1 //===-- ARMExpandPseudoInsts.cpp - Expand pseudo 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 // This file contains a pass that expands pseudo instructions into target 10 // instructions to allow proper scheduling, if-conversion, and other late 11 // optimizations. This pass should be run after register allocation but before 12 // the post-regalloc scheduling pass. 13 // 14 //===----------------------------------------------------------------------===// 15 16 #include "ARM.h" 17 #include "ARMBaseInstrInfo.h" 18 #include "ARMBaseRegisterInfo.h" 19 #include "ARMConstantPoolValue.h" 20 #include "ARMMachineFunctionInfo.h" 21 #include "ARMSubtarget.h" 22 #include "MCTargetDesc/ARMAddressingModes.h" 23 #include "llvm/CodeGen/LivePhysRegs.h" 24 #include "llvm/CodeGen/MachineFrameInfo.h" 25 #include "llvm/CodeGen/MachineFunctionPass.h" 26 #include "llvm/Support/Debug.h" 27 28 using namespace llvm; 29 30 #define DEBUG_TYPE "arm-pseudo" 31 32 static cl::opt<bool> 33 VerifyARMPseudo("verify-arm-pseudo-expand", cl::Hidden, 34 cl::desc("Verify machine code after expanding ARM pseudos")); 35 36 #define ARM_EXPAND_PSEUDO_NAME "ARM pseudo instruction expansion pass" 37 38 namespace { 39 class ARMExpandPseudo : public MachineFunctionPass { 40 public: 41 static char ID; 42 ARMExpandPseudo() : MachineFunctionPass(ID) {} 43 44 const ARMBaseInstrInfo *TII; 45 const TargetRegisterInfo *TRI; 46 const ARMSubtarget *STI; 47 ARMFunctionInfo *AFI; 48 49 bool runOnMachineFunction(MachineFunction &Fn) override; 50 51 MachineFunctionProperties getRequiredProperties() const override { 52 return MachineFunctionProperties().set( 53 MachineFunctionProperties::Property::NoVRegs); 54 } 55 56 StringRef getPassName() const override { 57 return ARM_EXPAND_PSEUDO_NAME; 58 } 59 60 private: 61 void TransferImpOps(MachineInstr &OldMI, 62 MachineInstrBuilder &UseMI, MachineInstrBuilder &DefMI); 63 bool ExpandMI(MachineBasicBlock &MBB, 64 MachineBasicBlock::iterator MBBI, 65 MachineBasicBlock::iterator &NextMBBI); 66 bool ExpandMBB(MachineBasicBlock &MBB); 67 void ExpandVLD(MachineBasicBlock::iterator &MBBI); 68 void ExpandVST(MachineBasicBlock::iterator &MBBI); 69 void ExpandLaneOp(MachineBasicBlock::iterator &MBBI); 70 void ExpandVTBL(MachineBasicBlock::iterator &MBBI, 71 unsigned Opc, bool IsExt); 72 void ExpandMOV32BitImm(MachineBasicBlock &MBB, 73 MachineBasicBlock::iterator &MBBI); 74 void CMSEClearGPRegs(MachineBasicBlock &MBB, 75 MachineBasicBlock::iterator MBBI, const DebugLoc &DL, 76 const SmallVectorImpl<unsigned> &ClearRegs, 77 unsigned ClobberReg); 78 MachineBasicBlock &CMSEClearFPRegs(MachineBasicBlock &MBB, 79 MachineBasicBlock::iterator MBBI); 80 MachineBasicBlock &CMSEClearFPRegsV8(MachineBasicBlock &MBB, 81 MachineBasicBlock::iterator MBBI, 82 const BitVector &ClearRegs); 83 MachineBasicBlock &CMSEClearFPRegsV81(MachineBasicBlock &MBB, 84 MachineBasicBlock::iterator MBBI, 85 const BitVector &ClearRegs); 86 void CMSESaveClearFPRegs(MachineBasicBlock &MBB, 87 MachineBasicBlock::iterator MBBI, DebugLoc &DL, 88 const LivePhysRegs &LiveRegs, 89 SmallVectorImpl<unsigned> &AvailableRegs); 90 void CMSESaveClearFPRegsV8(MachineBasicBlock &MBB, 91 MachineBasicBlock::iterator MBBI, DebugLoc &DL, 92 const LivePhysRegs &LiveRegs, 93 SmallVectorImpl<unsigned> &ScratchRegs); 94 void CMSESaveClearFPRegsV81(MachineBasicBlock &MBB, 95 MachineBasicBlock::iterator MBBI, DebugLoc &DL, 96 const LivePhysRegs &LiveRegs); 97 void CMSERestoreFPRegs(MachineBasicBlock &MBB, 98 MachineBasicBlock::iterator MBBI, DebugLoc &DL, 99 SmallVectorImpl<unsigned> &AvailableRegs); 100 void CMSERestoreFPRegsV8(MachineBasicBlock &MBB, 101 MachineBasicBlock::iterator MBBI, DebugLoc &DL, 102 SmallVectorImpl<unsigned> &AvailableRegs); 103 void CMSERestoreFPRegsV81(MachineBasicBlock &MBB, 104 MachineBasicBlock::iterator MBBI, DebugLoc &DL, 105 SmallVectorImpl<unsigned> &AvailableRegs); 106 bool ExpandCMP_SWAP(MachineBasicBlock &MBB, 107 MachineBasicBlock::iterator MBBI, unsigned LdrexOp, 108 unsigned StrexOp, unsigned UxtOp, 109 MachineBasicBlock::iterator &NextMBBI); 110 111 bool ExpandCMP_SWAP_64(MachineBasicBlock &MBB, 112 MachineBasicBlock::iterator MBBI, 113 MachineBasicBlock::iterator &NextMBBI); 114 }; 115 char ARMExpandPseudo::ID = 0; 116 } 117 118 INITIALIZE_PASS(ARMExpandPseudo, DEBUG_TYPE, ARM_EXPAND_PSEUDO_NAME, false, 119 false) 120 121 /// TransferImpOps - Transfer implicit operands on the pseudo instruction to 122 /// the instructions created from the expansion. 123 void ARMExpandPseudo::TransferImpOps(MachineInstr &OldMI, 124 MachineInstrBuilder &UseMI, 125 MachineInstrBuilder &DefMI) { 126 const MCInstrDesc &Desc = OldMI.getDesc(); 127 for (unsigned i = Desc.getNumOperands(), e = OldMI.getNumOperands(); 128 i != e; ++i) { 129 const MachineOperand &MO = OldMI.getOperand(i); 130 assert(MO.isReg() && MO.getReg()); 131 if (MO.isUse()) 132 UseMI.add(MO); 133 else 134 DefMI.add(MO); 135 } 136 } 137 138 namespace { 139 // Constants for register spacing in NEON load/store instructions. 140 // For quad-register load-lane and store-lane pseudo instructors, the 141 // spacing is initially assumed to be EvenDblSpc, and that is changed to 142 // OddDblSpc depending on the lane number operand. 143 enum NEONRegSpacing { 144 SingleSpc, 145 SingleLowSpc , // Single spacing, low registers, three and four vectors. 146 SingleHighQSpc, // Single spacing, high registers, four vectors. 147 SingleHighTSpc, // Single spacing, high registers, three vectors. 148 EvenDblSpc, 149 OddDblSpc 150 }; 151 152 // Entries for NEON load/store information table. The table is sorted by 153 // PseudoOpc for fast binary-search lookups. 