1 //===-- ARMExpandPseudoInsts.cpp - Expand pseudo instructions -------------===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 // This file contains a pass that expands pseudo instructions into target 11 // instructions to allow proper scheduling, if-conversion, and other late 12 // optimizations. This pass should be run after register allocation but before 13 // the post-regalloc scheduling pass. 14 // 15 //===----------------------------------------------------------------------===// 16 17 #include "ARM.h" 18 #include "ARMBaseInstrInfo.h" 19 #include "ARMBaseRegisterInfo.h" 20 #include "ARMConstantPoolValue.h" 21 #include "ARMMachineFunctionInfo.h" 22 #include "ARMSubtarget.h" 23 #include "MCTargetDesc/ARMAddressingModes.h" 24 #include "llvm/CodeGen/LivePhysRegs.h" 25 #include "llvm/CodeGen/MachineFrameInfo.h" 26 #include "llvm/CodeGen/MachineFunctionPass.h" 27 #include "llvm/CodeGen/MachineInstrBuilder.h" 28 #include "llvm/CodeGen/MachineInstrBundle.h" 29 #include "llvm/IR/GlobalValue.h" 30 #include "llvm/Support/CommandLine.h" 31 #include "llvm/Support/raw_ostream.h" // FIXME: for debug only. remove! 32 #include "llvm/Target/TargetFrameLowering.h" 33 #include "llvm/Target/TargetRegisterInfo.h" 34 35 using namespace llvm; 36 37 #define DEBUG_TYPE "arm-pseudo" 38 39 static cl::opt<bool> 40 VerifyARMPseudo("verify-arm-pseudo-expand", cl::Hidden, 41 cl::desc("Verify machine code after expanding ARM pseudos")); 42 43 namespace { 44 class ARMExpandPseudo : public MachineFunctionPass { 45 public: 46 static char ID; 47 ARMExpandPseudo() : MachineFunctionPass(ID) {} 48 49 const ARMBaseInstrInfo *TII; 50 const TargetRegisterInfo *TRI; 51 const ARMSubtarget *STI; 52 ARMFunctionInfo *AFI; 53 54 bool runOnMachineFunction(MachineFunction &Fn) override; 55 56 MachineFunctionProperties getRequiredProperties() const override { 57 return MachineFunctionProperties().set( 58 MachineFunctionProperties::Property::NoVRegs); 59 } 60 61 StringRef getPassName() const override { 62 return "ARM pseudo instruction expansion pass"; 63 } 64 65 private: 66 void TransferImpOps(MachineInstr &OldMI, 67 MachineInstrBuilder &UseMI, MachineInstrBuilder &DefMI); 68 bool ExpandMI(MachineBasicBlock &MBB, 69 MachineBasicBlock::iterator MBBI, 70 MachineBasicBlock::iterator &NextMBBI); 71 bool ExpandMBB(MachineBasicBlock &MBB); 72 void ExpandVLD(MachineBasicBlock::iterator &MBBI); 73 void ExpandVST(MachineBasicBlock::iterator &MBBI); 74 void ExpandLaneOp(MachineBasicBlock::iterator &MBBI); 75 void ExpandVTBL(MachineBasicBlock::iterator &MBBI, 76 unsigned Opc, bool IsExt); 77 void ExpandMOV32BitImm(MachineBasicBlock &MBB, 78 MachineBasicBlock::iterator &MBBI); 79 bool ExpandCMP_SWAP(MachineBasicBlock &MBB, 80 MachineBasicBlock::iterator MBBI, unsigned LdrexOp, 81 unsigned StrexOp, unsigned UxtOp, 82 MachineBasicBlock::iterator &NextMBBI); 83 84 bool ExpandCMP_SWAP_64(MachineBasicBlock &MBB, 85 MachineBasicBlock::iterator MBBI, 86 MachineBasicBlock::iterator &NextMBBI); 87 }; 88 char ARMExpandPseudo::ID = 0; 89 } 90 91 /// TransferImpOps - Transfer implicit operands on the pseudo instruction to 92 /// the instructions created from the expansion. 93 void ARMExpandPseudo::TransferImpOps(MachineInstr &OldMI, 94 MachineInstrBuilder &UseMI, 95 MachineInstrBuilder &DefMI) { 96 const MCInstrDesc &Desc = OldMI.getDesc(); 97 for (unsigned i = Desc.getNumOperands(), e = OldMI.getNumOperands(); 98 i != e; ++i) { 99 const MachineOperand &MO = OldMI.getOperand(i); 100 assert(MO.isReg() && MO.getReg()); 101 if (MO.isUse()) 102 UseMI.add(MO); 103 else 104 DefMI.add(MO); 105 } 106 } 107 108 namespace { 109 // Constants for register spacing in NEON load/store instructions. 110 // For quad-register load-lane and store-lane pseudo instructors, the 111 // spacing is initially assumed to be EvenDblSpc, and that is changed to 112 // OddDblSpc depending on the lane number operand. 113 enum NEONRegSpacing { 114 SingleSpc, 115 EvenDblSpc, 116 OddDblSpc 117 }; 118 119 // Entries for NEON load/store information table. The table is sorted by 120 // PseudoOpc for fast binary-search lookups. 121 struct NEONLdStTableEntry { 122 uint16_t PseudoOpc; 123 uint16_t RealOpc; 124 bool IsLoad; 125 bool isUpdating; 126 bool hasWritebackOperand; 127 uint8_t RegSpacing; // One of type NEONRegSpacing 128 uint8_t NumRegs; // D registers loaded or stored 129 uint8_t RegElts; // elements per D register; used for lane ops 130 // FIXME: Temporary flag to denote whether the real instruction takes 131 // a single register (like the encoding) or all of the registers in 132 // the list (like the asm syntax and the isel DAG). When all definitions 133 // are converted to take only the single encoded register, this will 134 // go away. 135 bool copyAllListRegs; 136 137 // Comparison methods for binary search of the table. 138 bool operator<(const NEONLdStTableEntry &TE) const { 139 return PseudoOpc < TE.PseudoOpc; 140 } 141 friend bool operator<(const NEONLdStTableEntry &TE, unsigned PseudoOpc) { 142 return TE.PseudoOpc < PseudoOpc; 143 } 144 friend bool LLVM_ATTRIBUTE_UNUSED operator<(unsigned PseudoOpc, 145 const NEONLdStTableEntry &TE) { 146 return PseudoOpc < TE.PseudoOpc; 147 } 148 }; 149 } 150 151 static const NEONLdStTableEntry NEONLdStTable[] = { 152 { ARM::VLD1LNq16Pseudo, ARM::VLD1LNd16, true, false, false, EvenDblSpc, 1, 4 ,true}, 153 { ARM::VLD1LNq16Pseudo_UPD, ARM::VLD1LNd16_UPD, true, true, true, EvenDblSpc, 1, 4 ,true}, 154 { ARM::VLD1LNq32Pseudo, ARM::VLD1LNd32, true, false, false, EvenDblSpc, 1, 2 ,true}, 155 { ARM::VLD1LNq32Pseudo_UPD, ARM::VLD1LNd32_UPD, true, true, true, EvenDblSpc, 1, 2 ,true}, 156 { ARM::VLD1LNq8Pseudo, ARM::VLD1LNd8, true, false, false, EvenDblSpc, 1, 8 ,true}, 157 { ARM::VLD1LNq8Pseudo_UPD, ARM::VLD1LNd8_UPD, true, true, true, EvenDblSpc, 1, 8 ,true}, 158 159 { ARM::VLD1d64QPseudo, ARM::VLD1d64Q, true, false, false, SingleSpc, 4, 1 ,false}, 160 { ARM::VLD1d64QPseudoWB_fixed, ARM::VLD1d64Qwb_fixed, true, true, false, SingleSpc, 4, 1 ,false}, 161 { ARM::VLD1d64TPseudo, ARM::VLD1d64T, true, false, false, SingleSpc, 3, 1 ,false}, 162 { ARM::VLD1d64TPseudoWB_fixed, ARM::VLD1d64Twb_fixed, true, true, false, SingleSpc, 3, 1 ,false}, 163 164 { ARM::VLD2LNd16Pseudo, ARM::VLD2LNd16, true, false, false, SingleSpc, 2, 4 ,true}, 165 { ARM::VLD2LNd16Pseudo_UPD, ARM::VLD2LNd16_UPD, true, true, true, SingleSpc, 2, 4 ,true}, 166 { ARM::VLD2LNd32Pseudo, ARM::VLD2LNd32, true, false, false, SingleSpc, 2, 2 ,true}, 167 { ARM::VLD2LNd32Pseudo_UPD, ARM::VLD2LNd32_UPD, true, true, true, SingleSpc, 2, 2 ,true}, 168 { ARM::VLD2LNd8Pseudo, ARM::VLD2LNd8, true, false, false, SingleSpc, 2, 8 ,true}, 169 { ARM::VLD2LNd8Pseudo_UPD, ARM::VLD2LNd8_UPD, true, true, true, SingleSpc, 2, 8 ,true}, 170 { ARM::VLD2LNq16Pseudo, ARM::VLD2LNq16, true, false, false, EvenDblSpc, 2, 4 ,true}, 171 { ARM::VLD2LNq16Pseudo_UPD, ARM::VLD2LNq16_UPD, true, true, true, EvenDblSpc, 2, 4 ,true}, 172 { ARM::VLD2LNq32Pseudo, ARM::VLD2LNq32, true, false, false, EvenDblSpc, 2, 2 ,true}, 173 { ARM::VLD2LNq32Pseudo_UPD, ARM::VLD2LNq32_UPD, true, true, true, EvenDblSpc, 2, 2 ,true}, 174 175 { ARM::VLD2q16Pseudo, ARM::VLD2q16, true, false, false, SingleSpc, 4, 4 ,false}, 176 { ARM::VLD2q16PseudoWB_fixed, ARM::VLD2q16wb_fixed, true, true, false, SingleSpc, 4, 4 ,false}, 177 { ARM::VLD2q16PseudoWB_register, ARM::VLD2q16wb_register, true, true, true, SingleSpc, 4, 4 ,false}, 178 { ARM::VLD2q32Pseudo, ARM::VLD2q32, true, false, false, SingleSpc, 4, 2 ,false}, 179 { ARM::VLD2q32PseudoWB_fixed, ARM::VLD2q32wb_fixed, true, true, false, SingleSpc, 4, 2 ,false}, 180 { ARM::VLD2q32PseudoWB_register, ARM::VLD2q32wb_register, true, true, true, SingleSpc, 4, 2 ,false}, 181 { ARM::VLD2q8Pseudo, ARM::VLD2q8, true, false, false, SingleSpc, 4, 8 ,false}, 182 { ARM::VLD2q8PseudoWB_fixed, ARM::VLD2q8wb_fixed, true, true, false, SingleSpc, 4, 8 ,false}, 183 { ARM::VLD2q8PseudoWB_register, ARM::VLD2q8wb_register, true, true, true, SingleSpc, 4, 8 ,false}, 184 185 { ARM::VLD3DUPd16Pseudo, ARM::VLD3DUPd16, true, false, false, SingleSpc, 3, 4,true}, 186 { ARM::VLD3DUPd16Pseudo_UPD, ARM::VLD3DUPd16_UPD, true, true, true, SingleSpc, 3, 4,true}, 187 { ARM::VLD3DUPd32Pseudo, ARM::VLD3DUPd32, true, false, false, SingleSpc, 3, 2,true}, 188 { ARM::VLD3DUPd32Pseudo_UPD, ARM::VLD3DUPd32_UPD, true, true, true, SingleSpc, 3, 2,true}, 189 { ARM::VLD3DUPd8Pseudo, ARM::VLD3DUPd8, true, false, false, SingleSpc, 3, 8,true}, 190 { ARM::VLD3DUPd8Pseudo_UPD, ARM::VLD3DUPd8_UPD, true, true, true, SingleSpc, 3, 8,true}, 191 192 { ARM::VLD3LNd16Pseudo, ARM::VLD3LNd16, true, false, false, SingleSpc, 3, 4 ,true}, 193 { ARM::VLD3LNd16Pseudo_UPD, ARM::VLD3LNd16_UPD, true, true, true, SingleSpc, 3, 4 ,true}, 194 { ARM::VLD3LNd32Pseudo, ARM::VLD3LNd32, true, false, false, SingleSpc, 3, 2 ,true}, 195 { ARM::VLD3LNd32Pseudo_UPD, ARM::VLD3LNd32_UPD, true, true, true, SingleSpc, 3, 2 ,true}, 196 { ARM::VLD3LNd8Pseudo, ARM::VLD3LNd8, true, false, false, SingleSpc, 3, 8 ,true}, 197 { ARM::VLD3LNd8Pseudo_UPD, ARM::VLD3LNd8_UPD, true, true, true, SingleSpc, 3, 8 ,true}, 198 { ARM::VLD3LNq16Pseudo, ARM::VLD3LNq16, true, false, false, EvenDblSpc, 3, 4 ,true}, 199 { ARM::VLD3LNq16Pseudo_UPD, ARM::VLD3LNq16_UPD, true, true, true, EvenDblSpc, 3, 4 ,true}, 200 { ARM::VLD3LNq32Pseudo, ARM::VLD3LNq32, true, false, false, EvenDblSpc, 3, 2 ,true}, 201 { ARM::VLD3LNq32Pseudo_UPD, ARM::VLD3LNq32_UPD, true, true, true, EvenDblSpc, 3, 2 ,true}, 202 203 { ARM::VLD3d16Pseudo, ARM::VLD3d16, true, false, false, SingleSpc, 3, 4 ,true}, 204 { ARM::VLD3d16Pseudo_UPD, ARM::VLD3d16_UPD, true, true, true, SingleSpc, 3, 4 ,true}, 205 { ARM::VLD3d32Pseudo, ARM::VLD3d32, true, false, false, SingleSpc, 3, 2 ,true}, 206 { ARM::VLD3d32Pseudo_UPD, ARM::VLD3d32_UPD, true, true, true, SingleSpc, 3, 2 ,true}, 207 { ARM::VLD3d8Pseudo, ARM::VLD3d8, true, false, false, SingleSpc, 3, 8 ,true}, 208 { ARM::VLD3d8Pseudo_UPD, ARM::VLD3d8_UPD, true, true, true, SingleSpc, 3, 8 ,true}, 209 210 { ARM::VLD3q16Pseudo_UPD, ARM::VLD3q16_UPD, true, true, true, EvenDblSpc, 3, 4 ,true}, 211 { ARM::VLD3q16oddPseudo, ARM::VLD3q16, true, false, false, OddDblSpc, 3, 4 ,true}, 212 { ARM::VLD3q16oddPseudo_UPD, ARM::VLD3q16_UPD, true, true, true, OddDblSpc, 3, 4 ,true}, 213 { ARM::VLD3q32Pseudo_UPD, ARM::VLD3q32_UPD, true, true, true, EvenDblSpc, 3, 2 ,true}, 214 { ARM::VLD3q32oddPseudo, ARM::VLD3q32, true, false, false, OddDblSpc, 3, 2 ,true}, 215 { ARM::VLD3q32oddPseudo_UPD, ARM::VLD3q32_UPD, true, true, true, OddDblSpc, 3, 2 ,true}, 216 { ARM::VLD3q8Pseudo_UPD, ARM::VLD3q8_UPD, true, true, true, EvenDblSpc, 3, 8 ,true}, 217 { ARM::VLD3q8oddPseudo, ARM::VLD3q8, true, false, false, OddDblSpc, 3, 8 ,true}, 218 { ARM::VLD3q8oddPseudo_UPD, ARM::VLD3q8_UPD, true, true, true, OddDblSpc, 3, 8 ,true}, 219 220 { ARM::VLD4DUPd16Pseudo, ARM::VLD4DUPd16, true, false, false, SingleSpc, 4, 4,true}, 221 { ARM::VLD4DUPd16Pseudo_UPD, ARM::VLD4DUPd16_UPD, true, true, true, SingleSpc, 4, 4,true}, 222 { ARM::VLD4DUPd32Pseudo, ARM::VLD4DUPd32, true, false, false, SingleSpc, 4, 2,true}, 223 { ARM::VLD4DUPd32Pseudo_UPD, ARM::VLD4DUPd32_UPD, true, true, true, SingleSpc, 4, 2,true}, 224 { ARM::VLD4DUPd8Pseudo, ARM::VLD4DUPd8, true, false, false, SingleSpc, 4, 8,true}, 225 { ARM::VLD4DUPd8Pseudo_UPD, ARM::VLD4DUPd8_UPD, true, true, true, SingleSpc, 4, 8,true}, 226 227 { ARM::VLD4LNd16Pseudo, ARM::VLD4LNd16, true, false, false, SingleSpc, 4, 4 ,true}, 228 { ARM::VLD4LNd16Pseudo_UPD, ARM::VLD4LNd16_UPD, true, true, true, SingleSpc, 4, 4 ,true}, 229 { ARM::VLD4LNd32Pseudo, ARM::VLD4LNd32, true, false, false, SingleSpc, 4, 2 ,true}, 230 { ARM::VLD4LNd32Pseudo_UPD, ARM::VLD4LNd32_UPD, true, true, true, SingleSpc, 4, 2 ,true}, 231 { ARM::VLD4LNd8Pseudo, ARM::VLD4LNd8, true, false, false, SingleSpc, 4, 8 ,true}, 232 { ARM::VLD4LNd8Pseudo_UPD, ARM::VLD4LNd8_UPD, true, true, true, SingleSpc, 4, 8 ,true}, 233 { ARM::VLD4LNq16Pseudo, ARM::VLD4LNq16, true, false, false, EvenDblSpc, 4, 4 ,true}, 234 { ARM::VLD4LNq16Pseudo_UPD, ARM::VLD4LNq16_UPD, true, true, true, EvenDblSpc, 4, 4 ,true}, 235 { ARM::VLD4LNq32Pseudo, ARM::VLD4LNq32, true, false, false, EvenDblSpc, 4, 2 ,true}, 236 { ARM::VLD4LNq32Pseudo_UPD, ARM::VLD4LNq32_UPD, true, true, true, EvenDblSpc, 4, 2 ,true}, 237 238 { ARM::VLD4d16Pseudo, ARM::VLD4d16, true, false, false, SingleSpc, 4, 4 ,true}, 239 { ARM::VLD4d16Pseudo_UPD, ARM::VLD4d16_UPD, true, true, true, SingleSpc, 4, 4 ,true}, 240 { ARM::VLD4d32Pseudo, ARM::VLD4d32, true, false, false, SingleSpc, 4, 2 ,true}, 241 { ARM::VLD4d32Pseudo_UPD, ARM::VLD4d32_UPD, true, true, true, SingleSpc, 4, 2 ,true}, 242 { ARM::VLD4d8Pseudo, ARM::VLD4d8, true, false, false, SingleSpc, 4, 8 ,true}, 243 { ARM::VLD4d8Pseudo_UPD, ARM::VLD4d8_UPD, true, true, true, SingleSpc, 4, 8 ,true}, 244 245 { ARM::VLD4q16Pseudo_UPD, ARM::VLD4q16_UPD, true, true, true, EvenDblSpc, 4, 4 ,true}, 246 { ARM::VLD4q16oddPseudo, ARM::VLD4q16, true, false, false, OddDblSpc, 4, 4 ,true}, 247 { ARM::VLD4q16oddPseudo_UPD, ARM::VLD4q16_UPD, true, true, true, OddDblSpc, 4, 4 ,true}, 248 { ARM::VLD4q32Pseudo_UPD, ARM::VLD4q32_UPD, true, true, true, EvenDblSpc, 4, 2 ,true}, 249 { ARM::VLD4q32oddPseudo, ARM::VLD4q32, true, false, false, OddDblSpc, 4, 2 ,true}, 250 { ARM::VLD4q32oddPseudo_UPD, ARM::VLD4q32_UPD, true, true, true, OddDblSpc, 4, 2 ,true}, 251 { ARM::VLD4q8Pseudo_UPD, ARM::VLD4q8_UPD, true, true, true, EvenDblSpc, 4, 8 ,true}, 252 { ARM::VLD4q8oddPseudo, ARM::VLD4q8, true, false, false, OddDblSpc, 4, 8 ,true}, 253 { ARM::VLD4q8oddPseudo_UPD, ARM::VLD4q8_UPD, true, true, true, OddDblSpc, 4, 8 ,true}, 254 255 { ARM::VST1LNq16Pseudo, ARM::VST1LNd16, false, false, false, EvenDblSpc, 1, 4 ,true}, 256 { ARM::VST1LNq16Pseudo_UPD, ARM::VST1LNd16_UPD, false, true, true, EvenDblSpc, 1, 4 ,true}, 257 { ARM::VST1LNq32Pseudo, ARM::VST1LNd32, false, false, false, EvenDblSpc, 1, 2 ,true}, 258 { ARM::VST1LNq32Pseudo_UPD, ARM::VST1LNd32_UPD, false, true, true, EvenDblSpc, 1, 2 ,true}, 259 { ARM::VST1LNq8Pseudo, ARM::VST1LNd8, false, false, false, EvenDblSpc, 1, 8 ,true}, 260 { ARM::VST1LNq8Pseudo_UPD, ARM::VST1LNd8_UPD, false, true, true, EvenDblSpc, 1, 8 ,true}, 261 262 { ARM::VST1d64QPseudo, ARM::VST1d64Q, false, false, false, SingleSpc, 4, 1 ,false}, 263 { ARM::VST1d64QPseudoWB_fixed, ARM::VST1d64Qwb_fixed, false, true, false, SingleSpc, 4, 1 ,false}, 264 { ARM::VST1d64QPseudoWB_register, ARM::VST1d64Qwb_register, false, true, true, SingleSpc, 4, 1 ,false}, 265 { ARM::VST1d64TPseudo, ARM::VST1d64T, false, false, false, SingleSpc, 3, 1 ,false}, 266 { ARM::VST1d64TPseudoWB_fixed, ARM::VST1d64Twb_fixed, false, true, false, SingleSpc, 3, 1 ,false}, 267 { ARM::VST1d64TPseudoWB_register, ARM::VST1d64Twb_register, false, true, true, SingleSpc, 3, 1 ,false}, 268 269 { ARM::VST2LNd16Pseudo, ARM::VST2LNd16, false, false, false, SingleSpc, 2, 4 ,true}, 270 { ARM::VST2LNd16Pseudo_UPD, ARM::VST2LNd16_UPD, false, true, true, SingleSpc, 2, 4 ,true}, 271 { ARM::VST2LNd32Pseudo, ARM::VST2LNd32, false, false, false, SingleSpc, 2, 2 ,true}, 272 { ARM::VST2LNd32Pseudo_UPD, ARM::VST2LNd32_UPD, false, true, true, SingleSpc, 2, 2 ,true}, 273 { ARM::VST2LNd8Pseudo, ARM::VST2LNd8, false, false, false, SingleSpc, 2, 8 ,true}, 274 { ARM::VST2LNd8Pseudo_UPD, ARM::VST2LNd8_UPD, false, true, true, SingleSpc, 2, 8 ,true}, 275 { ARM::VST2LNq16Pseudo, ARM::VST2LNq16, false, false, false, EvenDblSpc, 2, 4,true}, 276 { ARM::VST2LNq16Pseudo_UPD, ARM::VST2LNq16_UPD, false, true, true, EvenDblSpc, 2, 4,true}, 277 { ARM::VST2LNq32Pseudo, ARM::VST2LNq32, false, false, false, EvenDblSpc, 2, 2,true}, 278 { ARM::VST2LNq32Pseudo_UPD, ARM::VST2LNq32_UPD, false, true, true, EvenDblSpc, 2, 2,true}, 279 280 { ARM::VST2q16Pseudo, ARM::VST2q16, false, false, false, SingleSpc, 4, 4 ,false}, 281 { ARM::VST2q16PseudoWB_fixed, ARM::VST2q16wb_fixed, false, true, false, SingleSpc, 4, 4 ,false}, 282 { ARM::VST2q16PseudoWB_register, ARM::VST2q16wb_register, false, true, true, SingleSpc, 4, 4 ,false}, 283 { ARM::VST2q32Pseudo, ARM::VST2q32, false, false, false, SingleSpc, 4, 2 ,false}, 284 { ARM::VST2q32PseudoWB_fixed, ARM::VST2q32wb_fixed, false, true, false, SingleSpc, 4, 2 ,false}, 285 { ARM::VST2q32PseudoWB_register, ARM::VST2q32wb_register, false, true, true, SingleSpc, 4, 2 ,false}, 286 { ARM::VST2q8Pseudo, ARM::VST2q8, false, false, false, SingleSpc, 4, 8 ,false}, 287 { ARM::VST2q8PseudoWB_fixed, ARM::VST2q8wb_fixed, false, true, false, SingleSpc, 4, 8 ,false}, 288 { ARM::VST2q8PseudoWB_register, ARM::VST2q8wb_register, false, true, true, SingleSpc, 4, 8 ,false}, 289 290 { ARM::VST3LNd16Pseudo, ARM::VST3LNd16, false, false, false, SingleSpc, 3, 4 ,true}, 291 { ARM::VST3LNd16Pseudo_UPD, ARM::VST3LNd16_UPD, false, true, true, SingleSpc, 3, 4 ,true}, 292 { ARM::VST3LNd32Pseudo, ARM::VST3LNd32, false, false, false, SingleSpc, 3, 2 ,true}, 293 { ARM::VST3LNd32Pseudo_UPD, ARM::VST3LNd32_UPD, false, true, true, SingleSpc, 3, 2 ,true}, 294 { ARM::VST3LNd8Pseudo, ARM::VST3LNd8, false, false, false, SingleSpc, 3, 8 ,true}, 295 { ARM::VST3LNd8Pseudo_UPD, ARM::VST3LNd8_UPD, false, true, true, SingleSpc, 3, 8 ,true}, 296 { ARM::VST3LNq16Pseudo, ARM::VST3LNq16, false, false, false, EvenDblSpc, 3, 4,true}, 297 { ARM::VST3LNq16Pseudo_UPD, ARM::VST3LNq16_UPD, false, true, true, EvenDblSpc, 3, 4,true}, 298 { ARM::VST3LNq32Pseudo, ARM::VST3LNq32, false, false, false, EvenDblSpc, 3, 2,true}, 299 { ARM::VST3LNq32Pseudo_UPD, ARM::VST3LNq32_UPD, false, true, true, EvenDblSpc, 3, 2,true}, 300 301 { ARM::VST3d16Pseudo, ARM::VST3d16, false, false, false, SingleSpc, 3, 4 ,true}, 302 { ARM::VST3d16Pseudo_UPD, ARM::VST3d16_UPD, false, true, true, SingleSpc, 3, 4 ,true}, 