PBIOp) | |||
| 3777 | PBICond = Builder.CreateNot(PBICond, PBICond->getName() + ".not"); | |||
| 3778 | ||||
| 3779 | Value *BICond = BI->getCondition(); | |||
| 3780 | if (BIOp) | |||
| 3781 | BICond = Builder.CreateNot(BICond, BICond->getName() + ".not"); | |||
| 3782 | ||||
| 3783 | // Merge the conditions. | |||
| 3784 | Value *Cond = | |||
| 3785 | createLogicalOp(Builder, Instruction::Or, PBICond, BICond, "brmerge"); | |||
| 3786 | ||||
| 3787 | // Modify PBI to branch on the new condition to the new dests. | |||
| 3788 | PBI->setCondition(Cond); | |||
| 3789 | PBI->setSuccessor(0, CommonDest); | |||
| 3790 | PBI->setSuccessor(1, OtherDest); | |||
| 3791 | ||||
| 3792 | if (DTU) { | |||
| 3793 | Updates.push_back({DominatorTree::Insert, PBI->getParent(), OtherDest}); | |||
| 3794 | Updates.push_back({DominatorTree::Delete, PBI->getParent(), RemovedDest}); | |||
| 3795 | ||||
| 3796 | DTU->applyUpdates(Updates); | |||
| 3797 | } | |||
| 3798 | ||||
| 3799 | // Update branch weight for PBI. | |||
| 3800 | uint64_t PredTrueWeight, PredFalseWeight, SuccTrueWeight, SuccFalseWeight; | |||
| 3801 | uint64_t PredCommon, PredOther, SuccCommon, SuccOther; | |||
| 3802 | bool HasWeights = | |||
| 3803 | extractPredSuccWeights(PBI, BI, PredTrueWeight, PredFalseWeight, | |||
| 3804 | SuccTrueWeight, SuccFalseWeight); | |||
| 3805 | if (HasWeights) { | |||
| 3806 | PredCommon = PBIOp ? PredFalseWeight : PredTrueWeight; | |||
| 3807 | PredOther = PBIOp ? PredTrueWeight : PredFalseWeight; | |||
| 3808 | SuccCommon = BIOp ? SuccFalseWeight : SuccTrueWeight; | |||
| 3809 | SuccOther = BIOp ? SuccTrueWeight : SuccFalseWeight; | |||
| 3810 | // The weight to CommonDest should be PredCommon * SuccTotal + | |||
| 3811 | // PredOther * SuccCommon. | |||
| 3812 | // The weight to OtherDest should be PredOther * SuccOther. | |||
| 3813 | uint64_t NewWeights[2] = {PredCommon * (SuccCommon + SuccOther) + | |||
| 3814 | PredOther * SuccCommon, | |||
| 3815 | PredOther * SuccOther}; | |||
| 3816 | // Halve the weights if any of them cannot fit in an uint32_t | |||
| 3817 | FitWeights(NewWeights); | |||
| 3818 | ||||
| 3819 | setBranchWeights(PBI, NewWeights[0], NewWeights[1]); | |||
| 3820 | } | |||
| 3821 | ||||
| 3822 | // OtherDest may have phi nodes. If so, add an entry from PBI's | |||
| 3823 | // block that are identical to the entries for BI's block. | |||
| 3824 | AddPredecessorToBlock(OtherDest, PBI->getParent(), BB); | |||
| 3825 | ||||
| 3826 | // We know that the CommonDest already had an edge from PBI to | |||
| 3827 | // it. If it has PHIs though, the PHIs may have different | |||
| 3828 | // entries for BB and PBI's BB. If so, insert a select to make | |||
| 3829 | // them agree. | |||
| 3830 | for (PHINode &PN : CommonDest->phis()) { | |||
| 3831 | Value *BIV = PN.getIncomingValueForBlock(BB); | |||
| 3832 | unsigned PBBIdx = PN.getBasicBlockIndex(PBI->getParent()); | |||
| 3833 | Value *PBIV = PN.getIncomingValue(PBBIdx); | |||
| 3834 | if (BIV != PBIV) { | |||
| 3835 | // Insert a select in PBI to pick the right value. | |||
| 3836 | SelectInst *NV = cast<SelectInst>( | |||
| 3837 | Builder.CreateSelect(PBICond, PBIV, BIV, PBIV->getName() + ".mux")); | |||
| 3838 | PN.setIncomingValue(PBBIdx, NV); | |||
| 3839 | // Although the select has the same condition as PBI, the original branch | |||
| 3840 | // weights for PBI do not apply to the new select because the select's | |||
| 3841 | // 'logical' edges are incoming edges of the phi that is eliminated, not | |||
| 3842 | // the outgoing edges of PBI. | |||
| 3843 | if (HasWeights) { | |||
| 3844 | uint64_t PredCommon = PBIOp ? PredFalseWeight : PredTrueWeight; | |||
| 3845 | uint64_t PredOther = PBIOp ? PredTrueWeight : PredFalseWeight; | |||
| 3846 | uint64_t SuccCommon = BIOp ? SuccFalseWeight : SuccTrueWeight; | |||
| 3847 | uint64_t SuccOther = BIOp ? SuccTrueWeight : SuccFalseWeight; | |||
| 3848 | // The weight to PredCommonDest should be PredCommon * SuccTotal. | |||
| 3849 | // The weight to PredOtherDest should be PredOther * SuccCommon. | |||
| 3850 | uint64_t NewWeights[2] = {PredCommon * (SuccCommon + SuccOther), | |||
| 3851 | PredOther * SuccCommon}; | |||
| 3852 | ||||
| 3853 | FitWeights(NewWeights); | |||
| 3854 | ||||
| 3855 | setBranchWeights(NV, NewWeights[0], NewWeights[1]); | |||
| 3856 | } | |||
| 3857 | } | |||
| 3858 | } | |||
| 3859 | ||||
| 3860 | LLVM_DEBUG(dbgs() << "INTO: " << *PBI->getParent())do { } while (false); | |||
| 3861 | LLVM_DEBUG(dbgs() << *PBI->getParent()->getParent())do { } while (false); | |||
| 3862 | ||||
| 3863 | // This basic block is probably dead. We know it has at least | |||
| 3864 | // one fewer predecessor. | |||
| 3865 | return true; | |||
| 3866 | } | |||
| 3867 | ||||
| 3868 | // Simplifies a terminator by replacing it with a branch to TrueBB if Cond is | |||
| 3869 | // true or to FalseBB if Cond is false. | |||
| 3870 | // Takes care of updating the successors and removing the old terminator. | |||
| 3871 | // Also makes sure not to introduce new successors by assuming that edges to | |||
| 3872 | // non-successor TrueBBs and FalseBBs aren't reachable. | |||
| 3873 | bool SimplifyCFGOpt::SimplifyTerminatorOnSelect(Instruction *OldTerm, | |||
| 3874 | Value *Cond, BasicBlock *TrueBB, | |||
| 3875 | BasicBlock *FalseBB, | |||
| 3876 | uint32_t TrueWeight, | |||
| 3877 | uint32_t FalseWeight) { | |||
| 3878 | auto *BB = OldTerm->getParent(); | |||
| 3879 | // Remove any superfluous successor edges from the CFG. | |||
| 3880 | // First, figure out which successors to preserve. | |||
| 3881 | // If TrueBB and FalseBB are equal, only try to preserve one copy of that | |||
| 3882 | // successor. | |||
| 3883 | BasicBlock *KeepEdge1 = TrueBB; | |||
| 3884 | BasicBlock *KeepEdge2 = TrueBB != FalseBB ? FalseBB : nullptr; | |||
| 3885 | ||||
| 3886 | SmallPtrSet<BasicBlock *, 2> RemovedSuccessors; | |||
| 3887 | ||||
| 3888 | // Then remove the rest. | |||
| 3889 | for (BasicBlock *Succ : successors(OldTerm)) { | |||
| 3890 | // Make sure only to keep exactly one copy of each edge. | |||
| 3891 | if (Succ == KeepEdge1) | |||
| 3892 | KeepEdge1 = nullptr; | |||
| 3893 | else if (Succ == KeepEdge2) | |||
| 3894 | KeepEdge2 = nullptr; | |||
| 3895 | else { | |||
| 3896 | Succ->removePredecessor(BB, | |||
| 3897 | /*KeepOneInputPHIs=*/true); | |||
| 3898 | ||||
| 3899 | if (Succ != TrueBB && Succ != FalseBB) | |||
| 3900 | RemovedSuccessors.insert(Succ); | |||
| 3901 | } | |||
| 3902 | } | |||
| 3903 | ||||
| 3904 | IRBuilder<> Builder(OldTerm); | |||
| 3905 | Builder.SetCurrentDebugLocation(OldTerm->getDebugLoc()); | |||
| 3906 | ||||
| 3907 | // Insert an appropriate new terminator. | |||
| 3908 | if (!KeepEdge1 && !KeepEdge2) { | |||
| 3909 | if (TrueBB == FalseBB) { | |||
| 3910 | // We were only looking for one successor, and it was present. | |||
| 3911 | // Create an unconditional branch to it. | |||
| 3912 | Builder.CreateBr(TrueBB); | |||
| 3913 | } else { | |||
| 3914 | // We found both of the successors we were looking for. | |||
| 3915 | // Create a conditional branch sharing the condition of the select. | |||
| 3916 | BranchInst *NewBI = Builder.CreateCondBr(Cond, TrueBB, FalseBB); | |||
| 3917 | if (TrueWeight != FalseWeight) | |||
| 3918 | setBranchWeights(NewBI, TrueWeight, FalseWeight); | |||
| 3919 | } | |||
| 3920 | } else if (KeepEdge1 && (KeepEdge2 || TrueBB == FalseBB)) { | |||
| 3921 | // Neither of the selected blocks were successors, so this | |||
| 3922 | // terminator must be unreachable. | |||
| 3923 | new UnreachableInst(OldTerm->getContext(), OldTerm); | |||
| 3924 | } else { | |||
| 3925 | // One of the selected values was a successor, but the other wasn't. | |||
| 3926 | // Insert an unconditional branch to the one that was found; | |||
| 3927 | // the edge to the one that wasn't must be unreachable. | |||
| 3928 | if (!KeepEdge1) { | |||
| 3929 | // Only TrueBB was found. | |||
| 3930 | Builder.CreateBr(TrueBB); | |||
| 3931 | } else { | |||
| 3932 | // Only FalseBB was found. | |||
| 3933 | Builder.CreateBr(FalseBB); | |||
| 3934 | } | |||
| 3935 | } | |||
| 3936 | ||||
| 3937 | EraseTerminatorAndDCECond(OldTerm); | |||
| 3938 | ||||
| 3939 | if (DTU) { | |||
| 3940 | SmallVector<DominatorTree::UpdateType, 2> Updates; | |||
| 3941 | Updates.reserve(RemovedSuccessors.size()); | |||
| 3942 | for (auto *RemovedSuccessor : RemovedSuccessors) | |||
| 3943 | Updates.push_back({DominatorTree::Delete, BB, RemovedSuccessor}); | |||
| 3944 | DTU->applyUpdates(Updates); | |||
| 3945 | } | |||
| 3946 | ||||
| 3947 | return true; | |||
| 3948 | } | |||
| 3949 | ||||
| 3950 | // Replaces | |||
| 3951 | // (switch (select cond, X, Y)) on constant X, Y | |||
| 3952 | // with a branch - conditional if X and Y lead to distinct BBs, | |||
| 3953 | // unconditional otherwise. | |||
| 3954 | bool SimplifyCFGOpt::SimplifySwitchOnSelect(SwitchInst *SI, | |||
| 3955 | SelectInst *Select) { | |||
| 3956 | // Check for constant integer values in the select. | |||
| 3957 | ConstantInt *TrueVal = dyn_cast<ConstantInt>(Select->getTrueValue()); | |||
| 3958 | ConstantInt *FalseVal = dyn_cast<ConstantInt>(Select->getFalseValue()); | |||
| 3959 | if (!TrueVal || !FalseVal) | |||
| 3960 | return false; | |||
| 3961 | ||||
| 3962 | // Find the relevant condition and destinations. | |||
| 3963 | Value *Condition = Select->getCondition(); | |||
| 3964 | BasicBlock *TrueBB = SI->findCaseValue(TrueVal)->getCaseSuccessor(); | |||
| 3965 | BasicBlock *FalseBB = SI->findCaseValue(FalseVal)->getCaseSuccessor(); | |||
| 3966 | ||||
| 3967 | // Get weight for TrueBB and FalseBB. | |||
| 3968 | uint32_t TrueWeight = 0, FalseWeight = 0; | |||
| 3969 | SmallVector<uint64_t, 8> Weights; | |||
| 3970 | bool HasWeights = HasBranchWeights(SI); | |||
| 3971 | if (HasWeights) { | |||
| 3972 | GetBranchWeights(SI, Weights); | |||
| 3973 | if (Weights.size() == 1 + SI->getNumCases()) { | |||
| 3974 | TrueWeight = | |||
| 3975 | (uint32_t)Weights[SI->findCaseValue(TrueVal)->getSuccessorIndex()]; | |||
| 3976 | FalseWeight = | |||
| 3977 | (uint32_t)Weights[SI->findCaseValue(FalseVal)->getSuccessorIndex()]; | |||
| 3978 | } | |||
| 3979 | } | |||
| 3980 | ||||
| 3981 | // Perform the actual simplification. | |||
| 3982 | return SimplifyTerminatorOnSelect(SI, Condition, TrueBB, FalseBB, TrueWeight, | |||
| 3983 | FalseWeight); | |||
| 3984 | } | |||
| 3985 | ||||
| 3986 | // Replaces | |||
| 3987 | // (indirectbr (select cond, blockaddress(@fn, BlockA), | |||
| 3988 | // blockaddress(@fn, BlockB))) | |||
| 3989 | // with | |||
| 3990 | // (br cond, BlockA, BlockB). | |||
| 3991 | bool SimplifyCFGOpt::SimplifyIndirectBrOnSelect(IndirectBrInst *IBI, | |||
| 3992 | SelectInst *SI) { | |||
| 3993 | // Check that both operands of the select are block addresses. | |||
| 3994 | BlockAddress *TBA = dyn_cast<BlockAddress>(SI->getTrueValue()); | |||
| 3995 | BlockAddress *FBA = dyn_cast<BlockAddress>(SI->getFalseValue()); | |||
| 3996 | if (!TBA || !FBA) | |||
| 3997 | return false; | |||
| 3998 | ||||
| 3999 | // Extract the actual blocks. | |||
| 4000 | BasicBlock *TrueBB = TBA->getBasicBlock(); | |||
| 4001 | BasicBlock *FalseBB = FBA->getBasicBlock(); | |||
| 4002 | ||||
| 4003 | // Perform the actual simplification. | |||
| 4004 | return SimplifyTerminatorOnSelect(IBI, SI->getCondition(), TrueBB, FalseBB, 0, | |||
| 4005 | 0); | |||
| 4006 | } | |||
| 4007 | ||||
| 4008 | /// This is called when we find an icmp instruction | |||
| 4009 | /// (a seteq/setne with a constant) as the only instruction in a | |||
| 4010 | /// block that ends with an uncond branch. We are looking for a very specific | |||
| 4011 | /// pattern that occurs when "A == 1 || A == 2 || A == 3" gets simplified. In | |||
| 4012 | /// this case, we merge the first two "or's of icmp" into a switch, but then the | |||
| 4013 | /// default value goes to an uncond block with a seteq in it, we get something | |||
| 4014 | /// like: | |||
| 4015 | /// | |||
| 4016 | /// switch i8 %A, label %DEFAULT [ i8 1, label %end i8 2, label %end ] | |||
| 4017 | /// DEFAULT: | |||
| 4018 | /// %tmp = icmp eq i8 %A, 92 | |||
| 4019 | /// br label %end | |||
| 4020 | /// end: | |||
| 4021 | /// ... = phi i1 [ true, %entry ], [ %tmp, %DEFAULT ], [ true, %entry ] | |||
| 4022 | /// | |||
| 4023 | /// We prefer to split the edge to 'end' so that there is a true/false entry to | |||
| 4024 | /// the PHI, merging the third icmp into the switch. | |||
| 4025 | bool SimplifyCFGOpt::tryToSimplifyUncondBranchWithICmpInIt( | |||
| 4026 | ICmpInst *ICI, IRBuilder<> &Builder) { | |||
| 4027 | BasicBlock *BB = ICI->getParent(); | |||
| 4028 | ||||
| 4029 | // If the block has any PHIs in it or the icmp has multiple uses, it is too | |||
| 4030 | // complex. | |||
| 4031 | if (isa<PHINode>(BB->begin()) || !ICI->hasOneUse()) | |||
| 4032 | return false; | |||
| 4033 | ||||
| 4034 | Value *V = ICI->getOperand(0); | |||
| 4035 | ConstantInt *Cst = cast<ConstantInt>(ICI->getOperand(1)); | |||
| 4036 | ||||
| 4037 | // The pattern we're looking for is where our only predecessor is a switch on | |||
| 4038 | // 'V' and this block is the default case for the switch. In this case we can | |||
| 4039 | // fold the compared value into the switch to simplify things. | |||
| 4040 | BasicBlock *Pred = BB->getSinglePredecessor(); | |||
| 4041 | if (!Pred || !isa<SwitchInst>(Pred->getTerminator())) | |||
| 4042 | return false; | |||
| 4043 | ||||
| 4044 | SwitchInst *SI = cast<SwitchInst>(Pred->getTerminator()); | |||
| 4045 | if (SI->getCondition() != V) | |||
