16a9226d9SEugene Zelenko //===- PPCBoolRetToInt.cpp ------------------------------------------------===// 2a1c712faSKit Barton // 32946cd70SChandler Carruth // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 42946cd70SChandler Carruth // See https://llvm.org/LICENSE.txt for license information. 52946cd70SChandler Carruth // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6a1c712faSKit Barton // 7a1c712faSKit Barton //===----------------------------------------------------------------------===// 8a1c712faSKit Barton // 9f31c56dfSGuozhi Wei // This file implements converting i1 values to i32/i64 if they could be more 10a1c712faSKit Barton // profitably allocated as GPRs rather than CRs. This pass will become totally 11a1c712faSKit Barton // unnecessary if Register Bank Allocation and Global Instruction Selection ever 12a1c712faSKit Barton // go upstream. 13a1c712faSKit Barton // 14f31c56dfSGuozhi Wei // Presently, the pass converts i1 Constants, and Arguments to i32/i64 if the 15a1c712faSKit Barton // transitive closure of their uses includes only PHINodes, CallInsts, and 16a1c712faSKit Barton // ReturnInsts. The rational is that arguments are generally passed and returned 17f31c56dfSGuozhi Wei // in GPRs rather than CRs, so casting them to i32/i64 at the LLVM IR level will 18a1c712faSKit Barton // actually save casts at the Machine Instruction level. 19a1c712faSKit Barton // 20a1c712faSKit Barton // It might be useful to expand this pass to add bit-wise operations to the list 21a1c712faSKit Barton // of safe transitive closure types. Also, we miss some opportunities when LLVM 22a1c712faSKit Barton // represents logical AND and OR operations with control flow rather than data 23a1c712faSKit Barton // flow. For example by lowering the expression: return (A && B && C) 24a1c712faSKit Barton // 25a1c712faSKit Barton // as: return A ? true : B && C. 26a1c712faSKit Barton // 27a1c712faSKit Barton // There's code in SimplifyCFG that code be used to turn control flow in data 28a1c712faSKit Barton // flow using SelectInsts. Selects are slow on some architectures (P7/P8), so 29a1c712faSKit Barton // this probably isn't good in general, but for the special case of i1, the 30a1c712faSKit Barton // Selects could be further lowered to bit operations that are fast everywhere. 31a1c712faSKit Barton // 32a1c712faSKit Barton //===----------------------------------------------------------------------===// 33a1c712faSKit Barton 34a1c712faSKit Barton #include "PPC.h" 35f31c56dfSGuozhi Wei #include "PPCTargetMachine.h" 366a9226d9SEugene Zelenko #include "llvm/ADT/DenseMap.h" 376bda14b3SChandler Carruth #include "llvm/ADT/STLExtras.h" 38a1c712faSKit Barton #include "llvm/ADT/SmallPtrSet.h" 396a9226d9SEugene Zelenko #include "llvm/ADT/SmallVector.h" 40a1c712faSKit Barton #include "llvm/ADT/Statistic.h" 416a9226d9SEugene Zelenko #include "llvm/IR/Argument.h" 42a1c712faSKit Barton #include "llvm/IR/Constants.h" 43a1c712faSKit Barton #include "llvm/IR/Dominators.h" 446a9226d9SEugene Zelenko #include "llvm/IR/Function.h" 456a9226d9SEugene Zelenko #include "llvm/IR/Instruction.h" 46a1c712faSKit Barton #include "llvm/IR/Instructions.h" 