1 //===- GCNCreateVOPD.cpp - Create VOPD Instructions ----------------------===// 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 /// \file 10 /// Combine VALU pairs into VOPD instructions 11 /// Only works on wave32 12 /// Has register requirements, we reject creating VOPD if the requirements are 13 /// not met. 14 /// shouldCombineVOPD mutator in postRA machine scheduler puts candidate 15 /// instructions for VOPD back-to-back 16 /// 17 // 18 //===----------------------------------------------------------------------===// 19 20 #include "AMDGPU.h" 21 #include "GCNSubtarget.h" 22 #include "GCNVOPDUtils.h" 23 #include "SIInstrInfo.h" 24 #include "Utils/AMDGPUBaseInfo.h" 25 #include "llvm/ADT/SmallVector.h" 26 #include "llvm/ADT/Statistic.h" 27 #include "llvm/CodeGen/MachineBasicBlock.h" 28 #include "llvm/CodeGen/MachineInstr.h" 29 #include "llvm/CodeGen/MachineOperand.h" 30 #include "llvm/Support/Debug.h" 31 32 #define DEBUG_TYPE "gcn-create-vopd" 33 STATISTIC(NumVOPDCreated, "Number of VOPD Insts Created."); 34 35 using namespace llvm; 36 37 namespace { 38 39 class GCNCreateVOPD : public MachineFunctionPass { 40 private: 41 class VOPDCombineInfo { 42 public: 43 VOPDCombineInfo() = default; 44 VOPDCombineInfo(MachineInstr *First, MachineInstr *Second) 45 : FirstMI(First), SecondMI(Second) {} 46 47 MachineInstr *FirstMI; 48 MachineInstr *SecondMI; 49 }; 50 51 public: 52 static char ID; 53 const GCNSubtarget *ST = nullptr; 54 55 GCNCreateVOPD() : MachineFunctionPass(ID) {} 56 57 void getAnalysisUsage(AnalysisUsage &AU) const override { 58 AU.setPreservesCFG(); 59 MachineFunctionPass::getAnalysisUsage(AU); 60 } 61 62 StringRef getPassName() const override { 63 return "GCN Create VOPD Instructions"; 64 } 65 66 bool doReplace(const SIInstrInfo *SII, VOPDCombineInfo &CI) { 67 auto *FirstMI = CI.FirstMI; 68 auto *SecondMI = CI.SecondMI; 69 unsigned Opc1 = FirstMI->getOpcode(); 70 unsigned Opc2 = SecondMI->getOpcode(); 71 unsigned EncodingFamily = 72 AMDGPU::getVOPDEncodingFamily(SII->getSubtarget()); 73 int NewOpcode = 74 AMDGPU::getVOPDFull(AMDGPU::getVOPDOpcode(Opc1), 75 AMDGPU::getVOPDOpcode(Opc2), EncodingFamily); 76 assert(NewOpcode != -1 && 77 "Should have previously determined this as a possible VOPD\n"); 78 79 auto VOPDInst = BuildMI(*FirstMI->getParent(), FirstMI, 80 FirstMI->getDebugLoc(), SII->get(NewOpcode)) 81 .setMIFlags(FirstMI->getFlags() | SecondMI->getFlags()); 82 83 namespace VOPD = AMDGPU::VOPD; 84 MachineInstr *MI[] = {FirstMI, SecondMI}; 85 auto InstInfo = 86 AMDGPU::getVOPDInstInfo(FirstMI->getDesc(), SecondMI->getDesc()); 87 88 for (auto CompIdx : VOPD::COMPONENTS) { 89 auto MCOprIdx = InstInfo[CompIdx].getIndexOfDstInMCOperands(); 90 VOPDInst.add(MI[CompIdx]->getOperand(MCOprIdx)); 91 } 92 93 for (auto CompIdx : VOPD::COMPONENTS) { 94 auto CompSrcOprNum = InstInfo[CompIdx].getCompSrcOperandsNum(); 95 for (unsigned CompSrcIdx = 0; CompSrcIdx < CompSrcOprNum; ++CompSrcIdx) { 96 auto MCOprIdx = InstInfo[CompIdx].getIndexOfSrcInMCOperands(CompSrcIdx); 97 VOPDInst.add(MI[CompIdx]->getOperand(MCOprIdx)); 98 } 99 } 100 101 SII->fixImplicitOperands(*VOPDInst); 102 for (auto CompIdx : VOPD::COMPONENTS) 103 VOPDInst.copyImplicitOps(*MI[CompIdx]); 104 105 LLVM_DEBUG(dbgs() << "VOPD Fused: " << *VOPDInst << " from\tX: " 106 << *CI.FirstMI << "\tY: " << *CI.SecondMI << "\n"); 107 108 for (auto CompIdx : VOPD::COMPONENTS) 109 MI[CompIdx]->eraseFromParent(); 110 111 ++NumVOPDCreated; 112 return true; 113 } 114 115 bool runOnMachineFunction(MachineFunction &MF) override { 116 if (skipFunction(MF.getFunction())) 117 return false; 118 ST = &MF.getSubtarget<GCNSubtarget>(); 119 if (!AMDGPU::hasVOPD(*ST) || !ST->isWave32()) 120 return false; 121 LLVM_DEBUG(dbgs() << "CreateVOPD Pass:\n"); 122 123 const SIInstrInfo *SII = ST->getInstrInfo(); 124 bool Changed = false; 125 126 SmallVector<VOPDCombineInfo> ReplaceCandidates; 127 128 for (auto &MBB : MF) { 129 auto MII = MBB.begin(), E = MBB.end(); 130 while (MII != E) { 131 auto *FirstMI = &*MII; 132 MII = next_nodbg(MII, MBB.end()); 133 if (MII == MBB.end()) 134 break; 135 if (FirstMI->isDebugInstr()) 136 continue; 137 auto *SecondMI = &*MII; 138 unsigned Opc = FirstMI->getOpcode(); 139 unsigned Opc2 = SecondMI->getOpcode(); 140 llvm::AMDGPU::CanBeVOPD FirstCanBeVOPD = AMDGPU::getCanBeVOPD(Opc); 141 llvm::AMDGPU::CanBeVOPD SecondCanBeVOPD = AMDGPU::getCanBeVOPD(Opc2); 142 VOPDCombineInfo CI; 143 144 if (FirstCanBeVOPD.X && SecondCanBeVOPD.Y) 145 CI = VOPDCombineInfo(FirstMI, SecondMI); 146 else if (FirstCanBeVOPD.Y && SecondCanBeVOPD.X) 147 CI = VOPDCombineInfo(SecondMI, FirstMI); 148 else 149 continue; 150 // checkVOPDRegConstraints cares about program order, but doReplace 151 // cares about X-Y order in the constituted VOPD 152 if (llvm::checkVOPDRegConstraints(*SII, *FirstMI, *SecondMI)) { 153 ReplaceCandidates.push_back(CI); 154 ++MII; 155 } 156 } 157 } 158 for (auto &CI : ReplaceCandidates) { 159 Changed |= doReplace(SII, CI); 160 } 161 162 return Changed; 163 } 164 }; 165 166 } // namespace 167 168 char GCNCreateVOPD::ID = 0; 169 170 char &llvm::GCNCreateVOPDID = GCNCreateVOPD::ID; 171 172 INITIALIZE_PASS(GCNCreateVOPD, DEBUG_TYPE, "GCN Create VOPD Instructions", 173 false, false) 174