303 { ARM::VST3d32Pseudo, ARM::VST3d32, false, false, false, SingleSpc, 3, 2 ,true}, 304 { ARM::VST3d32Pseudo_UPD, ARM::VST3d32_UPD, false, true, true, SingleSpc, 3, 2 ,true}, 305 { ARM::VST3d8Pseudo, ARM::VST3d8, false, false, false, SingleSpc, 3, 8 ,true}, 306 { ARM::VST3d8Pseudo_UPD, ARM::VST3d8_UPD, false, true, true, SingleSpc, 3, 8 ,true}, 307 308 { ARM::VST3q16Pseudo_UPD, ARM::VST3q16_UPD, false, true, true, EvenDblSpc, 3, 4 ,true}, 309 { ARM::VST3q16oddPseudo, ARM::VST3q16, false, false, false, OddDblSpc, 3, 4 ,true}, 310 { ARM::VST3q16oddPseudo_UPD, ARM::VST3q16_UPD, false, true, true, OddDblSpc, 3, 4 ,true}, 311 { ARM::VST3q32Pseudo_UPD, ARM::VST3q32_UPD, false, true, true, EvenDblSpc, 3, 2 ,true}, 312 { ARM::VST3q32oddPseudo, ARM::VST3q32, false, false, false, OddDblSpc, 3, 2 ,true}, 313 { ARM::VST3q32oddPseudo_UPD, ARM::VST3q32_UPD, false, true, true, OddDblSpc, 3, 2 ,true}, 314 { ARM::VST3q8Pseudo_UPD, ARM::VST3q8_UPD, false, true, true, EvenDblSpc, 3, 8 ,true}, 315 { ARM::VST3q8oddPseudo, ARM::VST3q8, false, false, false, OddDblSpc, 3, 8 ,true}, 316 { ARM::VST3q8oddPseudo_UPD, ARM::VST3q8_UPD, false, true, true, OddDblSpc, 3, 8 ,true}, 317 318 { ARM::VST4LNd16Pseudo, ARM::VST4LNd16, false, false, false, SingleSpc, 4, 4 ,true}, 319 { ARM::VST4LNd16Pseudo_UPD, ARM::VST4LNd16_UPD, false, true, true, SingleSpc, 4, 4 ,true}, 320 { ARM::VST4LNd32Pseudo, ARM::VST4LNd32, false, false, false, SingleSpc, 4, 2 ,true}, 321 { ARM::VST4LNd32Pseudo_UPD, ARM::VST4LNd32_UPD, false, true, true, SingleSpc, 4, 2 ,true}, 322 { ARM::VST4LNd8Pseudo, ARM::VST4LNd8, false, false, false, SingleSpc, 4, 8 ,true}, 323 { ARM::VST4LNd8Pseudo_UPD, ARM::VST4LNd8_UPD, false, true, true, SingleSpc, 4, 8 ,true}, 324 { ARM::VST4LNq16Pseudo, ARM::VST4LNq16, false, false, false, EvenDblSpc, 4, 4,true}, 325 { ARM::VST4LNq16Pseudo_UPD, ARM::VST4LNq16_UPD, false, true, true, EvenDblSpc, 4, 4,true}, 326 { ARM::VST4LNq32Pseudo, ARM::VST4LNq32, false, false, false, EvenDblSpc, 4, 2,true}, 327 { ARM::VST4LNq32Pseudo_UPD, ARM::VST4LNq32_UPD, false, true, true, EvenDblSpc, 4, 2,true}, 328 329 { ARM::VST4d16Pseudo, ARM::VST4d16, false, false, false, SingleSpc, 4, 4 ,true}, 330 { ARM::VST4d16Pseudo_UPD, ARM::VST4d16_UPD, false, true, true, SingleSpc, 4, 4 ,true}, 331 { ARM::VST4d32Pseudo, ARM::VST4d32, false, false, false, SingleSpc, 4, 2 ,true}, 332 { ARM::VST4d32Pseudo_UPD, ARM::VST4d32_UPD, false, true, true, SingleSpc, 4, 2 ,true}, 333 { ARM::VST4d8Pseudo, ARM::VST4d8, false, false, false, SingleSpc, 4, 8 ,true}, 334 { ARM::VST4d8Pseudo_UPD, ARM::VST4d8_UPD, false, true, true, SingleSpc, 4, 8 ,true}, 335 336 { ARM::VST4q16Pseudo_UPD, ARM::VST4q16_UPD, false, true, true, EvenDblSpc, 4, 4 ,true}, 337 { ARM::VST4q16oddPseudo, ARM::VST4q16, false, false, false, OddDblSpc, 4, 4 ,true}, 338 { ARM::VST4q16oddPseudo_UPD, ARM::VST4q16_UPD, false, true, true, OddDblSpc, 4, 4 ,true}, 339 { ARM::VST4q32Pseudo_UPD, ARM::VST4q32_UPD, false, true, true, EvenDblSpc, 4, 2 ,true}, 340 { ARM::VST4q32oddPseudo, ARM::VST4q32, false, false, false, OddDblSpc, 4, 2 ,true}, 341 { ARM::VST4q32oddPseudo_UPD, ARM::VST4q32_UPD, false, true, true, OddDblSpc, 4, 2 ,true}, 342 { ARM::VST4q8Pseudo_UPD, ARM::VST4q8_UPD, false, true, true, EvenDblSpc, 4, 8 ,true}, 343 { ARM::VST4q8oddPseudo, ARM::VST4q8, false, false, false, OddDblSpc, 4, 8 ,true}, 344 { ARM::VST4q8oddPseudo_UPD, ARM::VST4q8_UPD, false, true, true, OddDblSpc, 4, 8 ,true} 345 }; 346 347 /// LookupNEONLdSt - Search the NEONLdStTable for information about a NEON 348 /// load or store pseudo instruction. 349 static const NEONLdStTableEntry *LookupNEONLdSt(unsigned Opcode) { 350 #ifndef NDEBUG 351 // Make sure the table is sorted. 352 static bool TableChecked = false; 353 if (!TableChecked) { 354 assert(std::is_sorted(std::begin(NEONLdStTable), std::end(NEONLdStTable)) && 355 "NEONLdStTable is not sorted!"); 356 TableChecked = true; 357 } 358 #endif 359 360 auto I = std::lower_bound(std::begin(NEONLdStTable), 361 std::end(NEONLdStTable), Opcode); 362 if (I != std::end(NEONLdStTable) && I->PseudoOpc == Opcode) 363 return I; 364 return nullptr; 365 } 366 367 /// GetDSubRegs - Get 4 D subregisters of a Q, QQ, or QQQQ register, 368 /// corresponding to the specified register spacing. Not all of the results 369 /// are necessarily valid, e.g., a Q register only has 2 D subregisters. 370 static void GetDSubRegs(unsigned Reg, NEONRegSpacing RegSpc, 371 const TargetRegisterInfo *TRI, unsigned &D0, 372 unsigned &D1, unsigned &D2, unsigned &D3) { 373 if (RegSpc == SingleSpc) { 374 D0 = TRI->getSubReg(Reg, ARM::dsub_0); 375 D1 = TRI->getSubReg(Reg, ARM::dsub_1); 376 D2 = TRI->getSubReg(Reg, ARM::dsub_2); 377 D3 = TRI->getSubReg(Reg, ARM::dsub_3); 378 } else if (RegSpc == EvenDblSpc) { 379 D0 = TRI->getSubReg(Reg, ARM::dsub_0); 380 D1 = TRI->getSubReg(Reg, ARM::dsub_2); 381 D2 = TRI->getSubReg(Reg, ARM::dsub_4); 382 D3 = TRI->getSubReg(Reg, ARM::dsub_6); 383 } else { 384 assert(RegSpc == OddDblSpc && "unknown register spacing"); 385 D0 = TRI->getSubReg(Reg, ARM::dsub_1); 386 D1 = TRI->getSubReg(Reg, ARM::dsub_3); 387 D2 = TRI->getSubReg(Reg, ARM::dsub_5); 388 D3 = TRI->getSubReg(Reg, ARM::dsub_7); 389 } 390 } 391 392 /// ExpandVLD - Translate VLD pseudo instructions with Q, QQ or QQQQ register 393 /// operands to real VLD instructions with D register operands. 394 void ARMExpandPseudo::ExpandVLD(MachineBasicBlock::iterator &MBBI) { 395 MachineInstr &MI = *MBBI; 396 MachineBasicBlock &MBB = *MI.getParent(); 397 398 const NEONLdStTableEntry *TableEntry = LookupNEONLdSt(MI.getOpcode()); 399 assert(TableEntry && TableEntry->IsLoad && "NEONLdStTable lookup failed"); 400 NEONRegSpacing RegSpc = (NEONRegSpacing)TableEntry->RegSpacing; 401 unsigned NumRegs = TableEntry->NumRegs; 402 403 MachineInstrBuilder MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(), 404 TII->get(TableEntry->RealOpc)); 405 unsigned OpIdx = 0; 406 407 bool DstIsDead = MI.getOperand(OpIdx).isDead(); 408 unsigned DstReg = MI.getOperand(OpIdx++).getReg(); 409 unsigned D0, D1, D2, D3; 410 GetDSubRegs(DstReg, RegSpc, TRI, D0, D1, D2, D3); 411 MIB.addReg(D0, RegState::Define | getDeadRegState(DstIsDead)); 412 if (NumRegs > 1 && TableEntry->copyAllListRegs) 413 MIB.addReg(D1, RegState::Define | getDeadRegState(DstIsDead)); 414 if (NumRegs > 2 && TableEntry->copyAllListRegs) 415 MIB.addReg(D2, RegState::Define | getDeadRegState(DstIsDead)); 416 if (NumRegs > 3 && TableEntry->copyAllListRegs) 417 MIB.addReg(D3, RegState::Define | getDeadRegState(DstIsDead)); 418 419 if (TableEntry->isUpdating) 420 MIB.add(MI.getOperand(OpIdx++)); 421 422 // Copy the addrmode6 operands. 423 MIB.add(MI.getOperand(OpIdx++)); 424 MIB.add(MI.getOperand(OpIdx++)); 425 // Copy the am6offset operand. 426 if (TableEntry->hasWritebackOperand) 427 MIB.add(MI.getOperand(OpIdx++)); 428 429 // For an instruction writing double-spaced subregs, the pseudo instruction 430 // has an extra operand that is a use of the super-register. Record the 431 // operand index and skip over it. 432 unsigned SrcOpIdx = 0; 433 if (RegSpc == EvenDblSpc || RegSpc == OddDblSpc) 434 SrcOpIdx = OpIdx++; 435 436 // Copy the predicate operands. 437 MIB.add(MI.getOperand(OpIdx++)); 438 MIB.add(MI.getOperand(OpIdx++)); 439 440 // Copy the super-register source operand used for double-spaced subregs over 441 // to the new instruction as an implicit operand. 442 if (SrcOpIdx != 0) { 443 MachineOperand MO = MI.getOperand(SrcOpIdx); 444 MO.setImplicit(true); 445 MIB.add(MO); 446 } 447 // Add an implicit def for the super-register. 448 MIB.addReg(DstReg, RegState::ImplicitDefine | getDeadRegState(DstIsDead)); 449 TransferImpOps(MI, MIB, MIB); 450 451 // Transfer memoperands. 452 MIB->setMemRefs(MI.memoperands_begin(), MI.memoperands_end()); 453 454 MI.eraseFromParent(); 455 } 456 457 /// ExpandVST - Translate VST pseudo instructions with Q, QQ or QQQQ register 458 /// operands to real VST instructions with D register operands. 459 void ARMExpandPseudo::ExpandVST(MachineBasicBlock::iterator &MBBI) { 460 MachineInstr &MI = *MBBI; 461 MachineBasicBlock &MBB = *MI.getParent(); 462 463 const NEONLdStTableEntry *TableEntry = LookupNEONLdSt(MI.getOpcode()); 464 assert(TableEntry && !TableEntry->IsLoad && "NEONLdStTable lookup failed"); 465 NEONRegSpacing RegSpc = (NEONRegSpacing)TableEntry->RegSpacing; 466 unsigned NumRegs = TableEntry->NumRegs; 467 468 MachineInstrBuilder MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(), 469 TII->get(TableEntry->RealOpc)); 470 unsigned OpIdx = 0; 471 if (TableEntry->isUpdating) 472 MIB.add(MI.getOperand(OpIdx++)); 473 474 // Copy the addrmode6 operands. 475 MIB.add(MI.getOperand(OpIdx++)); 476 MIB.add(MI.getOperand(OpIdx++)); 477 // Copy the am6offset operand. 