| 4046 | return false; | |||
| 4047 | ||||
| 4048 | // If BB is reachable on a non-default case, then we simply know the value of | |||
| 4049 | // V in this block. Substitute it and constant fold the icmp instruction | |||
| 4050 | // away. | |||
| 4051 | if (SI->getDefaultDest() != BB) { | |||
| 4052 | ConstantInt *VVal = SI->findCaseDest(BB); | |||
| 4053 | assert(VVal && "Should have a unique destination value")((void)0); | |||
| 4054 | ICI->setOperand(0, VVal); | |||
| 4055 | ||||
| 4056 | if (Value *V = SimplifyInstruction(ICI, {DL, ICI})) { | |||
| 4057 | ICI->replaceAllUsesWith(V); | |||
| 4058 | ICI->eraseFromParent(); | |||
| 4059 | } | |||
| 4060 | // BB is now empty, so it is likely to simplify away. | |||
| 4061 | return requestResimplify(); | |||
| 4062 | } | |||
| 4063 | ||||
| 4064 | // Ok, the block is reachable from the default dest. If the constant we're | |||
| 4065 | // comparing exists in one of the other edges, then we can constant fold ICI | |||
| 4066 | // and zap it. | |||
| 4067 | if (SI->findCaseValue(Cst) != SI->case_default()) { | |||
| 4068 | Value *V; | |||
| 4069 | if (ICI->getPredicate() == ICmpInst::ICMP_EQ) | |||
| 4070 | V = ConstantInt::getFalse(BB->getContext()); | |||
| 4071 | else | |||
| 4072 | V = ConstantInt::getTrue(BB->getContext()); | |||
| 4073 | ||||
| 4074 | ICI->replaceAllUsesWith(V); | |||
| 4075 | ICI->eraseFromParent(); | |||
| 4076 | // BB is now empty, so it is likely to simplify away. | |||
| 4077 | return requestResimplify(); | |||
| 4078 | } | |||
| 4079 | ||||
| 4080 | // The use of the icmp has to be in the 'end' block, by the only PHI node in | |||
| 4081 | // the block. | |||
| 4082 | BasicBlock *SuccBlock = BB->getTerminator()->getSuccessor(0); | |||
| 4083 | PHINode *PHIUse = dyn_cast<PHINode>(ICI->user_back()); | |||
| 4084 | if (PHIUse == nullptr || PHIUse != &SuccBlock->front() || | |||
| 4085 | isa<PHINode>(++BasicBlock::iterator(PHIUse))) | |||
| 4086 | return false; | |||
| 4087 | ||||
| 4088 | // If the icmp is a SETEQ, then the default dest gets false, the new edge gets | |||
| 4089 | // true in the PHI. | |||
| 4090 | Constant *DefaultCst = ConstantInt::getTrue(BB->getContext()); | |||
| 4091 | Constant *NewCst = ConstantInt::getFalse(BB->getContext()); | |||
| 4092 | ||||
| 4093 | if (ICI->getPredicate() == ICmpInst::ICMP_EQ) | |||
| 4094 | std::swap(DefaultCst, NewCst); | |||
| 4095 | ||||
| 4096 | // Replace ICI (which is used by the PHI for the default value) with true or | |||
| 4097 | // false depending on if it is EQ or NE. | |||
| 4098 | ICI->replaceAllUsesWith(DefaultCst); | |||
| 4099 | ICI->eraseFromParent(); | |||
| 4100 | ||||
| 4101 | SmallVector<DominatorTree::UpdateType, 2> Updates; | |||
| 4102 | ||||
| 4103 | // Okay, the switch goes to this block on a default value. Add an edge from | |||
| 4104 | // the switch to the merge point on the compared value. | |||
| 4105 | BasicBlock *NewBB = | |||
| 4106 | BasicBlock::Create(BB->getContext(), "switch.edge", BB->getParent(), BB); | |||
| 4107 | { | |||
| 4108 | SwitchInstProfUpdateWrapper SIW(*SI); | |||
| 4109 | auto W0 = SIW.getSuccessorWeight(0); | |||
| 4110 | SwitchInstProfUpdateWrapper::CaseWeightOpt NewW; | |||
| 4111 | if (W0) { | |||
| 4112 | NewW = ((uint64_t(*W0) + 1) >> 1); | |||
| 4113 | SIW.setSuccessorWeight(0, *NewW); | |||
| 4114 | } | |||
| 4115 | SIW.addCase(Cst, NewBB, NewW); | |||
| 4116 | if (DTU) | |||
| 4117 | Updates.push_back({DominatorTree::Insert, Pred, NewBB}); | |||
| 4118 | } | |||
| 4119 | ||||
| 4120 | // NewBB branches to the phi block, add the uncond branch and the phi entry. | |||
| 4121 | Builder.SetInsertPoint(NewBB); | |||
| 4122 | Builder.SetCurrentDebugLocation(SI->getDebugLoc()); | |||
| 4123 | Builder.CreateBr(SuccBlock); | |||
| 4124 | PHIUse->addIncoming(NewCst, NewBB); | |||
| 4125 | if (DTU) { | |||
| 4126 | Updates.push_back({DominatorTree::Insert, NewBB, SuccBlock}); | |||
| 4127 | DTU->applyUpdates(Updates); | |||
| 4128 | } | |||
| 4129 | return true; | |||
| 4130 | } | |||
| 4131 | ||||
| 4132 | /// The specified branch is a conditional branch. | |||
| 4133 | /// Check to see if it is branching on an or/and chain of icmp instructions, and | |||
| 4134 | /// fold it into a switch instruction if so. | |||
| 4135 | bool SimplifyCFGOpt::SimplifyBranchOnICmpChain(BranchInst *BI, | |||
| 4136 | IRBuilder<> &Builder, | |||
| 4137 | const DataLayout &DL) { | |||
| 4138 | Instruction *Cond = dyn_cast<Instruction>(BI->getCondition()); | |||
| 4139 | if (!Cond) | |||
| 4140 | return false; | |||
| 4141 | ||||
| 4142 | // Change br (X == 0 | X == 1), T, F into a switch instruction. | |||
| 4143 | // If this is a bunch of seteq's or'd together, or if it's a bunch of | |||
| 4144 | // 'setne's and'ed together, collect them. | |||
| 4145 | ||||
| 4146 | // Try to gather values from a chain of and/or to be turned into a switch | |||
| 4147 | ConstantComparesGatherer ConstantCompare(Cond, DL); | |||
| 4148 | // Unpack the result | |||
| 4149 | SmallVectorImpl<ConstantInt *> &Values = ConstantCompare.Vals; | |||
| 4150 | Value *CompVal = ConstantCompare.CompValue; | |||
| 4151 | unsigned UsedICmps = ConstantCompare.UsedICmps; | |||
| 4152 | Value *ExtraCase = ConstantCompare.Extra; | |||
| 4153 | ||||
| 4154 | // If we didn't have a multiply compared value, fail. | |||
| 4155 | if (!CompVal) | |||
| 4156 | return false; | |||
| 4157 | ||||
| 4158 | // Avoid turning single icmps into a switch. | |||
| 4159 | if (UsedICmps <= 1) | |||
| 4160 | return false; | |||
| 4161 | ||||
| 4162 | bool TrueWhenEqual = match(Cond, m_LogicalOr(m_Value(), m_Value())); | |||
| 4163 | ||||
| 4164 | // There might be duplicate constants in the list, which the switch | |||
| 4165 | // instruction can't handle, remove them now. | |||
| 4166 | array_pod_sort(Values.begin(), Values.end(), ConstantIntSortPredicate); | |||
| 4167 | Values.erase(std::unique(Values.begin(), Values.end()), Values.end()); | |||
| 4168 | ||||
| 4169 | // If Extra was used, we require at least two switch values to do the | |||
| 4170 | // transformation. A switch with one value is just a conditional branch. | |||
| 4171 | if (ExtraCase && Values.size() < 2) | |||
| 4172 | return false; | |||
| 4173 | ||||
| 4174 | // TODO: Preserve branch weight metadata, similarly to how | |||
| 4175 | // FoldValueComparisonIntoPredecessors preserves it. | |||
| 4176 | ||||
| 4177 | // Figure out which block is which destination. | |||
| 4178 | BasicBlock *DefaultBB = BI->getSuccessor(1); | |||
| 4179 | BasicBlock *EdgeBB = BI->getSuccessor(0); | |||
| 4180 | if (!TrueWhenEqual) | |||
| 4181 | std::swap(DefaultBB, EdgeBB); | |||
| 4182 | ||||
| 4183 | BasicBlock *BB = BI->getParent(); | |||
| 4184 | ||||
| 4185 | LLVM_DEBUG(dbgs() << "Converting 'icmp' chain with " << Values.size()do { } while (false) | |||
| 4186 | << " cases into SWITCH. BB is:\n"do { } while (false) | |||
| 4187 | << *BB)do { } while (false); | |||
| 4188 | ||||
| 4189 | SmallVector<DominatorTree::UpdateType, 2> Updates; | |||
| 4190 | ||||
| 4191 | // If there are any extra values that couldn't be folded into the switch | |||
| 4192 | // then we evaluate them with an explicit branch first. Split the block | |||
| 4193 | // right before the condbr to handle it. | |||
| 4194 | if (ExtraCase) { | |||
| 4195 | BasicBlock *NewBB = SplitBlock(BB, BI, DTU, /*LI=*/nullptr, | |||
| 4196 | /*MSSAU=*/nullptr, "switch.early.test"); | |||
| 4197 | ||||
| 4198 | // Remove the uncond branch added to the old block. | |||
| 4199 | Instruction *OldTI = BB->getTerminator(); | |||
| 4200 | Builder.SetInsertPoint(OldTI); | |||
| 4201 | ||||
| 4202 | // There can be an unintended UB if extra values are Poison. Before the | |||
| 4203 | // transformation, extra values may not be evaluated according to the | |||
| 4204 | // condition, and it will not raise UB. But after transformation, we are | |||
| 4205 | // evaluating extra values before checking the condition, and it will raise | |||
| 4206 | // UB. It can be solved by adding freeze instruction to extra values. | |||
| 4207 | AssumptionCache *AC = Options.AC; | |||
| 4208 | ||||
| 4209 | if (!isGuaranteedNotToBeUndefOrPoison(ExtraCase, AC, BI, nullptr)) | |||
| 4210 | ExtraCase = Builder.CreateFreeze(ExtraCase); | |||
| 4211 | ||||
| 4212 | if (TrueWhenEqual) | |||
| 4213 | Builder.CreateCondBr(ExtraCase, EdgeBB, NewBB); | |||
| 4214 | else | |||
| 4215 | Builder.CreateCondBr(ExtraCase, NewBB, EdgeBB); | |||
| 4216 | ||||
| 4217 | OldTI->eraseFromParent(); | |||
| 4218 | ||||
| 4219 | if (DTU) | |||
| 4220 | Updates.push_back({DominatorTree::Insert, BB, EdgeBB}); | |||
| 4221 | ||||
| 4222 | // If there are PHI nodes in EdgeBB, then we need to add a new entry to them | |||
| 4223 | // for the edge we just added. | |||
| 4224 | AddPredecessorToBlock(EdgeBB, BB, NewBB); | |||
| 4225 | ||||
| 4226 | LLVM_DEBUG(dbgs() << " ** 'icmp' chain unhandled condition: " << *ExtraCasedo { } while (false) | |||
| 4227 | << "\nEXTRABB = " << *BB)do { } while (false); | |||
| 4228 | BB = NewBB; | |||
| 4229 | } | |||
| 4230 | ||||
| 4231 | Builder.SetInsertPoint(BI); | |||
| 4232 | // Convert pointer to int before we switch. | |||
| 4233 | if (CompVal->getType()->isPointerTy()) { | |||
| 4234 | CompVal = Builder.CreatePtrToInt( | |||
| 4235 | CompVal, DL.getIntPtrType(CompVal->getType()), "magicptr"); | |||
| 4236 | } | |||
| 4237 | ||||
| 4238 | // Create the new switch instruction now. | |||
| 4239 | SwitchInst *New = Builder.CreateSwitch(CompVal, DefaultBB, Values.size()); | |||
| 4240 | ||||
| 4241 | // Add all of the 'cases' to the switch instruction. | |||
| 4242 | for (unsigned i = 0, e = Values.size(); i != e; ++i) | |||
| 4243 | New->addCase(Values[i], EdgeBB); | |||
| 4244 | ||||
| 4245 | // We added edges from PI to the EdgeBB. As such, if there were any | |||
| 4246 | // PHI nodes in EdgeBB, they need entries to be added corresponding to | |||
| 4247 | // the number of edges added. | |||
| 4248 | for (BasicBlock::iterator BBI = EdgeBB->begin(); isa<PHINode>(BBI); ++BBI) { | |||
| 4249 | PHINode *PN = cast<PHINode>(BBI); | |||
| 4250 | Value *InVal = PN->getIncomingValueForBlock(BB); | |||
| 4251 | for (unsigned i = 0, e = Values.size() - 1; i != e; ++i) | |||
| 4252 | PN->addIncoming(InVal, BB); | |||
| 4253 | } | |||
| 4254 | ||||
| 4255 | // Erase the old branch instruction. | |||
| 4256 | EraseTerminatorAndDCECond(BI); | |||
| 4257 | if (DTU) | |||
| 4258 | DTU->applyUpdates(Updates); | |||
| 4259 | ||||
| 4260 | LLVM_DEBUG(dbgs() << " ** 'icmp' chain result is:\n" << *BB << '\n')do { } while (false); | |||
| 4261 | return true; | |||
| 4262 | } | |||
| 4263 | ||||
| 4264 | bool SimplifyCFGOpt::simplifyResume(ResumeInst *RI, IRBuilder<> &Builder) { | |||
| 4265 | if (isa<PHINode>(RI->getValue())) | |||
| 4266 | return simplifyCommonResume(RI); | |||
| 4267 | else if (isa<LandingPadInst>(RI->getParent()->getFirstNonPHI()) && | |||
| 4268 | RI->getValue() == RI->getParent()->getFirstNonPHI()) | |||
| 4269 | // The resume must unwind the exception that caused control to branch here. | |||
| 4270 | return simplifySingleResume(RI); | |||
| 4271 | ||||
| 4272 | return false; | |||
| 4273 | } | |||
| 4274 | ||||
| 4275 | // Check if cleanup block is empty | |||
| 4276 | static bool isCleanupBlockEmpty(iterator_range<BasicBlock::iterator> R) { | |||
| 4277 | for (Instruction &I : R) { | |||
| 4278 | auto *II = dyn_cast<IntrinsicInst>(&I); | |||
| 4279 | if (!II) | |||
| 4280 | return false; | |||
| 4281 | ||||
| 4282 | Intrinsic::ID IntrinsicID = II->getIntrinsicID(); | |||
| 4283 | switch (IntrinsicID) { | |||
| 4284 | case Intrinsic::dbg_declare: | |||
| 4285 | case Intrinsic::dbg_value: | |||
| 4286 | case Intrinsic::dbg_label: | |||
| 4287 | case Intrinsic::lifetime_end: | |||
| 4288 | break; | |||
| 4289 | default: | |||
| 4290 | return false; | |||
| 4291 | } | |||
| 4292 | } | |||
| 4293 | return true; | |||
| 4294 | } | |||
| 4295 | ||||
| 4296 | // Simplify resume that is shared by several landing pads (phi of landing pad). | |||
| 4297 | bool SimplifyCFGOpt::simplifyCommonResume(ResumeInst *RI) { | |||
| 4298 | BasicBlock *BB = RI->getParent(); | |||
| 4299 | ||||
| 4300 | // Check that there are no other instructions except for debug and lifetime | |||
| 4301 | // intrinsics between the phi's and resume instruction. | |||
| 4302 | if (!isCleanupBlockEmpty( | |||
| 4303 | make_range(RI->getParent()->getFirstNonPHI(), BB->getTerminator()))) | |||
| 4304 | return false; | |||
| 4305 | ||||
| 4306 | SmallSetVector<BasicBlock *, 4> TrivialUnwindBlocks; | |||
| 4307 | auto *PhiLPInst = cast<PHINode>(RI->getValue()); | |||
| 4308 | ||||
| 4309 | // Check incoming blocks to see if any of them are trivial. | |||
| 4310 | for (unsigned Idx = 0, End = PhiLPInst->getNumIncomingValues(); Idx != End; | |||
| 4311 | Idx++) { | |||
| 4312 | auto *IncomingBB = PhiLPInst->getIncomingBlock(Idx); | |||
| 4313 | auto *IncomingValue = PhiLPInst->getIncomingValue(Idx); | |||
| 4314 | ||||
| 4315 | // If the block has other successors, we can not delete it because | |||
| 4316 | // it has other dependents. | |||
| 4317 | if (IncomingBB->getUniqueSuccessor() != BB) | |||
| 4318 | continue; | |||
| 4319 | ||||
| 4320 | auto *LandingPad = dyn_cast<LandingPadInst>(IncomingBB->getFirstNonPHI()); | |||
| 4321 | // Not the landing pad that caused the control to branch here. | |||
| 4322 | if (IncomingValue
| |||
| 4323 | continue; | |||
| 4324 | ||||
| 4325 | if (isCleanupBlockEmpty( | |||
| 4326 | make_range(LandingPad->getNextNode(), IncomingBB->getTerminator()))) | |||
| ||||