47a1c712faSKit Barton #include "llvm/IR/IntrinsicInst.h" 486a9226d9SEugene Zelenko #include "llvm/IR/OperandTraits.h" 496a9226d9SEugene Zelenko #include "llvm/IR/Type.h" 506a9226d9SEugene Zelenko #include "llvm/IR/Use.h" 516a9226d9SEugene Zelenko #include "llvm/IR/User.h" 526a9226d9SEugene Zelenko #include "llvm/IR/Value.h" 53a1c712faSKit Barton #include "llvm/Pass.h" 54f31c56dfSGuozhi Wei #include "llvm/CodeGen/TargetPassConfig.h" 556bda14b3SChandler Carruth #include "llvm/Support/Casting.h" 566a9226d9SEugene Zelenko #include <cassert> 57a1c712faSKit Barton 58a1c712faSKit Barton using namespace llvm; 59a1c712faSKit Barton 60a1c712faSKit Barton namespace { 61a1c712faSKit Barton 621bf4629fSArthur Eubanks #define DEBUG_TYPE "ppc-bool-ret-to-int" 63a1c712faSKit Barton 64a1c712faSKit Barton STATISTIC(NumBoolRetPromotion, 65a1c712faSKit Barton "Number of times a bool feeding a RetInst was promoted to an int"); 66a1c712faSKit Barton STATISTIC(NumBoolCallPromotion, 67a1c712faSKit Barton "Number of times a bool feeding a CallInst was promoted to an int"); 68a1c712faSKit Barton STATISTIC(NumBoolToIntPromotion, 69a1c712faSKit Barton "Total number of times a bool was promoted to an int"); 70a1c712faSKit Barton 71a1c712faSKit Barton class PPCBoolRetToInt : public FunctionPass { 72a1c712faSKit Barton static SmallPtrSet<Value *, 8> findAllDefs(Value *V) { 73a1c712faSKit Barton SmallPtrSet<Value *, 8> Defs; 74a1c712faSKit Barton SmallVector<Value *, 8> WorkList; 75a1c712faSKit Barton WorkList.push_back(V); 76a1c712faSKit Barton Defs.insert(V); 77a1c712faSKit Barton while (!WorkList.empty()) { 78a1c712faSKit Barton Value *Curr = WorkList.back(); 79a1c712faSKit Barton WorkList.pop_back(); 806a9226d9SEugene Zelenko auto *CurrUser = dyn_cast<User>(Curr); 81*cbea1756SKai Luo // Operands of CallInst/Constant are skipped because they may not be Bool 82*cbea1756SKai Luo // type. For CallInst, their positions are defined by ABI. 83*cbea1756SKai Luo if (CurrUser && !isa<CallInst>(Curr) && !isa<Constant>(Curr)) 84a1c712faSKit Barton for (auto &Op : CurrUser->operands()) 85a1c712faSKit Barton if (Defs.insert(Op).second) 86a1c712faSKit Barton WorkList.push_back(Op); 87a1c712faSKit Barton } 88a1c712faSKit Barton return Defs; 89a1c712faSKit Barton } 90a1c712faSKit Barton 91f31c56dfSGuozhi Wei // Translate a i1 value to an equivalent i32/i64 value: 92f31c56dfSGuozhi Wei Value *translate(Value *V) { 93*cbea1756SKai Luo assert(V->getType() == Type::getInt1Ty(V->getContext()) && 94*cbea1756SKai Luo "Expect an i1 value"); 95*cbea1756SKai Luo 96f31c56dfSGuozhi Wei Type *IntTy = ST->isPPC64() ? Type::getInt64Ty(V->getContext()) 97f31c56dfSGuozhi Wei : Type::getInt32Ty(V->getContext()); 98f31c56dfSGuozhi Wei 996a9226d9SEugene Zelenko if (auto *C = dyn_cast<Constant>(V)) 100f31c56dfSGuozhi Wei return ConstantExpr::getZExt(C, IntTy); 1016a9226d9SEugene Zelenko if (auto *P = dyn_cast<PHINode>(V)) { 102a1c712faSKit Barton // Temporarily set the operands to 0. We'll fix this later in 103a1c712faSKit Barton // runOnUse. 