1a1c712faSKit Barton //===- PPCBoolRetToInt.cpp - Convert bool literals to i32 if they are returned ==// 2a1c712faSKit Barton // 3a1c712faSKit Barton // The LLVM Compiler Infrastructure 4a1c712faSKit Barton // 5a1c712faSKit Barton // This file is distributed under the University of Illinois Open Source 6a1c712faSKit Barton // License. See LICENSE.TXT for details. 7a1c712faSKit Barton // 8a1c712faSKit Barton //===----------------------------------------------------------------------===// 9a1c712faSKit Barton // 10a1c712faSKit Barton // This file implements converting i1 values to i32 if they could be more 11a1c712faSKit Barton // profitably allocated as GPRs rather than CRs. This pass will become totally 12a1c712faSKit Barton // unnecessary if Register Bank Allocation and Global Instruction Selection ever 13a1c712faSKit Barton // go upstream. 14a1c712faSKit Barton // 15a1c712faSKit Barton // Presently, the pass converts i1 Constants, and Arguments to i32 if the 16a1c712faSKit Barton // transitive closure of their uses includes only PHINodes, CallInsts, and 17a1c712faSKit Barton // ReturnInsts. The rational is that arguments are generally passed and returned 18a1c712faSKit Barton // in GPRs rather than CRs, so casting them to i32 at the LLVM IR level will 19a1c712faSKit Barton // actually save casts at the Machine Instruction level. 20a1c712faSKit Barton // 21a1c712faSKit Barton // It might be useful to expand this pass to add bit-wise operations to the list 22a1c712faSKit Barton // of safe transitive closure types. Also, we miss some opportunities when LLVM 23a1c712faSKit Barton // represents logical AND and OR operations with control flow rather than data 24a1c712faSKit Barton // flow. For example by lowering the expression: return (A && B && C) 25a1c712faSKit Barton // 26a1c712faSKit Barton // as: return A ? true : B && C. 27a1c712faSKit Barton // 28a1c712faSKit Barton // There's code in SimplifyCFG that code be used to turn control flow in data 29a1c712faSKit Barton // flow using SelectInsts. Selects are slow on some architectures (P7/P8), so 30a1c712faSKit Barton // this probably isn't good in general, but for the special case of i1, the 31a1c712faSKit Barton // Selects could be further lowered to bit operations that are fast everywhere. 32a1c712faSKit Barton // 33a1c712faSKit Barton //===----------------------------------------------------------------------===// 34a1c712faSKit Barton 35a1c712faSKit Barton #include "PPC.h" 36a1c712faSKit Barton #include "llvm/Transforms/Scalar.h" 37a1c712faSKit Barton #include "llvm/ADT/SmallPtrSet.h" 38a1c712faSKit Barton #include "llvm/ADT/Statistic.h" 39a1c712faSKit Barton #include "llvm/IR/Constants.h" 40a1c712faSKit Barton #include "llvm/IR/Dominators.h" 41a1c712faSKit Barton #include "llvm/IR/Instructions.h" 42a1c712faSKit Barton #include "llvm/IR/IntrinsicInst.h" 43a1c712faSKit Barton #include "llvm/Support/raw_ostream.h" 44a1c712faSKit Barton #include "llvm/Pass.h" 45a1c712faSKit Barton 46a1c712faSKit Barton using namespace llvm; 47a1c712faSKit Barton 48a1c712faSKit Barton namespace { 49a1c712faSKit Barton 50a1c712faSKit Barton #define DEBUG_TYPE "bool-ret-to-int" 51a1c712faSKit Barton 52a1c712faSKit Barton STATISTIC(NumBoolRetPromotion, 53a1c712faSKit Barton "Number of times a bool feeding a RetInst was promoted to an int"); 54a1c712faSKit Barton STATISTIC(NumBoolCallPromotion, 55a1c712faSKit Barton "Number of times a bool feeding a CallInst was promoted to an int"); 56a1c712faSKit Barton STATISTIC(NumBoolToIntPromotion, 57a1c712faSKit Barton "Total number of times a bool was