| 4327 | TrivialUnwindBlocks.insert(IncomingBB); | |||
| 4328 | } | |||
| 4329 | ||||
| 4330 | // If no trivial unwind blocks, don't do any simplifications. | |||
| 4331 | if (TrivialUnwindBlocks.empty()) | |||
| 4332 | return false; | |||
| 4333 | ||||
| 4334 | // Turn all invokes that unwind here into calls. | |||
| 4335 | for (auto *TrivialBB : TrivialUnwindBlocks) { | |||
| 4336 | // Blocks that will be simplified should be removed from the phi node. | |||
| 4337 | // Note there could be multiple edges to the resume block, and we need | |||
| 4338 | // to remove them all. | |||
| 4339 | while (PhiLPInst->getBasicBlockIndex(TrivialBB) != -1) | |||
| 4340 | BB->removePredecessor(TrivialBB, true); | |||
| 4341 | ||||
| 4342 | for (BasicBlock *Pred : | |||
| 4343 | llvm::make_early_inc_range(predecessors(TrivialBB))) { | |||
| 4344 | removeUnwindEdge(Pred, DTU); | |||
| 4345 | ++NumInvokes; | |||
| 4346 | } | |||
| 4347 | ||||
| 4348 | // In each SimplifyCFG run, only the current processed block can be erased. | |||
| 4349 | // Otherwise, it will break the iteration of SimplifyCFG pass. So instead | |||
| 4350 | // of erasing TrivialBB, we only remove the branch to the common resume | |||
| 4351 | // block so that we can later erase the resume block since it has no | |||
| 4352 | // predecessors. | |||
| 4353 | TrivialBB->getTerminator()->eraseFromParent(); | |||
| 4354 | new UnreachableInst(RI->getContext(), TrivialBB); | |||
| 4355 | if (DTU) | |||
| 4356 | DTU->applyUpdates({{DominatorTree::Delete, TrivialBB, BB}}); | |||
| 4357 | } | |||
| 4358 | ||||
| 4359 | // Delete the resume block if all its predecessors have been removed. | |||
| 4360 | if (pred_empty(BB)) | |||
| 4361 | DeleteDeadBlock(BB, DTU); | |||
| 4362 | ||||
| 4363 | return !TrivialUnwindBlocks.empty(); | |||
| 4364 | } | |||
| 4365 | ||||
| 4366 | // Simplify resume that is only used by a single (non-phi) landing pad. | |||
| 4367 | bool SimplifyCFGOpt::simplifySingleResume(ResumeInst *RI) { | |||
| 4368 | BasicBlock *BB = RI->getParent(); | |||
| 4369 | auto *LPInst = cast<LandingPadInst>(BB->getFirstNonPHI()); | |||
| 4370 | assert(RI->getValue() == LPInst &&((void)0) | |||
| 4371 | "Resume must unwind the exception that caused control to here")((void)0); | |||
| 4372 | ||||
| 4373 | // Check that there are no other instructions except for debug intrinsics. | |||
| 4374 | if (!isCleanupBlockEmpty( | |||
| 4375 | make_range<Instruction *>(LPInst->getNextNode(), RI))) | |||
| 4376 | return false; | |||
| 4377 | ||||
| 4378 | // Turn all invokes that unwind here into calls and delete the basic block. | |||
| 4379 | for (BasicBlock *Pred : llvm::make_early_inc_range(predecessors(BB))) { | |||
| 4380 | removeUnwindEdge(Pred, DTU); | |||
| 4381 | ++NumInvokes; | |||
| 4382 | } | |||
| 4383 | ||||
| 4384 | // The landingpad is now unreachable. Zap it. | |||
| 4385 | DeleteDeadBlock(BB, DTU); | |||
| 4386 | return true; | |||
| 4387 | } | |||
| 4388 | ||||
| 4389 | static bool removeEmptyCleanup(CleanupReturnInst *RI, DomTreeUpdater *DTU) { | |||
| 4390 | // If this is a trivial cleanup pad that executes no instructions, it can be | |||
| 4391 | // eliminated. If the cleanup pad continues to the caller, any predecessor | |||
| 4392 | // that is an EH pad will be updated to continue to the caller and any | |||
| 4393 | // predecessor that terminates with an invoke instruction will have its invoke | |||
| 4394 | // instruction converted to a call instruction. If the cleanup pad being | |||
| 4395 | // simplified does not continue to the caller, each predecessor will be | |||
| 4396 | // updated to continue to the unwind destination of the cleanup pad being | |||
| 4397 | // simplified. | |||
| 4398 | BasicBlock *BB = RI->getParent(); | |||
| 4399 | CleanupPadInst *CPInst = RI->getCleanupPad(); | |||
| 4400 | if (CPInst->getParent() != BB) | |||
| 4401 | // This isn't an empty cleanup. | |||
| 4402 | return false; | |||
| 4403 | ||||
| 4404 | // We cannot kill the pad if it has multiple uses. This typically arises | |||
| 4405 | // from unreachable basic blocks. | |||
| 4406 | if (!CPInst->hasOneUse()) | |||
| 4407 | return false; | |||
| 4408 | ||||
| 4409 | // Check that there are no other instructions except for benign intrinsics. | |||
| 4410 | if (!isCleanupBlockEmpty( | |||
| 4411 | make_range<Instruction *>(CPInst->getNextNode(), RI))) | |||
| 4412 | return false; | |||
| 4413 | ||||
| 4414 | // If the cleanup return we are simplifying unwinds to the caller, this will | |||
| 4415 | // set UnwindDest to nullptr. | |||
| 4416 | BasicBlock *UnwindDest = RI->getUnwindDest(); | |||
| 4417 | Instruction *DestEHPad = UnwindDest ? UnwindDest->getFirstNonPHI() : nullptr; | |||
| 4418 | ||||
| 4419 | // We're about to remove BB from the control flow. Before we do, sink any | |||
| 4420 | // PHINodes into the unwind destination. Doing this before changing the | |||
| 4421 | // control flow avoids some potentially slow checks, since we can currently | |||
| 4422 | // be certain that UnwindDest and BB have no common predecessors (since they | |||
| 4423 | // are both EH pads). | |||
| 4424 | if (UnwindDest) { | |||
| 4425 | // First, go through the PHI nodes in UnwindDest and update any nodes that | |||
| 4426 | // reference the block we are removing | |||
| 4427 | for (PHINode &DestPN : UnwindDest->phis()) { | |||
| 4428 | int Idx = DestPN.getBasicBlockIndex(BB); | |||
| 4429 | // Since BB unwinds to UnwindDest, it has to be in the PHI node. | |||
| 4430 | assert(Idx != -1)((void)0); | |||
| 4431 | // This PHI node has an incoming value that corresponds to a control | |||
| 4432 | // path through the cleanup pad we are removing. If the incoming | |||
| 4433 | // value is in the cleanup pad, it must be a PHINode (because we | |||
| 4434 | // verified above that the block is otherwise empty). Otherwise, the | |||
| 4435 | // value is either a constant or a value that dominates the cleanup | |||
| 4436 | // pad being removed. | |||
| 4437 | // | |||
| 4438 | // Because BB and UnwindDest are both EH pads, all of their | |||
| 4439 | // predecessors must unwind to these blocks, and since no instruction | |||
| 4440 | // can have multiple unwind destinations, there will be no overlap in | |||
| 4441 | // incoming blocks between SrcPN and DestPN. | |||
| 4442 | Value *SrcVal = DestPN.getIncomingValue(Idx); | |||
| 4443 | PHINode *SrcPN = dyn_cast<PHINode>(SrcVal); | |||
| 4444 | ||||
| 4445 | bool NeedPHITranslation = SrcPN && SrcPN->getParent() == BB; | |||
| 4446 | for (auto *Pred : predecessors(BB)) { | |||
| 4447 | Value *Incoming = | |||
| 4448 | NeedPHITranslation ? SrcPN->getIncomingValueForBlock(Pred) : SrcVal; | |||
| 4449 | DestPN.addIncoming(Incoming, Pred); | |||
| 4450 | } | |||
| 4451 | } | |||
| 4452 | ||||
| 4453 | // Sink any remaining PHI nodes directly into UnwindDest. | |||
| 4454 | Instruction *InsertPt = DestEHPad; | |||
| 4455 | for (PHINode &PN : make_early_inc_range(BB->phis())) { | |||
| 4456 | if (PN.use_empty() || !PN.isUsedOutsideOfBlock(BB)) | |||
| 4457 | // If the PHI node has no uses or all of its uses are in this basic | |||
| 4458 | // block (meaning they are debug or lifetime intrinsics), just leave | |||
| 4459 | // it. It will be erased when we erase BB below. | |||
| 4460 | continue; | |||
| 4461 | ||||
| 4462 | // Otherwise, sink this PHI node into UnwindDest. | |||
| 4463 | // Any predecessors to UnwindDest which are not already represented | |||
| 4464 | // must be back edges which inherit the value from the path through | |||
| 4465 | // BB. In this case, the PHI value must reference itself. | |||
| 4466 | for (auto *pred : predecessors(UnwindDest)) | |||
| 4467 | if (pred != BB) | |||
| 4468 | PN.addIncoming(&PN, pred); | |||
| 4469 | PN.moveBefore(InsertPt); | |||
| 4470 | // Also, add a dummy incoming value for the original BB itself, | |||
| 4471 | // so that the PHI is well-formed until we drop said predecessor. | |||
| 4472 | PN.addIncoming(UndefValue::get(PN.getType()), BB); | |||
| 4473 | } | |||
| 4474 | } | |||
| 4475 | ||||
| 4476 | std::vector<DominatorTree::UpdateType> Updates; | |||
| 4477 | ||||
| 4478 | // We use make_early_inc_range here because we will remove all predecessors. | |||
| 4479 | for (BasicBlock *PredBB : llvm::make_early_inc_range(predecessors(BB))) { | |||
| 4480 | if (UnwindDest == nullptr) { | |||
| 4481 | if (DTU) { | |||
| 4482 | DTU->applyUpdates(Updates); | |||
| 4483 | Updates.clear(); | |||
| 4484 | } | |||
| 4485 | removeUnwindEdge(PredBB, DTU); | |||
| 4486 | ++NumInvokes; | |||
| 4487 | } else { | |||
| 4488 | BB->removePredecessor(PredBB); | |||
| 4489 | Instruction *TI = PredBB->getTerminator(); | |||
| 4490 | TI->replaceUsesOfWith(BB, UnwindDest); | |||
| 4491 | if (DTU) { | |||
| 4492 | Updates.push_back({DominatorTree::Insert, PredBB, UnwindDest}); | |||
| 4493 | Updates.push_back({DominatorTree::Delete, PredBB, BB}); | |||
| 4494 | } | |||
| 4495 | } | |||
| 4496 | } | |||
| 4497 | ||||
| 4498 | if (DTU) | |||
| 4499 | DTU->applyUpdates(Updates); | |||
| 4500 | ||||
| 4501 | DeleteDeadBlock(BB, DTU); | |||
| 4502 | ||||
| 4503 | return true; | |||
| 4504 | } | |||
| 4505 | ||||
| 4506 | // Try to merge two cleanuppads together. | |||
| 4507 | static bool mergeCleanupPad(CleanupReturnInst *RI) { | |||
| 4508 | // Skip any cleanuprets which unwind to caller, there is nothing to merge | |||
| 4509 | // with. | |||
| 4510 | BasicBlock *UnwindDest = RI->getUnwindDest(); | |||
| 4511 | if (!UnwindDest) | |||
| 4512 | return false; | |||
| 4513 | ||||
| 4514 | // This cleanupret isn't the only predecessor of this cleanuppad, it wouldn't | |||
| 4515 | // be safe to merge without code duplication. | |||
| 4516 | if (UnwindDest->getSinglePredecessor() != RI->getParent()) | |||
| 4517 | return false; | |||
| 4518 | ||||
| 4519 | // Verify that our cleanuppad's unwind destination is another cleanuppad. | |||
| 4520 | auto *SuccessorCleanupPad = dyn_cast<CleanupPadInst>(&UnwindDest->front()); | |||
| 4521 | if (!SuccessorCleanupPad) | |||
| 4522 | return false; | |||
| 4523 | ||||
| 4524 | CleanupPadInst *PredecessorCleanupPad = RI->getCleanupPad(); | |||
| 4525 | // Replace any uses of the successor cleanupad with the predecessor pad | |||
| 4526 | // The only cleanuppad uses should be this cleanupret, it's cleanupret and | |||
| 4527 | // funclet bundle operands. | |||
| 4528 | SuccessorCleanupPad->replaceAllUsesWith(PredecessorCleanupPad); | |||
| 4529 | // Remove the old cleanuppad. | |||
| 4530 | SuccessorCleanupPad->eraseFromParent(); | |||
| 4531 | // Now, we simply replace the cleanupret with a branch to the unwind | |||
| 4532 | // destination. | |||
| 4533 | BranchInst::Create(UnwindDest, RI->getParent()); | |||
| 4534 | RI->eraseFromParent(); | |||
| 4535 | ||||
| 4536 | return true; | |||
| 4537 | } | |||
| 4538 | ||||
| 4539 | bool SimplifyCFGOpt::simplifyCleanupReturn(CleanupReturnInst *RI) { | |||
| 4540 | // It is possible to transiantly have an undef cleanuppad operand because we | |||
| 4541 | // have deleted some, but not all, dead blocks. | |||
| 4542 | // Eventually, this block will be deleted. | |||
| 4543 | if (isa<UndefValue>(RI->getOperand(0))) | |||
| 4544 | return false; | |||
| 4545 | ||||
| 4546 | if (mergeCleanupPad(RI)) | |||
| 4547 | return true; | |||
| 4548 | ||||
| 4549 | if (removeEmptyCleanup(RI, DTU)) | |||
| 4550 | return true; | |||
| 4551 | ||||
| 4552 | return false; | |||
| 4553 | } | |||
| 4554 | ||||
| 4555 | // WARNING: keep in sync with InstCombinerImpl::visitUnreachableInst()! | |||
| 4556 | bool SimplifyCFGOpt::simplifyUnreachable(UnreachableInst *UI) { | |||
| 4557 | BasicBlock *BB = UI->getParent(); | |||
| 4558 | ||||
| 4559 | bool Changed = false; | |||
| 4560 | ||||
| 4561 | // If there are any instructions immediately before the unreachable that can | |||
| 4562 | // be removed, do so. | |||
| 4563 | while (UI->getIterator() != BB->begin()) { | |||
| 4564 | BasicBlock::iterator BBI = UI->getIterator(); | |||
| 4565 | --BBI; | |||
| 4566 | ||||
| 4567 | if (!isGuaranteedToTransferExecutionToSuccessor(&*BBI)) | |||
| 4568 | break; // Can not drop any more instructions. We're done here. | |||
| 4569 | // Otherwise, this instruction can be freely erased, | |||
| 4570 | // even if it is not side-effect free. | |||
| 4571 | ||||
| 4572 | // Note that deleting EH's here is in fact okay, although it involves a bit | |||
| 4573 | // of subtle reasoning. If this inst is an EH, all the predecessors of this | |||
| 4574 | // block will be the unwind edges of Invoke/CatchSwitch/CleanupReturn, | |||
| 4575 | // and we can therefore guarantee this block will be erased. | |||
| 4576 | ||||
| 4577 | // Delete this instruction (any uses are guaranteed to be dead) | |||
| 4578 | BBI->replaceAllUsesWith(PoisonValue::get(BBI->getType())); | |||
| 4579 | BBI->eraseFromParent(); | |||
| 4580 | Changed = true; | |||
| 4581 | } | |||
| 4582 | ||||
| 4583 | // If the unreachable instruction is the first in the block, take a gander | |||
| 4584 | // at all of the predecessors of this instruction, and simplify them. | |||
| 4585 | if (&BB->front() != UI) | |||
| 4586 | return Changed; | |||
| 4587 | ||||
| 4588 | std::vector<DominatorTree::UpdateType> Updates; | |||
| 4589 | ||||
| 4590 | SmallSetVector<BasicBlock *, 8> Preds(pred_begin(BB), pred_end(BB)); | |||
| 4591 | for (unsigned i = 0, e = Preds.size(); i != e; ++i) { | |||
| 4592 | auto *Predecessor = Preds[i]; | |||
| 4593 | Instruction *TI = Predecessor->getTerminator(); | |||
| 4594 | IRBuilder<> Builder(TI); | |||