478 if (TableEntry->hasWritebackOperand) 479 MIB.add(MI.getOperand(OpIdx++)); 480 481 bool SrcIsKill = MI.getOperand(OpIdx).isKill(); 482 bool SrcIsUndef = MI.getOperand(OpIdx).isUndef(); 483 unsigned SrcReg = MI.getOperand(OpIdx++).getReg(); 484 unsigned D0, D1, D2, D3; 485 GetDSubRegs(SrcReg, RegSpc, TRI, D0, D1, D2, D3); 486 MIB.addReg(D0, getUndefRegState(SrcIsUndef)); 487 if (NumRegs > 1 && TableEntry->copyAllListRegs) 488 MIB.addReg(D1, getUndefRegState(SrcIsUndef)); 489 if (NumRegs > 2 && TableEntry->copyAllListRegs) 490 MIB.addReg(D2, getUndefRegState(SrcIsUndef)); 491 if (NumRegs > 3 && TableEntry->copyAllListRegs) 492 MIB.addReg(D3, getUndefRegState(SrcIsUndef)); 493 494 // Copy the predicate operands. 495 MIB.add(MI.getOperand(OpIdx++)); 496 MIB.add(MI.getOperand(OpIdx++)); 497 498 if (SrcIsKill && !SrcIsUndef) // Add an implicit kill for the super-reg. 499 MIB->addRegisterKilled(SrcReg, TRI, true); 500 else if (!SrcIsUndef) 501 MIB.addReg(SrcReg, RegState::Implicit); // Add implicit uses for src reg. 502 TransferImpOps(MI, MIB, MIB); 503 504 // Transfer memoperands. 505 MIB->setMemRefs(MI.memoperands_begin(), MI.memoperands_end()); 506 507 MI.eraseFromParent(); 508 } 509 510 /// ExpandLaneOp - Translate VLD*LN and VST*LN instructions with Q, QQ or QQQQ 511 /// register operands to real instructions with D register operands. 512 void ARMExpandPseudo::ExpandLaneOp(MachineBasicBlock::iterator &MBBI) { 513 MachineInstr &MI = *MBBI; 514 MachineBasicBlock &MBB = *MI.getParent(); 515 516 const NEONLdStTableEntry *TableEntry = LookupNEONLdSt(MI.getOpcode()); 517 assert(TableEntry && "NEONLdStTable lookup failed"); 518 NEONRegSpacing RegSpc = (NEONRegSpacing)TableEntry->RegSpacing; 519 unsigned NumRegs = TableEntry->NumRegs; 520 unsigned RegElts = TableEntry->RegElts; 521 522 MachineInstrBuilder MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(), 523 TII->get(TableEntry->RealOpc)); 524 unsigned OpIdx = 0; 525 // The lane operand is always the 3rd from last operand, before the 2 526 // predicate operands. 527 unsigned Lane = MI.getOperand(MI.getDesc().getNumOperands() - 3).getImm(); 528 529 // Adjust the lane and spacing as needed for Q registers. 530 assert(RegSpc != OddDblSpc && "unexpected register spacing for VLD/VST-lane"); 531 if (RegSpc == EvenDblSpc && Lane >= RegElts) { 532 RegSpc = OddDblSpc; 533 Lane -= RegElts; 534 } 535 assert(Lane < RegElts && "out of range lane for VLD/VST-lane"); 536 537 unsigned D0 = 0, D1 = 0, D2 = 0, D3 = 0; 538 unsigned DstReg = 0; 539 bool DstIsDead = false; 540 if (TableEntry->IsLoad) { 541 DstIsDead = MI.getOperand(OpIdx).isDead(); 542 DstReg = MI.getOperand(OpIdx++).getReg(); 543 GetDSubRegs(DstReg, RegSpc, TRI, D0, D1, D2, D3); 544 MIB.addReg(D0, RegState::Define | getDeadRegState(DstIsDead)); 545 if (NumRegs > 1) 546 MIB.addReg(D1, RegState::Define | getDeadRegState(DstIsDead)); 547 if (NumRegs > 2) 548 MIB.addReg(D2, RegState::Define | getDeadRegState(DstIsDead)); 549 if (NumRegs > 3) 550 MIB.addReg(D3, RegState::Define | getDeadRegState(DstIsDead)); 551 } 552 553 if (TableEntry->isUpdating) 554 MIB.add(MI.getOperand(OpIdx++)); 555 556 // Copy the addrmode6 operands. 557 MIB.add(MI.getOperand(OpIdx++)); 558 MIB.add(MI.getOperand(OpIdx++)); 559 // Copy the am6offset operand. 560 if (TableEntry->hasWritebackOperand) 561 MIB.add(MI.getOperand(OpIdx++)); 562 563 // Grab the super-register source. 564 MachineOperand MO = MI.getOperand(OpIdx++); 565 if (!TableEntry->IsLoad) 566 GetDSubRegs(MO.getReg(), RegSpc, TRI, D0, D1, D2, D3); 567 568 // Add the subregs as sources of the new instruction. 569 unsigned SrcFlags = (getUndefRegState(MO.isUndef()) | 570 getKillRegState(MO.isKill())); 571 MIB.addReg(D0, SrcFlags); 572 if (NumRegs > 1) 573 MIB.addReg(D1, SrcFlags); 574 if (NumRegs > 2) 575 MIB.addReg(D2, SrcFlags); 576 if (NumRegs > 3) 577 MIB.addReg(D3, SrcFlags); 578 579 // Add the lane number operand. 580 MIB.addImm(Lane); 581 OpIdx += 1; 582 583 // Copy the predicate operands. 584 MIB.add(MI.getOperand(OpIdx++)); 585 MIB.add(MI.getOperand(OpIdx++)); 586 587 // Copy the super-register source to be an implicit source. 588 MO.setImplicit(true); 589 MIB.add(MO); 590 if (TableEntry->IsLoad) 591 // Add an implicit def for the super-register. 592 MIB.addReg(DstReg, RegState::ImplicitDefine | getDeadRegState(DstIsDead)); 593 TransferImpOps(MI, MIB, MIB); 594 // Transfer memoperands. 595 MIB->setMemRefs(MI.memoperands_begin(), MI.memoperands_end()); 596 MI.eraseFromParent(); 597 } 598 599 /// ExpandVTBL - Translate VTBL and VTBX pseudo instructions with Q or QQ 600 /// register operands to real instructions with D register operands. 601 void ARMExpandPseudo::ExpandVTBL(MachineBasicBlock::iterator &MBBI, 602 unsigned Opc, bool IsExt) { 603 MachineInstr &MI = *MBBI; 604 MachineBasicBlock &MBB = *MI.getParent(); 605 606 MachineInstrBuilder MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(Opc)); 607 unsigned OpIdx = 0; 608 609 // Transfer the destination register operand. 610 MIB.add(MI.getOperand(OpIdx++)); 611 if (IsExt) 612 MIB.add(MI.getOperand(OpIdx++)); 613 614 bool SrcIsKill = MI.getOperand(OpIdx).isKill(); 615 unsigned SrcReg = MI.getOperand(OpIdx++).getReg(); 616 unsigned D0, D1, D2, D3; 617 GetDSubRegs(SrcReg, SingleSpc, TRI, D0, D1, D2, D3); 618 MIB.addReg(D0); 619 620 // Copy the other source register operand. 621 MIB.add(MI.getOperand(OpIdx++)); 622 623 // Copy the predicate operands. 624 MIB.add(MI.getOperand(OpIdx++)); 625 MIB.add(MI.getOperand(OpIdx++)); 626 627 // Add an implicit kill and use for the super-reg. 628 MIB.addReg(SrcReg, RegState::Implicit | getKillRegState(SrcIsKill)); 629 TransferImpOps(MI, MIB, MIB); 630 MI.eraseFromParent(); 631 } 632 633 static bool IsAnAddressOperand(const MachineOperand &MO) { 634 // This check is overly conservative. Unless we are certain that the machine 635 // operand is not a symbol reference, we return that it is a symbol reference. 636 // This is important as the load pair may not be split up Windows. 637 switch (MO.getType()) { 638 case MachineOperand::MO_Register: 639 case MachineOperand::MO_Immediate: 640 case MachineOperand::MO_CImmediate: 641 case MachineOperand::MO_FPImmediate: 642 return false; 643 case MachineOperand::MO_MachineBasicBlock: 644 return true; 645 case MachineOperand::MO_FrameIndex: 646 return false; 647 case MachineOperand::MO_ConstantPoolIndex: 648 case MachineOperand::MO_TargetIndex: 649 case MachineOperand::MO_JumpTableIndex: 650 case MachineOperand::MO_ExternalSymbol: 651 case MachineOperand::MO_GlobalAddress: 652 case MachineOperand::MO_BlockAddress: 653 return true; 654 case MachineOperand::MO_RegisterMask: 655 case MachineOperand::MO_RegisterLiveOut: 656 return false; 657 case MachineOperand::MO_Metadata: 658 case MachineOperand::MO_MCSymbol: 659 return true; 660 case MachineOperand::MO_CFIIndex: 661 return false; 662 case MachineOperand::MO_IntrinsicID: 663 case MachineOperand::MO_Predicate: 664 llvm_unreachable("should not exist post-isel"); 665 } 666 llvm_unreachable("unhandled machine operand type"); 667 } 668 669 void ARMExpandPseudo::ExpandMOV32BitImm(MachineBasicBlock &MBB, 670 MachineBasicBlock::iterator &MBBI) { 671 MachineInstr &MI = *MBBI; 672 unsigned Opcode = MI.getOpcode(); 673 unsigned PredReg = 0; 674 ARMCC::CondCodes Pred = getInstrPredicate(MI, PredReg); 675 unsigned DstReg = MI.getOperand(0).getReg(); 676 bool DstIsDead = MI.getOperand(0).isDead(); 677 bool isCC = Opcode == ARM::MOVCCi32imm || Opcode == ARM::t2MOVCCi32imm; 678 const MachineOperand &MO = MI.getOperand(isCC ? 