104f31c56dfSGuozhi Wei Value *Zero = Constant::getNullValue(IntTy); 105a1c712faSKit Barton PHINode *Q = 106f31c56dfSGuozhi Wei PHINode::Create(IntTy, P->getNumIncomingValues(), P->getName(), P); 107a1c712faSKit Barton for (unsigned i = 0; i < P->getNumOperands(); ++i) 108a1c712faSKit Barton Q->addIncoming(Zero, P->getIncomingBlock(i)); 109a1c712faSKit Barton return Q; 110a1c712faSKit Barton } 111a1c712faSKit Barton 1126a9226d9SEugene Zelenko auto *A = dyn_cast<Argument>(V); 1136a9226d9SEugene Zelenko auto *I = dyn_cast<Instruction>(V); 114a1c712faSKit Barton assert((A || I) && "Unknown value type"); 115a1c712faSKit Barton 116a1c712faSKit Barton auto InstPt = 117a1c712faSKit Barton A ? &*A->getParent()->getEntryBlock().begin() : I->getNextNode(); 118f31c56dfSGuozhi Wei return new ZExtInst(V, IntTy, "", InstPt); 119a1c712faSKit Barton } 120a1c712faSKit Barton 121a1c712faSKit Barton typedef SmallPtrSet<const PHINode *, 8> PHINodeSet; 122a1c712faSKit Barton 123a1c712faSKit Barton // A PHINode is Promotable if: 124a1c712faSKit Barton // 1. Its type is i1 AND 125a1c712faSKit Barton // 2. All of its uses are ReturnInt, CallInst, PHINode, or DbgInfoIntrinsic 126a1c712faSKit Barton // AND 127a1c712faSKit Barton // 3. All of its operands are Constant or Argument or 128a1c712faSKit Barton // CallInst or PHINode AND 129a1c712faSKit Barton // 4. All of its PHINode uses are Promotable AND 130a1c712faSKit Barton // 5. All of its PHINode operands are Promotable 131a1c712faSKit Barton static PHINodeSet getPromotablePHINodes(const Function &F) { 132a1c712faSKit Barton PHINodeSet Promotable; 133a1c712faSKit Barton // Condition 1 134a1c712faSKit Barton for (auto &BB : F) 135a1c712faSKit Barton for (auto &I : BB) 1366a9226d9SEugene Zelenko if (const auto *P = dyn_cast<PHINode>(&I)) 137a1c712faSKit Barton if (P->getType()->isIntegerTy(1)) 138a1c712faSKit Barton Promotable.insert(P); 139a1c712faSKit Barton 140a1c712faSKit Barton SmallVector<const PHINode *, 8> ToRemove; 141451f54cfSBenjamin Kramer for (const PHINode *P : Promotable) { 142a1c712faSKit Barton // Condition 2 and 3 143a1c712faSKit Barton auto IsValidUser = [] (const Value *V) -> bool { 144a1c712faSKit Barton return isa<ReturnInst>(V) || isa<CallInst>(V) || isa<PHINode>(V) || 145a1c712faSKit Barton isa<DbgInfoIntrinsic>(V); 146a1c712faSKit Barton }; 147a1c712faSKit Barton auto IsValidOperand = [] (const Value *V) -> bool { 148a1c712faSKit Barton return isa<Constant>(V) || isa<Argument>(V) || isa<CallInst>(V) || 149a1c712faSKit Barton isa<PHINode>(V); 150a1c712faSKit Barton }; 151a1c712faSKit Barton const auto &Users = P->users(); 152a1c712faSKit Barton const auto &Operands = P->operands(); 1536a9226d9SEugene Zelenko if (!llvm::all_of(Users, IsValidUser) || 1546a9226d9SEugene Zelenko !llvm::all_of(Operands, IsValidOperand)) 155a1c712faSKit Barton ToRemove.push_back(P); 156a1c712faSKit Barton } 157a1c712faSKit Barton 158a1c712faSKit Barton // Iterate to convergence 159a1c712faSKit Barton auto IsPromotable = [&Promotable] (const Value *V) -> bool { 1606a9226d9SEugene Zelenko const auto *Phi = dyn_cast<PHINode>(V); 161a1c712faSKit Barton return !Phi || Promotable.count(Phi); 162a1c712faSKit Barton }; 163a1c712faSKit Barton while (!ToRemove.empty()) { 164a1c712faSKit Barton for (auto &User : ToRemove) 165a1c712faSKit Barton