| 4595 | if (auto *BI = dyn_cast<BranchInst>(TI)) { | |||
| 4596 | // We could either have a proper unconditional branch, | |||
| 4597 | // or a degenerate conditional branch with matching destinations. | |||
| 4598 | if (all_of(BI->successors(), | |||
| 4599 | [BB](auto *Successor) { return Successor == BB; })) { | |||
| 4600 | new UnreachableInst(TI->getContext(), TI); | |||
| 4601 | TI->eraseFromParent(); | |||
| 4602 | Changed = true; | |||
| 4603 | } else { | |||
| 4604 | assert(BI->isConditional() && "Can't get here with an uncond branch.")((void)0); | |||
| 4605 | Value* Cond = BI->getCondition(); | |||
| 4606 | assert(BI->getSuccessor(0) != BI->getSuccessor(1) &&((void)0) | |||
| 4607 | "The destinations are guaranteed to be different here.")((void)0); | |||
| 4608 | if (BI->getSuccessor(0) == BB) { | |||
| 4609 | Builder.CreateAssumption(Builder.CreateNot(Cond)); | |||
| 4610 | Builder.CreateBr(BI->getSuccessor(1)); | |||
| 4611 | } else { | |||
| 4612 | assert(BI->getSuccessor(1) == BB && "Incorrect CFG")((void)0); | |||
| 4613 | Builder.CreateAssumption(Cond); | |||
| 4614 | Builder.CreateBr(BI->getSuccessor(0)); | |||
| 4615 | } | |||
| 4616 | EraseTerminatorAndDCECond(BI); | |||
| 4617 | Changed = true; | |||
| 4618 | } | |||
| 4619 | if (DTU) | |||
| 4620 | Updates.push_back({DominatorTree::Delete, Predecessor, BB}); | |||
| 4621 | } else if (auto *SI = dyn_cast<SwitchInst>(TI)) { | |||
| 4622 | SwitchInstProfUpdateWrapper SU(*SI); | |||
| 4623 | for (auto i = SU->case_begin(), e = SU->case_end(); i != e;) { | |||
| 4624 | if (i->getCaseSuccessor() != BB) { | |||
| 4625 | ++i; | |||
| 4626 | continue; | |||
| 4627 | } | |||
| 4628 | BB->removePredecessor(SU->getParent()); | |||
| 4629 | i = SU.removeCase(i); | |||
| 4630 | e = SU->case_end(); | |||
| 4631 | Changed = true; | |||
| 4632 | } | |||
| 4633 | // Note that the default destination can't be removed! | |||
| 4634 | if (DTU && SI->getDefaultDest() != BB) | |||
| 4635 | Updates.push_back({DominatorTree::Delete, Predecessor, BB}); | |||
| 4636 | } else if (auto *II = dyn_cast<InvokeInst>(TI)) { | |||
| 4637 | if (II->getUnwindDest() == BB) { | |||
| 4638 | if (DTU) { | |||
| 4639 | DTU->applyUpdates(Updates); | |||
| 4640 | Updates.clear(); | |||
| 4641 | } | |||
| 4642 | removeUnwindEdge(TI->getParent(), DTU); | |||
| 4643 | Changed = true; | |||
| 4644 | } | |||
| 4645 | } else if (auto *CSI = dyn_cast<CatchSwitchInst>(TI)) { | |||
| 4646 | if (CSI->getUnwindDest() == BB) { | |||
| 4647 | if (DTU) { | |||
| 4648 | DTU->applyUpdates(Updates); | |||
| 4649 | Updates.clear(); | |||
| 4650 | } | |||
| 4651 | removeUnwindEdge(TI->getParent(), DTU); | |||
| 4652 | Changed = true; | |||
| 4653 | continue; | |||
| 4654 | } | |||
| 4655 | ||||
| 4656 | for (CatchSwitchInst::handler_iterator I = CSI->handler_begin(), | |||
| 4657 | E = CSI->handler_end(); | |||
| 4658 | I != E; ++I) { | |||
| 4659 | if (*I == BB) { | |||
| 4660 | CSI->removeHandler(I); | |||
| 4661 | --I; | |||
| 4662 | --E; | |||
| 4663 | Changed = true; | |||
| 4664 | } | |||
| 4665 | } | |||
| 4666 | if (DTU) | |||
| 4667 | Updates.push_back({DominatorTree::Delete, Predecessor, BB}); | |||
| 4668 | if (CSI->getNumHandlers() == 0) { | |||
| 4669 | if (CSI->hasUnwindDest()) { | |||
| 4670 | // Redirect all predecessors of the block containing CatchSwitchInst | |||
| 4671 | // to instead branch to the CatchSwitchInst's unwind destination. | |||
| 4672 | if (DTU) { | |||
| 4673 | for (auto *PredecessorOfPredecessor : predecessors(Predecessor)) { | |||
| 4674 | Updates.push_back({DominatorTree::Insert, | |||
| 4675 | PredecessorOfPredecessor, | |||
| 4676 | CSI->getUnwindDest()}); | |||
| 4677 | Updates.push_back({DominatorTree::Delete, | |||
| 4678 | PredecessorOfPredecessor, Predecessor}); | |||
| 4679 | } | |||
| 4680 | } | |||
| 4681 | Predecessor->replaceAllUsesWith(CSI->getUnwindDest()); | |||
| 4682 | } else { | |||
| 4683 | // Rewrite all preds to unwind to caller (or from invoke to call). | |||
| 4684 | if (DTU) { | |||
| 4685 | DTU->applyUpdates(Updates); | |||
| 4686 | Updates.clear(); | |||
| 4687 | } | |||
| 4688 | SmallVector<BasicBlock *, 8> EHPreds(predecessors(Predecessor)); | |||
| 4689 | for (BasicBlock *EHPred : EHPreds) | |||
| 4690 | removeUnwindEdge(EHPred, DTU); | |||
| 4691 | } | |||
| 4692 | // The catchswitch is no longer reachable. | |||
| 4693 | new UnreachableInst(CSI->getContext(), CSI); | |||
| 4694 | CSI->eraseFromParent(); | |||
| 4695 | Changed = true; | |||
| 4696 | } | |||
| 4697 | } else if (auto *CRI = dyn_cast<CleanupReturnInst>(TI)) { | |||
| 4698 | (void)CRI; | |||
| 4699 | assert(CRI->hasUnwindDest() && CRI->getUnwindDest() == BB &&((void)0) | |||
| 4700 | "Expected to always have an unwind to BB.")((void)0); | |||
| 4701 | if (DTU) | |||
| 4702 | Updates.push_back({DominatorTree::Delete, Predecessor, BB}); | |||
| 4703 | new UnreachableInst(TI->getContext(), TI); | |||
| 4704 | TI->eraseFromParent(); | |||
| 4705 | Changed = true; | |||
| 4706 | } | |||
| 4707 | } | |||
| 4708 | ||||
| 4709 | if (DTU) | |||
| 4710 | DTU->applyUpdates(Updates); | |||
| 4711 | ||||
| 4712 | // If this block is now dead, remove it. | |||
| 4713 | if (pred_empty(BB) && BB != &BB->getParent()->getEntryBlock()) { | |||
| 4714 | DeleteDeadBlock(BB, DTU); | |||
| 4715 | return true; | |||
| 4716 | } | |||
| 4717 | ||||
| 4718 | return Changed; | |||
| 4719 | } | |||
| 4720 | ||||
| 4721 | static bool CasesAreContiguous(SmallVectorImpl<ConstantInt *> &Cases) { | |||
| 4722 | assert(Cases.size() >= 1)((void)0); | |||
| 4723 | ||||
| 4724 | array_pod_sort(Cases.begin(), Cases.end(), ConstantIntSortPredicate); | |||
| 4725 | for (size_t I = 1, E = Cases.size(); I != E; ++I) { | |||
| 4726 | if (Cases[I - 1]->getValue() != Cases[I]->getValue() + 1) | |||
| 4727 | return false; | |||
| 4728 | } | |||
| 4729 | return true; | |||
| 4730 | } | |||
| 4731 | ||||
| 4732 | static void createUnreachableSwitchDefault(SwitchInst *Switch, | |||
| 4733 | DomTreeUpdater *DTU) { | |||
| 4734 | LLVM_DEBUG(dbgs() << "SimplifyCFG: switch default is dead.\n")do { } while (false); | |||
| 4735 | auto *BB = Switch->getParent(); | |||
| 4736 | BasicBlock *NewDefaultBlock = SplitBlockPredecessors( | |||
| 4737 | Switch->getDefaultDest(), Switch->getParent(), "", DTU); | |||
| 4738 | auto *OrigDefaultBlock = Switch->getDefaultDest(); | |||
| 4739 | Switch->setDefaultDest(&*NewDefaultBlock); | |||
| 4740 | if (DTU) | |||
| 4741 | DTU->applyUpdates({{DominatorTree::Insert, BB, &*NewDefaultBlock}, | |||
| 4742 | {DominatorTree::Delete, BB, OrigDefaultBlock}}); | |||
| 4743 | SplitBlock(&*NewDefaultBlock, &NewDefaultBlock->front(), DTU); | |||
| 4744 | SmallVector<DominatorTree::UpdateType, 2> Updates; | |||
| 4745 | if (DTU) | |||
| 4746 | for (auto *Successor : successors(NewDefaultBlock)) | |||
| 4747 | Updates.push_back({DominatorTree::Delete, NewDefaultBlock, Successor}); | |||
| 4748 | auto *NewTerminator = NewDefaultBlock->getTerminator(); | |||
| 4749 | new UnreachableInst(Switch->getContext(), NewTerminator); | |||
| 4750 | EraseTerminatorAndDCECond(NewTerminator); | |||
| 4751 | if (DTU) | |||
| 4752 | DTU->applyUpdates(Updates); | |||
| 4753 | } | |||
| 4754 | ||||
| 4755 | /// Turn a switch with two reachable destinations into an integer range | |||
| 4756 | /// comparison and branch. | |||
| 4757 | bool SimplifyCFGOpt::TurnSwitchRangeIntoICmp(SwitchInst *SI, | |||
| 4758 | IRBuilder<> &Builder) { | |||
| 4759 | assert(SI->getNumCases() > 1 && "Degenerate switch?")((void)0); | |||
| 4760 | ||||
| 4761 | bool HasDefault = | |||
| 4762 | !isa<UnreachableInst>(SI->getDefaultDest()->getFirstNonPHIOrDbg()); | |||
| 4763 | ||||
| 4764 | auto *BB = SI->getParent(); | |||
| 4765 | ||||
| 4766 | // Partition the cases into two sets with different destinations. | |||
| 4767 | BasicBlock *DestA = HasDefault ? SI->getDefaultDest() : nullptr; | |||
| 4768 | BasicBlock *DestB = nullptr; | |||
| 4769 | SmallVector<ConstantInt *, 16> CasesA; | |||
| 4770 | SmallVector<ConstantInt *, 16> CasesB; | |||
| 4771 | ||||
| 4772 | for (auto Case : SI->cases()) { | |||
| 4773 | BasicBlock *Dest = Case.getCaseSuccessor(); | |||
| 4774 | if (!DestA) | |||
| 4775 | DestA = Dest; | |||
| 4776 | if (Dest == DestA) { | |||
| 4777 | CasesA.push_back(Case.getCaseValue()); | |||
| 4778 | continue; | |||
| 4779 | } | |||
| 4780 | if (!DestB) | |||
| 4781 | DestB = Dest; | |||
| 4782 | if (Dest == DestB) { | |||
| 4783 | CasesB.push_back(Case.getCaseValue()); | |||
| 4784 | continue; | |||
| 4785 | } | |||
| 4786 | return false; // More than two destinations. | |||
| 4787 | } | |||
| 4788 | ||||
| 4789 | assert(DestA && DestB &&((void)0) | |||
| 4790 | "Single-destination switch should have been folded.")((void)0); | |||
| 4791 | assert(DestA != DestB)((void)0); | |||
| 4792 | assert(DestB != SI->getDefaultDest())((void)0); | |||
| 4793 | assert(!CasesB.empty() && "There must be non-default cases.")((void)0); | |||
| 4794 | assert(!CasesA.empty() || HasDefault)((void)0); | |||
| 4795 | ||||
| 4796 | // Figure out if one of the sets of cases form a contiguous range. | |||
| 4797 | SmallVectorImpl<ConstantInt *> *ContiguousCases = nullptr; | |||
| 4798 | BasicBlock *ContiguousDest = nullptr; | |||
| 4799 | BasicBlock *OtherDest = nullptr; | |||
| 4800 | if (!CasesA.empty() && CasesAreContiguous(CasesA)) { | |||
| 4801 | ContiguousCases = &CasesA; | |||
| 4802 | ContiguousDest = DestA; | |||
| 4803 | OtherDest = DestB; | |||
| 4804 | } else if (CasesAreContiguous(CasesB)) { | |||
| 4805 | ContiguousCases = &CasesB; | |||
| 4806 | ContiguousDest = DestB; | |||
| 4807 | OtherDest = DestA; | |||
| 4808 | } else | |||
| 4809 | return false; | |||
| 4810 | ||||
| 4811 | // Start building the compare and branch. | |||
| 4812 | ||||
| 4813 | Constant *Offset = ConstantExpr::getNeg(ContiguousCases->back()); | |||
| 4814 | Constant *NumCases = | |||
| 4815 | ConstantInt::get(Offset->getType(), ContiguousCases->size()); | |||
| 4816 | ||||
| 4817 | Value *Sub = SI->getCondition(); | |||
| 4818 | if (!Offset->isNullValue()) | |||
| 4819 | Sub = Builder.CreateAdd(Sub, Offset, Sub->getName() + ".off"); | |||
| 4820 | ||||
| 4821 | Value *Cmp; | |||
| 4822 | // If NumCases overflowed, then all possible values jump to the successor. | |||
| 4823 | if (NumCases->isNullValue() && !ContiguousCases->empty()) | |||
| 4824 | Cmp = ConstantInt::getTrue(SI->getContext()); | |||
| 4825 | else | |||
| 4826 | Cmp = Builder.CreateICmpULT(Sub, NumCases, "switch"); | |||
| 4827 | BranchInst *NewBI = Builder.CreateCondBr(Cmp, ContiguousDest, OtherDest); | |||
| 4828 | ||||
| 4829 | // Update weight for the newly-created conditional branch. | |||
| 4830 | if (HasBranchWeights(SI)) { | |||
| 4831 | SmallVector<uint64_t, 8> Weights; | |||
| 4832 | GetBranchWeights(SI, Weights); | |||
| 4833 | if (Weights.size() == 1 + SI->getNumCases()) { | |||
| 4834 | uint64_t TrueWeight = 0; | |||
| 4835 | uint64_t FalseWeight = 0; | |||
| 4836 | for (size_t I = 0, E = Weights.size(); I != E; ++I) { | |||
| 4837 | if (SI->getSuccessor(I) == ContiguousDest) | |||
| 4838 | TrueWeight += Weights[I]; | |||
| 4839 | else | |||
| 4840 | FalseWeight += Weights[I]; | |||
| 4841 | } | |||
| 4842 | while (TrueWeight > UINT32_MAX0xffffffffU || FalseWeight > UINT32_MAX0xffffffffU) { | |||
| 4843 | TrueWeight /= 2; | |||
| 4844 | FalseWeight /= 2; | |||
| 4845 | } | |||
| 4846 | setBranchWeights(NewBI, TrueWeight, FalseWeight); | |||
| 4847 | } | |||
| 4848 | } | |||
| 4849 | ||||
| 4850 | // Prune obsolete incoming values off the successors' PHI nodes. | |||
| 4851 | for (auto BBI = ContiguousDest->begin(); isa<PHINode>(BBI); ++BBI) { | |||
| 4852 | unsigned PreviousEdges = ContiguousCases->size(); | |||
| 4853 | if (ContiguousDest == SI->getDefaultDest()) | |||
| 4854 | ++PreviousEdges; | |||
| 4855 | for (unsigned I = 0, E = PreviousEdges - 1; I != E; ++I) | |||
| 4856 | cast<PHINode>(BBI)->removeIncomingValue(SI->getParent()); | |||
| 4857 | } | |||
| 4858 | for (auto BBI = OtherDest->begin(); isa<PHINode>(BBI); ++BBI) { | |||
| 4859 | unsigned PreviousEdges = SI->getNumCases() - ContiguousCases->size(); | |||
| 4860 | if (OtherDest == SI->getDefaultDest()) | |||
| 4861 | ++PreviousEdges; | |||
| 4862 | for (unsigned I = 0, E = PreviousEdges - 1; I != E; ++I) | |||
| 4863 | cast<PHINode>(BBI)->removeIncomingValue(SI->getParent()); | |||
| 4864 | } | |||
| 4865 | ||||
| 4866 | // Clean up the default block - it may have phis or other instructions before | |||
| 4867 | // the unreachable terminator. | |||
| 4868 | if (!HasDefault) | |||
| 4869 | createUnreachableSwitchDefault(SI, DTU); | |||
| 4870 | ||||
| 4871 | auto *UnreachableDefault = SI->getDefaultDest(); | |||
| 4872 | ||||
| 4873 | // Drop the switch. | |||
| 4874 | SI->eraseFromParent(); | |||
| 4875 | ||||
| 4876 | if (!HasDefault && DTU) | |||
| 4877 | DTU->applyUpdates({{DominatorTree::Delete, BB, UnreachableDefault}}); | |||
| 4878 | ||||
| 4879 | return true; | |||
| 4880 | } | |||
| 4881 | ||||
| 4882 | /// Compute masked bits for the condition of a switch | |||
| 4883 | /// and use it to remove dead cases. | |||
| 4884 | static bool eliminateDeadSwitchCases(SwitchInst *SI, DomTreeUpdater *DTU, | |||
| 4885 | AssumptionCache *AC, | |||
| 4886 | const DataLayout &DL) { | |||
| 4887 | Value *Cond = SI->getCondition(); | |||
| 4888 | unsigned Bits = Cond->getType()->getIntegerBitWidth(); | |||
| 4889 | KnownBits Known = computeKnownBits(Cond, DL, 0, AC, SI); | |||
| 4890 | ||||
| 4891 | // We can also eliminate cases by determining that their values are outside of | |||
