16a9226d9SEugene Zelenko //===- PPCBoolRetToInt.cpp ------------------------------------------------===// 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 // 10*f31c56dfSGuozhi Wei // This file implements converting i1 values to i32/i64 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 // 15*f31c56dfSGuozhi Wei // Presently, the pass converts i1 Constants, and Arguments to i32/i64 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 18*f31c56dfSGuozhi Wei // in GPRs rather than CRs, so casting them to i32/i64 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" 36*f31c56dfSGuozhi Wei #include "PPCTargetMachine.h" 376a9226d9SEugene Zelenko #include "llvm/ADT/DenseMap.h" 386bda14b3SChandler Carruth #include "llvm/ADT/STLExtras.h" 39a1c712faSKit Barton #include "llvm/ADT/SmallPtrSet.h" 406a9226d9SEugene Zelenko #include "llvm/ADT/SmallVector.h" 41a1c712faSKit Barton #include "llvm/ADT/Statistic.h" 426a9226d9SEugene Zelenko #include "llvm/IR/Argument.h" 43a1c712faSKit Barton #include "llvm/IR/Constants.h" 44a1c712faSKit Barton #include "llvm/IR/Dominators.h" 456a9226d9SEugene Zelenko #include "llvm/IR/Function.h" 466a9226d9SEugene Zelenko #include "llvm/IR/Instruction.h" 47a1c712faSKit Barton #include "llvm/IR/Instructions.h" 48a1c712faSKit Barton #include "llvm/IR/IntrinsicInst.h" 496a9226d9SEugene Zelenko #include "llvm/IR/OperandTraits.h" 506a9226d9SEugene Zelenko #include "llvm/IR/Type.h" 516a9226d9SEugene Zelenko #include "llvm/IR/Use.h" 526a9226d9SEugene Zelenko #include "llvm/IR/User.h" 536a9226d9SEugene Zelenko #include "llvm/IR/Value.h" 54a1c712faSKit Barton #include "llvm/Pass.h" 55*f31c56dfSGuozhi Wei #include "llvm/CodeGen/TargetPassConfig.h" 566bda14b3SChandler Carruth #include "llvm/Support/Casting.h" 576a9226d9SEugene Zelenko #include <cassert> 58a1c712faSKit Barton 59a1c712faSKit Barton using namespace llvm; 60a1c712faSKit Barton 61a1c712faSKit Barton namespace { 62a1c712faSKit Barton 63a1c712faSKit Barton #define DEBUG_TYPE "bool-ret-to-int" 64a1c712faSKit Barton 65a1c712faSKit Barton STATISTIC(NumBoolRetPromotion, 66a1c712faSKit Barton "Number of times a bool feeding a RetInst was promoted to an int"); 67a1c712faSKit Barton STATISTIC(NumBoolCallPromotion, 68a1c712faSKit Barton "Number of times a bool feeding a CallInst was promoted to an int"); 69a1c712faSKit Barton STATISTIC(NumBoolToIntPromotion, 70a1c712faSKit Barton "Total number of times a bool was promoted to an int"); 71a1c712faSKit Barton 72a1c712faSKit Barton class PPCBoolRetToInt : public FunctionPass { 73a1c712faSKit Barton static SmallPtrSet<Value *, 8> findAllDefs(Value *V) { 74a1c712faSKit Barton SmallPtrSet<Value *, 8> Defs; 75a1c712faSKit Barton SmallVector<Value *, 8> WorkList; 76a1c712faSKit Barton WorkList.push_back(V); 77a1c712faSKit Barton Defs.insert(V); 78a1c712faSKit Barton while (!WorkList.empty()) { 79a1c712faSKit Barton Value *Curr = WorkList.back(); 80a1c712faSKit Barton WorkList.pop_back(); 816a9226d9SEugene Zelenko auto *CurrUser = dyn_cast<User>(Curr); 829584d18dSGuozhi Wei // Operands of CallInst are skipped because they may not be Bool type, 839584d18dSGuozhi Wei // and their positions are defined by ABI. 849584d18dSGuozhi Wei if (CurrUser && !isa<CallInst>(Curr)) 85a1c712faSKit Barton for (auto &Op : CurrUser->operands()) 86a1c712faSKit Barton if (Defs.insert(Op).second) 87a1c712faSKit Barton WorkList.push_back(Op); 88a1c712faSKit Barton } 89a1c712faSKit Barton