Promotable.erase(User); 166a1c712faSKit Barton ToRemove.clear(); 167a1c712faSKit Barton 168451f54cfSBenjamin Kramer for (const PHINode *P : Promotable) { 169a1c712faSKit Barton // Condition 4 and 5 170a1c712faSKit Barton const auto &Users = P->users(); 171a1c712faSKit Barton const auto &Operands = P->operands(); 1726a9226d9SEugene Zelenko if (!llvm::all_of(Users, IsPromotable) || 1736a9226d9SEugene Zelenko !llvm::all_of(Operands, IsPromotable)) 174a1c712faSKit Barton ToRemove.push_back(P); 175a1c712faSKit Barton } 176a1c712faSKit Barton } 177a1c712faSKit Barton 178a1c712faSKit Barton return Promotable; 179a1c712faSKit Barton } 180a1c712faSKit Barton 181a1c712faSKit Barton typedef DenseMap<Value *, Value *> B2IMap; 182a1c712faSKit Barton 183a1c712faSKit Barton public: 184a1c712faSKit Barton static char ID; 1856a9226d9SEugene Zelenko 1869fd267c2SEric Christopher PPCBoolRetToInt() : FunctionPass(ID) { 187a1c712faSKit Barton initializePPCBoolRetToIntPass(*PassRegistry::getPassRegistry()); 188a1c712faSKit Barton } 189a1c712faSKit Barton 1906a9226d9SEugene Zelenko bool runOnFunction(Function &F) override { 191289bd5f6SAndrew Kaylor if (skipFunction(F)) 192289bd5f6SAndrew Kaylor return false; 193289bd5f6SAndrew Kaylor 194f31c56dfSGuozhi Wei auto *TPC = getAnalysisIfAvailable<TargetPassConfig>(); 195f31c56dfSGuozhi Wei if (!TPC) 196f31c56dfSGuozhi Wei return false; 197f31c56dfSGuozhi Wei 198f31c56dfSGuozhi Wei auto &TM = TPC->getTM<PPCTargetMachine>(); 199f31c56dfSGuozhi Wei ST = TM.getSubtargetImpl(F); 200f31c56dfSGuozhi Wei 201a1c712faSKit Barton PHINodeSet PromotablePHINodes = getPromotablePHINodes(F); 202a1c712faSKit Barton B2IMap Bool2IntMap; 203a1c712faSKit Barton bool Changed = false; 204a1c712faSKit Barton for (auto &BB : F) { 205a1c712faSKit Barton for (auto &I : BB) { 2066a9226d9SEugene Zelenko if (auto *R = dyn_cast<ReturnInst>(&I)) 207a1c712faSKit Barton if (F.getReturnType()->isIntegerTy(1)) 208a1c712faSKit Barton Changed |= 209a1c712faSKit Barton runOnUse(R->getOperandUse(0), PromotablePHINodes, Bool2IntMap); 210a1c712faSKit Barton 2116a9226d9SEugene Zelenko if (auto *CI = dyn_cast<CallInst>(&I)) 212a1c712faSKit Barton for (auto &U : CI->operands()) 213a1c712faSKit Barton if (U->getType()->isIntegerTy(1)) 214a1c712faSKit Barton Changed |= runOnUse(U, PromotablePHINodes, Bool2IntMap); 215a1c712faSKit Barton } 216a1c712faSKit Barton } 217a1c712faSKit Barton 218a1c712faSKit Barton return Changed; 219a1c712faSKit Barton } 220a1c712faSKit Barton 221f31c56dfSGuozhi Wei bool runOnUse(Use &U, const PHINodeSet &PromotablePHINodes, 222a1c712faSKit Barton B2IMap &BoolToIntMap) { 223a1c712faSKit Barton auto Defs = findAllDefs(U); 224a1c712faSKit Barton 225a1c712faSKit Barton // If the values are all Constants or Arguments, don't bother 22672d20b96SRahul Joshi if (llvm::none_of(Defs, [](Value *V) { return isa<Instruction>(V); })) 227a1c712faSKit Barton return false; 228a1c712faSKit Barton 2299584d18dSGuozhi Wei // Presently, we only know how to handle PHINode, Constant, Arguments and 2309584d18dSGuozhi Wei // CallInst. Potentially, bitwise operations (AND, OR, XOR, NOT) and sign 2319584d18dSGuozhi Wei // extension could also be handled in the future. 