154 struct NEONLdStTableEntry { 155 uint16_t PseudoOpc; 156 uint16_t RealOpc; 157 bool IsLoad; 158 bool isUpdating; 159 bool hasWritebackOperand; 160 uint8_t RegSpacing; // One of type NEONRegSpacing 161 uint8_t NumRegs; // D registers loaded or stored 162 uint8_t RegElts; // elements per D register; used for lane ops 163 // FIXME: Temporary flag to denote whether the real instruction takes 164 // a single register (like the encoding) or all of the registers in 165 // the list (like the asm syntax and the isel DAG). When all definitions 166 // are converted to take only the single encoded register, this will 167 // go away. 168 bool copyAllListRegs; 169 170 // Comparison methods for binary search of the table. 171 bool operator<(const NEONLdStTableEntry &TE) const { 172 return PseudoOpc < TE.PseudoOpc; 173 } 174 friend bool operator<(const NEONLdStTableEntry &TE, unsigned PseudoOpc) { 175 return TE.PseudoOpc < PseudoOpc; 176 } 177 friend bool LLVM_ATTRIBUTE_UNUSED operator<(unsigned PseudoOpc, 178 const NEONLdStTableEntry &TE) { 179 return PseudoOpc < TE.PseudoOpc; 180 } 181 }; 182 } 183 184 static const NEONLdStTableEntry NEONLdStTable[] = { 185 { ARM::VLD1LNq16Pseudo, ARM::VLD1LNd16, true, false, false, EvenDblSpc, 1, 4 ,true}, 186 { ARM::VLD1LNq16Pseudo_UPD, ARM::VLD1LNd16_UPD, true, true, true, EvenDblSpc, 1, 4 ,true}, 187 { ARM::VLD1LNq32Pseudo, ARM::VLD1LNd32, true, false, false, EvenDblSpc, 1, 2 ,true}, 188 { ARM::VLD1LNq32Pseudo_UPD, ARM::VLD1LNd32_UPD, true, true, true, EvenDblSpc, 1, 2 ,true}, 189 { ARM::VLD1LNq8Pseudo, ARM::VLD1LNd8, true, false, false, EvenDblSpc, 1, 8 ,true}, 190 { ARM::VLD1LNq8Pseudo_UPD, ARM::VLD1LNd8_UPD, true, true, true, EvenDblSpc, 1, 8 ,true}, 191 192 { ARM::VLD1d16QPseudo, ARM::VLD1d16Q, true, false, false, SingleSpc, 4, 4 ,false}, 193 { ARM::VLD1d16TPseudo, ARM::VLD1d16T, true, false, false, SingleSpc, 3, 4 ,false}, 194 { ARM::VLD1d32QPseudo, ARM::VLD1d32Q, true, false, false, SingleSpc, 4, 2 ,false}, 195 { ARM::VLD1d32TPseudo, ARM::VLD1d32T, true, false, false, SingleSpc, 3, 2 ,false}, 196 { ARM::VLD1d64QPseudo, ARM::VLD1d64Q, true, false, false, SingleSpc, 4, 1 ,false}, 197 { ARM::VLD1d64QPseudoWB_fixed, ARM::VLD1d64Qwb_fixed, true, true, false, SingleSpc, 4, 1 ,false}, 198 { ARM::VLD1d64QPseudoWB_register, ARM::VLD1d64Qwb_register, true, true, true, SingleSpc, 4, 1 ,false}, 199 { ARM::VLD1d64TPseudo, ARM::VLD1d64T, true, false, false, SingleSpc, 3, 1 ,false}, 200 { ARM::VLD1d64TPseudoWB_fixed, ARM::VLD1d64Twb_fixed, true, true, false, SingleSpc, 3, 1 ,false}, 201 { ARM::VLD1d64TPseudoWB_register, ARM::VLD1d64Twb_register, true, true, true, SingleSpc, 3, 1 ,false}, 202 { ARM::VLD1d8QPseudo, ARM::VLD1d8Q, true, false, false, SingleSpc, 4, 8 ,false}, 203 { ARM::VLD1d8TPseudo, ARM::VLD1d8T, true, false, false, SingleSpc, 3, 8 ,false}, 204 { ARM::VLD1q16HighQPseudo, ARM::VLD1d16Q, true, false, false, SingleHighQSpc, 4, 4 ,false}, 205 { ARM::VLD1q16HighTPseudo, ARM::VLD1d16T, true, false, false, SingleHighTSpc, 3, 4 ,false}, 206 { ARM::VLD1q16LowQPseudo_UPD, ARM::VLD1d16Qwb_fixed, true, true, true, SingleLowSpc, 4, 4 ,false}, 207 { ARM::VLD1q16LowTPseudo_UPD, ARM::VLD1d16Twb_fixed, true, true, true, SingleLowSpc, 3, 4 ,false}, 208 { ARM::VLD1q32HighQPseudo, ARM::VLD1d32Q, true, false, false, SingleHighQSpc, 4, 2 ,false}, 209 { ARM::VLD1q32HighTPseudo, ARM::VLD1d32T, true, false, false, SingleHighTSpc, 3, 2 ,false}, 210 { ARM::VLD1q32LowQPseudo_UPD, ARM::VLD1d32Qwb_fixed, true, true, true, SingleLowSpc, 4, 2 ,false}, 211 { ARM::VLD1q32LowTPseudo_UPD, ARM::VLD1d32Twb_fixed, true, true, true, SingleLowSpc, 3, 2 ,false}, 212 { ARM::VLD1q64HighQPseudo, ARM::VLD1d64Q, true, false, false, SingleHighQSpc, 4, 1 ,false}, 213 { ARM::VLD1q64HighTPseudo, ARM::VLD1d64T, true, false, false, SingleHighTSpc, 3, 1 ,false}, 214 { ARM::VLD1q64LowQPseudo_UPD, ARM::VLD1d64Qwb_fixed, true, true, true, SingleLowSpc, 4, 1 ,false}, 215 { ARM::VLD1q64LowTPseudo_UPD, ARM::VLD1d64Twb_fixed, true, true, true, SingleLowSpc, 3, 1 ,false}, 216 { ARM::VLD1q8HighQPseudo, ARM::VLD1d8Q, true, false, false, SingleHighQSpc, 4, 8 ,false}, 217 { ARM::VLD1q8HighTPseudo, ARM::VLD1d8T, true, false, false, SingleHighTSpc, 3, 8 ,false}, 218 { ARM::VLD1q8LowQPseudo_UPD, ARM::VLD1d8Qwb_fixed, true, true, true, SingleLowSpc, 4, 8 ,false}, 219 { ARM::VLD1q8LowTPseudo_UPD, ARM::VLD1d8Twb_fixed, true, true, true, SingleLowSpc, 3, 8 ,false}, 220 221 { ARM::VLD2DUPq16EvenPseudo, ARM::VLD2DUPd16x2, true, false, false, EvenDblSpc, 2, 4 ,false}, 222 { ARM::VLD2DUPq16OddPseudo, ARM::VLD2DUPd16x2, true, false, false, OddDblSpc, 2, 4 ,false}, 223 { ARM::VLD2DUPq32EvenPseudo, ARM::VLD2DUPd32x2, true, false, false, EvenDblSpc, 2, 2 ,false}, 224 { ARM::VLD2DUPq32OddPseudo, ARM::VLD2DUPd32x2, true, false, false, OddDblSpc, 2, 2 ,false}, 225 { ARM::VLD2DUPq8EvenPseudo, ARM::VLD2DUPd8x2, true, false, false, EvenDblSpc, 2, 8 ,false}, 226 { ARM::VLD2DUPq8OddPseudo, ARM::VLD2DUPd8x2, true, false, false, OddDblSpc, 2, 8 ,false}, 227 228 { ARM::VLD2LNd16Pseudo, ARM::VLD2LNd16, true, false, false, SingleSpc, 2, 4 ,true}, 229 { ARM::VLD2LNd16Pseudo_UPD, ARM::VLD2LNd16_UPD, true, true, true, SingleSpc, 2, 4 ,true}, 230 { ARM::VLD2LNd32Pseudo, ARM::VLD2LNd32, true, false, false, SingleSpc, 2, 2 ,true}, 231 { ARM::VLD2LNd32Pseudo_UPD, ARM::VLD2LNd32_UPD, true, true, true, SingleSpc, 2, 2 ,true}, 232 { ARM::VLD2LNd8Pseudo, ARM::VLD2LNd8, true, false, false, SingleSpc, 2, 8 ,true}, 233 { ARM::VLD2LNd8Pseudo_UPD, ARM::VLD2LNd8_UPD, true, true, true, SingleSpc, 2, 8 ,true}, 234 { ARM::VLD2LNq16Pseudo, ARM::VLD2LNq16, true, false, false, EvenDblSpc, 2, 4 ,true}, 235 { ARM::VLD2LNq16Pseudo_UPD, ARM::VLD2LNq16_UPD, true, true, true, EvenDblSpc, 2, 4 ,true}, 236 { ARM::VLD2LNq32Pseudo, ARM::VLD2LNq32, true, false, false, EvenDblSpc, 2, 2 ,true}, 237 { ARM::VLD2LNq32Pseudo_UPD, ARM::VLD2LNq32_UPD, true, true, true, EvenDblSpc, 2, 2 ,true}, 238 