return Defs; 90a1c712faSKit Barton } 91a1c712faSKit Barton 92*f31c56dfSGuozhi Wei // Translate a i1 value to an equivalent i32/i64 value: 93*f31c56dfSGuozhi Wei Value *translate(Value *V) { 94*f31c56dfSGuozhi Wei Type *IntTy = ST->isPPC64() ? Type::getInt64Ty(V->getContext()) 95*f31c56dfSGuozhi Wei : Type::getInt32Ty(V->getContext()); 96*f31c56dfSGuozhi Wei 976a9226d9SEugene Zelenko if (auto *C = dyn_cast<Constant>(V)) 98*f31c56dfSGuozhi Wei return ConstantExpr::getZExt(C, IntTy); 996a9226d9SEugene Zelenko if (auto *P = dyn_cast<PHINode>(V)) { 100a1c712faSKit Barton // Temporarily set the operands to 0. We'll fix this later in 101a1c712faSKit Barton // runOnUse. 102*f31c56dfSGuozhi Wei Value *Zero = Constant::getNullValue(IntTy); 103a1c712faSKit Barton PHINode *Q = 104*f31c56dfSGuozhi Wei PHINode::Create(IntTy, P->getNumIncomingValues(), P->getName(), P); 105a1c712faSKit Barton for (unsigned i = 0; i < P->getNumOperands(); ++i) 106a1c712faSKit Barton Q->addIncoming(Zero, P->getIncomingBlock(i)); 107a1c712faSKit Barton return Q; 108a1c712faSKit Barton } 109a1c712faSKit Barton 1106a9226d9SEugene Zelenko auto *A = dyn_cast<Argument>(V); 1116a9226d9SEugene Zelenko auto *I = dyn_cast<Instruction>(V); 112a1c712faSKit Barton assert((A || I) && "Unknown value type"); 113a1c712faSKit Barton 114a1c712faSKit Barton auto InstPt = 115a1c712faSKit Barton A ? &*A->getParent()->getEntryBlock().begin() : I->getNextNode(); 116*f31c56dfSGuozhi Wei return new ZExtInst(V, IntTy, "", InstPt); 117a1c712faSKit Barton } 118a1c712faSKit Barton 119a1c712faSKit Barton typedef SmallPtrSet<const PHINode *, 8> PHINodeSet; 120a1c712faSKit Barton 121a1c712faSKit Barton // A PHINode is Promotable if: 122a1c712faSKit Barton // 1. Its type is i1 AND 123a1c712faSKit Barton // 2. All of its uses are ReturnInt, CallInst, PHINode, or DbgInfoIntrinsic 124a1c712faSKit Barton // AND 125a1c712faSKit Barton // 3. All of its operands are Constant or Argument or 126a1c712faSKit Barton // CallInst or PHINode AND 127a1c712faSKit Barton // 4. All of its PHINode uses are Promotable AND 128a1c712faSKit Barton // 5. All of its PHINode operands are Promotable 129a1c712faSKit Barton static PHINodeSet getPromotablePHINodes(const Function &F) { 130a1c712faSKit Barton PHINodeSet Promotable; 131a1c712faSKit Barton // Condition 1 132a1c712faSKit Barton for (auto &BB : F) 133a1c712faSKit Barton for (auto &I : BB) 1346a9226d9SEugene Zelenko if (const auto *P = dyn_cast<PHINode>(&I)) 135a1c712faSKit Barton if (P->getType()->isIntegerTy(1)) 136a1c712faSKit Barton Promotable.insert(P); 137a1c712faSKit Barton 138a1c712faSKit Barton SmallVector<const PHINode *, 8> ToRemove; 139451f54cfSBenjamin Kramer for (const PHINode *P : Promotable) { 140a1c712faSKit Barton // Condition 2 and 3 141a1c712faSKit Barton auto IsValidUser = [] (const Value *V) -> bool { 142a1c712faSKit Barton return isa<ReturnInst>(V) || isa<CallInst>(V) || isa<PHINode>(V) || 143a1c712faSKit Barton isa<DbgInfoIntrinsic>(V); 144a1c712faSKit Barton }; 145a1c712faSKit Barton auto IsValidOperand = [] (const Value *V) -> bool { 146a1c712faSKit Barton return isa<Constant>(V) || isa<Argument>(V) || isa<CallInst>(V) || 147a1c712faSKit Barton isa<PHINode>(V); 148a1c712faSKit Barton }; 149a1c712faSKit Barton const auto &Users = P->users(); 150a1c712faSKit Barton const auto &Operands = P->operands(); 1516a9226d9SEugene Zelenko if (!llvm::all_of(Users, IsValidUser) || 1526a9226d9SEugene