232451f54cfSBenjamin Kramer for (Value *V : Defs) 2339584d18dSGuozhi Wei if (!isa<PHINode>(V) && !isa<Constant>(V) && 2349584d18dSGuozhi Wei !isa<Argument>(V) && !isa<CallInst>(V)) 235a1c712faSKit Barton return false; 236a1c712faSKit Barton 237451f54cfSBenjamin Kramer for (Value *V : Defs) 2386a9226d9SEugene Zelenko if (const auto *P = dyn_cast<PHINode>(V)) 239a1c712faSKit Barton if (!PromotablePHINodes.count(P)) 240a1c712faSKit Barton return false; 241a1c712faSKit Barton 242a1c712faSKit Barton if (isa<ReturnInst>(U.getUser())) 243a1c712faSKit Barton ++NumBoolRetPromotion; 244a1c712faSKit Barton if (isa<CallInst>(U.getUser())) 245a1c712faSKit Barton ++NumBoolCallPromotion; 246a1c712faSKit Barton ++NumBoolToIntPromotion; 247a1c712faSKit Barton 248451f54cfSBenjamin Kramer for (Value *V : Defs) 249a1c712faSKit Barton if (!BoolToIntMap.count(V)) 250a1c712faSKit Barton BoolToIntMap[V] = translate(V); 251a1c712faSKit Barton 2529584d18dSGuozhi Wei // Replace the operands of the translated instructions. They were set to 253a1c712faSKit Barton // zero in the translate function. 254a1c712faSKit Barton for (auto &Pair : BoolToIntMap) { 2556a9226d9SEugene Zelenko auto *First = dyn_cast<User>(Pair.first); 2566a9226d9SEugene Zelenko auto *Second = dyn_cast<User>(Pair.second); 257a1c712faSKit Barton assert((!First || Second) && "translated from user to non-user!?"); 258*cbea1756SKai Luo // Operands of CallInst/Constant are skipped because they may not be Bool 259*cbea1756SKai Luo // type. For CallInst, their positions are defined by ABI. 260*cbea1756SKai Luo if (First && !isa<CallInst>(First) && !isa<Constant>(First)) 261a1c712faSKit Barton for (unsigned i = 0; i < First->getNumOperands(); ++i) 262a1c712faSKit Barton Second->setOperand(i, BoolToIntMap[First->getOperand(i)]); 263a1c712faSKit Barton } 264a1c712faSKit Barton 265a1c712faSKit Barton Value *IntRetVal = BoolToIntMap[U]; 266a1c712faSKit Barton Type *Int1Ty = Type::getInt1Ty(U->getContext()); 2676a9226d9SEugene Zelenko auto *I = cast<Instruction>(U.getUser()); 268a1c712faSKit Barton Value *BackToBool = new TruncInst(IntRetVal, Int1Ty, "backToBool", I); 269a1c712faSKit Barton U.set(BackToBool); 270a1c712faSKit Barton 271a1c712faSKit Barton return true; 272a1c712faSKit Barton } 273a1c712faSKit Barton 2746a9226d9SEugene Zelenko void getAnalysisUsage(AnalysisUsage &AU) const override { 275a1c712faSKit Barton AU.addPreserved<DominatorTreeWrapperPass>(); 276a1c712faSKit Barton FunctionPass::getAnalysisUsage(AU); 277a1c712faSKit Barton } 278f31c56dfSGuozhi Wei 279f31c56dfSGuozhi Wei private: 280f31c56dfSGuozhi Wei const PPCSubtarget *ST; 281a1c712faSKit Barton }; 2826a9226d9SEugene Zelenko 2836a9226d9SEugene Zelenko } // end anonymous namespace 284a1c712faSKit Barton 285a1c712faSKit Barton char PPCBoolRetToInt::ID = 0; 2861bf4629fSArthur Eubanks INITIALIZE_PASS(PPCBoolRetToInt, "ppc-bool-ret-to-int", 2871bf4629fSArthur Eubanks "Convert i1 constants to i32/i64 if they are returned", false, 2881bf4629fSArthur Eubanks false) 289a1c712faSKit Barton 2909fd267c2SEric Christopher FunctionPass *llvm::createPPCBoolRetToIntPass() { return new PPCBoolRetToInt(); } 291