239 { ARM::VLD2q16Pseudo, ARM::VLD2q16, true, false, false, SingleSpc, 4, 4 ,false}, 240 { ARM::VLD2q16PseudoWB_fixed, ARM::VLD2q16wb_fixed, true, true, false, SingleSpc, 4, 4 ,false}, 241 { ARM::VLD2q16PseudoWB_register, ARM::VLD2q16wb_register, true, true, true, SingleSpc, 4, 4 ,false}, 242 { ARM::VLD2q32Pseudo, ARM::VLD2q32, true, false, false, SingleSpc, 4, 2 ,false}, 243 { ARM::VLD2q32PseudoWB_fixed, ARM::VLD2q32wb_fixed, true, true, false, SingleSpc, 4, 2 ,false}, 244 { ARM::VLD2q32PseudoWB_register, ARM::VLD2q32wb_register, true, true, true, SingleSpc, 4, 2 ,false}, 245 { ARM::VLD2q8Pseudo, ARM::VLD2q8, true, false, false, SingleSpc, 4, 8 ,false}, 246 { ARM::VLD2q8PseudoWB_fixed, ARM::VLD2q8wb_fixed, true, true, false, SingleSpc, 4, 8 ,false}, 247 { ARM::VLD2q8PseudoWB_register, ARM::VLD2q8wb_register, true, true, true, SingleSpc, 4, 8 ,false}, 248 249 { ARM::VLD3DUPd16Pseudo, ARM::VLD3DUPd16, true, false, false, SingleSpc, 3, 4,true}, 250 { ARM::VLD3DUPd16Pseudo_UPD, ARM::VLD3DUPd16_UPD, true, true, true, SingleSpc, 3, 4,true}, 251 { ARM::VLD3DUPd32Pseudo, ARM::VLD3DUPd32, true, false, false, SingleSpc, 3, 2,true}, 252 { ARM::VLD3DUPd32Pseudo_UPD, ARM::VLD3DUPd32_UPD, true, true, true, SingleSpc, 3, 2,true}, 253 { ARM::VLD3DUPd8Pseudo, ARM::VLD3DUPd8, true, false, false, SingleSpc, 3, 8,true}, 254 { ARM::VLD3DUPd8Pseudo_UPD, ARM::VLD3DUPd8_UPD, true, true, true, SingleSpc, 3, 8,true}, 255 { ARM::VLD3DUPq16EvenPseudo, ARM::VLD3DUPq16, true, false, false, EvenDblSpc, 3, 4 ,true}, 256 { ARM::VLD3DUPq16OddPseudo, ARM::VLD3DUPq16, true, false, false, OddDblSpc, 3, 4 ,true}, 257 { ARM::VLD3DUPq32EvenPseudo, ARM::VLD3DUPq32, true, false, false, EvenDblSpc, 3, 2 ,true}, 258 { ARM::VLD3DUPq32OddPseudo, ARM::VLD3DUPq32, true, false, false, OddDblSpc, 3, 2 ,true}, 259 { ARM::VLD3DUPq8EvenPseudo, ARM::VLD3DUPq8, true, false, false, EvenDblSpc, 3, 8 ,true}, 260 { ARM::VLD3DUPq8OddPseudo, ARM::VLD3DUPq8, true, false, false, OddDblSpc, 3, 8 ,true}, 261 262 { ARM::VLD3LNd16Pseudo, ARM::VLD3LNd16, true, false, false, SingleSpc, 3, 4 ,true}, 263 { ARM::VLD3LNd16Pseudo_UPD, ARM::VLD3LNd16_UPD, true, true, true, SingleSpc, 3, 4 ,true}, 264 { ARM::VLD3LNd32Pseudo, ARM::VLD3LNd32, true, false, false, SingleSpc, 3, 2 ,true}, 265 { ARM::VLD3LNd32Pseudo_UPD, ARM::VLD3LNd32_UPD, true, true, true, SingleSpc, 3, 2 ,true}, 266 { ARM::VLD3LNd8Pseudo, ARM::VLD3LNd8, true, false, false, SingleSpc, 3, 8 ,true}, 267 { ARM::VLD3LNd8Pseudo_UPD, ARM::VLD3LNd8_UPD, true, true, true, SingleSpc, 3, 8 ,true}, 268 { ARM::VLD3LNq16Pseudo, ARM::VLD3LNq16, true, false, false, EvenDblSpc, 3, 4 ,true}, 269 { ARM::VLD3LNq16Pseudo_UPD, ARM::VLD3LNq16_UPD, true, true, true, EvenDblSpc, 3, 4 ,true}, 270 { ARM::VLD3LNq32Pseudo, ARM::VLD3LNq32, true, false, false, EvenDblSpc, 3, 2 ,true}, 271 { ARM::VLD3LNq32Pseudo_UPD, ARM::VLD3LNq32_UPD, true, true, true, EvenDblSpc, 3, 2 ,true}, 272 273 { ARM::VLD3d16Pseudo, ARM::VLD3d16, true, false, false, SingleSpc, 3, 4 ,true}, 274 { ARM::VLD3d16Pseudo_UPD, ARM::VLD3d16_UPD, true, true, true, SingleSpc, 3, 4 ,true}, 275 { ARM::VLD3d32Pseudo, ARM::VLD3d32, true, false, false, SingleSpc, 3, 2 ,true}, 276 { ARM::VLD3d32Pseudo_UPD, ARM::VLD3d32_UPD, true, true, true, SingleSpc, 3, 2 ,true}, 277 { ARM::VLD3d8Pseudo, ARM::VLD3d8, true, false, false, SingleSpc, 3, 8 ,true}, 278 { ARM::VLD3d8Pseudo_UPD, ARM::VLD3d8_UPD, true, true, true, SingleSpc, 3, 8 ,true}, 279 280 { ARM::VLD3q16Pseudo_UPD, ARM::VLD3q16_UPD, true, true, true, EvenDblSpc, 3, 4 ,true}, 281 { ARM::VLD3q16oddPseudo, ARM::VLD3q16, true, false, false, OddDblSpc, 3, 4 ,true}, 282 { ARM::VLD3q16oddPseudo_UPD, ARM::VLD3q16_UPD, true, true, true, OddDblSpc, 3, 4 ,true}, 283 { ARM::VLD3q32Pseudo_UPD, ARM::VLD3q32_UPD, true, true, true, EvenDblSpc, 3, 2 ,true}, 284 { ARM::VLD3q32oddPseudo, ARM::VLD3q32, true, false, false, OddDblSpc, 3, 2 ,true}, 285 { ARM::VLD3q32oddPseudo_UPD, ARM::VLD3q32_UPD, true, true, true, OddDblSpc, 3, 2 ,true}, 286 { ARM::VLD3q8Pseudo_UPD, ARM::VLD3q8_UPD, true, true, true, EvenDblSpc, 3, 8 ,true}, 287 { ARM::VLD3q8oddPseudo, ARM::VLD3q8, true, false, false, OddDblSpc, 3, 8 ,true}, 288 { ARM::VLD3q8oddPseudo_UPD, ARM::VLD3q8_UPD, true, true, true, OddDblSpc, 3, 8 ,true}, 289 290 { ARM::VLD4DUPd16Pseudo, ARM::VLD4DUPd16, true, false, false, SingleSpc, 4, 4,true}, 291 { ARM::VLD4DUPd16Pseudo_UPD, ARM::VLD4DUPd16_UPD, true, true, true, SingleSpc, 4, 4,true}, 292 { ARM::VLD4DUPd32Pseudo, ARM::VLD4DUPd32, true, false, false, SingleSpc, 4, 2,true}, 293 { ARM::VLD4DUPd32Pseudo_UPD, ARM::VLD4DUPd32_UPD, true, true, true, SingleSpc, 4, 2,true}, 294 { ARM::VLD4DUPd8Pseudo, ARM::VLD4DUPd8, true, false, false, SingleSpc, 4, 8,true}, 295 { ARM::VLD4DUPd8Pseudo_UPD, ARM::VLD4DUPd8_UPD, true, true, true, SingleSpc, 4, 8,true}, 296 { ARM::VLD4DUPq16EvenPseudo, ARM::VLD4DUPq16, true, false, false, EvenDblSpc, 4, 4 ,true}, 297 { ARM::VLD4DUPq16OddPseudo, ARM::VLD4DUPq16, true, false, false, OddDblSpc, 4, 4 ,true}, 298 { ARM::VLD4DUPq32EvenPseudo, ARM::VLD4DUPq32, true, false, false, EvenDblSpc, 4, 2 ,true}, 299 { ARM::VLD4DUPq32OddPseudo, ARM::VLD4DUPq32, true, false, false, OddDblSpc, 4, 2 ,true}, 300 { ARM::VLD4DUPq8EvenPseudo, ARM::VLD4DUPq8, true, false, false, EvenDblSpc, 4, 8 ,true}, 301 { ARM::VLD4DUPq8OddPseudo, ARM::VLD4DUPq8, true, false, false, OddDblSpc, 4, 8 ,true}, 302 303 { ARM::VLD4LNd16Pseudo, ARM::VLD4LNd16, true, false, false, SingleSpc, 4, 4 ,true}, 304 { ARM::VLD4LNd16Pseudo_UPD, ARM::VLD4LNd16_UPD, true, true, true, SingleSpc, 4, 4 ,true}, 305 { ARM::VLD4LNd32Pseudo, ARM::VLD4LNd32, true, false, false, SingleSpc, 4, 2 ,true}, 306 { ARM::VLD4LNd32Pseudo_UPD, ARM::VLD4LNd32_UPD, true, true, true, SingleSpc, 4, 2 ,true}, 307 { ARM::VLD4LNd8Pseudo, ARM::VLD4LNd8, true, false, false, SingleSpc, 4, 8 ,true}, 308 { ARM::VLD4LNd8Pseudo_UPD, ARM::VLD4LNd8_UPD, true, true, true, SingleSpc, 4, 8 ,true}, 309 { ARM::VLD4LNq16Pseudo, ARM::VLD4LNq16, true, false, false, EvenDblSpc, 4, 4 ,true}, 310 { ARM::VLD4LNq16Pseudo_UPD, ARM::VLD4LNq16_UPD, true, true, true, EvenDblSpc, 4, 4 ,true}, 311 { ARM::VLD4LNq32Pseudo, ARM::VLD4LNq32, true, false, false, EvenDblSpc, 4, 2 ,true}, 312 { ARM::VLD4LNq32Pseudo_UPD, ARM::VLD4LNq32_UPD, true, true, true, EvenDblSpc, 4, 2 ,true}, 313 314 { ARM::VLD4d16Pseudo, ARM::VLD4d16, true, false, false, SingleSpc, 4, 4 ,true}, 315 { ARM::VLD4d16Pseudo_UPD, ARM::VLD4d16_UPD, true, true, true, SingleSpc, 4, 4 ,true}, 316 { ARM::VLD4d32Pseudo, ARM::VLD4d32, true, false, false, SingleSpc, 4, 2 ,true}, 317 { ARM::VLD4d32Pseudo_UPD, ARM::VLD4d32_UPD, true, true, true, SingleSpc, 4, 2 ,true}, 318 { ARM::VLD4d8Pseudo, ARM::VLD4d8, true, false, false, SingleSpc, 4, 8 ,true}, 319 { ARM::VLD4d8Pseudo_UPD, ARM::VLD4d8_UPD, true, true, true, SingleSpc, 4, 8 ,true}, 320 321 { ARM::VLD4q16Pseudo_UPD, ARM::VLD4q16_UPD, true, true, true, EvenDblSpc, 4, 4 ,true}, 322 { ARM::VLD4q16oddPseudo, ARM::VLD4q16, true, false, false, OddDblSpc, 4, 4 ,true}, 323 { ARM::VLD4q16oddPseudo_UPD, ARM::VLD4q16_UPD, true, true, true, OddDblSpc, 4, 4 ,true}, 324 { ARM::VLD4q32Pseudo_UPD, ARM::VLD4q32_UPD, true, true, true, EvenDblSpc, 4, 2 ,true}, 325 { ARM::VLD4q32oddPseudo, ARM::VLD4q32, true, false, false, OddDblSpc, 4, 2 ,true}, 326 { ARM::VLD4q32oddPseudo_UPD, ARM::VLD4q32_UPD, true, true, true, OddDblSpc, 4, 2 ,true}, 327 { ARM::VLD4q8Pseudo_UPD, ARM::VLD4q8_UPD, true, true, true, EvenDblSpc, 4, 8 ,true}, 328 { ARM::VLD4q8oddPseudo, ARM::VLD4q8, true, false, false, OddDblSpc, 4, 8 ,true}, 329 { ARM::VLD4q8oddPseudo_UPD, ARM::VLD4q8_UPD, true, true, true, OddDblSpc, 4, 8 ,true}, 330 331 { ARM::VST1LNq16Pseudo, ARM::VST1LNd16, false, false, false, EvenDblSpc, 1, 4 ,true}, 332 { ARM::VST1LNq16Pseudo_UPD, ARM::VST1LNd16_UPD, false, true, true, EvenDblSpc, 1, 4 ,true}, 333 { ARM::VST1LNq32Pseudo, ARM::VST1LNd32, false, false, false, EvenDblSpc, 1, 2 ,true}, 334 { ARM::VST1LNq32Pseudo_UPD, ARM::VST1LNd32_UPD, false, true, true, EvenDblSpc, 1, 2 ,true}, 335 { ARM::VST1LNq8Pseudo, ARM::VST1LNd8, false, false, false, EvenDblSpc, 1, 8 ,true}, 336 { ARM::VST1LNq8Pseudo_UPD, ARM::VST1LNd8_UPD, false, true, true, EvenDblSpc, 1, 8 ,true}, 337 338 { ARM::VST1d16QPseudo, ARM::VST1d16Q, false, false, false, SingleSpc, 4, 4 ,false}, 339 { ARM::VST1d16TPseudo, ARM::VST1d16T, false, false, false, SingleSpc, 3, 4 ,false}, 340 { ARM::VST1d32QPseudo, ARM::VST1d32Q, false, false, false, SingleSpc, 4, 2 ,false}, 341 { ARM::VST1d32TPseudo, ARM::VST1d32T, false, false, false, SingleSpc, 3, 2 ,false}, 342 { ARM::VST1d64QPseudo, ARM::VST1d64Q, false, false, false, SingleSpc, 4, 1 ,false}, 343 { ARM::VST1d64QPseudoWB_fixed, ARM::VST1d64Qwb_fixed, false, true, false, SingleSpc, 4, 1 ,false}, 344 { ARM::VST1d64QPseudoWB_register, ARM::VST1d64Qwb_register, false, true, true, SingleSpc, 4, 1 ,false}, 345 { ARM::VST1d64TPseudo, ARM::VST1d64T, false, false, false, SingleSpc, 3, 1 ,false}, 346 { ARM::VST1d64TPseudoWB_fixed, ARM::VST1d64Twb_fixed, false, true, false, SingleSpc, 3, 1 ,false}, 347 { ARM::VST1d64TPseudoWB_register, ARM::VST1d64Twb_register, false, true, true, SingleSpc, 3, 1 ,false}, 348 { ARM::VST1d8QPseudo, ARM::VST1d8Q, false, false, false, SingleSpc, 4, 8 ,false}, 349 { ARM::VST1d8TPseudo, ARM::VST1d8T, false, false, false, SingleSpc, 3, 8 ,false}, 350 { ARM::VST1q16HighQPseudo, ARM::VST1d16Q, false, false, false, SingleHighQSpc, 4, 4 ,false}, 351 { ARM::VST1q16HighTPseudo, ARM::VST1d16T, false, false, false, SingleHighTSpc, 3, 4 ,false}, 352 { ARM::VST1q16LowQPseudo_UPD, ARM::VST1d16Qwb_fixed, false, true, true, SingleLowSpc, 4, 4 ,false}, 353 { ARM::VST1q16LowTPseudo_UPD, ARM::VST1d16Twb_fixed, false, true, true, SingleLowSpc, 3, 4 ,false}, 354 { ARM::VST1q32HighQPseudo, ARM::VST1d32Q, false, false, false, SingleHighQSpc, 4, 2 ,false}, 355 { ARM::VST1q32HighTPseudo, ARM::VST1d32T, false, false, false, SingleHighTSpc, 3, 2 ,false}, 356 { ARM::VST1q32LowQPseudo_UPD, ARM::VST1d32Qwb_fixed, false, true, true, SingleLowSpc, 4, 2 ,false}, 357 { ARM::VST1q32LowTPseudo_UPD, ARM::VST1d32Twb_fixed, false, true, true, SingleLowSpc, 3, 2 ,false}, 358 { ARM::VST1q64HighQPseudo, ARM::VST1d64Q, false, false, false, SingleHighQSpc, 4, 1 ,false}, 359 { ARM::VST1q64HighTPseudo, ARM::VST1d64T, false, false, false, SingleHighTSpc, 3, 1 ,false}, 360 { ARM::VST1q64LowQPseudo_UPD, ARM::VST1d64Qwb_fixed, false, true, true, SingleLowSpc, 4, 1 ,false}, 361 { ARM::VST1q64LowTPseudo_UPD, ARM::VST1d64Twb_fixed, false, true, true, SingleLowSpc, 3, 1 ,false}, 362 { ARM::VST1q8HighQPseudo, ARM::VST1d8Q, false, false, false, SingleHighQSpc, 4, 8 ,false}, 363 { ARM::VST1q8HighTPseudo, ARM::VST1d8T, false, false, false, SingleHighTSpc, 3, 8 ,false}, 364 { ARM::VST1q8LowQPseudo_UPD, ARM::VST1d8Qwb_fixed, false, true, true, SingleLowSpc, 4, 8 ,false}, 365 { ARM::VST1q8LowTPseudo_UPD, ARM::VST1d8Twb_fixed, false, true, true, SingleLowSpc, 3, 8 ,false}, 366 367 { ARM::VST2LNd16Pseudo, ARM::VST2LNd16, false, false, false, SingleSpc, 2, 4 ,true}, 368 { ARM::VST2LNd16Pseudo_UPD, ARM::VST2LNd16_UPD, false, true, true, SingleSpc, 2, 4 ,true}, 369 { ARM::VST2LNd32Pseudo, ARM::VST2LNd32, false, false, false, SingleSpc, 2, 2 ,true}, 370 { ARM::VST2LNd32Pseudo_UPD, ARM::VST2LNd32_UPD, false, true, true, SingleSpc, 2, 2 ,true}, 371 { ARM::VST2LNd8Pseudo, ARM::VST2LNd8, false, false, false, SingleSpc, 2, 8 ,true}, 372 { ARM::VST2LNd8Pseudo_UPD, ARM::VST2LNd8_UPD, false, true, true, SingleSpc, 2, 8 ,true}, 373 { ARM::VST2LNq16Pseudo, ARM::VST2LNq16, false, false, false, EvenDblSpc, 2, 4,true}, 374 { ARM::VST2LNq16Pseudo_UPD, ARM::VST2LNq16_UPD, false, true, true, EvenDblSpc, 