Zelenko !llvm::all_of(Operands, IsValidOperand)) 153a1c712faSKit Barton ToRemove.push_back(P); 154a1c712faSKit Barton } 155a1c712faSKit Barton 156a1c712faSKit Barton // Iterate to convergence 157a1c712faSKit Barton auto IsPromotable = [&Promotable] (const Value *V) -> bool { 1586a9226d9SEugene Zelenko const auto *Phi = dyn_cast<PHINode>(V); 159a1c712faSKit Barton return !Phi || Promotable.count(Phi); 160a1c712faSKit Barton }; 161a1c712faSKit Barton while (!ToRemove.empty()) { 162a1c712faSKit Barton for (auto &User : ToRemove) 163a1c712faSKit Barton Promotable.erase(User); 164a1c712faSKit Barton ToRemove.clear(); 165a1c712faSKit Barton 166451f54cfSBenjamin Kramer for (const PHINode *P : Promotable) { 167a1c712faSKit Barton // Condition 4 and 5 168a1c712faSKit Barton const auto &Users = P->users(); 169a1c712faSKit Barton const auto &Operands = P->operands(); 1706a9226d9SEugene Zelenko if (!llvm::all_of(Users, IsPromotable) || 1716a9226d9SEugene Zelenko !llvm::all_of(Operands, IsPromotable)) 172a1c712faSKit Barton ToRemove.push_back(P); 173a1c712faSKit Barton } 174a1c712faSKit Barton } 175a1c712faSKit Barton 176a1c712faSKit Barton return Promotable; 177a1c712faSKit Barton } 178a1c712faSKit Barton 179a1c712faSKit Barton typedef DenseMap<Value *, Value *> B2IMap; 180a1c712faSKit Barton 181a1c712faSKit Barton public: 182a1c712faSKit Barton static char ID; 1836a9226d9SEugene Zelenko 1849fd267c2SEric Christopher PPCBoolRetToInt() : FunctionPass(ID) { 185a1c712faSKit Barton initializePPCBoolRetToIntPass(*PassRegistry::getPassRegistry()); 186a1c712faSKit Barton } 187a1c712faSKit Barton 1886a9226d9SEugene Zelenko bool runOnFunction(Function &F) override { 189289bd5f6SAndrew Kaylor if (skipFunction(F)) 190289bd5f6SAndrew Kaylor return false; 191289bd5f6SAndrew Kaylor 192*f31c56dfSGuozhi Wei auto *TPC = getAnalysisIfAvailable<TargetPassConfig>(); 193*f31c56dfSGuozhi Wei if (!TPC) 194*f31c56dfSGuozhi Wei return false; 195*f31c56dfSGuozhi Wei 196*f31c56dfSGuozhi Wei auto &TM = TPC->getTM<PPCTargetMachine>(); 197*f31c56dfSGuozhi Wei ST = TM.getSubtargetImpl(F); 198*f31c56dfSGuozhi Wei 199a1c712faSKit Barton PHINodeSet PromotablePHINodes = getPromotablePHINodes(F); 200a1c712faSKit Barton B2IMap Bool2IntMap; 201a1c712faSKit Barton bool Changed = false; 202a1c712faSKit Barton for (auto &BB : F) { 203a1c712faSKit Barton for (auto &I : BB) { 2046a9226d9SEugene Zelenko if (auto *R = dyn_cast<ReturnInst>(&I)) 205a1c712faSKit Barton if (F.getReturnType()->isIntegerTy(1)) 206a1c712faSKit Barton Changed |= 207a1c712faSKit Barton runOnUse(R->getOperandUse(0), PromotablePHINodes, Bool2IntMap); 208a1c712faSKit Barton 2096a9226d9SEugene Zelenko if (auto *CI = dyn_cast<CallInst>(&I)) 210a1c712faSKit Barton for (auto &U : CI->operands()) 211a1c712faSKit Barton if (U->getType()->isIntegerTy(1)) 212a1c712faSKit Barton Changed |= runOnUse(U, PromotablePHINodes, Bool2IntMap); 213a1c712faSKit Barton } 214a1c712faSKit Barton } 215a1c712faSKit Barton 216a1c712faSKit Barton return Changed; 217a1c712faSKit Barton } 218a1c712faSKit Barton 219*f31c56dfSGuozhi Wei bool runOnUse(Use &U, const PHINodeSet &PromotablePHINodes, 220a1c712faSKit Barton B2IMap &BoolToIntMap) { 221a1c712faSKit Barton auto Defs = findAllDefs(U); 222a1c712faSKit Barton 223a1c712faSKit Barton // If the values are all Constants or Arguments, don't bother 2246a9226d9SEugene Zelenko if (llvm::none_of(Defs, isa<Instruction, Value *>)) 225a1c712faSKit Barton return false; 226a1c712faSKit Barton 2279584d18dSGuozhi Wei // Presently, we only know how to handle PHINode, Constant, Arguments and 2289584d18dSGuozhi Wei // CallInst. Potentially, bitwise operations (AND, OR, XOR, NOT) and sign 2299584d18dSGuozhi Wei // extension could also be handled in the future. 