lib/Transforms/Utils/LowerSwitch.cpp

   1 //===- LowerSwitch.cpp - Eliminate Switch instructions --------------------===//
   2 //
   3 //                     The LLVM Compiler Infrastructure
   4 //
   5 // This file is distributed under the University of Illinois Open Source
   6 // License. See LICENSE.TXT for details.
   7 //
   8 //===----------------------------------------------------------------------===//
   9 //
  10 // The LowerSwitch transformation rewrites switch instructions with a sequence
  11 // of branches, which allows targets to get away with not implementing the
  12 // switch instruction until it is convenient.
  13 //
  14 //===----------------------------------------------------------------------===//
  15
  16 #include "llvm/ADT/STLExtras.h"
  17 #include "llvm/IR/CFG.h"
  18 #include "llvm/IR/Constants.h"
  19 #include "llvm/IR/Function.h"
  20 #include "llvm/IR/Instructions.h"
  21 #include "llvm/IR/LLVMContext.h"
  22 #include "llvm/Pass.h"
  23 #include "llvm/Support/Compiler.h"
  24 #include "llvm/Support/Debug.h"
  25 #include "llvm/Support/raw_ostream.h"
  26 #include "llvm/Transforms/Scalar.h"
  27 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
  28 #include "llvm/Transforms/Utils/UnifyFunctionExitNodes.h"
  29 #include <algorithm>
  30 using namespace llvm;
  31
  32 #define DEBUG_TYPE "lower-switch"
  33
  34 namespace {
  35   struct IntRange {
  36     int64_t Low, High;
  37   };
  38   // Return true iff R is covered by Ranges.
  39   static bool IsInRanges(const IntRange &R,
  40                          const std::vector<IntRange> &Ranges) {
  41     // Note: Ranges must be sorted, non-overlapping and non-adjacent.
  42
  43     // Find the first range whose High field is >= R.High,
  44     // then check if the Low field is <= R.Low. If so, we
  45     // have a Range that covers R.
  46     auto I = std::lower_bound(
  47         Ranges.begin(), Ranges.end(), R,
  48         [](const IntRange &A, const IntRange &B) { return A.High < B.High; });
  49     return I != Ranges.end() && I->Low <= R.Low;
  50   }
  51
  52   /// Replace all SwitchInst instructions with chained branch instructions.
  53   class LowerSwitch : public FunctionPass {
  54   public:
  55     static char ID; // Pass identification, replacement for typeid
  56     LowerSwitch() : FunctionPass(ID) {
  57       initializeLowerSwitchPass(*PassRegistry::getPassRegistry());
  58     }
  59
  60     bool runOnFunction(Function &F) override;
  61
  62     struct CaseRange {
  63       ConstantInt* Low;
  64       ConstantInt* High;
  65       BasicBlock* BB;
  66
  67       CaseRange(ConstantInt *low, ConstantInt *high, BasicBlock *bb)
  68           : Low(low), High(high), BB(bb) {}
  69     };
  70
  71     typedef std::vector<CaseRange> CaseVector;
  72     typedef std::vector<CaseRange>::iterator CaseItr;
  73   private:
  74     void processSwitchInst(SwitchInst *SI, SmallPtrSetImpl<BasicBlock*> &DeleteList);
  75
  76     BasicBlock *switchConvert(CaseItr Begin, CaseItr End,
  77                               ConstantInt *LowerBound, ConstantInt *UpperBound,
  78                               Value *Val, BasicBlock *Predecessor,
  79                               BasicBlock *OrigBlock, BasicBlock *Default,
  80                               const std::vector<IntRange> &UnreachableRanges);
  81     BasicBlock *newLeafBlock(CaseRange &Leaf, Value *Val, BasicBlock *OrigBlock,
  82                              BasicBlock *Default);
  83     unsigned Clusterify(CaseVector &Cases, SwitchInst *SI);
  84   };
  85
  86   /// The comparison function for sorting the switch case values in the vector.
  87   /// WARNING: Case ranges should be disjoint!