2, 4,true}, 375 { ARM::VST2LNq32Pseudo, ARM::VST2LNq32, false, false, false, EvenDblSpc, 2, 2,true}, 376 { ARM::VST2LNq32Pseudo_UPD, ARM::VST2LNq32_UPD, false, true, true, EvenDblSpc, 2, 2,true}, 377 378 { ARM::VST2q16Pseudo, ARM::VST2q16, false, false, false, SingleSpc, 4, 4 ,false}, 379 { ARM::VST2q16PseudoWB_fixed, ARM::VST2q16wb_fixed, false, true, false, SingleSpc, 4, 4 ,false}, 380 { ARM::VST2q16PseudoWB_register, ARM::VST2q16wb_register, false, true, true, SingleSpc, 4, 4 ,false}, 381 { ARM::VST2q32Pseudo, ARM::VST2q32, false, false, false, SingleSpc, 4, 2 ,false}, 382 { ARM::VST2q32PseudoWB_fixed, ARM::VST2q32wb_fixed, false, true, false, SingleSpc, 4, 2 ,false}, 383 { ARM::VST2q32PseudoWB_register, ARM::VST2q32wb_register, false, true, true, SingleSpc, 4, 2 ,false}, 384 { ARM::VST2q8Pseudo, ARM::VST2q8, false, false, false, SingleSpc, 4, 8 ,false}, 385 { ARM::VST2q8PseudoWB_fixed, ARM::VST2q8wb_fixed, false, true, false, SingleSpc, 4, 8 ,false}, 386 { ARM::VST2q8PseudoWB_register, ARM::VST2q8wb_register, false, true, true, SingleSpc, 4, 8 ,false}, 387 388 { ARM::VST3LNd16Pseudo, ARM::VST3LNd16, false, false, false, SingleSpc, 3, 4 ,true}, 389 { ARM::VST3LNd16Pseudo_UPD, ARM::VST3LNd16_UPD, false, true, true, SingleSpc, 3, 4 ,true}, 390 { ARM::VST3LNd32Pseudo, ARM::VST3LNd32, false, false, false, SingleSpc, 3, 2 ,true}, 391 { ARM::VST3LNd32Pseudo_UPD, ARM::VST3LNd32_UPD, false, true, true, SingleSpc, 3, 2 ,true}, 392 { ARM::VST3LNd8Pseudo, ARM::VST3LNd8, false, false, false, SingleSpc, 3, 8 ,true}, 393 { ARM::VST3LNd8Pseudo_UPD, ARM::VST3LNd8_UPD, false, true, true, SingleSpc, 3, 8 ,true}, 394 { ARM::VST3LNq16Pseudo, ARM::VST3LNq16, false, false, false, EvenDblSpc, 3, 4,true}, 395 { ARM::VST3LNq16Pseudo_UPD, ARM::VST3LNq16_UPD, false, true, true, EvenDblSpc, 3, 4,true}, 396 { ARM::VST3LNq32Pseudo, ARM::VST3LNq32, false, false, false, EvenDblSpc, 3, 2,true}, 397 { ARM::VST3LNq32Pseudo_UPD, ARM::VST3LNq32_UPD, false, true, true, EvenDblSpc, 3, 2,true}, 398 399 { ARM::VST3d16Pseudo, ARM::VST3d16, false, false, false, SingleSpc, 3, 4 ,true}, 400 { ARM::VST3d16Pseudo_UPD, ARM::VST3d16_UPD, false, true, true, SingleSpc, 3, 4 ,true}, 401 { ARM::VST3d32Pseudo, ARM::VST3d32, false, false, false, SingleSpc, 3, 2 ,true}, 402 { ARM::VST3d32Pseudo_UPD, ARM::VST3d32_UPD, false, true, true, SingleSpc, 3, 2 ,true}, 403 { ARM::VST3d8Pseudo, ARM::VST3d8, false, false, false, SingleSpc, 3, 8 ,true}, 404 { ARM::VST3d8Pseudo_UPD, ARM::VST3d8_UPD, false, true, true, SingleSpc, 3, 8 ,true}, 405 406 { ARM::VST3q16Pseudo_UPD, ARM::VST3q16_UPD, false, true, true, EvenDblSpc, 3, 4 ,true}, 407 { ARM::VST3q16oddPseudo, ARM::VST3q16, false, false, false, OddDblSpc, 3, 4 ,true}, 408 { ARM::VST3q16oddPseudo_UPD, ARM::VST3q16_UPD, false, true, true, OddDblSpc, 3, 4 ,true}, 409 { ARM::VST3q32Pseudo_UPD, ARM::VST3q32_UPD, false, true, true, EvenDblSpc, 3, 2 ,true}, 410 { ARM::VST3q32oddPseudo, ARM::VST3q32, false, false, false, OddDblSpc, 3, 2 ,true}, 411 { ARM::VST3q32oddPseudo_UPD, ARM::VST3q32_UPD, false, true, true, OddDblSpc, 3, 2 ,true}, 412 { ARM::VST3q8Pseudo_UPD, ARM::VST3q8_UPD, false, true, true, EvenDblSpc, 3, 8 ,true}, 413 { ARM::VST3q8oddPseudo, ARM::VST3q8, false, false, false, OddDblSpc, 3, 8 ,true}, 414 { ARM::VST3q8oddPseudo_UPD, ARM::VST3q8_UPD, false, true, true, OddDblSpc, 3, 8 ,true}, 415 416 { ARM::VST4LNd16Pseudo, ARM::VST4LNd16, false, false, false, SingleSpc, 4, 4 ,true}, 417 { ARM::VST4LNd16Pseudo_UPD, ARM::VST4LNd16_UPD, false, true, true, SingleSpc, 4, 4 ,true}, 418 { ARM::VST4LNd32Pseudo, ARM::VST4LNd32, false, false, false, SingleSpc, 4, 2 ,true}, 419 { ARM::VST4LNd32Pseudo_UPD, ARM::VST4LNd32_UPD, false, true, true, SingleSpc, 4, 2 ,true}, 420 { ARM::VST4LNd8Pseudo, ARM::VST4LNd8, false, false, false, SingleSpc, 4, 8 ,true}, 421 { ARM::VST4LNd8Pseudo_UPD, ARM::VST4LNd8_UPD, false, true, true, SingleSpc, 4, 8 ,true}, 422 { ARM::VST4LNq16Pseudo, ARM::VST4LNq16, false, false, false, EvenDblSpc, 4, 4,true}, 423 { ARM::VST4LNq16Pseudo_UPD, ARM::VST4LNq16_UPD, false, true, true, EvenDblSpc, 4, 4,true}, 424 { ARM::VST4LNq32Pseudo, ARM::VST4LNq32, false, false, false, EvenDblSpc, 4, 2,true}, 425 { ARM::VST4LNq32Pseudo_UPD, ARM::VST4LNq32_UPD, false, true, true, EvenDblSpc, 4, 2,true}, 426 427 { ARM::VST4d16Pseudo, ARM::VST4d16, false, false, false, SingleSpc, 4, 4 ,true}, 428 { ARM::VST4d16Pseudo_UPD, ARM::VST4d16_UPD, false, true, true, SingleSpc, 4, 4 ,true}, 429 { ARM::VST4d32Pseudo, ARM::VST4d32, false, false, false, SingleSpc, 4, 2 ,true}, 430 { ARM::VST4d32Pseudo_UPD, ARM::VST4d32_UPD, false, true, true, SingleSpc, 4, 2 ,true}, 431 { ARM::VST4d8Pseudo, ARM::VST4d8, false, false, false, SingleSpc, 4, 8 ,true}, 432 { ARM::VST4d8Pseudo_UPD, ARM::VST4d8_UPD, false, true, true, SingleSpc, 4, 8 ,true}, 433 434 { ARM::VST4q16Pseudo_UPD, ARM::VST4q16_UPD, false, true, true, EvenDblSpc, 4, 4 ,true}, 435 { ARM::VST4q16oddPseudo, ARM::VST4q16, false, false, false, OddDblSpc, 4, 4 ,true}, 436 { ARM::VST4q16oddPseudo_UPD, ARM::VST4q16_UPD, false, true, true, OddDblSpc, 4, 4 ,true}, 437 { ARM::VST4q32Pseudo_UPD, ARM::VST4q32_UPD, false, true, true, EvenDblSpc, 4, 2 ,true}, 438 { ARM::VST4q32oddPseudo, ARM::VST4q32, false, false, false, OddDblSpc, 4, 2 ,true}, 439 { ARM::VST4q32oddPseudo_UPD, ARM::VST4q32_UPD, false, true, true, OddDblSpc, 4, 2 ,true}, 440 { ARM::VST4q8Pseudo_UPD, ARM::VST4q8_UPD, false, true, true, EvenDblSpc, 4, 8 ,true}, 441 { ARM::VST4q8oddPseudo, ARM::VST4q8, false, false, false, OddDblSpc, 4, 8 ,true}, 442 { ARM::VST4q8oddPseudo_UPD, ARM::VST4q8_UPD, false, true, true, OddDblSpc, 4, 8 ,true} 443 }; 444 445 /// LookupNEONLdSt - Search the NEONLdStTable for information about a NEON 446 /// load or store pseudo instruction. 