230451f54cfSBenjamin Kramer for (Value *V : Defs) 2319584d18dSGuozhi Wei if (!isa<PHINode>(V) && !isa<Constant>(V) && 2329584d18dSGuozhi Wei !isa<Argument>(V) && !isa<CallInst>(V)) 233a1c712faSKit Barton return false; 234a1c712faSKit Barton 235451f54cfSBenjamin Kramer for (Value *V : Defs) 2366a9226d9SEugene Zelenko if (const auto *P = dyn_cast<PHINode>(V)) 237a1c712faSKit Barton if (!PromotablePHINodes.count(P)) 238a1c712faSKit Barton return false; 239a1c712faSKit Barton 240a1c712faSKit Barton if (isa<ReturnInst>(U.getUser())) 241a1c712faSKit Barton ++NumBoolRetPromotion; 242a1c712faSKit Barton if (isa<CallInst>(U.getUser())) 243a1c712faSKit Barton ++NumBoolCallPromotion; 244a1c712faSKit Barton ++NumBoolToIntPromotion; 245a1c712faSKit Barton 246451f54cfSBenjamin Kramer for (Value *V : Defs) 247a1c712faSKit Barton if (!BoolToIntMap.count(V)) 248a1c712faSKit Barton BoolToIntMap[V] = translate(V); 249a1c712faSKit Barton 2509584d18dSGuozhi Wei // Replace the operands of the translated instructions. They were set to 251a1c712faSKit Barton // zero in the translate function. 252a1c712faSKit Barton for (auto &Pair : BoolToIntMap) { 2536a9226d9SEugene Zelenko auto *First = dyn_cast<User>(Pair.first); 2546a9226d9SEugene Zelenko auto *Second = dyn_cast<User>(Pair.second); 255a1c712faSKit Barton assert((!First || Second) && "translated from user to non-user!?"); 2569584d18dSGuozhi Wei // Operands of CallInst are skipped because they may not be Bool type, 2579584d18dSGuozhi Wei // and their positions are defined by ABI. 2589584d18dSGuozhi Wei if (First && !isa<CallInst>(First)) 259a1c712faSKit Barton for (unsigned i = 0; i < First->getNumOperands(); ++i) 260a1c712faSKit Barton Second->setOperand(i, BoolToIntMap[First->getOperand(i)]); 261a1c712faSKit Barton } 262a1c712faSKit Barton 263a1c712faSKit Barton Value *IntRetVal = BoolToIntMap[U]; 264a1c712faSKit Barton Type *Int1Ty = Type::getInt1Ty(U->getContext()); 2656a9226d9SEugene Zelenko auto *I = cast<Instruction>(U.getUser()); 266a1c712faSKit Barton Value *BackToBool = new TruncInst(IntRetVal, Int1Ty, "backToBool", I); 267a1c712faSKit Barton U.set(BackToBool); 268a1c712faSKit Barton 269a1c712faSKit Barton return true; 270a1c712faSKit Barton } 271a1c712faSKit Barton 2726a9226d9SEugene Zelenko void getAnalysisUsage(AnalysisUsage &AU) const override { 273a1c712faSKit Barton AU.addPreserved<DominatorTreeWrapperPass>(); 274a1c712faSKit Barton FunctionPass::getAnalysisUsage(AU); 275a1c712faSKit Barton } 276*f31c56dfSGuozhi Wei 277*f31c56dfSGuozhi Wei private: 278*f31c56dfSGuozhi Wei const PPCSubtarget *ST; 279a1c712faSKit Barton }; 2806a9226d9SEugene Zelenko 2816a9226d9SEugene Zelenko } // end anonymous namespace 282a1c712faSKit Barton 283a1c712faSKit Barton char PPCBoolRetToInt::ID = 0; 2849fd267c2SEric Christopher INITIALIZE_PASS(PPCBoolRetToInt, "bool-ret-to-int", 285*f31c56dfSGuozhi Wei "Convert i1 constants to i32/i64 if they are returned", 286a1c712faSKit Barton false, false) 287a1c712faSKit Barton 2889fd267c2SEric Christopher FunctionPass *llvm::createPPCBoolRetToIntPass() { return new PPCBoolRetToInt(); } 289