  88   struct CaseCmp {
  89     bool operator () (const LowerSwitch::CaseRange& C1,
  90                       const LowerSwitch::CaseRange& C2) {
  91
  92       const ConstantInt* CI1 = cast<const ConstantInt>(C1.Low);
  93       const ConstantInt* CI2 = cast<const ConstantInt>(C2.High);
  94       return CI1->getValue().slt(CI2->getValue());
  95     }
  96   };
  97 }
  98
  99 char LowerSwitch::ID = 0;
 100 INITIALIZE_PASS(LowerSwitch, "lowerswitch",
 101                 "Lower SwitchInst's to branches", false, false)
 102
 103 // Publicly exposed interface to pass...
 104 char &llvm::LowerSwitchID = LowerSwitch::ID;
 105 // createLowerSwitchPass - Interface to this file...
 106 FunctionPass *llvm::createLowerSwitchPass() {
 107   return new LowerSwitch();
 108 }
 109
 110 bool LowerSwitch::runOnFunction(Function &F) {
 111   bool Changed = false;
 112   SmallPtrSet<BasicBlock*, 8> DeleteList;
 113
 114   for (Function::iterator I = F.begin(), E = F.end(); I != E; ) {
 115     BasicBlock *Cur = &*I++; // Advance over block so we don't traverse new blocks
 116
 117     // If the block is a dead Default block that will be deleted later, don't
 118     // waste time processing it.
 119     if (DeleteList.count(Cur))
 120       continue;
 121
 122     if (SwitchInst *SI = dyn_cast<SwitchInst>(Cur->getTerminator())) {
 123       Changed = true;
 124       processSwitchInst(SI, DeleteList);
 125     }
 126   }
 127
 128   for (BasicBlock* BB: DeleteList) {
 129     DeleteDeadBlock(BB);
 130   }
 131
 132   return Changed;
 133 }
 134
 135 /// Used for debugging purposes.
 136 static raw_ostream& operator<<(raw_ostream &O,
 137                                const LowerSwitch::CaseVector &C)
 138     LLVM_ATTRIBUTE_USED;
 139 static raw_ostream& operator<<(raw_ostream &O,
 140                                const LowerSwitch::CaseVector &C) {
 141   O << "[";
 142
 143   for (LowerSwitch::CaseVector::const_iterator B = C.begin(),
 144          E = C.end(); B != E; ) {
 145     O << *B->Low << " -" << *B->High;
 146     if (++B != E) O << ", ";
 147   }
 148
 149   return O << "]";
 150 }
 151
 152 /// \brief Update the first occurrence of the "switch statement" BB in the PHI
 153 /// node with the "new" BB. The other occurrences will:
 154 ///
 155 /// 1) Be updated by subsequent calls to this function.  Switch statements may
 156 /// have more than one outcoming edge into the same BB if they all have the same
 157 /// value. When the switch statement is converted these incoming edges are now
 158 /// coming from multiple BBs.
 159 /// 2) Removed if subsequent incoming values now share the same case, i.e.,
 160 /// multiple outcome edges are condensed into one. This is necessary to keep the
 161 /// number of phi values equal to the number of branches to SuccBB.
 162 static void fixPhis(BasicBlock *SuccBB, BasicBlock *OrigBB, BasicBlock *NewBB,
 163                     unsigned NumMergedCases) {
 164   for (BasicBlock::iterator I = SuccBB->begin(),
 165                             IE = SuccBB->getFirstNonPHI()->getIterator();
 166        I != IE; ++I) {
 167     PHINode *PN = cast<PHINode>(I);
 168
 169     // Only update the first occurrence.
 170     unsigned Idx = 0, E = PN->getNumIncomingValues();
 171     unsigned LocalNumMergedCases = NumMergedCases;
 172     for (; Idx != E; ++Idx) {
 173       if (PN->getIncomingBlock(Idx) == OrigBB) {
 174         PN->setIncomingBlock(Idx, NewBB);
 175         break;
 176       }
 177     }
 178
 179     // Remove additional occurrences coming from condensed cases and keep the
 180     // number of incoming values equal to the number of branches to SuccBB.