| 4892 | // the limited range of the condition based on how many significant (non-sign) | |||
| 4893 | // bits are in the condition value. | |||
| 4894 | unsigned ExtraSignBits = ComputeNumSignBits(Cond, DL, 0, AC, SI) - 1; | |||
| 4895 | unsigned MaxSignificantBitsInCond = Bits - ExtraSignBits; | |||
| 4896 | ||||
| 4897 | // Gather dead cases. | |||
| 4898 | SmallVector<ConstantInt *, 8> DeadCases; | |||
| 4899 | SmallDenseMap<BasicBlock *, int, 8> NumPerSuccessorCases; | |||
| 4900 | for (auto &Case : SI->cases()) { | |||
| 4901 | auto *Successor = Case.getCaseSuccessor(); | |||
| 4902 | if (DTU) | |||
| 4903 | ++NumPerSuccessorCases[Successor]; | |||
| 4904 | const APInt &CaseVal = Case.getCaseValue()->getValue(); | |||
| 4905 | if (Known.Zero.intersects(CaseVal) || !Known.One.isSubsetOf(CaseVal) || | |||
| 4906 | (CaseVal.getMinSignedBits() > MaxSignificantBitsInCond)) { | |||
| 4907 | DeadCases.push_back(Case.getCaseValue()); | |||
| 4908 | if (DTU) | |||
| 4909 | --NumPerSuccessorCases[Successor]; | |||
| 4910 | LLVM_DEBUG(dbgs() << "SimplifyCFG: switch case " << CaseValdo { } while (false) | |||
| 4911 | << " is dead.\n")do { } while (false); | |||
| 4912 | } | |||
| 4913 | } | |||
| 4914 | ||||
| 4915 | // If we can prove that the cases must cover all possible values, the | |||
| 4916 | // default destination becomes dead and we can remove it. If we know some | |||
| 4917 | // of the bits in the value, we can use that to more precisely compute the | |||
| 4918 | // number of possible unique case values. | |||
| 4919 | bool HasDefault = | |||
| 4920 | !isa<UnreachableInst>(SI->getDefaultDest()->getFirstNonPHIOrDbg()); | |||
| 4921 | const unsigned NumUnknownBits = | |||
| 4922 | Bits - (Known.Zero | Known.One).countPopulation(); | |||
| 4923 | assert(NumUnknownBits <= Bits)((void)0); | |||
| 4924 | if (HasDefault && DeadCases.empty() && | |||
| 4925 | NumUnknownBits < 64 /* avoid overflow */ && | |||
| 4926 | SI->getNumCases() == (1ULL << NumUnknownBits)) { | |||
| 4927 | createUnreachableSwitchDefault(SI, DTU); | |||
| 4928 | return true; | |||
| 4929 | } | |||
| 4930 | ||||
| 4931 | if (DeadCases.empty()) | |||
| 4932 | return false; | |||
| 4933 | ||||
| 4934 | SwitchInstProfUpdateWrapper SIW(*SI); | |||
| 4935 | for (ConstantInt *DeadCase : DeadCases) { | |||
| 4936 | SwitchInst::CaseIt CaseI = SI->findCaseValue(DeadCase); | |||
| 4937 | assert(CaseI != SI->case_default() &&((void)0) | |||
| 4938 | "Case was not found. Probably mistake in DeadCases forming.")((void)0); | |||
| 4939 | // Prune unused values from PHI nodes. | |||
| 4940 | CaseI->getCaseSuccessor()->removePredecessor(SI->getParent()); | |||
| 4941 | SIW.removeCase(CaseI); | |||
| 4942 | } | |||
| 4943 | ||||
| 4944 | if (DTU) { | |||
| 4945 | std::vector<DominatorTree::UpdateType> Updates; | |||
| 4946 | for (const std::pair<BasicBlock *, int> &I : NumPerSuccessorCases) | |||
| 4947 | if (I.second == 0) | |||
| 4948 | Updates.push_back({DominatorTree::Delete, SI->getParent(), I.first}); | |||
| 4949 | DTU->applyUpdates(Updates); | |||
| 4950 | } | |||
| 4951 | ||||
| 4952 | return true; | |||
| 4953 | } | |||
| 4954 | ||||
| 4955 | /// If BB would be eligible for simplification by | |||
| 4956 | /// TryToSimplifyUncondBranchFromEmptyBlock (i.e. it is empty and terminated | |||
| 4957 | /// by an unconditional branch), look at the phi node for BB in the successor | |||
| 4958 | /// block and see if the incoming value is equal to CaseValue. If so, return | |||
| 4959 | /// the phi node, and set PhiIndex to BB's index in the phi node. | |||
| 4960 | static PHINode *FindPHIForConditionForwarding(ConstantInt *CaseValue, | |||
| 4961 | BasicBlock *BB, int *PhiIndex) { | |||
| 4962 | if (BB->getFirstNonPHIOrDbg() != BB->getTerminator()) | |||
| 4963 | return nullptr; // BB must be empty to be a candidate for simplification. | |||
| 4964 | if (!BB->getSinglePredecessor()) | |||
| 4965 | return nullptr; // BB must be dominated by the switch. | |||
| 4966 | ||||
| 4967 | BranchInst *Branch = dyn_cast<BranchInst>(BB->getTerminator()); | |||
| 4968 | if (!Branch || !Branch->isUnconditional()) | |||
| 4969 | return nullptr; // Terminator must be unconditional branch. | |||
| 4970 | ||||
| 4971 | BasicBlock *Succ = Branch->getSuccessor(0); | |||
| 4972 | ||||
| 4973 | for (PHINode &PHI : Succ->phis()) { | |||
| 4974 | int Idx = PHI.getBasicBlockIndex(BB); | |||
| 4975 | assert(Idx >= 0 && "PHI has no entry for predecessor?")((void)0); | |||
| 4976 | ||||
| 4977 | Value *InValue = PHI.getIncomingValue(Idx); | |||
| 4978 | if (InValue != CaseValue) | |||
| 4979 | continue; | |||
| 4980 | ||||
| 4981 | *PhiIndex = Idx; | |||
| 4982 | return &PHI; | |||
| 4983 | } | |||
| 4984 | ||||
| 4985 | return nullptr; | |||
| 4986 | } | |||
| 4987 | ||||
| 4988 | /// Try to forward the condition of a switch instruction to a phi node | |||
| 4989 | /// dominated by the switch, if that would mean that some of the destination | |||
| 4990 | /// blocks of the switch can be folded away. Return true if a change is made. | |||
| 4991 | static bool ForwardSwitchConditionToPHI(SwitchInst *SI) { | |||
| 4992 | using ForwardingNodesMap = DenseMap<PHINode *, SmallVector<int, 4>>; | |||
| 4993 | ||||
| 4994 | ForwardingNodesMap ForwardingNodes; | |||
| 4995 | BasicBlock *SwitchBlock = SI->getParent(); | |||
| 4996 | bool Changed = false; | |||
| 4997 | for (auto &Case : SI->cases()) { | |||
| 4998 | ConstantInt *CaseValue = Case.getCaseValue(); | |||
| 4999 | BasicBlock *CaseDest = Case.getCaseSuccessor(); | |||
| 5000 | ||||
| 5001 | // Replace phi operands in successor blocks that are using the constant case | |||
| 5002 | // value rather than the switch condition variable: | |||
| 5003 | // switchbb: | |||
| 5004 | // switch i32 %x, label %default [ | |||
| 5005 | // i32 17, label %succ | |||
| 5006 | // ... | |||
| 5007 | // succ: | |||
| 5008 | // %r = phi i32 ... [ 17, %switchbb ] ... | |||
| 5009 | // --> | |||
| 5010 | // %r = phi i32 ... [ %x, %switchbb ] ... | |||
| 5011 | ||||
| 5012 | for (PHINode &Phi : CaseDest->phis()) { | |||
| 5013 | // This only works if there is exactly 1 incoming edge from the switch to | |||
| 5014 | // a phi. If there is >1, that means multiple cases of the switch map to 1 | |||
| 5015 | // value in the phi, and that phi value is not the switch condition. Thus, | |||
| 5016 | // this transform would not make sense (the phi would be invalid because | |||
| 5017 | // a phi can't have different incoming values from the same block). | |||
| 5018 | int SwitchBBIdx = Phi.getBasicBlockIndex(SwitchBlock); | |||
| 5019 | if (Phi.getIncomingValue(SwitchBBIdx) == CaseValue && | |||
| 5020 | count(Phi.blocks(), SwitchBlock) == 1) { | |||
| 5021 | Phi.setIncomingValue(SwitchBBIdx, SI->getCondition()); | |||
| 5022 | Changed = true; | |||
| 5023 | } | |||
| 5024 | } | |||
| 5025 | ||||
| 5026 | // Collect phi nodes that are indirectly using this switch's case constants. | |||
| 5027 | int PhiIdx; | |||
| 5028 | if (auto *Phi = FindPHIForConditionForwarding(CaseValue, CaseDest, &PhiIdx)) | |||
| 5029 | ForwardingNodes[Phi].push_back(PhiIdx); | |||
| 5030 | } | |||
| 5031 | ||||
| 5032 | for (auto &ForwardingNode : ForwardingNodes) { | |||
| 5033 | PHINode *Phi = ForwardingNode.first; | |||
| 5034 | SmallVectorImpl<int> &Indexes = ForwardingNode.second; | |||
| 5035 | if (Indexes.size() < 2) | |||
| 5036 | continue; | |||
| 5037 | ||||
| 5038 | for (int Index : Indexes) | |||
| 5039 | Phi->setIncomingValue(Index, SI->getCondition()); | |||
| 5040 | Changed = true; | |||
| 5041 | } | |||
| 5042 | ||||
| 5043 | return Changed; | |||
| 5044 | } | |||
| 5045 | ||||
| 5046 | /// Return true if the backend will be able to handle | |||
| 5047 | /// initializing an array of constants like C. | |||
| 5048 | static bool ValidLookupTableConstant(Constant *C, const TargetTransformInfo &TTI) { | |||
| 5049 | if (C->isThreadDependent()) | |||
| 5050 | return false; | |||
| 5051 | if (C->isDLLImportDependent()) | |||
| 5052 | return false; | |||
| 5053 | ||||
| 5054 | if (!isa<ConstantFP>(C) && !isa<ConstantInt>(C) && | |||
| 5055 | !isa<ConstantPointerNull>(C) && !isa<GlobalValue>(C) && | |||
| 5056 | !isa<UndefValue>(C) && !isa<ConstantExpr>(C)) | |||
| 5057 | return false; | |||
| 5058 | ||||
| 5059 | if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C)) { | |||
| 5060 | if (!CE->isGEPWithNoNotionalOverIndexing()) | |||
| 5061 | return false; | |||
| 5062 | if (!ValidLookupTableConstant(CE->getOperand(0), TTI)) | |||
| 5063 | return false; | |||
| 5064 | } | |||
| 5065 | ||||
| 5066 | if (!TTI.shouldBuildLookupTablesForConstant(C)) | |||
| 5067 | return false; | |||
| 5068 | ||||
| 5069 | return true; | |||
| 5070 | } | |||
| 5071 | ||||
| 5072 | /// If V is a Constant, return it. Otherwise, try to look up | |||
| 5073 | /// its constant value in ConstantPool, returning 0 if it's not there. | |||
| 5074 | static Constant * | |||
| 5075 | LookupConstant(Value *V, | |||
| 5076 | const SmallDenseMap<Value *, Constant *> &ConstantPool) { | |||
| 5077 | if (Constant *C = dyn_cast<Constant>(V)) | |||
| 5078 | return C; | |||
| 5079 | return ConstantPool.lookup(V); | |||
| 5080 | } | |||
| 5081 | ||||
| 5082 | /// Try to fold instruction I into a constant. This works for | |||
| 5083 | /// simple instructions such as binary operations where both operands are | |||
| 5084 | /// constant or can be replaced by constants from the ConstantPool. Returns the | |||
| 5085 | /// resulting constant on success, 0 otherwise. | |||
| 5086 | static Constant * | |||
| 5087 | ConstantFold(Instruction *I, const DataLayout &DL, | |||
| 5088 | const SmallDenseMap<Value *, Constant *> &ConstantPool) { | |||
| 5089 | if (SelectInst *Select = dyn_cast<SelectInst>(I)) { | |||
| 5090 | Constant *A = LookupConstant(Select->getCondition(), ConstantPool); | |||
| 5091 | if (!A) | |||
| 5092 | return nullptr; | |||
| 5093 | if (A->isAllOnesValue()) | |||
| 5094 | return LookupConstant(Select->getTrueValue(), ConstantPool); | |||
| 5095 | if (A->isNullValue()) | |||
| 5096 | return LookupConstant(Select->getFalseValue(), ConstantPool); | |||
| 5097 | return nullptr; | |||
| 5098 | } | |||
| 5099 | ||||
| 5100 | SmallVector<Constant *, 4> COps; | |||
| 5101 | for (unsigned N = 0, E = I->getNumOperands(); N != E; ++N) { | |||
| 5102 | if (Constant *A = LookupConstant(I->getOperand(N), ConstantPool)) | |||
| 5103 | COps.push_back(A); | |||
| 5104 | else | |||
| 5105 | return nullptr; | |||
| 5106 | } | |||
| 5107 | ||||
| 5108 | if (CmpInst *Cmp = dyn_cast<CmpInst>(I)) { | |||
| 5109 | return ConstantFoldCompareInstOperands(Cmp->getPredicate(), COps[0], | |||
| 5110 | COps[1], DL); | |||
| 5111 | } | |||
| 5112 | ||||
| 5113 | return ConstantFoldInstOperands(I, COps, DL); | |||
| 5114 | } | |||
| 5115 | ||||
| 5116 | /// Try to determine the resulting constant values in phi nodes | |||
| 5117 | /// at the common destination basic block, *CommonDest, for one of the case | |||
| 5118 | /// destionations CaseDest corresponding to value CaseVal (0 for the default | |||
| 5119 | /// case), of a switch instruction SI. | |||
| 5120 | static bool | |||
| 5121 | GetCaseResults(SwitchInst *SI, ConstantInt *CaseVal, BasicBlock *CaseDest, | |||
| 5122 | BasicBlock **CommonDest, | |||
| 5123 | SmallVectorImpl<std::pair<PHINode *, Constant *>> &Res, | |||
| 5124 | const DataLayout &DL, const TargetTransformInfo &TTI) { | |||
| 5125 | // The block from which we enter the common destination. | |||
| 5126 | BasicBlock *Pred = SI->getParent(); | |||
| 5127 | ||||
| 5128 | // If CaseDest is empty except for some side-effect free instructions through | |||
| 5129 | // which we can constant-propagate the CaseVal, continue to its successor. | |||
| 5130 | SmallDenseMap<Value *, Constant *> ConstantPool; | |||
| 5131 | ConstantPool.insert(std::make_pair(SI->getCondition(), CaseVal)); | |||
| 5132 | for (Instruction &I :CaseDest->instructionsWithoutDebug()) { | |||
| 5133 | if (I.isTerminator()) { | |||
| 5134 | // If the terminator is a simple branch, continue to the next block. | |||
| 5135 | if (I.getNumSuccessors() != 1 || I.isExceptionalTerminator()) | |||
| 5136 | return false; | |||
| 5137 | Pred = CaseDest; | |||
| 5138 | CaseDest = I.getSuccessor(0); | |||
| 5139 | } else if (Constant *C = ConstantFold(&I, DL, ConstantPool)) { | |||
| 5140 | // Instruction is side-effect free and constant. | |||
| 5141 | ||||
| 5142 | // If the instruction has uses outside this block or a phi node slot for | |||
| 5143 | // the block, it is not safe to bypass the instruction since it would then | |||
| 5144 | // no longer dominate all its uses. | |||
| 5145 | for (auto &Use : I.uses()) { | |||
| 5146 | User *User = Use.getUser(); | |||
| 5147 | if (Instruction *I = dyn_cast<Instruction>(User)) | |||
| 5148 | if (I->getParent() == CaseDest) | |||
| 5149 | continue; | |||
| 5150 | if (PHINode *Phi = dyn_cast<PHINode>(User)) | |||
| 5151 | if (Phi->getIncomingBlock(Use) == CaseDest) | |||
| 5152 | continue; | |||
| 5153 | return false; | |||
| 5154 | } | |||
| 5155 | ||||
| 5156 | ConstantPool.insert(std::make_pair(&I, C)); | |||
| 5157 | } else { | |||
| 5158 | break; | |||
| 5159 | } | |||
| 5160 | } | |||
| 5161 | ||||
| 5162 | // If we did not have a CommonDest before, use the current one. | |||
| 5163 | if (!*CommonDest) | |||
| 5164 | *CommonDest = CaseDest; | |||
| 5165 | // If the destination isn't the common one, abort. | |||
| 5166 | if (CaseDest != *CommonDest) | |||
| 5167 | return false; | |||
| 5168 | ||||
| 5169 | // Get the values for this case from phi nodes in the destination block. | |||