447 static const NEONLdStTableEntry *LookupNEONLdSt(unsigned Opcode) { 448 #ifndef NDEBUG 449 // Make sure the table is sorted. 450 static std::atomic<bool> TableChecked(false); 451 if (!TableChecked.load(std::memory_order_relaxed)) { 452 assert(llvm::is_sorted(NEONLdStTable) && "NEONLdStTable is not sorted!"); 453 TableChecked.store(true, std::memory_order_relaxed); 454 } 455 #endif 456 457 auto I = llvm::lower_bound(NEONLdStTable, Opcode); 458 if (I != std::end(NEONLdStTable) && I->PseudoOpc == Opcode) 459 return I; 460 return nullptr; 461 } 462 463 /// GetDSubRegs - Get 4 D subregisters of a Q, QQ, or QQQQ register, 464 /// corresponding to the specified register spacing. Not all of the results 465 /// are necessarily valid, e.g., a Q register only has 2 D subregisters. 466 static void GetDSubRegs(unsigned Reg, NEONRegSpacing RegSpc, 467 const TargetRegisterInfo *TRI, unsigned &D0, 468 unsigned &D1, unsigned &D2, unsigned &D3) { 469 if (RegSpc == SingleSpc || RegSpc == SingleLowSpc) { 470 D0 = TRI->getSubReg(Reg, ARM::dsub_0); 471 D1 = TRI->getSubReg(Reg, ARM::dsub_1); 472 D2 = TRI->getSubReg(Reg, ARM::dsub_2); 473 D3 = TRI->getSubReg(Reg, ARM::dsub_3); 474 } else if (RegSpc == SingleHighQSpc) { 475 D0 = TRI->getSubReg(Reg, ARM::dsub_4); 476 D1 = TRI->getSubReg(Reg, ARM::dsub_5); 477 D2 = TRI->getSubReg(Reg, ARM::dsub_6); 478 D3 = TRI->getSubReg(Reg, ARM::dsub_7); 479 } else if (RegSpc == SingleHighTSpc) { 480 D0 = TRI->getSubReg(Reg, ARM::dsub_3); 481 D1 = TRI->getSubReg(Reg, ARM::dsub_4); 482 D2 = TRI->getSubReg(Reg, ARM::dsub_5); 483 D3 = TRI->getSubReg(Reg, ARM::dsub_6); 484 } else if (RegSpc == EvenDblSpc) { 485 D0 = TRI->getSubReg(Reg, ARM::dsub_0); 486 D1 = TRI->getSubReg(Reg, ARM::dsub_2); 487 D2 = TRI->getSubReg(Reg, ARM::dsub_4); 488 D3 = TRI->getSubReg(Reg, ARM::dsub_6); 489 } else { 490 assert(RegSpc == OddDblSpc && "unknown register spacing"); 491 D0 = TRI->getSubReg(Reg, ARM::dsub_1); 492 D1 = TRI->getSubReg(Reg, ARM::dsub_3); 493 D2 = TRI->getSubReg(Reg, ARM::dsub_5); 494 D3 = TRI->getSubReg(Reg, ARM::dsub_7); 495 } 496 } 497 498 /// ExpandVLD - Translate VLD pseudo instructions with Q, QQ or QQQQ register 499 /// operands to real VLD instructions with D register operands. 500 void ARMExpandPseudo::ExpandVLD(MachineBasicBlock::iterator &MBBI) { 501 MachineInstr &MI = *MBBI; 502 MachineBasicBlock &MBB = *MI.getParent(); 503 LLVM_DEBUG(dbgs() << "Expanding: "; MI.dump()); 504 505 const NEONLdStTableEntry *TableEntry = LookupNEONLdSt(MI.getOpcode()); 506 assert(TableEntry && TableEntry->IsLoad && "NEONLdStTable lookup failed"); 507 NEONRegSpacing RegSpc = (NEONRegSpacing)TableEntry->RegSpacing; 508 unsigned NumRegs = TableEntry->NumRegs; 509 510 MachineInstrBuilder MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(), 511 TII->get(TableEntry->RealOpc)); 512 unsigned OpIdx = 0; 513 514 bool DstIsDead = MI.getOperand(OpIdx).isDead(); 515 Register DstReg = MI.getOperand(OpIdx++).getReg(); 516 if(TableEntry->RealOpc == ARM::VLD2DUPd8x2 || 517 TableEntry->RealOpc == ARM::VLD2DUPd16x2 || 518 TableEntry->RealOpc == ARM::VLD2DUPd32x2) { 519 unsigned SubRegIndex; 520 if (RegSpc == EvenDblSpc) { 521 SubRegIndex = ARM::dsub_0; 522 } else { 523 assert(RegSpc == OddDblSpc && "Unexpected spacing!"); 524 SubRegIndex = ARM::dsub_1; 525 } 526 Register SubReg = TRI->getSubReg(DstReg, SubRegIndex); 527 unsigned DstRegPair = TRI->getMatchingSuperReg(SubReg, ARM::dsub_0, 528 &ARM::DPairSpcRegClass); 529 MIB.addReg(DstRegPair, RegState::Define | getDeadRegState(DstIsDead)); 530 } else { 531 unsigned D0, D1, D2, D3; 532 GetDSubRegs(DstReg, RegSpc, TRI, D0, D1, D2, D3); 533 MIB.addReg(D0, RegState::Define | getDeadRegState(DstIsDead)); 534 if (NumRegs > 1 && TableEntry->copyAllListRegs) 535 MIB.addReg(D1, RegState::Define | getDeadRegState(DstIsDead)); 536 if (NumRegs > 2 && TableEntry->copyAllListRegs) 537 MIB.addReg(D2, RegState::Define | getDeadRegState(DstIsDead)); 538 if (NumRegs > 3 && TableEntry->copyAllListRegs) 539 MIB.addReg(D3, RegState::Define | getDeadRegState(DstIsDead)); 540 } 541 542 if (TableEntry->isUpdating) 543 MIB.add(MI.getOperand(OpIdx++)); 544 545 // Copy the addrmode6 operands. 546 MIB.add(MI.getOperand(OpIdx++)); 547 MIB.add(MI.getOperand(OpIdx++)); 548 549 // Copy the am6offset operand. 550 if (TableEntry->hasWritebackOperand) { 551 // TODO: The writing-back pseudo instructions we translate here are all 552 // defined to take am6offset nodes that are capable to represent both fixed 553 // and register forms. Some real instructions, however, do not rely on 554 // am6offset and have separate definitions for such forms. When this is the 555 // case, fixed forms do not take any offset nodes, so here we skip them for 556 // such instructions. Once all real and pseudo writing-back instructions are 557 // rewritten without use of am6offset nodes, this code will go away. 558 const MachineOperand &AM6Offset = MI.getOperand(OpIdx++); 559 if (TableEntry->RealOpc == ARM::VLD1d8Qwb_fixed || 560 TableEntry->RealOpc == ARM::VLD1d16Qwb_fixed || 561 TableEntry->RealOpc == ARM::VLD1d32Qwb_fixed || 562 TableEntry->RealOpc == ARM::VLD1d64Qwb_fixed || 563 TableEntry->RealOpc == ARM::VLD1d8Twb_fixed || 564 TableEntry->RealOpc == ARM::VLD1d16Twb_fixed || 565 TableEntry->RealOpc == ARM::VLD1d32Twb_fixed || 566 TableEntry->RealOpc == ARM::VLD1d64Twb_fixed) { 567 assert(AM6Offset.getReg() == 0 && 568 "A fixed writing-back pseudo instruction provides an offset " 569 "register!"); 570 } else { 571 MIB.add(AM6Offset); 572 } 573 } 574 575 // For an instruction writing double-spaced subregs, the pseudo instruction 576 // has an extra operand that is a use of the super-register. Record the 577 // operand index and skip over it. 578 unsigned SrcOpIdx = 0; 579 if(TableEntry->RealOpc != ARM::VLD2DUPd8x2 && 580 TableEntry->RealOpc != ARM::VLD2DUPd16x2 && 581 TableEntry->RealOpc != ARM::VLD2DUPd32x2) { 582 if (RegSpc == EvenDblSpc || RegSpc == OddDblSpc || 583 RegSpc == SingleLowSpc || RegSpc == SingleHighQSpc || 584 RegSpc == SingleHighTSpc) 585 SrcOpIdx = OpIdx++; 586 } 587 588 // Copy the predicate operands. 