 181     SmallVector<unsigned, 8> Indices;
 182     for (++Idx; LocalNumMergedCases > 0 && Idx < E; ++Idx)
 183       if (PN->getIncomingBlock(Idx) == OrigBB) {
 184         Indices.push_back(Idx);
 185         LocalNumMergedCases--;
 186       }
 187     // Remove incoming values in the reverse order to prevent invalidating
 188     // *successive* index.
 189     for (unsigned III : reverse(Indices))
 190       PN->removeIncomingValue(III);
 191   }
 192 }
 193
 194 /// Convert the switch statement into a binary lookup of the case values.
 195 /// The function recursively builds this tree. LowerBound and UpperBound are
 196 /// used to keep track of the bounds for Val that have already been checked by
 197 /// a block emitted by one of the previous calls to switchConvert in the call
 198 /// stack.
 199 BasicBlock *
 200 LowerSwitch::switchConvert(CaseItr Begin, CaseItr End, ConstantInt *LowerBound,
 201                            ConstantInt *UpperBound, Value *Val,
 202                            BasicBlock *Predecessor, BasicBlock *OrigBlock,
 203                            BasicBlock *Default,
 204                            const std::vector<IntRange> &UnreachableRanges) {
 205   unsigned Size = End - Begin;
 206
 207   if (Size == 1) {
 208     // Check if the Case Range is perfectly squeezed in between
 209     // already checked Upper and Lower bounds. If it is then we can avoid
 210     // emitting the code that checks if the value actually falls in the range
 211     // because the bounds already tell us so.
 212     if (Begin->Low == LowerBound && Begin->High == UpperBound) {
 213       unsigned NumMergedCases = 0;
 214       if (LowerBound && UpperBound)
 215         NumMergedCases =
 216             UpperBound->getSExtValue() - LowerBound->getSExtValue();
 217       fixPhis(Begin->BB, OrigBlock, Predecessor, NumMergedCases);
 218       return Begin->BB;
 219     }
 220     return newLeafBlock(*Begin, Val, OrigBlock, Default);
 221   }
 222
 223   unsigned Mid = Size / 2;
 224   std::vector<CaseRange> LHS(Begin, Begin + Mid);
 225   DEBUG(dbgs() << "LHS: " << LHS << "\n");
 226   std::vector<CaseRange> RHS(Begin + Mid, End);
 227   DEBUG(dbgs() << "RHS: " << RHS << "\n");
 228
 229   CaseRange &Pivot = *(Begin + Mid);
 230   DEBUG(dbgs() << "Pivot ==> "
 231                << Pivot.Low->getValue()
 232                << " -" << Pivot.High->getValue() << "\n");
 233
 234   // NewLowerBound here should never be the integer minimal value.
 235   // This is because it is computed from a case range that is never
 236   // the smallest, so there is always a case range that has at least
 237   // a smaller value.
 238   ConstantInt *NewLowerBound = Pivot.Low;
 239
 240   // Because NewLowerBound is never the smallest representable integer
 241   // it is safe here to subtract one.
 242   ConstantInt *NewUpperBound = ConstantInt::get(NewLowerBound->getContext(),
 243                                                 NewLowerBound->getValue() - 1);
 244
 245   if (!UnreachableRanges.empty()) {
 246     // Check if the gap between LHS's highest and NewLowerBound is unreachable.
 247     int64_t GapLow = LHS.back().High->getSExtValue() + 1;
 248     int64_t GapHigh = NewLowerBound->getSExtValue() - 1;
 249     IntRange Gap = { GapLow, GapHigh };
 250     if (GapHigh >= GapLow && IsInRanges(Gap, UnreachableRanges))
 251       NewUpperBound = LHS.back().High;
 252   }
 253
 254   DEBUG(dbgs() << "LHS Bounds ==> ";
 255         if (LowerBound) {
 256           dbgs() << LowerBound->getSExtValue();
 257         } else {
 258           dbgs() << "NONE";
 259         }
 260         dbgs() << " - " << NewUpperBound->getSExtValue() << "\n";
 261         dbgs() << "RHS Bounds ==> ";
 262         dbgs() << NewLowerBound->getSExtValue() << " - ";
 263         if (UpperBound) {
 264           dbgs() << UpperBound->getSExtValue() << "\n";
 265         } else {
 266           dbgs() << "NONE\n";
 267         });
 268
 269   // Create a new node that checks if the value is < pivot. Go to the
 270   // left branch if it is and right branch if not.