| 5170 | for (PHINode &PHI : (*CommonDest)->phis()) { | |||
| 5171 | int Idx = PHI.getBasicBlockIndex(Pred); | |||
| 5172 | if (Idx == -1) | |||
| 5173 | continue; | |||
| 5174 | ||||
| 5175 | Constant *ConstVal = | |||
| 5176 | LookupConstant(PHI.getIncomingValue(Idx), ConstantPool); | |||
| 5177 | if (!ConstVal) | |||
| 5178 | return false; | |||
| 5179 | ||||
| 5180 | // Be conservative about which kinds of constants we support. | |||
| 5181 | if (!ValidLookupTableConstant(ConstVal, TTI)) | |||
| 5182 | return false; | |||
| 5183 | ||||
| 5184 | Res.push_back(std::make_pair(&PHI, ConstVal)); | |||
| 5185 | } | |||
| 5186 | ||||
| 5187 | return Res.size() > 0; | |||
| 5188 | } | |||
| 5189 | ||||
| 5190 | // Helper function used to add CaseVal to the list of cases that generate | |||
| 5191 | // Result. Returns the updated number of cases that generate this result. | |||
| 5192 | static uintptr_t MapCaseToResult(ConstantInt *CaseVal, | |||
| 5193 | SwitchCaseResultVectorTy &UniqueResults, | |||
| 5194 | Constant *Result) { | |||
| 5195 | for (auto &I : UniqueResults) { | |||
| 5196 | if (I.first == Result) { | |||
| 5197 | I.second.push_back(CaseVal); | |||
| 5198 | return I.second.size(); | |||
| 5199 | } | |||
| 5200 | } | |||
| 5201 | UniqueResults.push_back( | |||
| 5202 | std::make_pair(Result, SmallVector<ConstantInt *, 4>(1, CaseVal))); | |||
| 5203 | return 1; | |||
| 5204 | } | |||
| 5205 | ||||
| 5206 | // Helper function that initializes a map containing | |||
| 5207 | // results for the PHI node of the common destination block for a switch | |||
| 5208 | // instruction. Returns false if multiple PHI nodes have been found or if | |||
| 5209 | // there is not a common destination block for the switch. | |||
| 5210 | static bool | |||
| 5211 | InitializeUniqueCases(SwitchInst *SI, PHINode *&PHI, BasicBlock *&CommonDest, | |||
| 5212 | SwitchCaseResultVectorTy &UniqueResults, | |||
| 5213 | Constant *&DefaultResult, const DataLayout &DL, | |||
| 5214 | const TargetTransformInfo &TTI, | |||
| 5215 | uintptr_t MaxUniqueResults, uintptr_t MaxCasesPerResult) { | |||
| 5216 | for (auto &I : SI->cases()) { | |||
| 5217 | ConstantInt *CaseVal = I.getCaseValue(); | |||
| 5218 | ||||
| 5219 | // Resulting value at phi nodes for this case value. | |||
| 5220 | SwitchCaseResultsTy Results; | |||
| 5221 | if (!GetCaseResults(SI, CaseVal, I.getCaseSuccessor(), &CommonDest, Results, | |||
| 5222 | DL, TTI)) | |||
| 5223 | return false; | |||
| 5224 | ||||
| 5225 | // Only one value per case is permitted. | |||
| 5226 | if (Results.size() > 1) | |||
| 5227 | return false; | |||
| 5228 | ||||
| 5229 | // Add the case->result mapping to UniqueResults. | |||
| 5230 | const uintptr_t NumCasesForResult = | |||
| 5231 | MapCaseToResult(CaseVal, UniqueResults, Results.begin()->second); | |||
| 5232 | ||||
| 5233 | // Early out if there are too many cases for this result. | |||
| 5234 | if (NumCasesForResult > MaxCasesPerResult) | |||
| 5235 | return false; | |||
| 5236 | ||||
| 5237 | // Early out if there are too many unique results. | |||
| 5238 | if (UniqueResults.size() > MaxUniqueResults) | |||
| 5239 | return false; | |||
| 5240 | ||||
| 5241 | // Check the PHI consistency. | |||
| 5242 | if (!PHI) | |||
| 5243 | PHI = Results[0].first; | |||
| 5244 | else if (PHI != Results[0].first) | |||
| 5245 | return false; | |||
| 5246 | } | |||
| 5247 | // Find the default result value. | |||
| 5248 | SmallVector<std::pair<PHINode *, Constant *>, 1> DefaultResults; | |||
| 5249 | BasicBlock *DefaultDest = SI->getDefaultDest(); | |||
| 5250 | GetCaseResults(SI, nullptr, SI->getDefaultDest(), &CommonDest, DefaultResults, | |||
| 5251 | DL, TTI); | |||
| 5252 | // If the default value is not found abort unless the default destination | |||
| 5253 | // is unreachable. | |||
| 5254 | DefaultResult = | |||
| 5255 | DefaultResults.size() == 1 ? DefaultResults.begin()->second : nullptr; | |||
| 5256 | if ((!DefaultResult && | |||
| 5257 | !isa<UnreachableInst>(DefaultDest->getFirstNonPHIOrDbg()))) | |||
| 5258 | return false; | |||
| 5259 | ||||
| 5260 | return true; | |||
| 5261 | } | |||
| 5262 | ||||
| 5263 | // Helper function that checks if it is possible to transform a switch with only | |||
| 5264 | // two cases (or two cases + default) that produces a result into a select. | |||
| 5265 | // Example: | |||
| 5266 | // switch (a) { | |||
| 5267 | // case 10: %0 = icmp eq i32 %a, 10 | |||
| 5268 | // return 10; %1 = select i1 %0, i32 10, i32 4 | |||
| 5269 | // case 20: ----> %2 = icmp eq i32 %a, 20 | |||
| 5270 | // return 2; %3 = select i1 %2, i32 2, i32 %1 | |||
| 5271 | // default: | |||
| 5272 | // return 4; | |||
| 5273 | // } | |||
| 5274 | static Value *ConvertTwoCaseSwitch(const SwitchCaseResultVectorTy &ResultVector, | |||
| 5275 | Constant *DefaultResult, Value *Condition, | |||
| 5276 | IRBuilder<> &Builder) { | |||
| 5277 | // If we are selecting between only two cases transform into a simple | |||
| 5278 | // select or a two-way select if default is possible. | |||
| 5279 | if (ResultVector.size() == 2 && ResultVector[0].second.size() == 1 && | |||
| 5280 | ResultVector[1].second.size() == 1) { | |||
| 5281 | ConstantInt *const FirstCase = ResultVector[0].second[0]; | |||
| 5282 | ConstantInt *const SecondCase = ResultVector[1].second[0]; | |||
| 5283 | ||||
| 5284 | bool DefaultCanTrigger = DefaultResult; | |||
| 5285 | Value *SelectValue = ResultVector[1].first; | |||
| 5286 | if (DefaultCanTrigger) { | |||
| 5287 | Value *const ValueCompare = | |||
| 5288 | Builder.CreateICmpEQ(Condition, SecondCase, "switch.selectcmp"); | |||
| 5289 | SelectValue = Builder.CreateSelect(ValueCompare, ResultVector[1].first, | |||
| 5290 | DefaultResult, "switch.select"); | |||
| 5291 | } | |||
| 5292 | Value *const ValueCompare = | |||
| 5293 | Builder.CreateICmpEQ(Condition, FirstCase, "switch.selectcmp"); | |||
| 5294 | return Builder.CreateSelect(ValueCompare, ResultVector[0].first, | |||
| 5295 | SelectValue, "switch.select"); | |||
| 5296 | } | |||
| 5297 | ||||
| 5298 | // Handle the degenerate case where two cases have the same value. | |||
| 5299 | if (ResultVector.size() == 1 && ResultVector[0].second.size() == 2 && | |||
| 5300 | DefaultResult) { | |||
| 5301 | Value *Cmp1 = Builder.CreateICmpEQ( | |||
| 5302 | Condition, ResultVector[0].second[0], "switch.selectcmp.case1"); | |||
| 5303 | Value *Cmp2 = Builder.CreateICmpEQ( | |||
| 5304 | Condition, ResultVector[0].second[1], "switch.selectcmp.case2"); | |||
| 5305 | Value *Cmp = Builder.CreateOr(Cmp1, Cmp2, "switch.selectcmp"); | |||
| 5306 | return Builder.CreateSelect(Cmp, ResultVector[0].first, DefaultResult); | |||
| 5307 | } | |||
| 5308 | ||||
| 5309 | return nullptr; | |||
| 5310 | } | |||
| 5311 | ||||
| 5312 | // Helper function to cleanup a switch instruction that has been converted into | |||
| 5313 | // a select, fixing up PHI nodes and basic blocks. | |||
| 5314 | static void RemoveSwitchAfterSelectConversion(SwitchInst *SI, PHINode *PHI, | |||
| 5315 | Value *SelectValue, | |||
| 5316 | IRBuilder<> &Builder, | |||
| 5317 | DomTreeUpdater *DTU) { | |||
| 5318 | std::vector<DominatorTree::UpdateType> Updates; | |||
| 5319 | ||||
| 5320 | BasicBlock *SelectBB = SI->getParent(); | |||
| 5321 | BasicBlock *DestBB = PHI->getParent(); | |||
| 5322 | ||||
| 5323 | if (DTU && !is_contained(predecessors(DestBB), SelectBB)) | |||
| 5324 | Updates.push_back({DominatorTree::Insert, SelectBB, DestBB}); | |||
| 5325 | Builder.CreateBr(DestBB); | |||
| 5326 | ||||
| 5327 | // Remove the switch. | |||
| 5328 | ||||
| 5329 | while (PHI->getBasicBlockIndex(SelectBB) >= 0) | |||
| 5330 | PHI->removeIncomingValue(SelectBB); | |||
| 5331 | PHI->addIncoming(SelectValue, SelectBB); | |||
| 5332 | ||||
| 5333 | SmallPtrSet<BasicBlock *, 4> RemovedSuccessors; | |||
| 5334 | for (unsigned i = 0, e = SI->getNumSuccessors(); i < e; ++i) { | |||
| 5335 | BasicBlock *Succ = SI->getSuccessor(i); | |||
| 5336 | ||||
| 5337 | if (Succ == DestBB) | |||
| 5338 | continue; | |||
| 5339 | Succ->removePredecessor(SelectBB); | |||
| 5340 | if (DTU && RemovedSuccessors.insert(Succ).second) | |||
| 5341 | Updates.push_back({DominatorTree::Delete, SelectBB, Succ}); | |||
| 5342 | } | |||
| 5343 | SI->eraseFromParent(); | |||
| 5344 | if (DTU) | |||
| 5345 | DTU->applyUpdates(Updates); | |||
| 5346 | } | |||
| 5347 | ||||
| 5348 | /// If the switch is only used to initialize one or more | |||
| 5349 | /// phi nodes in a common successor block with only two different | |||
| 5350 | /// constant values, replace the switch with select. | |||
| 5351 | static bool switchToSelect(SwitchInst *SI, IRBuilder<> &Builder, | |||
| 5352 | DomTreeUpdater *DTU, const DataLayout &DL, | |||
| 5353 | const TargetTransformInfo &TTI) { | |||
| 5354 | Value *const Cond = SI->getCondition(); | |||
| 5355 | PHINode *PHI = nullptr; | |||
| 5356 | BasicBlock *CommonDest = nullptr; | |||
| 5357 | Constant *DefaultResult; | |||
| 5358 | SwitchCaseResultVectorTy UniqueResults; | |||
| 5359 | // Collect all the cases that will deliver the same value from the switch. | |||
| 5360 | if (!InitializeUniqueCases(SI, PHI, CommonDest, UniqueResults, DefaultResult, | |||
| 5361 | DL, TTI, /*MaxUniqueResults*/2, | |||
| 5362 | /*MaxCasesPerResult*/2)) | |||
| 5363 | return false; | |||
| 5364 | assert(PHI != nullptr && "PHI for value select not found")((void)0); | |||
| 5365 | ||||
| 5366 | Builder.SetInsertPoint(SI); | |||
| 5367 | Value *SelectValue = | |||
| 5368 | ConvertTwoCaseSwitch(UniqueResults, DefaultResult, Cond, Builder); | |||
| 5369 | if (SelectValue) { | |||
| 5370 | RemoveSwitchAfterSelectConversion(SI, PHI, SelectValue, Builder, DTU); | |||
| 5371 | return true; | |||
| 5372 | } | |||
| 5373 | // The switch couldn't be converted into a select. | |||
| 5374 | return false; | |||
| 5375 | } | |||
| 5376 | ||||
| 5377 | namespace { | |||
| 5378 | ||||
| 5379 | /// This class represents a lookup table that can be used to replace a switch. | |||
| 5380 | class SwitchLookupTable { | |||
| 5381 | public: | |||
| 5382 | /// Create a lookup table to use as a switch replacement with the contents | |||
| 5383 | /// of Values, using DefaultValue to fill any holes in the table. | |||
| 5384 | SwitchLookupTable( | |||
| 5385 | Module &M, uint64_t TableSize, ConstantInt *Offset, | |||
| 5386 | const SmallVectorImpl<std::pair<ConstantInt *, Constant *>> &Values, | |||
| 5387 | Constant *DefaultValue, const DataLayout &DL, const StringRef &FuncName); | |||
| 5388 | ||||
| 5389 | /// Build instructions with Builder to retrieve the value at | |||
| 5390 | /// the position given by Index in the lookup table. | |||
| 5391 | Value *BuildLookup(Value *Index, IRBuilder<> &Builder); | |||
| 5392 | ||||
| 5393 | /// Return true if a table with TableSize elements of | |||
| 5394 | /// type ElementType would fit in a target-legal register. | |||
| 5395 | static bool WouldFitInRegister(const DataLayout &DL, uint64_t TableSize, | |||
| 5396 | Type *ElementType); | |||
| 5397 | ||||
| 5398 | private: | |||
| 5399 | // Depending on the contents of the table, it can be represented in | |||
| 5400 | // different ways. | |||
| 5401 | enum { | |||
| 5402 | // For tables where each element contains the same value, we just have to | |||
| 5403 | // store that single value and return it for each lookup. | |||
| 5404 | SingleValueKind, | |||
| 5405 | ||||
| 5406 | // For tables where there is a linear relationship between table index | |||
| 5407 | // and values. We calculate the result with a simple multiplication | |||
| 5408 | // and addition instead of a table lookup. | |||
| 5409 | LinearMapKind, | |||
| 5410 | ||||
| 5411 | // For small tables with integer elements, we can pack them into a bitmap | |||
| 5412 | // that fits into a target-legal register. Values are retrieved by | |||
| 5413 | // shift and mask operations. | |||
| 5414 | BitMapKind, | |||
| 5415 | ||||
| 5416 | // The table is stored as an array of values. Values are retrieved by load | |||
| 5417 | // instructions from the table. | |||
| 5418 | ArrayKind | |||
| 5419 | } Kind; | |||
| 5420 | ||||
| 5421 | // For SingleValueKind, this is the single value. | |||
| 5422 | Constant *SingleValue = nullptr; | |||
| 5423 | ||||
| 5424 | // For BitMapKind, this is the bitmap. | |||
| 5425 | ConstantInt *BitMap = nullptr; | |||
| 5426 | IntegerType *BitMapElementTy = nullptr; | |||
| 5427 | ||||
| 5428 | // For LinearMapKind, these are the constants used to derive the value. | |||
| 5429 | ConstantInt *LinearOffset = nullptr; | |||
| 5430 | ConstantInt *LinearMultiplier = nullptr; | |||
| 5431 | ||||
| 5432 | // For ArrayKind, this is the array. | |||
| 5433 | GlobalVariable *Array = nullptr; | |||
| 5434 | }; | |||
| 5435 | ||||
| 5436 | } // end anonymous namespace | |||
| 5437 | ||||
| 5438 | SwitchLookupTable::SwitchLookupTable( | |||
| 5439 | Module &M, uint64_t TableSize, ConstantInt *Offset, | |||
| 5440 | const SmallVectorImpl<std::pair<ConstantInt *, Constant *>> &Values, | |||
| 5441 | Constant *DefaultValue, const DataLayout &DL, const StringRef &FuncName) { | |||
| 5442 | assert(Values.size() && "Can't build lookup table without values!")((void)0); | |||
| 5443 | assert(TableSize >= Values.size() && "Can't fit values in table!")((void)0); | |||
| 5444 | ||||
| 5445 | // If all values in the table are equal, this is that value. | |||
| 5446 | SingleValue = Values.begin()->second; | |||
| 5447 | ||||
| 5448 | Type *ValueType = Values.begin()->second->getType(); | |||
| 5449 | ||||
| 5450 | // Build up the table contents. | |||