promoted to an int"); 58a1c712faSKit Barton 59a1c712faSKit Barton class PPCBoolRetToInt : public FunctionPass { 60a1c712faSKit Barton 61a1c712faSKit Barton static SmallPtrSet<Value *, 8> findAllDefs(Value *V) { 62a1c712faSKit Barton SmallPtrSet<Value *, 8> Defs; 63a1c712faSKit Barton SmallVector<Value *, 8> WorkList; 64a1c712faSKit Barton WorkList.push_back(V); 65a1c712faSKit Barton Defs.insert(V); 66a1c712faSKit Barton while (!WorkList.empty()) { 67a1c712faSKit Barton Value *Curr = WorkList.back(); 68a1c712faSKit Barton WorkList.pop_back(); 69*9584d18dSGuozhi Wei User *CurrUser = dyn_cast<User>(Curr); 70*9584d18dSGuozhi Wei // Operands of CallInst are skipped because they may not be Bool type, 71*9584d18dSGuozhi Wei // and their positions are defined by ABI. 72*9584d18dSGuozhi Wei if (CurrUser && !isa<CallInst>(Curr)) 73a1c712faSKit Barton for (auto &Op : CurrUser->operands()) 74a1c712faSKit Barton if (Defs.insert(Op).second) 75a1c712faSKit Barton WorkList.push_back(Op); 76a1c712faSKit Barton } 77a1c712faSKit Barton return Defs; 78a1c712faSKit Barton } 79a1c712faSKit Barton 80a1c712faSKit Barton // Translate a i1 value to an equivalent i32 value: 81a1c712faSKit Barton static Value *translate(Value *V) { 82a1c712faSKit Barton Type *Int32Ty = Type::getInt32Ty(V->getContext()); 83a1c712faSKit Barton if (Constant *C = dyn_cast<Constant>(V)) 84a1c712faSKit Barton return ConstantExpr::getZExt(C, Int32Ty); 85a1c712faSKit Barton if (PHINode *P = dyn_cast<PHINode>(V)) { 86a1c712faSKit Barton // Temporarily set the operands to 0. We'll fix this later in 87a1c712faSKit Barton // runOnUse. 88a1c712faSKit Barton Value *Zero = Constant::getNullValue(Int32Ty); 89a1c712faSKit Barton PHINode *Q = 90a1c712faSKit Barton PHINode::Create(Int32Ty, P->getNumIncomingValues(), P->getName(), P); 91a1c712faSKit Barton for (unsigned i = 0; i < P->getNumOperands(); ++i) 92a1c712faSKit Barton Q->addIncoming(Zero, P->getIncomingBlock(i)); 93a1c712faSKit Barton return Q; 94a1c712faSKit Barton } 95a1c712faSKit Barton 96a1c712faSKit Barton Argument *A = dyn_cast<Argument>(V); 97a1c712faSKit Barton Instruction *I = dyn_cast<Instruction>(V); 98a1c712faSKit Barton assert((A || I) && "Unknown value type"); 99a1c712faSKit Barton 100a1c712faSKit Barton auto InstPt = 101a1c712faSKit Barton A ? &*A->getParent()->getEntryBlock().begin() : I->getNextNode(); 102a1c712faSKit Barton return new ZExtInst(V, Int32Ty, "", InstPt); 103a1c712faSKit Barton } 104a1c712faSKit Barton 105a1c712faSKit Barton typedef SmallPtrSet<const PHINode *, 8> PHINodeSet; 106a1c712faSKit Barton 107a1c712faSKit Barton // A PHINode is Promotable if: 108a1c712faSKit Barton // 1. Its type is i1 AND 109a1c712faSKit Barton // 2. All of its uses are ReturnInt, CallInst, PHINode, or DbgInfoIntrinsic 110a1c712faSKit Barton // AND 111a1c712faSKit Barton // 3. All of its operands are Constant or Argument or 112a1c712faSKit Barton // CallInst or PHINode AND 113a1c712faSKit Barton // 4. All of its PHINode uses are Promotable AND 114a1c712faSKit Barton // 5. All of its PHINode operands are Promotable 115a1c712faSKit Barton static PHINodeSet getPromotablePHINodes(const Function &F) { 116a1c712faSKit Barton PHINodeSet Promotable; 117a1c712faSKit Barton // Condition 1 118a1c712faSKit Barton for (auto &BB : F) 119a1c712faSKit Barton for (auto &I : BB) 120a1c712faSKit Barton if (const PHINode *P = dyn_cast<PHINode>(&I)) 121a1c712faSKit Barton if (P->getType()->isIntegerTy(1)) 122a1c712faSKit Barton Promotable.insert(P); 123a1c712faSKit Barton 124a1c712faSKit Barton SmallVector<const PHINode *, 8> ToRemove; 