2 : 1); 679 bool RequiresBundling = STI->isTargetWindows() && IsAnAddressOperand(MO); 680 MachineInstrBuilder LO16, HI16; 681 682 if (!STI->hasV6T2Ops() && 683 (Opcode == ARM::MOVi32imm || Opcode == ARM::MOVCCi32imm)) { 684 // FIXME Windows CE supports older ARM CPUs 685 assert(!STI->isTargetWindows() && "Windows on ARM requires ARMv7+"); 686 687 // Expand into a movi + orr. 688 LO16 = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVi), DstReg); 689 HI16 = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::ORRri)) 690 .addReg(DstReg, RegState::Define | getDeadRegState(DstIsDead)) 691 .addReg(DstReg); 692 693 assert (MO.isImm() && "MOVi32imm w/ non-immediate source operand!"); 694 unsigned ImmVal = (unsigned)MO.getImm(); 695 unsigned SOImmValV1 = ARM_AM::getSOImmTwoPartFirst(ImmVal); 696 unsigned SOImmValV2 = ARM_AM::getSOImmTwoPartSecond(ImmVal); 697 LO16 = LO16.addImm(SOImmValV1); 698 HI16 = HI16.addImm(SOImmValV2); 699 LO16->setMemRefs(MI.memoperands_begin(), MI.memoperands_end()); 700 HI16->setMemRefs(MI.memoperands_begin(), MI.memoperands_end()); 701 LO16.addImm(Pred).addReg(PredReg).add(condCodeOp()); 702 HI16.addImm(Pred).addReg(PredReg).add(condCodeOp()); 703 TransferImpOps(MI, LO16, HI16); 704 MI.eraseFromParent(); 705 return; 706 } 707 708 unsigned LO16Opc = 0; 709 unsigned HI16Opc = 0; 710 if (Opcode == ARM::t2MOVi32imm || Opcode == ARM::t2MOVCCi32imm) { 711 LO16Opc = ARM::t2MOVi16; 712 HI16Opc = ARM::t2MOVTi16; 713 } else { 714 LO16Opc = ARM::MOVi16; 715 HI16Opc = ARM::MOVTi16; 716 } 717 718 LO16 = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(LO16Opc), DstReg); 719 HI16 = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(HI16Opc)) 720 .addReg(DstReg, RegState::Define | getDeadRegState(DstIsDead)) 721 .addReg(DstReg); 722 723 switch (MO.getType()) { 724 case MachineOperand::MO_Immediate: { 725 unsigned Imm = MO.getImm(); 726 unsigned Lo16 = Imm & 0xffff; 727 unsigned Hi16 = (Imm >> 16) & 0xffff; 728 LO16 = LO16.addImm(Lo16); 729 HI16 = HI16.addImm(Hi16); 730 break; 731 } 732 case MachineOperand::MO_ExternalSymbol: { 733 const char *ES = MO.getSymbolName(); 734 unsigned TF = MO.getTargetFlags(); 735 LO16 = LO16.addExternalSymbol(ES, TF | ARMII::MO_LO16); 736 HI16 = HI16.addExternalSymbol(ES, TF | ARMII::MO_HI16); 737 break; 738 } 739 default: { 740 const GlobalValue *GV = MO.getGlobal(); 741 unsigned TF = MO.getTargetFlags(); 742 LO16 = LO16.addGlobalAddress(GV, MO.getOffset(), TF | ARMII::MO_LO16); 743 HI16 = HI16.addGlobalAddress(GV, MO.getOffset(), TF | ARMII::MO_HI16); 744 break; 745 } 746 } 747 748 LO16->setMemRefs(MI.memoperands_begin(), MI.memoperands_end()); 749 HI16->setMemRefs(MI.memoperands_begin(), MI.memoperands_end()); 750 LO16.addImm(Pred).addReg(PredReg); 751 HI16.addImm(Pred).addReg(PredReg); 752 753 if (RequiresBundling) 754 finalizeBundle(MBB, LO16->getIterator(), MBBI->getIterator()); 755 756 TransferImpOps(MI, LO16, HI16); 757 MI.eraseFromParent(); 758 } 759 760 /// Expand a CMP_SWAP pseudo-inst to an ldrex/strex loop as simply as 761 /// possible. This only gets used at -O0 so we don't care about efficiency of 762 /// the generated code. 763 bool ARMExpandPseudo::ExpandCMP_SWAP(MachineBasicBlock &MBB, 764 MachineBasicBlock::iterator MBBI, 765 unsigned LdrexOp, unsigned StrexOp, 766 unsigned UxtOp, 767 MachineBasicBlock::iterator &NextMBBI) { 768 bool IsThumb = STI->isThumb(); 769 MachineInstr &MI = *MBBI; 770 DebugLoc DL = MI.getDebugLoc(); 771 const MachineOperand &Dest = MI.getOperand(0); 772 unsigned TempReg = MI.getOperand(1).getReg(); 773 // Duplicating undef operands into 2 instructions does not guarantee the same 774 // value on both; However undef should be replaced by xzr anyway. 775 assert(!MI.getOperand(2).isUndef() && "cannot handle undef"); 776 unsigned AddrReg = MI.getOperand(2).getReg(); 777 unsigned DesiredReg = MI.getOperand(3).getReg(); 778 unsigned NewReg = MI.getOperand(4).getReg(); 779 780 MachineFunction *MF = MBB.getParent(); 781 auto LoadCmpBB = MF->CreateMachineBasicBlock(MBB.getBasicBlock()); 782 auto StoreBB = MF->CreateMachineBasicBlock(MBB.getBasicBlock()); 783 auto DoneBB = MF->CreateMachineBasicBlock(MBB.getBasicBlock()); 784 785 MF->insert(++MBB.getIterator(), LoadCmpBB); 786 MF->insert(++LoadCmpBB->getIterator(), StoreBB); 787 MF->insert(++StoreBB->getIterator(), DoneBB); 788 789 if (UxtOp) { 790 MachineInstrBuilder MIB = 791 BuildMI(MBB, MBBI, DL, TII->get(UxtOp), DesiredReg) 792 .addReg(DesiredReg, RegState::Kill); 793 if (!IsThumb) 794 MIB.addImm(0); 795 MIB.add(predOps(ARMCC::AL)); 796 } 797 798 // .Lloadcmp: 799 // ldrex rDest, [rAddr] 800 // cmp rDest, rDesired 801 // bne .Ldone 802 803 MachineInstrBuilder MIB; 804 MIB = BuildMI(LoadCmpBB, DL, TII->get(LdrexOp), Dest.getReg()); 805 MIB.addReg(AddrReg); 806 if (LdrexOp == ARM::t2LDREX) 807 MIB.addImm(0); // a 32-bit Thumb ldrex (only) allows an offset. 808 MIB.add(predOps(ARMCC::AL)); 809 810 unsigned CMPrr = IsThumb ? ARM::tCMPhir : ARM::CMPrr; 811 BuildMI(LoadCmpBB, DL, TII->get(CMPrr)) 812 .addReg(Dest.getReg(), getKillRegState(Dest.isDead())) 813 .addReg(DesiredReg) 814 .add(predOps(ARMCC::AL)); 815 unsigned Bcc = IsThumb ? ARM::tBcc : ARM::Bcc; 816 BuildMI(LoadCmpBB, DL, TII->get(Bcc)) 817 .addMBB(DoneBB) 818 .addImm(ARMCC::NE) 819 .addReg(ARM::CPSR, RegState::Kill); 820 LoadCmpBB->addSuccessor(DoneBB); 821 LoadCmpBB->addSuccessor(StoreBB); 822 823 // .Lstore: 824 // strex rTempReg, rNew, [rAddr] 825 // cmp rTempReg, #0 826 // bne .Lloadcmp 827 MIB = BuildMI(StoreBB, DL, TII->get(StrexOp), TempReg) 828 .addReg(NewReg) 829 .addReg(AddrReg); 830 if (StrexOp == ARM::t2STREX) 831 MIB.addImm(0); // a 32-bit Thumb strex (only) allows an offset. 832 MIB.add(predOps(ARMCC::AL)); 833 834 unsigned CMPri = IsThumb ? ARM::t2CMPri : ARM::CMPri; 835 BuildMI(StoreBB, DL, TII->get(CMPri)) 836 .addReg(TempReg, RegState::Kill) 837 .addImm(0) 838 .add(predOps(ARMCC::AL)); 839 BuildMI(StoreBB, DL, TII->get(Bcc)) 840 .addMBB(LoadCmpBB) 841 .addImm(ARMCC::NE) 842 .addReg(ARM::CPSR, RegState::Kill); 843 StoreBB->addSuccessor(LoadCmpBB); 844 StoreBB->addSuccessor(DoneBB); 845 846 DoneBB->splice(DoneBB->end(), &MBB, MI, MBB.end()); 847 DoneBB->transferSuccessors(&MBB); 848 849 MBB.addSuccessor(LoadCmpBB); 850 851 NextMBBI = MBB.end(); 852 MI.eraseFromParent(); 853 854 // Recompute livein lists. 855 const MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo(); 856 LivePhysRegs LiveRegs; 857 computeLiveIns(LiveRegs, MRI, *DoneBB); 858 computeLiveIns(LiveRegs, MRI, *StoreBB); 859 computeLiveIns(LiveRegs, MRI, *LoadCmpBB); 860 // Do an extra pass around the loop to get loop carried registers right. 861 StoreBB->clearLiveIns(); 862 computeLiveIns(LiveRegs, MRI, *StoreBB); 863 LoadCmpBB->clearLiveIns(); 864 computeLiveIns(LiveRegs, MRI, *LoadCmpBB); 865 866 return true; 867 } 868 869 /// ARM's ldrexd/strexd take a consecutive register pair (represented as a 870 /// single GPRPair register), Thumb's take two separate registers so we need to 871 /// extract the subregs from the pair. 872 static void addExclusiveRegPair(MachineInstrBuilder &MIB, MachineOperand &Reg, 873 unsigned Flags, bool IsThumb, 874 const TargetRegisterInfo *TRI) { 875 if (IsThumb) { 876 unsigned RegLo = TRI->getSubReg(Reg.getReg(), ARM::gsub_0); 877 unsigned RegHi = TRI->getSubReg(Reg.getReg(), ARM::gsub_1); 878 MIB.addReg(RegLo, Flags | getKillRegState(Reg.isDead())); 879 MIB.addReg(RegHi, Flags | getKillRegState(Reg.isDead())); 880 } else 881 MIB.addReg(Reg.getReg(), Flags | getKillRegState(Reg.isDead())); 882 } 883 884 /// Expand a 64-bit CMP_SWAP to an ldrexd/strexd loop. 885 bool ARMExpandPseudo::ExpandCMP_SWAP_64(MachineBasicBlock &MBB, 886 MachineBasicBlock::iterator MBBI, 887 MachineBasicBlock::iterator &NextMBBI) { 888 bool IsThumb = STI->isThumb(); 889 MachineInstr &MI = *MBBI; 890 DebugLoc DL = MI.getDebugLoc(); 891 MachineOperand &Dest = MI.getOperand(0); 892 unsigned TempReg = MI.getOperand(1).getReg(); 893 // Duplicating undef operands into 2 instructions does not guarantee the same 894 // value on both; However undef should be replaced by xzr anyway. 895 assert(!MI.getOperand(2).isUndef() && "cannot handle undef"); 896 unsigned AddrReg = MI.getOperand(2).getReg(); 897 unsigned DesiredReg = MI.getOperand(3).getReg(); 898 MachineOperand New = MI.getOperand(4); 899 New.setIsKill(false); 900 901 unsigned DestLo = TRI->getSubReg(Dest.getReg(), ARM::gsub_0); 902 unsigned DestHi = TRI->getSubReg(Dest.getReg(), ARM::gsub_1); 903 unsigned DesiredLo = TRI->getSubReg(DesiredReg, ARM::gsub_0); 904 unsigned DesiredHi = TRI->getSubReg(DesiredReg, ARM::gsub_1); 905 906 MachineFunction *MF = MBB.getParent(); 907 auto LoadCmpBB = MF->CreateMachineBasicBlock(MBB.getBasicBlock()); 908 auto StoreBB = MF->CreateMachineBasicBlock(MBB.getBasicBlock()); 909 auto DoneBB = MF->CreateMachineBasicBlock(MBB.getBasicBlock()); 910 911 MF->insert(++MBB.getIterator(), LoadCmpBB); 912 MF->insert(++LoadCmpBB->getIterator(), StoreBB); 913 MF->insert(++StoreBB->getIterator(), DoneBB); 914 915 // .Lloadcmp: 916 // ldrexd rDestLo, rDestHi, [rAddr] 917 // cmp rDestLo, rDesiredLo 918 // sbcs rTempReg<dead>, rDestHi, rDesiredHi 919 // bne .Ldone 920 unsigned LDREXD = IsThumb ? ARM::t2LDREXD : ARM::LDREXD; 921 MachineInstrBuilder MIB; 922 MIB = BuildMI(LoadCmpBB, DL, TII->get(LDREXD)); 923 addExclusiveRegPair(MIB, Dest, RegState::Define, IsThumb, TRI); 924 MIB.addReg(AddrReg).add(predOps(ARMCC::AL)); 925 926 unsigned CMPrr = IsThumb ? ARM::tCMPhir : ARM::CMPrr; 927 BuildMI(LoadCmpBB, DL, TII->get(CMPrr)) 928 .addReg(DestLo, getKillRegState(Dest.isDead())) 929 .addReg(DesiredLo) 930 .add(predOps(ARMCC::AL)); 931 932 BuildMI(LoadCmpBB, DL, TII->get(CMPrr)) 933 .addReg(DestHi, getKillRegState(Dest.isDead())) 934 .addReg(DesiredHi) 935 .addImm(ARMCC::EQ).addReg(ARM::CPSR, RegState::Kill); 936 937 unsigned Bcc = IsThumb ? ARM::tBcc : ARM::Bcc; 938 BuildMI(LoadCmpBB, DL, TII->get(Bcc)) 939 .addMBB(DoneBB) 940 .addImm(ARMCC::NE) 941 .addReg(ARM::CPSR, RegState::Kill); 942 LoadCmpBB->addSuccessor(DoneBB); 943 LoadCmpBB->addSuccessor(StoreBB); 944 945 // .Lstore: 946 // strexd rTempReg, rNewLo, rNewHi, [rAddr] 947 // cmp rTempReg, #0 948 // bne .Lloadcmp 949 unsigned STREXD = IsThumb ? ARM::t2STREXD : ARM::STREXD; 950 MIB = BuildMI(StoreBB, DL, TII->get(STREXD), TempReg); 951 addExclusiveRegPair(MIB, New, 0, IsThumb, TRI); 952 MIB.addReg(AddrReg).add(predOps(ARMCC::AL)); 953 954 unsigned CMPri = IsThumb ? ARM::t2CMPri : ARM::CMPri; 955 BuildMI(StoreBB, DL, TII->get(CMPri)) 956 .addReg(TempReg, RegState::Kill) 957 .addImm(0) 958 .add(predOps(ARMCC::AL)); 959 BuildMI(StoreBB, DL, TII->get(Bcc)) 960 .addMBB(LoadCmpBB) 961 .addImm(ARMCC::NE) 962 .addReg(ARM::CPSR, RegState::Kill); 963 StoreBB->addSuccessor(LoadCmpBB); 964 StoreBB->addSuccessor(DoneBB); 965 966 DoneBB->splice(DoneBB->end(), &MBB, MI, MBB.end()); 967 DoneBB->transferSuccessors(&MBB); 968 969 MBB.addSuccessor(LoadCmpBB); 970 971 NextMBBI = MBB.end(); 972 MI.eraseFromParent(); 973 974 // Recompute livein lists. 