| 5451 | SmallVector<Constant *, 64> TableContents(TableSize); | |||
| 5452 | for (size_t I = 0, E = Values.size(); I != E; ++I) { | |||
| 5453 | ConstantInt *CaseVal = Values[I].first; | |||
| 5454 | Constant *CaseRes = Values[I].second; | |||
| 5455 | assert(CaseRes->getType() == ValueType)((void)0); | |||
| 5456 | ||||
| 5457 | uint64_t Idx = (CaseVal->getValue() - Offset->getValue()).getLimitedValue(); | |||
| 5458 | TableContents[Idx] = CaseRes; | |||
| 5459 | ||||
| 5460 | if (CaseRes != SingleValue) | |||
| 5461 | SingleValue = nullptr; | |||
| 5462 | } | |||
| 5463 | ||||
| 5464 | // Fill in any holes in the table with the default result. | |||
| 5465 | if (Values.size() < TableSize) { | |||
| 5466 | assert(DefaultValue &&((void)0) | |||
| 5467 | "Need a default value to fill the lookup table holes.")((void)0); | |||
| 5468 | assert(DefaultValue->getType() == ValueType)((void)0); | |||
| 5469 | for (uint64_t I = 0; I < TableSize; ++I) { | |||
| 5470 | if (!TableContents[I]) | |||
| 5471 | TableContents[I] = DefaultValue; | |||
| 5472 | } | |||
| 5473 | ||||
| 5474 | if (DefaultValue != SingleValue) | |||
| 5475 | SingleValue = nullptr; | |||
| 5476 | } | |||
| 5477 | ||||
| 5478 | // If each element in the table contains the same value, we only need to store | |||
| 5479 | // that single value. | |||
| 5480 | if (SingleValue) { | |||
| 5481 | Kind = SingleValueKind; | |||
| 5482 | return; | |||
| 5483 | } | |||
| 5484 | ||||
| 5485 | // Check if we can derive the value with a linear transformation from the | |||
| 5486 | // table index. | |||
| 5487 | if (isa<IntegerType>(ValueType)) { | |||
| 5488 | bool LinearMappingPossible = true; | |||
| 5489 | APInt PrevVal; | |||
| 5490 | APInt DistToPrev; | |||
| 5491 | assert(TableSize >= 2 && "Should be a SingleValue table.")((void)0); | |||
| 5492 | // Check if there is the same distance between two consecutive values. | |||
| 5493 | for (uint64_t I = 0; I < TableSize; ++I) { | |||
| 5494 | ConstantInt *ConstVal = dyn_cast<ConstantInt>(TableContents[I]); | |||
| 5495 | if (!ConstVal) { | |||
| 5496 | // This is an undef. We could deal with it, but undefs in lookup tables | |||
| 5497 | // are very seldom. It's probably not worth the additional complexity. | |||
| 5498 | LinearMappingPossible = false; | |||
| 5499 | break; | |||
| 5500 | } | |||
| 5501 | const APInt &Val = ConstVal->getValue(); | |||
| 5502 | if (I != 0) { | |||
| 5503 | APInt Dist = Val - PrevVal; | |||
| 5504 | if (I == 1) { | |||
| 5505 | DistToPrev = Dist; | |||
| 5506 | } else if (Dist != DistToPrev) { | |||
| 5507 | LinearMappingPossible = false; | |||
| 5508 | break; | |||
| 5509 | } | |||
| 5510 | } | |||
| 5511 | PrevVal = Val; | |||
| 5512 | } | |||
| 5513 | if (LinearMappingPossible) { | |||
| 5514 | LinearOffset = cast<ConstantInt>(TableContents[0]); | |||
| 5515 | LinearMultiplier = ConstantInt::get(M.getContext(), DistToPrev); | |||
| 5516 | Kind = LinearMapKind; | |||
| 5517 | ++NumLinearMaps; | |||
| 5518 | return; | |||
| 5519 | } | |||
| 5520 | } | |||
| 5521 | ||||
| 5522 | // If the type is integer and the table fits in a register, build a bitmap. | |||
| 5523 | if (WouldFitInRegister(DL, TableSize, ValueType)) { | |||
| 5524 | IntegerType *IT = cast<IntegerType>(ValueType); | |||
| 5525 | APInt TableInt(TableSize * IT->getBitWidth(), 0); | |||
| 5526 | for (uint64_t I = TableSize; I > 0; --I) { | |||
| 5527 | TableInt <<= IT->getBitWidth(); | |||
| 5528 | // Insert values into the bitmap. Undef values are set to zero. | |||
| 5529 | if (!isa<UndefValue>(TableContents[I - 1])) { | |||
| 5530 | ConstantInt *Val = cast<ConstantInt>(TableContents[I - 1]); | |||
| 5531 | TableInt |= Val->getValue().zext(TableInt.getBitWidth()); | |||
| 5532 | } | |||
| 5533 | } | |||
| 5534 | BitMap = ConstantInt::get(M.getContext(), TableInt); | |||
| 5535 | BitMapElementTy = IT; | |||
| 5536 | Kind = BitMapKind; | |||
| 5537 | ++NumBitMaps; | |||
| 5538 | return; | |||
| 5539 | } | |||
| 5540 | ||||
| 5541 | // Store the table in an array. | |||
| 5542 | ArrayType *ArrayTy = ArrayType::get(ValueType, TableSize); | |||
| 5543 | Constant *Initializer = ConstantArray::get(ArrayTy, TableContents); | |||
| 5544 | ||||
| 5545 | Array = new GlobalVariable(M, ArrayTy, /*isConstant=*/true, | |||
| 5546 | GlobalVariable::PrivateLinkage, Initializer, | |||
| 5547 | "switch.table." + FuncName); | |||
| 5548 | Array->setUnnamedAddr(GlobalValue::UnnamedAddr::Global); | |||
| 5549 | // Set the alignment to that of an array items. We will be only loading one | |||
| 5550 | // value out of it. | |||
| 5551 | Array->setAlignment(Align(DL.getPrefTypeAlignment(ValueType))); | |||
| 5552 | Kind = ArrayKind; | |||
| 5553 | } | |||
| 5554 | ||||
| 5555 | Value *SwitchLookupTable::BuildLookup(Value *Index, IRBuilder<> &Builder) { | |||
| 5556 | switch (Kind) { | |||
| 5557 | case SingleValueKind: | |||
| 5558 | return SingleValue; | |||
| 5559 | case LinearMapKind: { | |||
| 5560 | // Derive the result value from the input value. | |||
| 5561 | Value *Result = Builder.CreateIntCast(Index, LinearMultiplier->getType(), | |||
| 5562 | false, "switch.idx.cast"); | |||
| 5563 | if (!LinearMultiplier->isOne()) | |||
| 5564 | Result = Builder.CreateMul(Result, LinearMultiplier, "switch.idx.mult"); | |||
| 5565 | if (!LinearOffset->isZero()) | |||
| 5566 | Result = Builder.CreateAdd(Result, LinearOffset, "switch.offset"); | |||
| 5567 | return Result; | |||
| 5568 | } | |||
| 5569 | case BitMapKind: { | |||
| 5570 | // Type of the bitmap (e.g. i59). | |||
| 5571 | IntegerType *MapTy = BitMap->getType(); | |||
| 5572 | ||||
| 5573 | // Cast Index to the same type as the bitmap. | |||
| 5574 | // Note: The Index is <= the number of elements in the table, so | |||
| 5575 | // truncating it to the width of the bitmask is safe. | |||
| 5576 | Value *ShiftAmt = Builder.CreateZExtOrTrunc(Index, MapTy, "switch.cast"); | |||
| 5577 | ||||
| 5578 | // Multiply the shift amount by the element width. | |||
| 5579 | ShiftAmt = Builder.CreateMul( | |||
| 5580 | ShiftAmt, ConstantInt::get(MapTy, BitMapElementTy->getBitWidth()), | |||
| 5581 | "switch.shiftamt"); | |||
| 5582 | ||||
| 5583 | // Shift down. | |||
| 5584 | Value *DownShifted = | |||
| 5585 | Builder.CreateLShr(BitMap, ShiftAmt, "switch.downshift"); | |||
| 5586 | // Mask off. | |||
| 5587 | return Builder.CreateTrunc(DownShifted, BitMapElementTy, "switch.masked"); | |||
| 5588 | } | |||
| 5589 | case ArrayKind: { | |||
| 5590 | // Make sure the table index will not overflow when treated as signed. | |||
| 5591 | IntegerType *IT = cast<IntegerType>(Index->getType()); | |||
| 5592 | uint64_t TableSize = | |||
| 5593 | Array->getInitializer()->getType()->getArrayNumElements(); | |||
| 5594 | if (TableSize > (1ULL << (IT->getBitWidth() - 1))) | |||
| 5595 | Index = Builder.CreateZExt( | |||
| 5596 | Index, IntegerType::get(IT->getContext(), IT->getBitWidth() + 1), | |||
| 5597 | "switch.tableidx.zext"); | |||
| 5598 | ||||
| 5599 | Value *GEPIndices[] = {Builder.getInt32(0), Index}; | |||
| 5600 | Value *GEP = Builder.CreateInBoundsGEP(Array->getValueType(), Array, | |||
| 5601 | GEPIndices, "switch.gep"); | |||
| 5602 | return Builder.CreateLoad( | |||
| 5603 | cast<ArrayType>(Array->getValueType())->getElementType(), GEP, | |||
| 5604 | "switch.load"); | |||
| 5605 | } | |||
| 5606 | } | |||
| 5607 | llvm_unreachable("Unknown lookup table kind!")__builtin_unreachable(); | |||
| 5608 | } | |||
| 5609 | ||||
| 5610 | bool SwitchLookupTable::WouldFitInRegister(const DataLayout &DL, | |||
| 5611 | uint64_t TableSize, | |||
| 5612 | Type *ElementType) { | |||
| 5613 | auto *IT = dyn_cast<IntegerType>(ElementType); | |||
| 5614 | if (!IT) | |||
| 5615 | return false; | |||
| 5616 | // FIXME: If the type is wider than it needs to be, e.g. i8 but all values | |||
| 5617 | // are <= 15, we could try to narrow the type. | |||
| 5618 | ||||
| 5619 | // Avoid overflow, fitsInLegalInteger uses unsigned int for the width. | |||
| 5620 | if (TableSize >= UINT_MAX(2147483647 *2U +1U) / IT->getBitWidth()) | |||
| 5621 | return false; | |||
| 5622 | return DL.fitsInLegalInteger(TableSize * IT->getBitWidth()); | |||
| 5623 | } | |||
| 5624 | ||||
| 5625 | /// Determine whether a lookup table should be built for this switch, based on | |||
| 5626 | /// the number of cases, size of the table, and the types of the results. | |||
| 5627 | static bool | |||
| 5628 | ShouldBuildLookupTable(SwitchInst *SI, uint64_t TableSize, | |||
| 5629 | const TargetTransformInfo &TTI, const DataLayout &DL, | |||
| 5630 | const SmallDenseMap<PHINode *, Type *> &ResultTypes) { | |||
| 5631 | if (SI->getNumCases() > TableSize || TableSize >= UINT64_MAX0xffffffffffffffffULL / 10) | |||
| 5632 | return false; // TableSize overflowed, or mul below might overflow. | |||
| 5633 | ||||
| 5634 | bool AllTablesFitInRegister = true; | |||
| 5635 | bool HasIllegalType = false; | |||
| 5636 | for (const auto &I : ResultTypes) { | |||
| 5637 | Type *Ty = I.second; | |||
| 5638 | ||||
| 5639 | // Saturate this flag to true. | |||
| 5640 | HasIllegalType = HasIllegalType || !TTI.isTypeLegal(Ty); | |||
| 5641 | ||||
| 5642 | // Saturate this flag to false. | |||
| 5643 | AllTablesFitInRegister = | |||
| 5644 | AllTablesFitInRegister && | |||
| 5645 | SwitchLookupTable::WouldFitInRegister(DL, TableSize, Ty); | |||
| 5646 | ||||
| 5647 | // If both flags saturate, we're done. NOTE: This *only* works with | |||
| 5648 | // saturating flags, and all flags have to saturate first due to the | |||
| 5649 | // non-deterministic behavior of iterating over a dense map. | |||
| 5650 | if (HasIllegalType && !AllTablesFitInRegister) | |||
| 5651 | break; | |||
| 5652 | } | |||
| 5653 | ||||
| 5654 | // If each table would fit in a register, we should build it anyway. | |||
| 5655 | if (AllTablesFitInRegister) | |||
| 5656 | return true; | |||
| 5657 | ||||
| 5658 | // Don't build a table that doesn't fit in-register if it has illegal types. | |||
| 5659 | if (HasIllegalType) | |||
| 5660 | return false; | |||
| 5661 | ||||
| 5662 | // The table density should be at least 40%. This is the same criterion as for | |||
| 5663 | // jump tables, see SelectionDAGBuilder::handleJTSwitchCase. | |||
| 5664 | // FIXME: Find the best cut-off. | |||
| 5665 | return SI->getNumCases() * 10 >= TableSize * 4; | |||
| 5666 | } | |||
| 5667 | ||||
| 5668 | /// Try to reuse the switch table index compare. Following pattern: | |||
| 5669 | /// \code | |||
| 5670 | /// if (idx < tablesize) | |||
| 5671 | /// r = table[idx]; // table does not contain default_value | |||
| 5672 | /// else | |||
| 5673 | /// r = default_value; | |||
| 5674 | /// if (r != default_value) | |||
| 5675 | /// ... | |||
| 5676 | /// \endcode | |||
| 5677 | /// Is optimized to: | |||
| 5678 | /// \code | |||
| 5679 | /// cond = idx < tablesize; | |||
| 5680 | /// if (cond) | |||
| 5681 | /// r = table[idx]; | |||
| 5682 | /// else | |||
| 5683 | /// r = default_value; | |||
| 5684 | /// if (cond) | |||
| 5685 | /// ... | |||
| 5686 | /// \endcode | |||
| 5687 | /// Jump threading will then eliminate the second if(cond). | |||
| 5688 | static void reuseTableCompare( | |||
| 5689 | User *PhiUser, BasicBlock *PhiBlock, BranchInst *RangeCheckBranch, | |||
| 5690 | Constant *DefaultValue, | |||
| 5691 | const SmallVectorImpl<std::pair<ConstantInt *, Constant *>> &Values) { | |||
| 5692 | ICmpInst *CmpInst = dyn_cast<ICmpInst>(PhiUser); | |||
| 5693 | if (!CmpInst) | |||
| 5694 | return; | |||
| 5695 | ||||
| 5696 | // We require that the compare is in the same block as the phi so that jump | |||
| 5697 | // threading can do its work afterwards. | |||
| 5698 | if (CmpInst->getParent() != PhiBlock) | |||
| 5699 | return; | |||
| 5700 | ||||
| 5701 | Constant *CmpOp1 = dyn_cast<Constant>(CmpInst->getOperand(1)); | |||
| 5702 | if (!CmpOp1) | |||
| 5703 | return; | |||
| 5704 | ||||
| 5705 | Value *RangeCmp = RangeCheckBranch->getCondition(); | |||
| 5706 | Constant *TrueConst = ConstantInt::getTrue(RangeCmp->getType()); | |||
| 5707 | Constant *FalseConst = ConstantInt::getFalse(RangeCmp->getType()); | |||
| 5708 | ||||
| 5709 | // Check if the compare with the default value is constant true or false. | |||
| 5710 | Constant *DefaultConst = ConstantExpr::getICmp(CmpInst->getPredicate(), | |||
| 5711 | DefaultValue, CmpOp1, true); | |||
| 5712 | if (DefaultConst != TrueConst && DefaultConst != FalseConst) | |||
| 5713 | return; | |||
| 5714 | ||||
| 5715 | // Check if the compare with the case values is distinct from the default | |||
| 5716 | // compare result. | |||
| 5717 | for (auto ValuePair : Values) { | |||
| 5718 | Constant *CaseConst = ConstantExpr::getICmp(CmpInst->getPredicate(), | |||
| 5719 | ValuePair.second, CmpOp1, true); | |||
| 5720 | if (!CaseConst || CaseConst == DefaultConst || isa<UndefValue>(CaseConst)) | |||
| 5721 | return; | |||
| 5722 | assert((CaseConst == TrueConst || CaseConst == FalseConst) &&((void)0) | |||
| 5723 | "Expect true or false as compare result.")((void)0); | |||
| 5724 | } | |||
| 5725 | ||||
| 5726 | // Check if the branch instruction dominates the phi node. It's a simple | |||
| 5727 | // dominance check, but sufficient for our needs. | |||
| 5728 | // Although this check is invariant in the calling loops, it's better to do it | |||
| 5729 | // at this late stage. Practically we do it at most once for a switch. | |||
| 5730 | BasicBlock *BranchBlock = RangeCheckBranch->getParent(); | |||
| 5731 | for (BasicBlock *Pred : predecessors(PhiBlock)) { | |||
