1 //===--- ExpandLargeDivRem.cpp - Expand large div/rem ---------------------===// 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 // This pass expands div/rem instructions with a bitwidth above a threshold 10 // into a call to auto-generated functions. 11 // This is useful for targets like x86_64 that cannot lower divisions 12 // with more than 128 bits or targets like x86_32 that cannot lower divisions 13 // with more than 64 bits. 14 // 15 //===----------------------------------------------------------------------===// 16 17 #include "llvm/CodeGen/ExpandLargeDivRem.h" 18 #include "llvm/ADT/SmallVector.h" 19 #include "llvm/Analysis/GlobalsModRef.h" 20 #include "llvm/CodeGen/Passes.h" 21 #include "llvm/CodeGen/TargetLowering.h" 22 #include "llvm/CodeGen/TargetPassConfig.h" 23 #include "llvm/CodeGen/TargetSubtargetInfo.h" 24 #include "llvm/IR/IRBuilder.h" 25 #include "llvm/IR/InstIterator.h" 26 #include "llvm/IR/PassManager.h" 27 #include "llvm/InitializePasses.h" 28 #include "llvm/Pass.h" 29 #include "llvm/Support/CommandLine.h" 30 #include "llvm/Target/TargetMachine.h" 31 #include "llvm/Transforms/Utils/IntegerDivision.h" 32 33 using namespace llvm; 34 35 static cl::opt<unsigned> 36 ExpandDivRemBits("expand-div-rem-bits", cl::Hidden, 37 cl::init(llvm::IntegerType::MAX_INT_BITS), 38 cl::desc("div and rem instructions on integers with " 39 "more than <N> bits are expanded.")); 40 41 static bool isConstantPowerOfTwo(llvm::Value *V, bool SignedOp) { 42 auto *C = dyn_cast<ConstantInt>(V); 43 if (!C) 44 return false; 45 46 APInt Val = C->getValue(); 47 if (SignedOp && Val.isNegative()) 48 Val = -Val; 49 return Val.isPowerOf2(); 50 } 51 52 static bool isSigned(unsigned int Opcode) { 53 return Opcode == Instruction::SDiv || Opcode == Instruction::SRem; 54 } 55 56 static void scalarize(BinaryOperator *BO, 57 SmallVectorImpl<BinaryOperator *> &Replace) { 58 VectorType *VTy = cast<FixedVectorType>(BO->getType()); 59 60 IRBuilder<> Builder(BO); 61 62 unsigned NumElements = VTy->getElementCount().getFixedValue(); 63 Value *Result = PoisonValue::get(VTy); 64 for (unsigned Idx = 0; Idx < NumElements; ++Idx) { 65 Value *LHS = Builder.CreateExtractElement(BO->getOperand(0), Idx); 66 Value *RHS = Builder.CreateExtractElement(BO->getOperand(1), Idx); 67 Value *Op = Builder.CreateBinOp(BO->getOpcode(), LHS, RHS); 68 Result = Builder.CreateInsertElement(Result, Op, Idx); 69 if (auto *NewBO = dyn_cast<BinaryOperator>(Op)) { 70 NewBO->copyIRFlags(Op, true); 71 Replace.push_back(NewBO); 72 } 73 } 74 BO->replaceAllUsesWith(Result); 75 BO->dropAllReferences(); 76 BO->eraseFromParent(); 77 } 78 79 static bool runImpl(Function &F, const TargetLowering &TLI) { 80 SmallVector<BinaryOperator *, 4> Replace; 81 SmallVector<BinaryOperator *, 4> ReplaceVector; 82 bool Modified = false; 83 84 unsigned MaxLegalDivRemBitWidth = TLI.getMaxDivRemBitWidthSupported(); 85 if (ExpandDivRemBits != llvm::IntegerType::MAX_INT_BITS) 86 MaxLegalDivRemBitWidth = ExpandDivRemBits; 87 88 if (MaxLegalDivRemBitWidth >= llvm::IntegerType::MAX_INT_BITS) 89 return false; 90 91 for (auto &I : instructions(F)) { 92 switch (I.getOpcode()) { 93 case Instruction::UDiv: 94 case Instruction::SDiv: 95 case Instruction::URem: 96 case Instruction::SRem: { 97 // TODO: This pass doesn't handle scalable vectors. 98 if (I.getOperand(0)->getType()->isScalableTy()) 99 continue; 100 101 auto *IntTy = dyn_cast<IntegerType>(I.getType()->getScalarType()); 102 if (!IntTy || IntTy->getIntegerBitWidth() <= MaxLegalDivRemBitWidth) 103 continue; 104 105 // The backend has peephole optimizations for powers of two. 106 // TODO: We don't consider vectors here. 107 if (isConstantPowerOfTwo(I.getOperand(1), isSigned(I.getOpcode()))) 108 continue; 109 110 if (I.getOperand(0)->getType()->isVectorTy()) 111 ReplaceVector.push_back(&cast<BinaryOperator>(I)); 112 else 113 Replace.push_back(&cast<BinaryOperator>(I)); 114 Modified = true; 115 break; 116 } 117 default: 118 break; 119 } 120 } 121 122 while (!ReplaceVector.empty()) { 123 BinaryOperator *BO = ReplaceVector.pop_back_val(); 124 scalarize(BO, Replace); 125 } 126 127 if (Replace.empty()) 128 return false; 129 130 while (!Replace.empty()) { 131 BinaryOperator *I = Replace.pop_back_val(); 132 133 if (I->getOpcode() == Instruction::UDiv || 134 I->getOpcode() == Instruction::SDiv) { 135 expandDivision(I); 136 } else { 137 expandRemainder(I); 138 } 139 } 140 141 return Modified; 142 } 143 144 namespace { 145 class ExpandLargeDivRemLegacyPass : public FunctionPass { 146 public: 147 static char ID; 148 149 ExpandLargeDivRemLegacyPass() : FunctionPass(ID) { 150 initializeExpandLargeDivRemLegacyPassPass(*PassRegistry::getPassRegistry()); 151 } 152 153 bool runOnFunction(Function &F) override { 154 auto *TM = &getAnalysis<TargetPassConfig>().getTM<TargetMachine>(); 155 auto *TLI = TM->getSubtargetImpl(F)->getTargetLowering(); 156 return runImpl(F, *TLI); 157 } 158 159 void getAnalysisUsage(AnalysisUsage &AU) const override { 160 AU.addRequired<TargetPassConfig>(); 161 AU.addPreserved<AAResultsWrapperPass>(); 162 AU.addPreserved<GlobalsAAWrapperPass>(); 163 } 164 }; 165 } // namespace 166 167 PreservedAnalyses ExpandLargeDivRemPass::run(Function &F, 168 FunctionAnalysisManager &FAM) { 169 const TargetSubtargetInfo *STI = TM->getSubtargetImpl(F); 170 return runImpl(F, *STI->getTargetLowering()) ? PreservedAnalyses::none() 171 : PreservedAnalyses::all(); 172 } 173 174 char ExpandLargeDivRemLegacyPass::ID = 0; 175 INITIALIZE_PASS_BEGIN(ExpandLargeDivRemLegacyPass, "expand-large-div-rem", 176 "Expand large div/rem", false, false) 177 INITIALIZE_PASS_END(ExpandLargeDivRemLegacyPass, "expand-large-div-rem", 178 "Expand large div/rem", false, false) 179 180 FunctionPass *llvm::createExpandLargeDivRemPass() { 181 return new ExpandLargeDivRemLegacyPass(); 182 } 183