975 const MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo(); 976 LivePhysRegs LiveRegs; 977 computeLiveIns(LiveRegs, MRI, *DoneBB); 978 computeLiveIns(LiveRegs, MRI, *StoreBB); 979 computeLiveIns(LiveRegs, MRI, *LoadCmpBB); 980 // Do an extra pass around the loop to get loop carried registers right. 981 StoreBB->clearLiveIns(); 982 computeLiveIns(LiveRegs, MRI, *StoreBB); 983 LoadCmpBB->clearLiveIns(); 984 computeLiveIns(LiveRegs, MRI, *LoadCmpBB); 985 986 return true; 987 } 988 989 990 bool ARMExpandPseudo::ExpandMI(MachineBasicBlock &MBB, 991 MachineBasicBlock::iterator MBBI, 992 MachineBasicBlock::iterator &NextMBBI) { 993 MachineInstr &MI = *MBBI; 994 unsigned Opcode = MI.getOpcode(); 995 switch (Opcode) { 996 default: 997 return false; 998 999 case ARM::TCRETURNdi: 1000 case ARM::TCRETURNri: { 1001 MachineBasicBlock::iterator MBBI = MBB.getLastNonDebugInstr(); 1002 assert(MBBI->isReturn() && 1003 "Can only insert epilog into returning blocks"); 1004 unsigned RetOpcode = MBBI->getOpcode(); 1005 DebugLoc dl = MBBI->getDebugLoc(); 1006 const ARMBaseInstrInfo &TII = *static_cast<const ARMBaseInstrInfo *>( 1007 MBB.getParent()->getSubtarget().getInstrInfo()); 1008 1009 // Tail call return: adjust the stack pointer and jump to callee. 1010 MBBI = MBB.getLastNonDebugInstr(); 1011 MachineOperand &JumpTarget = MBBI->getOperand(0); 1012 1013 // Jump to label or value in register. 1014 if (RetOpcode == ARM::TCRETURNdi) { 1015 unsigned TCOpcode = 1016 STI->isThumb() 1017 ? (STI->isTargetMachO() ? ARM::tTAILJMPd : ARM::tTAILJMPdND) 1018 : ARM::TAILJMPd; 1019 MachineInstrBuilder MIB = BuildMI(MBB, MBBI, dl, TII.get(TCOpcode)); 1020 if (JumpTarget.isGlobal()) 1021 MIB.addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset(), 1022 JumpTarget.getTargetFlags()); 1023 else { 1024 assert(JumpTarget.isSymbol()); 1025 MIB.addExternalSymbol(JumpTarget.getSymbolName(), 1026 JumpTarget.getTargetFlags()); 1027 } 1028 1029 // Add the default predicate in Thumb mode. 1030 if (STI->isThumb()) 1031 MIB.add(predOps(ARMCC::AL)); 1032 } else if (RetOpcode == ARM::TCRETURNri) { 1033 BuildMI(MBB, MBBI, dl, 1034 TII.get(STI->isThumb() ? ARM::tTAILJMPr : ARM::TAILJMPr)) 1035 .addReg(JumpTarget.getReg(), RegState::Kill); 1036 } 1037 1038 auto NewMI = std::prev(MBBI); 1039 for (unsigned i = 1, e = MBBI->getNumOperands(); i != e; ++i) 1040 NewMI->addOperand(MBBI->getOperand(i)); 1041 1042 // Delete the pseudo instruction TCRETURN. 1043 MBB.erase(MBBI); 1044 MBBI = NewMI; 1045 return true; 1046 } 1047 case ARM::VMOVScc: 1048 case ARM::VMOVDcc: { 1049 unsigned newOpc = Opcode == ARM::VMOVScc ? ARM::VMOVS : ARM::VMOVD; 1050 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(newOpc), 1051 MI.getOperand(1).getReg()) 1052 .add(MI.getOperand(2)) 1053 .addImm(MI.getOperand(3).getImm()) // 'pred' 1054 .add(MI.getOperand(4)); 1055 1056 MI.eraseFromParent(); 1057 return true; 1058 } 1059 case ARM::t2MOVCCr: 1060 case ARM::MOVCCr: { 1061 unsigned Opc = AFI->isThumbFunction() ? ARM::t2MOVr : ARM::MOVr; 1062 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(Opc), 1063 MI.getOperand(1).getReg()) 1064 .add(MI.getOperand(2)) 1065 .addImm(MI.getOperand(3).getImm()) // 'pred' 1066 .add(MI.getOperand(4)) 1067 .add(condCodeOp()); // 's' bit 1068 1069 MI.eraseFromParent(); 1070 return true; 1071 } 1072 case ARM::MOVCCsi: { 1073 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVsi), 1074 (MI.getOperand(1).getReg())) 1075 .add(MI.getOperand(2)) 1076 .addImm(MI.getOperand(3).getImm()) 1077 .addImm(MI.getOperand(4).getImm()) // 'pred' 1078 .add(MI.getOperand(5)) 1079 .add(condCodeOp()); // 's' bit 1080 1081 MI.eraseFromParent(); 1082 return true; 1083 } 1084 case ARM::MOVCCsr: { 1085 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVsr), 1086 (MI.getOperand(1).getReg())) 1087 .add(MI.getOperand(2)) 1088 .add(MI.getOperand(3)) 1089 .addImm(MI.getOperand(4).getImm()) 1090 .addImm(MI.getOperand(5).getImm()) // 'pred' 1091 .add(MI.getOperand(6)) 1092 .add(condCodeOp()); // 's' bit 1093 1094 MI.eraseFromParent(); 1095 return true; 1096 } 1097 case ARM::t2MOVCCi16: 1098 case ARM::MOVCCi16: { 1099 unsigned NewOpc = AFI->isThumbFunction() ? ARM::t2MOVi16 : ARM::MOVi16; 1100 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(NewOpc), 1101 MI.getOperand(1).getReg()) 1102 .addImm(MI.getOperand(2).getImm()) 1103 .addImm(MI.getOperand(3).getImm()) // 'pred' 1104 .add(MI.getOperand(4)); 1105 MI.eraseFromParent(); 1106 return true; 1107 } 1108 case ARM::t2MOVCCi: 1109 case ARM::MOVCCi: { 1110 unsigned Opc = AFI->isThumbFunction() ? ARM::t2MOVi : ARM::MOVi; 1111 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(Opc), 1112 MI.getOperand(1).getReg()) 1113 .addImm(MI.getOperand(2).getImm()) 1114 .addImm(MI.getOperand(3).getImm()) // 'pred' 1115 .add(MI.getOperand(4)) 1116 .add(condCodeOp()); // 's' bit 1117 1118 MI.eraseFromParent(); 1119 return true; 1120 } 1121 case ARM::t2MVNCCi: 1122 case ARM::MVNCCi: { 1123 unsigned Opc = AFI->isThumbFunction() ? ARM::t2MVNi : ARM::MVNi; 1124 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(Opc), 1125 MI.getOperand(1).getReg()) 1126 .addImm(MI.getOperand(2).getImm()) 1127 .addImm(MI.getOperand(3).getImm()) // 'pred' 1128 .add(MI.getOperand(4)) 1129 .add(condCodeOp()); // 's' bit 1130 1131 MI.eraseFromParent(); 1132 return true; 1133 } 1134 case ARM::t2MOVCClsl: 1135 case ARM::t2MOVCClsr: 1136 case ARM::t2MOVCCasr: 1137 case ARM::t2MOVCCror: { 1138 unsigned NewOpc; 1139 switch (Opcode) { 1140 case ARM::t2MOVCClsl: NewOpc = ARM::t2LSLri; break; 1141 case ARM::t2MOVCClsr: NewOpc = ARM::t2LSRri; break; 1142 case ARM::t2MOVCCasr: NewOpc = ARM::t2ASRri; break; 1143 case ARM::t2MOVCCror: NewOpc = ARM::t2RORri; break; 1144 default: llvm_unreachable("unexpeced conditional move"); 1145 } 1146 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(NewOpc), 1147 MI.getOperand(1).getReg()) 1148 .add(MI.getOperand(2)) 1149 .addImm(MI.getOperand(3).getImm()) 1150 .addImm(MI.getOperand(4).getImm()) // 'pred' 1151 .add(MI.getOperand(5)) 1152 .add(condCodeOp()); // 's' bit 1153 MI.eraseFromParent(); 1154 return true; 1155 } 1156 case ARM::Int_eh_sjlj_dispatchsetup: { 1157 MachineFunction &MF = *MI.getParent()->getParent(); 1158 const ARMBaseInstrInfo *AII = 1159 static_cast<const ARMBaseInstrInfo*>(TII); 1160 const ARMBaseRegisterInfo &RI = AII->getRegisterInfo(); 1161 // For functions using a base pointer, we rematerialize it (via the frame 1162 // pointer) here since eh.sjlj.setjmp and eh.sjlj.longjmp don't do it 1163 // for us. Otherwise, expand to nothing. 1164 if (RI.hasBasePointer(MF)) { 1165 int32_t NumBytes = AFI->getFramePtrSpillOffset(); 1166 unsigned FramePtr = RI.getFrameRegister(MF); 1167 assert(MF.getSubtarget().getFrameLowering()->hasFP(MF) && 1168 "base pointer without frame pointer?"); 1169 1170 if (AFI->isThumb2Function()) { 1171 emitT2RegPlusImmediate(MBB, MBBI, MI.getDebugLoc(), ARM::R6, 1172 FramePtr, -NumBytes, ARMCC::AL, 0, *TII); 1173 } else if (AFI->isThumbFunction()) { 1174 emitThumbRegPlusImmediate(MBB, MBBI, MI.getDebugLoc(), ARM::R6, 1175 FramePtr, -NumBytes, *TII, RI); 1176 } else { 1177 emitARMRegPlusImmediate(MBB, MBBI, MI.getDebugLoc(), ARM::R6, 1178 FramePtr, -NumBytes, ARMCC::AL, 0, 1179 *TII); 1180 } 1181 // If there's dynamic realignment, adjust for it. 1182 if (RI.needsStackRealignment(MF)) { 1183 MachineFrameInfo &MFI = MF.getFrameInfo(); 1184 unsigned MaxAlign = MFI.getMaxAlignment(); 1185 assert (!AFI->isThumb1OnlyFunction()); 1186 // Emit bic r6, r6, MaxAlign 1187 assert(MaxAlign <= 256 && "The BIC instruction cannot encode " 1188 "immediates larger than 256 with all lower " 1189 "bits set."); 1190 unsigned bicOpc = AFI->isThumbFunction() ? 1191 ARM::t2BICri : ARM::BICri; 1192 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(bicOpc), ARM::R6) 1193 .addReg(ARM::R6, RegState::Kill) 1194 .addImm(MaxAlign - 1) 1195 .add(predOps(ARMCC::AL)) 1196 .add(condCodeOp()); 1197 } 1198 1199 } 1200 MI.eraseFromParent(); 1201 return true; 1202 } 1203 1204 case ARM::MOVsrl_flag: 1205 case ARM::MOVsra_flag: { 1206 // These are just fancy MOVs instructions. 