 271   Function* F = OrigBlock->getParent();
 272   BasicBlock* NewNode = BasicBlock::Create(Val->getContext(), "NodeBlock");
 273
 274   ICmpInst* Comp = new ICmpInst(ICmpInst::ICMP_SLT,
 275                                 Val, Pivot.Low, "Pivot");
 276
 277   BasicBlock *LBranch = switchConvert(LHS.begin(), LHS.end(), LowerBound,
 278                                       NewUpperBound, Val, NewNode, OrigBlock,
 279                                       Default, UnreachableRanges);
 280   BasicBlock *RBranch = switchConvert(RHS.begin(), RHS.end(), NewLowerBound,
 281                                       UpperBound, Val, NewNode, OrigBlock,
 282                                       Default, UnreachableRanges);
 283
 284   F->getBasicBlockList().insert(++OrigBlock->getIterator(), NewNode);
 285   NewNode->getInstList().push_back(Comp);
 286
 287   BranchInst::Create(LBranch, RBranch, Comp, NewNode);
 288   return NewNode;
 289 }
 290
 291 /// Create a new leaf block for the binary lookup tree. It checks if the
 292 /// switch's value == the case's value. If not, then it jumps to the default
 293 /// branch. At this point in the tree, the value can't be another valid case
 294 /// value, so the jump to the "default" branch is warranted.
 295 BasicBlock* LowerSwitch::newLeafBlock(CaseRange& Leaf, Value* Val,
 296                                       BasicBlock* OrigBlock,
 297                                       BasicBlock* Default)
 298 {
 299   Function* F = OrigBlock->getParent();
 300   BasicBlock* NewLeaf = BasicBlock::Create(Val->getContext(), "LeafBlock");
 301   F->getBasicBlockList().insert(++OrigBlock->getIterator(), NewLeaf);
 302
 303   // Emit comparison
 304   ICmpInst* Comp = nullptr;
 305   if (Leaf.Low == Leaf.High) {
 306     // Make the seteq instruction...
 307     Comp = new ICmpInst(*NewLeaf, ICmpInst::ICMP_EQ, Val,
 308                         Leaf.Low, "SwitchLeaf");
 309   } else {
 310     // Make range comparison
 311     if (Leaf.Low->isMinValue(true /*isSigned*/)) {
 312       // Val >= Min && Val <= Hi --> Val <= Hi
 313       Comp = new ICmpInst(*NewLeaf, ICmpInst::ICMP_SLE, Val, Leaf.High,
 314                           "SwitchLeaf");
 315     } else if (Leaf.Low->isZero()) {
 316       // Val >= 0 && Val <= Hi --> Val <=u Hi
 317       Comp = new ICmpInst(*NewLeaf, ICmpInst::ICMP_ULE, Val, Leaf.High,
 318                           "SwitchLeaf");
 319     } else {
 320       // Emit V-Lo <=u Hi-Lo
 321       Constant* NegLo = ConstantExpr::getNeg(Leaf.Low);
 322       Instruction* Add = BinaryOperator::CreateAdd(Val, NegLo,
 323                                                    Val->getName()+".off",
 324                                                    NewLeaf);
 325       Constant *UpperBound = ConstantExpr::getAdd(NegLo, Leaf.High);
 326       Comp = new ICmpInst(*NewLeaf, ICmpInst::ICMP_ULE, Add, UpperBound,
 327                           "SwitchLeaf");
 328     }
 329   }
 330
 331   // Make the conditional branch...
 332   BasicBlock* Succ = Leaf.BB;
 333   BranchInst::Create(Succ, Default, Comp, NewLeaf);
 334
 335   // If there were any PHI nodes in this successor, rewrite one entry
 336   // from OrigBlock to come from NewLeaf.