| 5732 | if (Pred != BranchBlock && Pred->getUniquePredecessor() != BranchBlock) | |||
| 5733 | return; | |||
| 5734 | } | |||
| 5735 | ||||
| 5736 | if (DefaultConst == FalseConst) { | |||
| 5737 | // The compare yields the same result. We can replace it. | |||
| 5738 | CmpInst->replaceAllUsesWith(RangeCmp); | |||
| 5739 | ++NumTableCmpReuses; | |||
| 5740 | } else { | |||
| 5741 | // The compare yields the same result, just inverted. We can replace it. | |||
| 5742 | Value *InvertedTableCmp = BinaryOperator::CreateXor( | |||
| 5743 | RangeCmp, ConstantInt::get(RangeCmp->getType(), 1), "inverted.cmp", | |||
| 5744 | RangeCheckBranch); | |||
| 5745 | CmpInst->replaceAllUsesWith(InvertedTableCmp); | |||
| 5746 | ++NumTableCmpReuses; | |||
| 5747 | } | |||
| 5748 | } | |||
| 5749 | ||||
| 5750 | /// If the switch is only used to initialize one or more phi nodes in a common | |||
| 5751 | /// successor block with different constant values, replace the switch with | |||
| 5752 | /// lookup tables. | |||
| 5753 | static bool SwitchToLookupTable(SwitchInst *SI, IRBuilder<> &Builder, | |||
| 5754 | DomTreeUpdater *DTU, const DataLayout &DL, | |||
| 5755 | const TargetTransformInfo &TTI) { | |||
| 5756 | assert(SI->getNumCases() > 1 && "Degenerate switch?")((void)0); | |||
| 5757 | ||||
| 5758 | BasicBlock *BB = SI->getParent(); | |||
| 5759 | Function *Fn = BB->getParent(); | |||
| 5760 | // Only build lookup table when we have a target that supports it or the | |||
| 5761 | // attribute is not set. | |||
| 5762 | if (!TTI.shouldBuildLookupTables() || | |||
| 5763 | (Fn->getFnAttribute("no-jump-tables").getValueAsBool())) | |||
| 5764 | return false; | |||
| 5765 | ||||
| 5766 | // FIXME: If the switch is too sparse for a lookup table, perhaps we could | |||
| 5767 | // split off a dense part and build a lookup table for that. | |||
| 5768 | ||||
| 5769 | // FIXME: This creates arrays of GEPs to constant strings, which means each | |||
| 5770 | // GEP needs a runtime relocation in PIC code. We should just build one big | |||
| 5771 | // string and lookup indices into that. | |||
| 5772 | ||||
| 5773 | // Ignore switches with less than three cases. Lookup tables will not make | |||
| 5774 | // them faster, so we don't analyze them. | |||
| 5775 | if (SI->getNumCases() < 3) | |||
| 5776 | return false; | |||
| 5777 | ||||
| 5778 | // Figure out the corresponding result for each case value and phi node in the | |||
| 5779 | // common destination, as well as the min and max case values. | |||
| 5780 | assert(!SI->cases().empty())((void)0); | |||
| 5781 | SwitchInst::CaseIt CI = SI->case_begin(); | |||
| 5782 | ConstantInt *MinCaseVal = CI->getCaseValue(); | |||
| 5783 | ConstantInt *MaxCaseVal = CI->getCaseValue(); | |||
| 5784 | ||||
| 5785 | BasicBlock *CommonDest = nullptr; | |||
| 5786 | ||||
| 5787 | using ResultListTy = SmallVector<std::pair<ConstantInt *, Constant *>, 4>; | |||
| 5788 | SmallDenseMap<PHINode *, ResultListTy> ResultLists; | |||
| 5789 | ||||
| 5790 | SmallDenseMap<PHINode *, Constant *> DefaultResults; | |||
| 5791 | SmallDenseMap<PHINode *, Type *> ResultTypes; | |||
| 5792 | SmallVector<PHINode *, 4> PHIs; | |||
| 5793 | ||||
| 5794 | for (SwitchInst::CaseIt E = SI->case_end(); CI != E; ++CI) { | |||
| 5795 | ConstantInt *CaseVal = CI->getCaseValue(); | |||
| 5796 | if (CaseVal->getValue().slt(MinCaseVal->getValue())) | |||
| 5797 | MinCaseVal = CaseVal; | |||
| 5798 | if (CaseVal->getValue().sgt(MaxCaseVal->getValue())) | |||
| 5799 | MaxCaseVal = CaseVal; | |||
| 5800 | ||||
| 5801 | // Resulting value at phi nodes for this case value. | |||
| 5802 | using ResultsTy = SmallVector<std::pair<PHINode *, Constant *>, 4>; | |||
| 5803 | ResultsTy Results; | |||
| 5804 | if (!GetCaseResults(SI, CaseVal, CI->getCaseSuccessor(), &CommonDest, | |||
| 5805 | Results, DL, TTI)) | |||
| 5806 | return false; | |||
| 5807 | ||||
| 5808 | // Append the result from this case to the list for each phi. | |||
| 5809 | for (const auto &I : Results) { | |||
| 5810 | PHINode *PHI = I.first; | |||
| 5811 | Constant *Value = I.second; | |||
| 5812 | if (!ResultLists.count(PHI)) | |||
| 5813 | PHIs.push_back(PHI); | |||
| 5814 | ResultLists[PHI].push_back(std::make_pair(CaseVal, Value)); | |||
| 5815 | } | |||
| 5816 | } | |||
| 5817 | ||||
| 5818 | // Keep track of the result types. | |||
| 5819 | for (PHINode *PHI : PHIs) { | |||
| 5820 | ResultTypes[PHI] = ResultLists[PHI][0].second->getType(); | |||
| 5821 | } | |||
| 5822 | ||||
| 5823 | uint64_t NumResults = ResultLists[PHIs[0]].size(); | |||
| 5824 | APInt RangeSpread = MaxCaseVal->getValue() - MinCaseVal->getValue(); | |||
| 5825 | uint64_t TableSize = RangeSpread.getLimitedValue() + 1; | |||
| 5826 | bool TableHasHoles = (NumResults < TableSize); | |||
| 5827 | ||||
| 5828 | // If the table has holes, we need a constant result for the default case | |||
| 5829 | // or a bitmask that fits in a register. | |||
| 5830 | SmallVector<std::pair<PHINode *, Constant *>, 4> DefaultResultsList; | |||
| 5831 | bool HasDefaultResults = | |||
| 5832 | GetCaseResults(SI, nullptr, SI->getDefaultDest(), &CommonDest, | |||
| 5833 | DefaultResultsList, DL, TTI); | |||
| 5834 | ||||
| 5835 | bool NeedMask = (TableHasHoles && !HasDefaultResults); | |||
| 5836 | if (NeedMask) { | |||
| 5837 | // As an extra penalty for the validity test we require more cases. | |||
| 5838 | if (SI->getNumCases() < 4) // FIXME: Find best threshold value (benchmark). | |||
| 5839 | return false; | |||
| 5840 | if (!DL.fitsInLegalInteger(TableSize)) | |||
| 5841 | return false; | |||
| 5842 | } | |||
| 5843 | ||||
| 5844 | for (const auto &I : DefaultResultsList) { | |||
| 5845 | PHINode *PHI = I.first; | |||
| 5846 | Constant *Result = I.second; | |||
| 5847 | DefaultResults[PHI] = Result; | |||
| 5848 | } | |||
| 5849 | ||||
| 5850 | if (!ShouldBuildLookupTable(SI, TableSize, TTI, DL, ResultTypes)) | |||
| 5851 | return false; | |||
| 5852 | ||||
| 5853 | std::vector<DominatorTree::UpdateType> Updates; | |||
| 5854 | ||||
| 5855 | // Create the BB that does the lookups. | |||
| 5856 | Module &Mod = *CommonDest->getParent()->getParent(); | |||
| 5857 | BasicBlock *LookupBB = BasicBlock::Create( | |||
| 5858 | Mod.getContext(), "switch.lookup", CommonDest->getParent(), CommonDest); | |||
| 5859 | ||||
| 5860 | // Compute the table index value. | |||
| 5861 | Builder.SetInsertPoint(SI); | |||
| 5862 | Value *TableIndex; | |||
| 5863 | if (MinCaseVal->isNullValue()) | |||
| 5864 | TableIndex = SI->getCondition(); | |||
| 5865 | else | |||
| 5866 | TableIndex = Builder.CreateSub(SI->getCondition(), MinCaseVal, | |||
| 5867 | "switch.tableidx"); | |||
| 5868 | ||||
| 5869 | // Compute the maximum table size representable by the integer type we are | |||
| 5870 | // switching upon. | |||
| 5871 | unsigned CaseSize = MinCaseVal->getType()->getPrimitiveSizeInBits(); | |||
| 5872 | uint64_t MaxTableSize = CaseSize > 63 ? UINT64_MAX0xffffffffffffffffULL : 1ULL << CaseSize; | |||
| 5873 | assert(MaxTableSize >= TableSize &&((void)0) | |||
| 5874 | "It is impossible for a switch to have more entries than the max "((void)0) | |||
| 5875 | "representable value of its input integer type's size.")((void)0); | |||
| 5876 | ||||
| 5877 | // If the default destination is unreachable, or if the lookup table covers | |||
| 5878 | // all values of the conditional variable, branch directly to the lookup table | |||
| 5879 | // BB. Otherwise, check that the condition is within the case range. | |||
| 5880 | const bool DefaultIsReachable = | |||
| 5881 | !isa<UnreachableInst>(SI->getDefaultDest()->getFirstNonPHIOrDbg()); | |||
| 5882 | const bool GeneratingCoveredLookupTable = (MaxTableSize == TableSize); | |||
| 5883 | BranchInst *RangeCheckBranch = nullptr; | |||
| 5884 | ||||
| 5885 | if (!DefaultIsReachable || GeneratingCoveredLookupTable) { | |||
| 5886 | Builder.CreateBr(LookupBB); | |||
| 5887 | if (DTU) | |||
| 5888 | Updates.push_back({DominatorTree::Insert, BB, LookupBB}); | |||
| 5889 | // Note: We call removeProdecessor later since we need to be able to get the | |||
| 5890 | // PHI value for the default case in case we're using a bit mask. | |||
| 5891 | } else { | |||
| 5892 | Value *Cmp = Builder.CreateICmpULT( | |||
| 5893 | TableIndex, ConstantInt::get(MinCaseVal->getType(), TableSize)); | |||
| 5894 | RangeCheckBranch = | |||
| 5895 | Builder.CreateCondBr(Cmp, LookupBB, SI->getDefaultDest()); | |||
| 5896 | if (DTU) | |||
| 5897 | Updates.push_back({DominatorTree::Insert, BB, LookupBB}); | |||
| 5898 | } | |||
| 5899 | ||||
| 5900 | // Populate the BB that does the lookups. | |||
| 5901 | Builder.SetInsertPoint(LookupBB); | |||
| 5902 | ||||
| 5903 | if (NeedMask) { | |||
| 5904 | // Before doing the lookup, we do the hole check. The LookupBB is therefore | |||
| 5905 | // re-purposed to do the hole check, and we create a new LookupBB. | |||
| 5906 | BasicBlock *MaskBB = LookupBB; | |||
| 5907 | MaskBB->setName("switch.hole_check"); | |||
| 5908 | LookupBB = BasicBlock::Create(Mod.getContext(), "switch.lookup", | |||
| 5909 | CommonDest->getParent(), CommonDest); | |||
| 5910 | ||||
| 5911 | // Make the mask's bitwidth at least 8-bit and a power-of-2 to avoid | |||
| 5912 | // unnecessary illegal types. | |||
| 5913 | uint64_t TableSizePowOf2 = NextPowerOf2(std::max(7ULL, TableSize - 1ULL)); | |||
| 5914 | APInt MaskInt(TableSizePowOf2, 0); | |||
| 5915 | APInt One(TableSizePowOf2, 1); | |||
| 5916 | // Build bitmask; fill in a 1 bit for every case. | |||
| 5917 | const ResultListTy &ResultList = ResultLists[PHIs[0]]; | |||
| 5918 | for (size_t I = 0, E = ResultList.size(); I != E; ++I) { | |||
| 5919 | uint64_t Idx = (ResultList[I].first->getValue() - MinCaseVal->getValue()) | |||
| 5920 | .getLimitedValue(); | |||
| 5921 | MaskInt |= One << Idx; | |||
| 5922 | } | |||
| 5923 | ConstantInt *TableMask = ConstantInt::get(Mod.getContext(), MaskInt); | |||
| 5924 | ||||
| 5925 | // Get the TableIndex'th bit of the bitmask. | |||
| 5926 | // If this bit is 0 (meaning hole) jump to the default destination, | |||
| 5927 | // else continue with table lookup. | |||
| 5928 | IntegerType *MapTy = TableMask->getType(); | |||
| 5929 | Value *MaskIndex = | |||
| 5930 | Builder.CreateZExtOrTrunc(TableIndex, MapTy, "switch.maskindex"); | |||
| 5931 | Value *Shifted = Builder.CreateLShr(TableMask, MaskIndex, "switch.shifted"); | |||
| 5932 | Value *LoBit = Builder.CreateTrunc( | |||
| 5933 | Shifted, Type::getInt1Ty(Mod.getContext()), "switch.lobit"); | |||
| 5934 | Builder.CreateCondBr(LoBit, LookupBB, SI->getDefaultDest()); | |||
| 5935 | if (DTU) { | |||
| 5936 | Updates.push_back({DominatorTree::Insert, MaskBB, LookupBB}); | |||
| 5937 | Updates.push_back({DominatorTree::Insert, MaskBB, SI->getDefaultDest()}); | |||
| 5938 | } | |||
| 5939 | Builder.SetInsertPoint(LookupBB); | |||
| 5940 | AddPredecessorToBlock(SI->getDefaultDest(), MaskBB, BB); | |||
| 5941 | } | |||
| 5942 | ||||
| 5943 | if (!DefaultIsReachable || GeneratingCoveredLookupTable) { | |||
| 5944 | // We cached PHINodes in PHIs. To avoid accessing deleted PHINodes later, | |||
| 5945 | // do not delete PHINodes here. | |||
| 5946 | SI->getDefaultDest()->removePredecessor(BB, | |||
| 5947 | /*KeepOneInputPHIs=*/true); | |||
| 5948 | if (DTU) | |||
| 5949 | Updates.push_back({DominatorTree::Delete, BB, SI->getDefaultDest()}); | |||
| 5950 | } | |||
| 5951 | ||||
| 5952 | for (PHINode *PHI : PHIs) { | |||
| 5953 | const ResultListTy &ResultList = ResultLists[PHI]; | |||
| 5954 | ||||
| 5955 | // If using a bitmask, use any value to fill the lookup table holes. | |||
| 5956 | Constant *DV = NeedMask ? ResultLists[PHI][0].second : DefaultResults[PHI]; | |||
| 5957 | StringRef FuncName = Fn->getName(); | |||
| 5958 | SwitchLookupTable Table(Mod, TableSize, MinCaseVal, ResultList, DV, DL, | |||
| 5959 | FuncName); | |||
| 5960 | ||||
| 5961 | Value *Result = Table.BuildLookup(TableIndex, Builder); | |||
| 5962 | ||||
| 5963 | // Do a small peephole optimization: re-use the switch table compare if | |||
| 5964 | // possible. | |||
| 5965 | if (!TableHasHoles && HasDefaultResults && RangeCheckBranch) { | |||
| 5966 | BasicBlock *PhiBlock = PHI->getParent(); | |||
| 5967 | // Search for compare instructions which use the phi. | |||
| 5968 | for (auto *User : PHI->users()) { | |||
| 5969 | reuseTableCompare(User, PhiBlock, RangeCheckBranch, DV, ResultList); | |||
| 5970 | } | |||
| 5971 | } | |||
| 5972 | ||||
| 5973 | PHI->addIncoming(Result, LookupBB); | |||
| 5974 | } | |||
| 5975 | ||||
| 5976 | Builder.CreateBr(CommonDest); | |||
| 5977 | if (DTU) | |||
| 5978 | Updates.push_back({DominatorTree::Insert, LookupBB, CommonDest}); | |||
| 5979 | ||||
| 5980 | // Remove the switch. | |||
| 5981 | SmallPtrSet<BasicBlock *, 8> RemovedSuccessors; | |||
| 5982 | for (unsigned i = 0, e = SI->getNumSuccessors(); i < e; ++i) { | |||
| 5983 | BasicBlock *Succ = SI->getSuccessor(i); | |||
| 5984 | ||||
| 5985 | if (Succ == SI->getDefaultDest()) | |||
| 5986 | continue; | |||
| 5987 | Succ->removePredecessor(BB); | |||
| 5988 | RemovedSuccessors.insert(Succ); | |||
| 5989 | } | |||
| 5990 | SI->eraseFromParent(); | |||
| 5991 | ||||
| 5992 | if (DTU) { | |||
| 5993 | for (BasicBlock *RemovedSuccessor : RemovedSuccessors) | |||
| 5994 | Updates.push_back({DominatorTree::Delete, BB, RemovedSuccessor}); | |||
| 5995 | DTU->applyUpdates(Updates); | |||
| 5996 | } | |||
| 5997 | ||||
| 5998 | ++NumLookupTables; | |||
| 5999 | if (NeedMask) | |||
| 6000 | ++NumLookupTablesHoles; | |||
| 6001 | return true; | |||
| 6002 | } | |||
| 6003 | ||||
| 6004 |