589 MIB.add(MI.getOperand(OpIdx++)); 590 MIB.add(MI.getOperand(OpIdx++)); 591 592 // Copy the super-register source operand used for double-spaced subregs over 593 // to the new instruction as an implicit operand. 594 if (SrcOpIdx != 0) { 595 MachineOperand MO = MI.getOperand(SrcOpIdx); 596 MO.setImplicit(true); 597 MIB.add(MO); 598 } 599 // Add an implicit def for the super-register. 600 MIB.addReg(DstReg, RegState::ImplicitDefine | getDeadRegState(DstIsDead)); 601 TransferImpOps(MI, MIB, MIB); 602 603 // Transfer memoperands. 604 MIB.cloneMemRefs(MI); 605 MI.eraseFromParent(); 606 LLVM_DEBUG(dbgs() << "To: "; MIB.getInstr()->dump();); 607 } 608 609 /// ExpandVST - Translate VST pseudo instructions with Q, QQ or QQQQ register 610 /// operands to real VST instructions with D register operands. 611 void ARMExpandPseudo::ExpandVST(MachineBasicBlock::iterator &MBBI) { 612 MachineInstr &MI = *MBBI; 613 MachineBasicBlock &MBB = *MI.getParent(); 614 LLVM_DEBUG(dbgs() << "Expanding: "; MI.dump()); 615 616 const NEONLdStTableEntry *TableEntry = LookupNEONLdSt(MI.getOpcode()); 617 assert(TableEntry && !TableEntry->IsLoad && "NEONLdStTable lookup failed"); 618 NEONRegSpacing RegSpc = (NEONRegSpacing)TableEntry->RegSpacing; 619 unsigned NumRegs = TableEntry->NumRegs; 620 621 MachineInstrBuilder MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(), 622 TII->get(TableEntry->RealOpc)); 623 unsigned OpIdx = 0; 624 if (TableEntry->isUpdating) 625 MIB.add(MI.getOperand(OpIdx++)); 626 627 // Copy the addrmode6 operands. 628 MIB.add(MI.getOperand(OpIdx++)); 629 MIB.add(MI.getOperand(OpIdx++)); 630 631 if (TableEntry->hasWritebackOperand) { 632 // TODO: The writing-back pseudo instructions we translate here are all 633 // defined to take am6offset nodes that are capable to represent both fixed 634 // and register forms. Some real instructions, however, do not rely on 635 // am6offset and have separate definitions for such forms. When this is the 636 // case, fixed forms do not take any offset nodes, so here we skip them for 637 // such instructions. Once all real and pseudo writing-back instructions are 638 // rewritten without use of am6offset nodes, this code will go away. 639 const MachineOperand &AM6Offset = MI.getOperand(OpIdx++); 640 if (TableEntry->RealOpc == ARM::VST1d8Qwb_fixed || 641 TableEntry->RealOpc == ARM::VST1d16Qwb_fixed || 642 TableEntry->RealOpc == ARM::VST1d32Qwb_fixed || 643 TableEntry->RealOpc == ARM::VST1d64Qwb_fixed || 644 TableEntry->RealOpc == ARM::VST1d8Twb_fixed || 645 TableEntry->RealOpc == ARM::VST1d16Twb_fixed || 646 TableEntry->RealOpc == ARM::VST1d32Twb_fixed || 647 TableEntry->RealOpc == ARM::VST1d64Twb_fixed) { 648 assert(AM6Offset.getReg() == 0 && 649 "A fixed writing-back pseudo instruction provides an offset " 650 "register!"); 651 } else { 652 MIB.add(AM6Offset); 653 } 654 } 655 656 bool SrcIsKill = MI.getOperand(OpIdx).isKill(); 657 bool SrcIsUndef = MI.getOperand(OpIdx).isUndef(); 658 Register SrcReg = MI.getOperand(OpIdx++).getReg(); 659 unsigned D0, D1, D2, D3; 660 GetDSubRegs(SrcReg, RegSpc, TRI, D0, D1, D2, D3); 661 MIB.addReg(D0, getUndefRegState(SrcIsUndef)); 662 if (NumRegs > 1 && TableEntry->copyAllListRegs) 663 MIB.addReg(D1, getUndefRegState(SrcIsUndef)); 664 if (NumRegs > 2 && TableEntry->copyAllListRegs) 665 MIB.addReg(D2, getUndefRegState(SrcIsUndef)); 666 if (NumRegs > 3 && TableEntry->copyAllListRegs) 667 MIB.addReg(D3, getUndefRegState(SrcIsUndef)); 668 669 // Copy the predicate operands. 670 MIB.add(MI.getOperand(OpIdx++)); 671 MIB.add(MI.getOperand(OpIdx++)); 672 673 if (SrcIsKill && !SrcIsUndef) // Add an implicit kill for the super-reg. 674 MIB->addRegisterKilled(SrcReg, TRI, true); 675 else if (!SrcIsUndef) 676 MIB.addReg(SrcReg, RegState::Implicit); // Add implicit uses for src reg. 677 TransferImpOps(MI, MIB, MIB); 678 679 // Transfer memoperands. 680 MIB.cloneMemRefs(MI); 681 MI.eraseFromParent(); 682 LLVM_DEBUG(dbgs() << "To: "; MIB.getInstr()->dump();); 683 } 684 685 /// ExpandLaneOp - Translate VLD*LN and VST*LN instructions with Q, QQ or QQQQ 686 /// register operands to real instructions with D register operands. 687 void ARMExpandPseudo::ExpandLaneOp(MachineBasicBlock::iterator &MBBI) { 688 MachineInstr &MI = *MBBI; 689 MachineBasicBlock &MBB = *MI.getParent(); 690 LLVM_DEBUG(dbgs() << "Expanding: "; MI.dump()); 691 692 const NEONLdStTableEntry *TableEntry = LookupNEONLdSt(MI.getOpcode()); 693 assert(TableEntry && "NEONLdStTable lookup failed"); 694 NEONRegSpacing RegSpc = (NEONRegSpacing)TableEntry->RegSpacing; 695 unsigned NumRegs = TableEntry->NumRegs; 696 unsigned RegElts = TableEntry->RegElts; 697 698 MachineInstrBuilder MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(), 699 TII->get(TableEntry->RealOpc)); 700 unsigned OpIdx = 0; 701 // The lane operand is always the 3rd from last operand, before the 2 702 // predicate operands. 703 unsigned Lane = MI.getOperand(MI.getDesc().getNumOperands() - 3).getImm(); 704 705 // Adjust the lane and spacing as needed for Q registers. 