1207 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVsi), 1208 MI.getOperand(0).getReg()) 1209 .add(MI.getOperand(1)) 1210 .addImm(ARM_AM::getSORegOpc( 1211 (Opcode == ARM::MOVsrl_flag ? ARM_AM::lsr : ARM_AM::asr), 1)) 1212 .add(predOps(ARMCC::AL)) 1213 .addReg(ARM::CPSR, RegState::Define); 1214 MI.eraseFromParent(); 1215 return true; 1216 } 1217 case ARM::RRX: { 1218 // This encodes as "MOVs Rd, Rm, rrx 1219 MachineInstrBuilder MIB = 1220 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVsi), 1221 MI.getOperand(0).getReg()) 1222 .add(MI.getOperand(1)) 1223 .addImm(ARM_AM::getSORegOpc(ARM_AM::rrx, 0)) 1224 .add(predOps(ARMCC::AL)) 1225 .add(condCodeOp()); 1226 TransferImpOps(MI, MIB, MIB); 1227 MI.eraseFromParent(); 1228 return true; 1229 } 1230 case ARM::tTPsoft: 1231 case ARM::TPsoft: { 1232 const bool Thumb = Opcode == ARM::tTPsoft; 1233 1234 MachineInstrBuilder MIB; 1235 if (STI->genLongCalls()) { 1236 MachineFunction *MF = MBB.getParent(); 1237 MachineConstantPool *MCP = MF->getConstantPool(); 1238 unsigned PCLabelID = AFI->createPICLabelUId(); 1239 MachineConstantPoolValue *CPV = 1240 ARMConstantPoolSymbol::Create(MF->getFunction()->getContext(), 1241 "__aeabi_read_tp", PCLabelID, 0); 1242 unsigned Reg = MI.getOperand(0).getReg(); 1243 MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(), 1244 TII->get(Thumb ? ARM::tLDRpci : ARM::LDRi12), Reg) 1245 .addConstantPoolIndex(MCP->getConstantPoolIndex(CPV, 4)); 1246 if (!Thumb) 1247 MIB.addImm(0); 1248 MIB.add(predOps(ARMCC::AL)); 1249 1250 MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(), 1251 TII->get(Thumb ? ARM::tBLXr : ARM::BLX)); 1252 if (Thumb) 1253 MIB.add(predOps(ARMCC::AL)); 1254 MIB.addReg(Reg, RegState::Kill); 1255 } else { 1256 MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(), 1257 TII->get(Thumb ? ARM::tBL : ARM::BL)); 1258 if (Thumb) 1259 MIB.add(predOps(ARMCC::AL)); 1260 MIB.addExternalSymbol("__aeabi_read_tp", 0); 1261 } 1262 1263 MIB->setMemRefs(MI.memoperands_begin(), MI.memoperands_end()); 1264 TransferImpOps(MI, MIB, MIB); 1265 MI.eraseFromParent(); 1266 return true; 1267 } 1268 case ARM::tLDRpci_pic: 1269 case ARM::t2LDRpci_pic: { 1270 unsigned NewLdOpc = (Opcode == ARM::tLDRpci_pic) 1271 ? ARM::tLDRpci : ARM::t2LDRpci; 1272 unsigned DstReg = MI.getOperand(0).getReg(); 1273 bool DstIsDead = MI.getOperand(0).isDead(); 1274 MachineInstrBuilder MIB1 = 1275 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(NewLdOpc), DstReg) 1276 .add(MI.getOperand(1)) 1277 .add(predOps(ARMCC::AL)); 1278 MIB1->setMemRefs(MI.memoperands_begin(), MI.memoperands_end()); 1279 MachineInstrBuilder MIB2 = 1280 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::tPICADD)) 1281 .addReg(DstReg, RegState::Define | getDeadRegState(DstIsDead)) 1282 .addReg(DstReg) 1283 .add(MI.getOperand(2)); 1284 TransferImpOps(MI, MIB1, MIB2); 1285 MI.eraseFromParent(); 1286 return true; 1287 } 1288 1289 case ARM::LDRLIT_ga_abs: 1290 case ARM::LDRLIT_ga_pcrel: 1291 case ARM::LDRLIT_ga_pcrel_ldr: 1292 case ARM::tLDRLIT_ga_abs: 1293 case ARM::tLDRLIT_ga_pcrel: { 1294 unsigned DstReg = MI.getOperand(0).getReg(); 1295 bool DstIsDead = MI.getOperand(0).isDead(); 1296 const MachineOperand &MO1 = MI.getOperand(1); 1297 const GlobalValue *GV = MO1.getGlobal(); 1298 bool IsARM = 1299 Opcode != ARM::tLDRLIT_ga_pcrel && Opcode != ARM::tLDRLIT_ga_abs; 1300 bool IsPIC = 1301 Opcode != ARM::LDRLIT_ga_abs && Opcode != ARM::tLDRLIT_ga_abs; 1302 unsigned LDRLITOpc = IsARM ? ARM::LDRi12 : ARM::tLDRpci; 1303 unsigned PICAddOpc = 1304 IsARM 1305 ? (Opcode == ARM::LDRLIT_ga_pcrel_ldr ? ARM::PICLDR : ARM::PICADD) 1306 : ARM::tPICADD; 1307 1308 // We need a new const-pool entry to load from. 1309 MachineConstantPool *MCP = MBB.getParent()->getConstantPool(); 1310 unsigned ARMPCLabelIndex = 0; 1311 MachineConstantPoolValue *CPV; 1312 1313 if (IsPIC) { 1314 unsigned PCAdj = IsARM ? 8 : 4; 1315 ARMPCLabelIndex = AFI->createPICLabelUId(); 1316 CPV = ARMConstantPoolConstant::Create(GV, ARMPCLabelIndex, 1317 ARMCP::CPValue, PCAdj); 1318 } else 1319 CPV = ARMConstantPoolConstant::Create(GV, ARMCP::no_modifier); 1320 1321 MachineInstrBuilder MIB = 1322 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(LDRLITOpc), DstReg) 1323 .addConstantPoolIndex(MCP->getConstantPoolIndex(CPV, 4)); 1324 if (IsARM) 1325 MIB.addImm(0); 1326 MIB.add(predOps(ARMCC::AL)); 1327 1328 if (IsPIC) { 1329 MachineInstrBuilder MIB = 1330 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(PICAddOpc)) 1331 .addReg(DstReg, RegState::Define | getDeadRegState(DstIsDead)) 1332 .addReg(DstReg) 1333 .addImm(ARMPCLabelIndex); 1334 1335 if (IsARM) 1336 MIB.add(predOps(ARMCC::AL)); 1337 } 1338 1339 MI.eraseFromParent(); 1340 return true; 1341 } 1342 case ARM::MOV_ga_pcrel: 1343 case ARM::MOV_ga_pcrel_ldr: 1344 case ARM::t2MOV_ga_pcrel: { 1345 // Expand into movw + movw. Also "add pc" / ldr [pc] in PIC mode. 1346 unsigned LabelId = AFI->createPICLabelUId(); 1347 unsigned DstReg = MI.getOperand(0).getReg(); 1348 bool DstIsDead = MI.getOperand(0).isDead(); 1349 const MachineOperand &MO1 = MI.getOperand(1); 1350 const GlobalValue *GV = MO1.getGlobal(); 1351 unsigned TF = MO1.getTargetFlags(); 1352 bool isARM = Opcode != ARM::t2MOV_ga_pcrel; 1353 unsigned LO16Opc = isARM ? ARM::MOVi16_ga_pcrel : ARM::t2MOVi16_ga_pcrel; 1354 unsigned HI16Opc = isARM ? ARM::MOVTi16_ga_pcrel :ARM::t2MOVTi16_ga_pcrel; 1355 unsigned LO16TF = TF | ARMII::MO_LO16; 1356 unsigned HI16TF = TF | ARMII::MO_HI16; 1357 unsigned PICAddOpc = isARM 1358 ? (Opcode == ARM::MOV_ga_pcrel_ldr ? ARM::PICLDR : ARM::PICADD) 1359 : ARM::tPICADD; 1360 MachineInstrBuilder MIB1 = BuildMI(MBB, MBBI, MI.getDebugLoc(), 1361 TII->get(LO16Opc), DstReg) 1362 .addGlobalAddress(GV, MO1.getOffset(), TF | LO16TF) 1363 .addImm(LabelId); 1364 1365 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(HI16Opc), DstReg) 1366 .addReg(DstReg) 1367 .addGlobalAddress(GV, MO1.getOffset(), TF | HI16TF) 1368 .addImm(LabelId); 1369 1370 MachineInstrBuilder MIB3 = BuildMI(MBB, MBBI, MI.getDebugLoc(), 1371 TII->get(PICAddOpc)) 1372 .addReg(DstReg, RegState::Define | getDeadRegState(DstIsDead)) 1373 .addReg(DstReg).addImm(LabelId); 1374 if (isARM) { 1375 MIB3.add(predOps(ARMCC::AL)); 1376 if (Opcode == ARM::MOV_ga_pcrel_ldr) 1377 MIB3->setMemRefs(MI.memoperands_begin(), MI.memoperands_end()); 1378 } 1379 TransferImpOps(MI, MIB1, MIB3); 1380 MI.eraseFromParent(); 1381 return true; 1382 } 1383 1384 case ARM::MOVi32imm: 1385 case ARM::MOVCCi32imm: 1386 case ARM::t2MOVi32imm: 1387 case ARM::t2MOVCCi32imm: 1388 ExpandMOV32BitImm(MBB, MBBI); 1389 return true; 1390 1391 case ARM::SUBS_PC_LR: { 1392 MachineInstrBuilder MIB = 1393 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::SUBri), ARM::PC) 1394 .addReg(ARM::LR) 1395 .add(MI.getOperand(0)) 1396 .add(MI.getOperand(1)) 1397 .add(MI.getOperand(2)) 1398 .addReg(ARM::CPSR, RegState::Undef); 1399 TransferImpOps(MI, MIB, MIB); 1400 MI.eraseFromParent(); 1401 return true; 1402 } 1403 case ARM::VLDMQIA: { 1404 unsigned NewOpc = ARM::VLDMDIA; 1405 MachineInstrBuilder MIB = 1406 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(NewOpc)); 1407 unsigned OpIdx = 0; 1408 1409 // Grab the Q register destination. 1410 bool DstIsDead = MI.getOperand(OpIdx).isDead(); 1411 unsigned DstReg = MI.getOperand(OpIdx++).getReg(); 1412 1413 // Copy the source register. 1414 MIB.add(MI.getOperand(OpIdx++)); 1415 1416 // Copy the predicate operands. 1417 MIB.add(MI.getOperand(OpIdx++)); 1418 MIB.add(MI.getOperand(OpIdx++)); 1419 1420 // Add the destination operands (D subregs). 1421 unsigned D0 = TRI->getSubReg(DstReg, ARM::dsub_0); 1422 unsigned D1 = TRI->getSubReg(DstReg, ARM::dsub_1); 1423 MIB.addReg(D0, RegState::Define | getDeadRegState(DstIsDead)) 1424 .addReg(D1, RegState::Define | getDeadRegState(DstIsDead)); 1425 1426 // Add an implicit def for the super-register. 1427 MIB.addReg(DstReg, RegState::ImplicitDefine | getDeadRegState(DstIsDead)); 1428 TransferImpOps(MI, MIB, MIB); 1429 MIB.setMemRefs(MI.memoperands_begin(), MI.memoperands_end()); 1430 MI.eraseFromParent(); 1431 return true; 1432 } 1433 1434 case ARM::VSTMQIA: { 1435 unsigned NewOpc = ARM::VSTMDIA; 1436 MachineInstrBuilder MIB = 1437 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(NewOpc)); 1438 unsigned OpIdx = 0; 1439 1440 // Grab the Q register source. 1441 bool SrcIsKill = MI.getOperand(OpIdx).isKill(); 1442 unsigned SrcReg = MI.getOperand(OpIdx++).getReg(); 1443 1444 // Copy the destination register. 1445 MIB.add(MI.getOperand(OpIdx++)); 1446 1447 // Copy the predicate operands. 