 337   for (BasicBlock::iterator I = Succ->begin(); isa<PHINode>(I); ++I) {
 338     PHINode* PN = cast<PHINode>(I);
 339     // Remove all but one incoming entries from the cluster
 340     uint64_t Range = Leaf.High->getSExtValue() -
 341                      Leaf.Low->getSExtValue();
 342     for (uint64_t j = 0; j < Range; ++j) {
 343       PN->removeIncomingValue(OrigBlock);
 344     }
 345
 346     int BlockIdx = PN->getBasicBlockIndex(OrigBlock);
 347     assert(BlockIdx != -1 && "Switch didn't go to this successor??");
 348     PN->setIncomingBlock((unsigned)BlockIdx, NewLeaf);
 349   }
 350
 351   return NewLeaf;
 352 }
 353
 354 /// Transform simple list of Cases into list of CaseRange's.
 355 unsigned LowerSwitch::Clusterify(CaseVector& Cases, SwitchInst *SI) {
 356   unsigned numCmps = 0;
 357
 358   // Start with "simple" cases
 359   for (auto Case : SI->cases())
 360     Cases.push_back(CaseRange(Case.getCaseValue(), Case.getCaseValue(),
 361                               Case.getCaseSuccessor()));
 362
 363   std::sort(Cases.begin(), Cases.end(), CaseCmp());
 364
 365   // Merge case into clusters
 366   if (Cases.size() >= 2) {
 367     CaseItr I = Cases.begin();
 368     for (CaseItr J = std::next(I), E = Cases.end(); J != E; ++J) {
 369       int64_t nextValue = J->Low->getSExtValue();
 370       int64_t currentValue = I->High->getSExtValue();
 371       BasicBlock* nextBB = J->BB;
 372       BasicBlock* currentBB = I->BB;
 373
 374       // If the two neighboring cases go to the same destination, merge them
 375       // into a single case.
 376       assert(nextValue > currentValue && "Cases should be strictly ascending");
 377       if ((nextValue == currentValue + 1) && (currentBB == nextBB)) {
 378         I->High = J->High;
 379         // FIXME: Combine branch weights.
 380       } else if (++I != J) {
 381         *I = *J;
 382       }
 383     }
 384     Cases.erase(std::next(I), Cases.end());
 385   }
 386
 387   for (CaseItr I=Cases.begin(), E=Cases.end(); I!=E; ++I, ++numCmps) {
 388     if (I->Low != I->High)
 389       // A range counts double, since it requires two compares.
 390       ++numCmps;
 391   }
 392
 393   return numCmps;
 394 }
 395
 396 /// Replace the specified switch instruction with a sequence of chained if-then
 397 /// insts in a balanced binary search.
 398 void LowerSwitch::processSwitchInst(SwitchInst *SI,
 399                                     SmallPtrSetImpl<BasicBlock*> &DeleteList) {
 400   BasicBlock *CurBlock = SI->getParent();
 401   BasicBlock *OrigBlock = CurBlock;
 402   Function *F = CurBlock->getParent();
 403   Value *Val = SI->getCondition();  // The value we are switching on...
 404   BasicBlock* Default = SI->getDefaultDest();
 405
 406   // Don't handle unreachable blocks. If there are successors with phis, this
 407   // would leave them behind with missing predecessors.
 408   if ((CurBlock != &F->getEntryBlock() && pred_empty(CurBlock)) ||
 409       CurBlock->getSinglePredecessor() == CurBlock) {
 410     DeleteList.insert(CurBlock);
 411     return;
 412   }
 413
 414   // If there is only the default destination, just branch.
 415   if (!SI->getNumCases()) {
 416     BranchInst::Create(Default, CurBlock);
 417     SI->eraseFromParent();
 418     return;
 419   }
 420
 421   // Prepare cases vector.
 422   CaseVector Cases;
 423   unsigned numCmps = Clusterify(Cases, SI);
 424   DEBUG(dbgs() << "Clusterify finished. Total clusters: " << Cases.size()
 425                << ". Total compares: " << numCmps << "\n");
 426   DEBUG(dbgs() << "Cases: " << Cases << "\n");
 427   (void)numCmps;
 428
 429   ConstantInt *LowerBound = nullptr;
 430   ConstantInt *UpperBound = nullptr;
 431   std::vector<IntRange> UnreachableRanges;
 432
 433   if (isa<UnreachableInst>(Default->getFirstNonPHIOrDbg())) {
 434     // Make the bounds tightly fitted around the case value range, because we
 435     // know that the value passed to the switch must be exactly one of the case
 436     // values.