125451f54cfSBenjamin Kramer for (const PHINode *P : Promotable) { 126a1c712faSKit Barton // Condition 2 and 3 127a1c712faSKit Barton auto IsValidUser = [] (const Value *V) -> bool { 128a1c712faSKit Barton return isa<ReturnInst>(V) || isa<CallInst>(V) || isa<PHINode>(V) || 129a1c712faSKit Barton isa<DbgInfoIntrinsic>(V); 130a1c712faSKit Barton }; 131a1c712faSKit Barton auto IsValidOperand = [] (const Value *V) -> bool { 132a1c712faSKit Barton return isa<Constant>(V) || isa<Argument>(V) || isa<CallInst>(V) || 133a1c712faSKit Barton isa<PHINode>(V); 134a1c712faSKit Barton }; 135a1c712faSKit Barton const auto &Users = P->users(); 136a1c712faSKit Barton const auto &Operands = P->operands(); 137a1c712faSKit Barton if (!std::all_of(Users.begin(), Users.end(), IsValidUser) || 138a1c712faSKit Barton !std::all_of(Operands.begin(), Operands.end(), IsValidOperand)) 139a1c712faSKit Barton ToRemove.push_back(P); 140a1c712faSKit Barton } 141a1c712faSKit Barton 142a1c712faSKit Barton // Iterate to convergence 143a1c712faSKit Barton auto IsPromotable = [&Promotable] (const Value *V) -> bool { 144a1c712faSKit Barton const PHINode *Phi = dyn_cast<PHINode>(V); 145a1c712faSKit Barton return !Phi || Promotable.count(Phi); 146a1c712faSKit Barton }; 147a1c712faSKit Barton while (!ToRemove.empty()) { 148a1c712faSKit Barton for (auto &User : ToRemove) 149a1c712faSKit Barton Promotable.erase(User); 150a1c712faSKit Barton ToRemove.clear(); 151a1c712faSKit Barton 152451f54cfSBenjamin Kramer for (const PHINode *P : Promotable) { 153a1c712faSKit Barton // Condition 4 and 5 154a1c712faSKit Barton const auto &Users = P->users(); 155a1c712faSKit Barton const auto &Operands = P->operands(); 156a1c712faSKit Barton if (!std::all_of(Users.begin(), Users.end(), IsPromotable) || 157a1c712faSKit Barton !std::all_of(Operands.begin(), Operands.end(), IsPromotable)) 158a1c712faSKit Barton ToRemove.push_back(P); 159a1c712faSKit Barton } 160a1c712faSKit Barton } 161a1c712faSKit Barton 162a1c712faSKit Barton return Promotable; 163a1c712faSKit Barton } 164a1c712faSKit Barton 165a1c712faSKit Barton typedef DenseMap<Value *, Value *> B2IMap; 166a1c712faSKit Barton 167a1c712faSKit Barton public: 168a1c712faSKit Barton static char ID; 169a1c712faSKit Barton PPCBoolRetToInt() : FunctionPass(ID) { 170a1c712faSKit Barton initializePPCBoolRetToIntPass(*PassRegistry::getPassRegistry()); 171a1c712faSKit Barton } 172a1c712faSKit Barton 173a1c712faSKit Barton bool runOnFunction(Function &F) { 174289bd5f6SAndrew Kaylor if (skipFunction(F)) 175289bd5f6SAndrew Kaylor return false; 176289bd5f6SAndrew Kaylor 177a1c712faSKit Barton PHINodeSet PromotablePHINodes = getPromotablePHINodes(F); 178a1c712faSKit Barton B2IMap Bool2IntMap; 179a1c712faSKit Barton bool Changed = false; 180a1c712faSKit Barton for (auto &BB : F) { 181a1c712faSKit Barton for (auto &I : BB) { 182a1c712faSKit Barton if (ReturnInst *R = dyn_cast<ReturnInst>(&I)) 183a1c712faSKit Barton if (F.getReturnType()->isIntegerTy(1)) 184a1c712faSKit Barton Changed |= 185a1c712faSKit Barton runOnUse(R->getOperandUse(0), PromotablePHINodes, Bool2IntMap); 186a1c712faSKit Barton 187a1c712faSKit Barton if (CallInst *CI = dyn_cast<CallInst>(&I)) 188a1c712faSKit Barton for (auto &U : CI->operands()) 189a1c712faSKit Barton if (U->getType()->isIntegerTy(1)) 190a1c712faSKit Barton Changed |= runOnUse(U, PromotablePHINodes, Bool2IntMap); 191a1c712faSKit Barton } 192a1c712faSKit Barton } 193a1c712faSKit Barton 194a1c712faSKit Barton return Changed; 195a1c712faSKit Barton } 196a1c712faSKit Barton 197a1c712faSKit Barton static bool runOnUse(Use &U, const PHINodeSet &PromotablePHINodes, 198a1c712faSKit Barton B2IMap &BoolToIntMap) { 199a1c712faSKit Barton auto Defs = findAllDefs(U); 200a1c712faSKit Barton 201a1c712faSKit Barton // If the values are all Constants or Arguments, don't bother 202a1c712faSKit Barton if (!std::any_of(Defs.begin(), Defs.end(), isa<Instruction, Value *>)) 203a1c712faSKit Barton return false; 204a1c712faSKit Barton 205*9584d18dSGuozhi Wei // Presently, we only know how to handle PHINode, Constant, Arguments and 206*9584d18dSGuozhi Wei // CallInst. Potentially, bitwise operations (AND, OR, XOR, NOT) and sign 207*9584d18dSGuozhi Wei // extension could also be handled in the future. 208451f54cfSBenjamin Kramer for (Value *V : Defs) 209*9584d18dSGuozhi Wei if (!isa<PHINode>(V) && !isa<Constant>(V) && 210*9584d18dSGuozhi Wei !isa<Argument>(V) && !isa<CallInst>(V)) 211a1c712faSKit Barton return false; 212a1c712faSKit Barton 213451f54cfSBenjamin Kramer for (Value *V : Defs) 214a1c712faSKit Barton if (const PHINode *P = dyn_cast<PHINode>(V)) 215a1c712faSKit Barton if (!PromotablePHINodes.count(P)) 216a1c712faSKit Barton return false; 217a1c712faSKit Barton 218a1c712faSKit Barton if (isa<ReturnInst>(U.getUser())) 219a1c712faSKit Barton ++NumBoolRetPromotion; 220a1c712faSKit Barton if (isa<CallInst>(U.getUser())) 221a1c712faSKit Barton ++NumBoolCallPromotion; 222a1c712faSKit Barton ++NumBoolToIntPromotion; 223a1c712faSKit Barton 224451f54cfSBenjamin Kramer for (Value *V : Defs) 225a1c712faSKit Barton if (!BoolToIntMap.count(V)) 226a1c712faSKit Barton BoolToIntMap[V] = translate(V); 227a1c712faSKit Barton 228*9584d18dSGuozhi Wei // Replace the operands of the translated instructions. They were set to 229a1c712faSKit Barton // zero in the translate function. 230a1c712faSKit Barton for (auto &Pair : BoolToIntMap) { 231a1c712faSKit Barton User *First = dyn_cast<User>(Pair.first); 232a1c712faSKit Barton User *Second = dyn_cast<User>(Pair.second); 233a1c712faSKit Barton assert((!First || Second) && "translated from user to non-user!?"); 234*9584d18dSGuozhi Wei // Operands of CallInst are skipped because they may not be Bool type, 235*9584d18dSGuozhi Wei // and their positions are defined by ABI. 236*9584d18dSGuozhi Wei if (First && !isa<CallInst>(First)) 237a1c712faSKit Barton for (unsigned i = 0; i < First->getNumOperands(); ++i) 238a1c712faSKit Barton Second->setOperand(i, BoolToIntMap[First->getOperand(i)]); 239a1c712faSKit Barton } 240a1c712faSKit Barton 241a1c712faSKit Barton Value *IntRetVal = BoolToIntMap[U]; 242a1c712faSKit Barton Type *Int1Ty = Type::getInt1Ty(U->getContext()); 243a1c712faSKit Barton Instruction *I = cast<Instruction>(U.getUser()); 244a1c712faSKit Barton Value *BackToBool = new TruncInst(IntRetVal, Int1Ty, "backToBool", I); 245a1c712faSKit Barton U.set(BackToBool); 246a1c712faSKit Barton 247a1c712faSKit Barton return true; 248a1c712faSKit Barton } 249a1c712faSKit Barton 250a1c712faSKit Barton void getAnalysisUsage(AnalysisUsage &AU) const { 251a1c712faSKit Barton AU.addPreserved<DominatorTreeWrapperPass>(); 252a1c712faSKit Barton FunctionPass::getAnalysisUsage(AU); 253a1c712faSKit Barton } 254a1c712faSKit Barton }; 255a1c712faSKit Barton } 256a1c712faSKit Barton 257a1c712faSKit Barton char PPCBoolRetToInt::ID = 0; 258a1c712faSKit Barton INITIALIZE_PASS(PPCBoolRetToInt, "bool-ret-to-int", 259a1c712faSKit Barton "Convert i1 constants to i32 if they are returned", 260a1c712faSKit Barton false, false) 261a1c712faSKit Barton 262a1c712faSKit Barton FunctionPass *llvm::createPPCBoolRetToIntPass() { return new PPCBoolRetToInt(); } 263