1448 MIB.add(MI.getOperand(OpIdx++)); 1449 MIB.add(MI.getOperand(OpIdx++)); 1450 1451 // Add the source operands (D subregs). 1452 unsigned D0 = TRI->getSubReg(SrcReg, ARM::dsub_0); 1453 unsigned D1 = TRI->getSubReg(SrcReg, ARM::dsub_1); 1454 MIB.addReg(D0, SrcIsKill ? RegState::Kill : 0) 1455 .addReg(D1, SrcIsKill ? RegState::Kill : 0); 1456 1457 if (SrcIsKill) // Add an implicit kill for the Q register. 1458 MIB->addRegisterKilled(SrcReg, TRI, true); 1459 1460 TransferImpOps(MI, MIB, MIB); 1461 MIB.setMemRefs(MI.memoperands_begin(), MI.memoperands_end()); 1462 MI.eraseFromParent(); 1463 return true; 1464 } 1465 1466 case ARM::VLD2q8Pseudo: 1467 case ARM::VLD2q16Pseudo: 1468 case ARM::VLD2q32Pseudo: 1469 case ARM::VLD2q8PseudoWB_fixed: 1470 case ARM::VLD2q16PseudoWB_fixed: 1471 case ARM::VLD2q32PseudoWB_fixed: 1472 case ARM::VLD2q8PseudoWB_register: 1473 case ARM::VLD2q16PseudoWB_register: 1474 case ARM::VLD2q32PseudoWB_register: 1475 case ARM::VLD3d8Pseudo: 1476 case ARM::VLD3d16Pseudo: 1477 case ARM::VLD3d32Pseudo: 1478 case ARM::VLD1d64TPseudo: 1479 case ARM::VLD1d64TPseudoWB_fixed: 1480 case ARM::VLD3d8Pseudo_UPD: 1481 case ARM::VLD3d16Pseudo_UPD: 1482 case ARM::VLD3d32Pseudo_UPD: 1483 case ARM::VLD3q8Pseudo_UPD: 1484 case ARM::VLD3q16Pseudo_UPD: 1485 case ARM::VLD3q32Pseudo_UPD: 1486 case ARM::VLD3q8oddPseudo: 1487 case ARM::VLD3q16oddPseudo: 1488 case ARM::VLD3q32oddPseudo: 1489 case ARM::VLD3q8oddPseudo_UPD: 1490 case ARM::VLD3q16oddPseudo_UPD: 1491 case ARM::VLD3q32oddPseudo_UPD: 1492 case ARM::VLD4d8Pseudo: 1493 case ARM::VLD4d16Pseudo: 1494 case ARM::VLD4d32Pseudo: 1495 case ARM::VLD1d64QPseudo: 1496 case ARM::VLD1d64QPseudoWB_fixed: 1497 case ARM::VLD4d8Pseudo_UPD: 1498 case ARM::VLD4d16Pseudo_UPD: 1499 case ARM::VLD4d32Pseudo_UPD: 1500 case ARM::VLD4q8Pseudo_UPD: 1501 case ARM::VLD4q16Pseudo_UPD: 1502 case ARM::VLD4q32Pseudo_UPD: 1503 case ARM::VLD4q8oddPseudo: 1504 case ARM::VLD4q16oddPseudo: 1505 case ARM::VLD4q32oddPseudo: 1506 case ARM::VLD4q8oddPseudo_UPD: 1507 case ARM::VLD4q16oddPseudo_UPD: 1508 case ARM::VLD4q32oddPseudo_UPD: 1509 case ARM::VLD3DUPd8Pseudo: 1510 case ARM::VLD3DUPd16Pseudo: 1511 case ARM::VLD3DUPd32Pseudo: 1512 case ARM::VLD3DUPd8Pseudo_UPD: 1513 case ARM::VLD3DUPd16Pseudo_UPD: 1514 case ARM::VLD3DUPd32Pseudo_UPD: 1515 case ARM::VLD4DUPd8Pseudo: 1516 case ARM::VLD4DUPd16Pseudo: 1517 case ARM::VLD4DUPd32Pseudo: 1518 case ARM::VLD4DUPd8Pseudo_UPD: 1519 case ARM::VLD4DUPd16Pseudo_UPD: 1520 case ARM::VLD4DUPd32Pseudo_UPD: 1521 ExpandVLD(MBBI); 1522 return true; 1523 1524 case ARM::VST2q8Pseudo: 1525 case ARM::VST2q16Pseudo: 1526 case ARM::VST2q32Pseudo: 1527 case ARM::VST2q8PseudoWB_fixed: 1528 case ARM::VST2q16PseudoWB_fixed: 1529 case ARM::VST2q32PseudoWB_fixed: 1530 case ARM::VST2q8PseudoWB_register: 1531 case ARM::VST2q16PseudoWB_register: 1532 case ARM::VST2q32PseudoWB_register: 1533 case ARM::VST3d8Pseudo: 1534 case ARM::VST3d16Pseudo: 1535 case ARM::VST3d32Pseudo: 1536 case ARM::VST1d64TPseudo: 1537 case ARM::VST3d8Pseudo_UPD: 1538 case ARM::VST3d16Pseudo_UPD: 1539 case ARM::VST3d32Pseudo_UPD: 1540 case ARM::VST1d64TPseudoWB_fixed: 1541 case ARM::VST1d64TPseudoWB_register: 1542 case ARM::VST3q8Pseudo_UPD: 1543 case ARM::VST3q16Pseudo_UPD: 1544 case ARM::VST3q32Pseudo_UPD: 1545 case ARM::VST3q8oddPseudo: 1546 case ARM::VST3q16oddPseudo: 1547 case ARM::VST3q32oddPseudo: 1548 case ARM::VST3q8oddPseudo_UPD: 1549 case ARM::VST3q16oddPseudo_UPD: 1550 case ARM::VST3q32oddPseudo_UPD: 1551 case ARM::VST4d8Pseudo: 1552 case ARM::VST4d16Pseudo: 1553 case ARM::VST4d32Pseudo: 1554 case ARM::VST1d64QPseudo: 1555 case ARM::VST4d8Pseudo_UPD: 1556 case ARM::VST4d16Pseudo_UPD: 1557 case ARM::VST4d32Pseudo_UPD: 1558 case ARM::VST1d64QPseudoWB_fixed: 1559 case ARM::VST1d64QPseudoWB_register: 1560 case ARM::VST4q8Pseudo_UPD: 1561 case ARM::VST4q16Pseudo_UPD: 1562 case ARM::VST4q32Pseudo_UPD: 1563 case ARM::VST4q8oddPseudo: 1564 case ARM::VST4q16oddPseudo: 1565 case ARM::VST4q32oddPseudo: 1566 case ARM::VST4q8oddPseudo_UPD: 1567 case ARM::VST4q16oddPseudo_UPD: 1568 case ARM::VST4q32oddPseudo_UPD: 1569 ExpandVST(MBBI); 1570 return true; 1571 1572 case ARM::VLD1LNq8Pseudo: 1573 case ARM::VLD1LNq16Pseudo: 1574 case ARM::VLD1LNq32Pseudo: 1575 case ARM::VLD1LNq8Pseudo_UPD: 1576 case ARM::VLD1LNq16Pseudo_UPD: 1577 case ARM::VLD1LNq32Pseudo_UPD: 1578 case ARM::VLD2LNd8Pseudo: 1579 case ARM::VLD2LNd16Pseudo: 1580 case ARM::VLD2LNd32Pseudo: 1581 case ARM::VLD2LNq16Pseudo: 1582 case ARM::VLD2LNq32Pseudo: 1583 case ARM::VLD2LNd8Pseudo_UPD: 1584 case ARM::VLD2LNd16Pseudo_UPD: 1585 case ARM::VLD2LNd32Pseudo_UPD: 1586 case ARM::VLD2LNq16Pseudo_UPD: 1587 case ARM::VLD2LNq32Pseudo_UPD: 1588 case ARM::VLD3LNd8Pseudo: 1589 case ARM::VLD3LNd16Pseudo: 1590 case ARM::VLD3LNd32Pseudo: 1591 case ARM::VLD3LNq16Pseudo: 1592 case ARM::VLD3LNq32Pseudo: 1593 case ARM::VLD3LNd8Pseudo_UPD: 1594 case ARM::VLD3LNd16Pseudo_UPD: 1595 case ARM::VLD3LNd32Pseudo_UPD: 1596 case ARM::VLD3LNq16Pseudo_UPD: 1597 case ARM::VLD3LNq32Pseudo_UPD: 1598 case ARM::VLD4LNd8Pseudo: 1599 case ARM::VLD4LNd16Pseudo: 1600 case ARM::VLD4LNd32Pseudo: 1601 case ARM::VLD4LNq16Pseudo: 1602 case ARM::VLD4LNq32Pseudo: 1603 case ARM::VLD4LNd8Pseudo_UPD: 1604 case ARM::VLD4LNd16Pseudo_UPD: 1605 case ARM::VLD4LNd32Pseudo_UPD: 1606 case ARM::VLD4LNq16Pseudo_UPD: 1607 case ARM::VLD4LNq32Pseudo_UPD: 1608 case ARM::VST1LNq8Pseudo: 1609 case ARM::VST1LNq16Pseudo: 1610 case ARM::VST1LNq32Pseudo: 1611 case ARM::VST1LNq8Pseudo_UPD: 1612 case ARM::VST1LNq16Pseudo_UPD: 1613 case ARM::VST1LNq32Pseudo_UPD: 1614 case ARM::VST2LNd8Pseudo: 1615 case ARM::VST2LNd16Pseudo: 1616 case ARM::VST2LNd32Pseudo: 1617 case ARM::VST2LNq16Pseudo: 1618 case ARM::VST2LNq32Pseudo: 1619 case ARM::VST2LNd8Pseudo_UPD: 1620 case ARM::VST2LNd16Pseudo_UPD: 1621 case ARM::VST2LNd32Pseudo_UPD: 1622 case ARM::VST2LNq16Pseudo_UPD: 1623 case ARM::VST2LNq32Pseudo_UPD: 1624 case ARM::VST3LNd8Pseudo: 1625 case ARM::VST3LNd16Pseudo: 1626 case ARM::VST3LNd32Pseudo: 1627 case ARM::VST3LNq16Pseudo: 1628 case ARM::VST3LNq32Pseudo: 1629 case ARM::VST3LNd8Pseudo_UPD: 1630 case ARM::VST3LNd16Pseudo_UPD: 1631 case ARM::VST3LNd32Pseudo_UPD: 1632 case ARM::VST3LNq16Pseudo_UPD: 1633 case ARM::VST3LNq32Pseudo_UPD: 1634 case ARM::VST4LNd8Pseudo: 1635 case ARM::VST4LNd16Pseudo: 1636 case ARM::VST4LNd32Pseudo: 1637 case ARM::VST4LNq16Pseudo: 1638 case ARM::VST4LNq32Pseudo: 1639 case ARM::VST4LNd8Pseudo_UPD: 1640 case ARM::VST4LNd16Pseudo_UPD: 1641 case ARM::VST4LNd32Pseudo_UPD: 1642 case ARM::VST4LNq16Pseudo_UPD: 1643 case ARM::VST4LNq32Pseudo_UPD: 1644 ExpandLaneOp(MBBI); 1645 return true; 1646 1647 case ARM::VTBL3Pseudo: ExpandVTBL(MBBI, ARM::VTBL3, false); return true; 1648 case ARM::VTBL4Pseudo: ExpandVTBL(MBBI, ARM::VTBL4, false); return true; 1649 case ARM::VTBX3Pseudo: ExpandVTBL(MBBI, ARM::VTBX3, true); return true; 1650 case ARM::VTBX4Pseudo: ExpandVTBL(MBBI, ARM::VTBX4, true); return true; 1651 1652 case ARM::CMP_SWAP_8: 1653 if (STI->isThumb()) 1654 return ExpandCMP_SWAP(MBB, MBBI, ARM::t2LDREXB, ARM::t2STREXB, 1655 ARM::tUXTB, NextMBBI); 1656 else 1657 return ExpandCMP_SWAP(MBB, MBBI, ARM::LDREXB, ARM::STREXB, 1658 ARM::UXTB, NextMBBI); 1659 case ARM::CMP_SWAP_16: 1660 if (STI->isThumb()) 1661 return ExpandCMP_SWAP(MBB, MBBI, ARM::t2LDREXH, ARM::t2STREXH, 1662 ARM::tUXTH, NextMBBI); 1663 else 1664 return ExpandCMP_SWAP(MBB, MBBI, ARM::LDREXH, ARM::STREXH, 1665 ARM::UXTH, NextMBBI); 1666 case ARM::CMP_SWAP_32: 1667 if (STI->isThumb()) 1668 return ExpandCMP_SWAP(MBB, MBBI, ARM::t2LDREX, ARM::t2STREX, 0, 1669 NextMBBI); 1670 else 1671 return ExpandCMP_SWAP(MBB, MBBI, ARM::LDREX, ARM::STREX, 0, NextMBBI); 1672 1673 case ARM::CMP_SWAP_64: 1674 return ExpandCMP_SWAP_64(MBB, MBBI, NextMBBI); 1675 } 1676 } 1677 1678 bool ARMExpandPseudo::ExpandMBB(MachineBasicBlock &MBB) { 1679 bool Modified = false; 1680 1681 MachineBasicBlock::iterator MBBI = MBB.begin(), E = MBB.end(); 1682 while (MBBI != E) { 1683 MachineBasicBlock::iterator NMBBI = std::next(MBBI); 1684 Modified |= ExpandMI(MBB, MBBI, NMBBI); 1685 MBBI = NMBBI; 1686 } 1687 1688 return Modified; 1689 } 1690 1691 bool ARMExpandPseudo::runOnMachineFunction(MachineFunction &MF) { 1692 STI = &static_cast<const ARMSubtarget &>(MF.getSubtarget()); 1693 TII = STI->getInstrInfo(); 1694 TRI = STI->getRegisterInfo(); 1695 AFI = MF.getInfo<ARMFunctionInfo>(); 1696 1697 bool Modified = false; 1698 for (MachineFunction::iterator MFI = MF.begin(), E = MF.end(); MFI != E; 1699 ++MFI) 1700 Modified |= ExpandMBB(*MFI); 1701 if (VerifyARMPseudo) 1702 MF.verify(this, "After expanding ARM pseudo instructions."); 1703 return Modified; 1704 } 1705 1706 /// createARMExpandPseudoPass - returns an instance of the pseudo instruction 1707 /// expansion pass. 1708 FunctionPass *llvm::createARMExpandPseudoPass() { 1709 return new ARMExpandPseudo(); 1710 } 1711