 437     assert(!Cases.empty());
 438     LowerBound = Cases.front().Low;
 439     UpperBound = Cases.back().High;
 440
 441     DenseMap<BasicBlock *, unsigned> Popularity;
 442     unsigned MaxPop = 0;
 443     BasicBlock *PopSucc = nullptr;
 444
 445     IntRange R = { INT64_MIN, INT64_MAX };
 446     UnreachableRanges.push_back(R);
 447     for (const auto &I : Cases) {
 448       int64_t Low = I.Low->getSExtValue();
 449       int64_t High = I.High->getSExtValue();
 450
 451       IntRange &LastRange = UnreachableRanges.back();
 452       if (LastRange.Low == Low) {
 453         // There is nothing left of the previous range.
 454         UnreachableRanges.pop_back();
 455       } else {
 456         // Terminate the previous range.
 457         assert(Low > LastRange.Low);
 458         LastRange.High = Low - 1;
 459       }
 460       if (High != INT64_MAX) {
 461         IntRange R = { High + 1, INT64_MAX };
 462         UnreachableRanges.push_back(R);
 463       }
 464
 465       // Count popularity.
 466       int64_t N = High - Low + 1;
 467       unsigned &Pop = Popularity[I.BB];
 468       if ((Pop += N) > MaxPop) {
 469         MaxPop = Pop;
 470         PopSucc = I.BB;
 471       }
 472     }
 473 #ifndef NDEBUG
 474     /* UnreachableRanges should be sorted and the ranges non-adjacent. */
 475     for (auto I = UnreachableRanges.begin(), E = UnreachableRanges.end();
 476          I != E; ++I) {
 477       assert(I->Low <= I->High);
 478       auto Next = I + 1;
 479       if (Next != E) {
 480         assert(Next->Low > I->High);
 481       }
 482     }
 483 #endif
 484
 485     // Use the most popular block as the new default, reducing the number of
 486     // cases.
 487     assert(MaxPop > 0 && PopSucc);
 488     Default = PopSucc;
 489     Cases.erase(
 490         remove_if(Cases,
 491                   [PopSucc](const CaseRange &R) { return R.BB == PopSucc; }),
 492         Cases.end());
 493
 494     // If there are no cases left, just branch.
 495     if (Cases.empty()) {
 496       BranchInst::Create(Default, CurBlock);
 497       SI->eraseFromParent();
 498       return;
 499     }
 500   }
 501
 502   // Create a new, empty default block so that the new hierarchy of
 503   // if-then statements go to this and the PHI nodes are happy.
 504   BasicBlock *NewDefault = BasicBlock::Create(SI->getContext(), "NewDefault");
 505   F->getBasicBlockList().insert(Default->getIterator(), NewDefault);
 506   BranchInst::Create(Default, NewDefault);
 507
 508   // If there is an entry in any PHI nodes for the default edge, make sure
 509   // to update them as well.
 510   for (BasicBlock::iterator I = Default->begin(); isa<PHINode>(I); ++I) {
 511     PHINode *PN = cast<PHINode>(I);
 512     int BlockIdx = PN->getBasicBlockIndex(OrigBlock);
 513     assert(BlockIdx != -1 && "Switch didn't go to this successor??");
 514     PN->setIncomingBlock((unsigned)BlockIdx, NewDefault);
 515   }
 516
 517   BasicBlock *SwitchBlock =
 518       switchConvert(Cases.begin(), Cases.end(), LowerBound, UpperBound, Val,
 519                     OrigBlock, OrigBlock, NewDefault, UnreachableRanges);
 520
 521   // Branch to our shiny new if-then stuff...
 522   BranchInst::Create(SwitchBlock, OrigBlock);
 523
 524   // We are now done with the switch instruction, delete it.
 525   BasicBlock *OldDefault = SI->getDefaultDest();
 526   CurBlock->getInstList().erase(SI);
 527
 528   // If the Default block has no more predecessors just add it to DeleteList.
 529   if (pred_begin(OldDefault) == pred_end(OldDefault))
 530     DeleteList.insert(OldDefault);
 531 }