706 assert(RegSpc != OddDblSpc && "unexpected register spacing for VLD/VST-lane"); 707 if (RegSpc == EvenDblSpc && Lane >= RegElts) { 708 RegSpc = OddDblSpc; 709 Lane -= RegElts; 710 } 711 assert(Lane < RegElts && "out of range lane for VLD/VST-lane"); 712 713 unsigned D0 = 0, D1 = 0, D2 = 0, D3 = 0; 714 unsigned DstReg = 0; 715 bool DstIsDead = false; 716 if (TableEntry->IsLoad) { 717 DstIsDead = MI.getOperand(OpIdx).isDead(); 718 DstReg = MI.getOperand(OpIdx++).getReg(); 719 GetDSubRegs(DstReg, RegSpc, TRI, D0, D1, D2, D3); 720 MIB.addReg(D0, RegState::Define | getDeadRegState(DstIsDead)); 721 if (NumRegs > 1) 722 MIB.addReg(D1, RegState::Define | getDeadRegState(DstIsDead)); 723 if (NumRegs > 2) 724 MIB.addReg(D2, RegState::Define | getDeadRegState(DstIsDead)); 725 if (NumRegs > 3) 726 MIB.addReg(D3, RegState::Define | getDeadRegState(DstIsDead)); 727 } 728 729 if (TableEntry->isUpdating) 730 MIB.add(MI.getOperand(OpIdx++)); 731 732 // Copy the addrmode6 operands. 733 MIB.add(MI.getOperand(OpIdx++)); 734 MIB.add(MI.getOperand(OpIdx++)); 735 // Copy the am6offset operand. 736 if (TableEntry->hasWritebackOperand) 737 MIB.add(MI.getOperand(OpIdx++)); 738 739 // Grab the super-register source. 740 MachineOperand MO = MI.getOperand(OpIdx++); 741 if (!TableEntry->IsLoad) 742 GetDSubRegs(MO.getReg(), RegSpc, TRI, D0, D1, D2, D3); 743 744 // Add the subregs as sources of the new instruction. 745 unsigned SrcFlags = (getUndefRegState(MO.isUndef()) | 746 getKillRegState(MO.isKill())); 747 MIB.addReg(D0, SrcFlags); 748 if (NumRegs > 1) 749 MIB.addReg(D1, SrcFlags); 750 if (NumRegs > 2) 751 MIB.addReg(D2, SrcFlags); 752 if (NumRegs > 3) 753 MIB.addReg(D3, SrcFlags); 754 755 // Add the lane number operand. 756 MIB.addImm(Lane); 757 OpIdx += 1; 758 759 // Copy the predicate operands. 760 MIB.add(MI.getOperand(OpIdx++)); 761 MIB.add(MI.getOperand(OpIdx++)); 762 763 // Copy the super-register source to be an implicit source. 764 MO.setImplicit(true); 765 MIB.add(MO); 766 if (TableEntry->IsLoad) 767 // Add an implicit def for the super-register. 768 MIB.addReg(DstReg, RegState::ImplicitDefine | getDeadRegState(DstIsDead)); 769 TransferImpOps(MI, MIB, MIB); 770 // Transfer memoperands. 771 MIB.cloneMemRefs(MI); 772 MI.eraseFromParent(); 773 } 774 775 /// ExpandVTBL - Translate VTBL and VTBX pseudo instructions with Q or QQ 776 /// register operands to real instructions with D register operands. 777 void ARMExpandPseudo::ExpandVTBL(MachineBasicBlock::iterator &MBBI, 778 unsigned Opc, bool IsExt) { 779 MachineInstr &MI = *MBBI; 780 MachineBasicBlock &MBB = *MI.getParent(); 781 LLVM_DEBUG(dbgs() << "Expanding: "; MI.dump()); 782 783 MachineInstrBuilder MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(Opc)); 784 unsigned OpIdx = 0; 785 786 // Transfer the destination register operand. 787 MIB.add(MI.getOperand(OpIdx++)); 788 if (IsExt) { 789 MachineOperand VdSrc(MI.getOperand(OpIdx++)); 790 MIB.add(VdSrc); 791 } 792 793 bool SrcIsKill = MI.getOperand(OpIdx).isKill(); 794 Register SrcReg = MI.getOperand(OpIdx++).getReg(); 795 unsigned D0, D1, D2, D3; 796 GetDSubRegs(SrcReg, SingleSpc, TRI, D0, D1, D2, D3); 797 MIB.addReg(D0); 798 799 // Copy the other source register operand. 800 MachineOperand VmSrc(MI.getOperand(OpIdx++)); 801 MIB.add(VmSrc); 802 803 // Copy the predicate operands. 804 MIB.add(MI.getOperand(OpIdx++)); 805 MIB.add(MI.getOperand(OpIdx++)); 806 807 // Add an implicit kill and use for the super-reg. 808 MIB.addReg(SrcReg, RegState::Implicit | getKillRegState(SrcIsKill)); 809 TransferImpOps(MI, MIB, MIB); 810 MI.eraseFromParent(); 811 LLVM_DEBUG(dbgs() << "To: "; MIB.getInstr()->dump();); 812 } 813 814 static bool IsAnAddressOperand(const MachineOperand &MO) { 815 // This check is overly conservative. Unless we are certain that the machine 816 // operand is not a symbol reference, we return that it is a symbol reference. 817 // This is important as the load pair may not be split up Windows. 818 switch (MO.getType()) { 819 case MachineOperand::MO_Register: 820 case MachineOperand::MO_Immediate: 821 case MachineOperand::MO_CImmediate: 822 case MachineOperand::MO_FPImmediate: 823 case MachineOperand::MO_ShuffleMask: 824 return false; 825 case MachineOperand::MO_MachineBasicBlock: 826 return true; 827 case MachineOperand::MO_FrameIndex: 828 return false; 829 case MachineOperand::MO_ConstantPoolIndex: 830 case MachineOperand::MO_TargetIndex: 831 case MachineOperand::MO_JumpTableIndex: 832 case MachineOperand::MO_ExternalSymbol: 833 case MachineOperand::MO_GlobalAddress: 834 case MachineOperand::MO_BlockAddress: 835 return true; 836 case MachineOperand::MO_RegisterMask: 837 case MachineOperand::MO_RegisterLiveOut: 838 return false; 839 case MachineOperand::MO_Metadata: 840 case MachineOperand::MO_MCSymbol: 841 return true; 842 case MachineOperand::MO_CFIIndex: 843 return false; 844 case MachineOperand::MO_IntrinsicID: 845 case MachineOperand::MO_Predicate: 846 llvm_unreachable("should not exist post-isel"); 847 } 848 llvm_unreachable("unhandled machine operand type"); 849 } 850 851 static MachineOperand makeImplicit(const MachineOperand &MO) { 852 MachineOperand NewMO = MO; 853 NewMO.setImplicit(); 854 return NewMO; 855 } 856 857 void ARMExpandPseudo::ExpandMOV32BitImm(MachineBasicBlock &MBB, 858 MachineBasicBlock::iterator &MBBI) { 859 MachineInstr &MI = *MBBI; 860 unsigned Opcode = MI.getOpcode(); 861 Register PredReg; 862 ARMCC::CondCodes Pred = getInstrPredicate(MI, PredReg); 863 Register DstReg = MI.getOperand(0).getReg(); 864 bool DstIsDead = MI.getOperand(0).isDead(); 865 bool isCC = Opcode == ARM::MOVCCi32imm || Opcode == ARM::t2MOVCCi32imm; 866 const MachineOperand &MO = MI.getOperand(isCC ? 2 : 1); 867 bool RequiresBundling = STI->isTargetWindows() && IsAnAddressOperand(MO); 868 MachineInstrBuilder LO16, HI16; 869 LLVM_DEBUG(dbgs() << "Expanding: "; MI.dump()); 870 871 if (!STI->hasV6T2Ops() && 872 (Opcode == ARM::MOVi32imm || Opcode == ARM::MOVCCi32imm)) { 873 // FIXME Windows CE supports older ARM CPUs 874 assert(!STI->isTargetWindows() && "Windows on ARM requires ARMv7+"); 875 876 assert (MO.isImm() && "MOVi32imm w/ non-immediate source operand!"); 877 unsigned ImmVal = (unsigned)MO.getImm(); 878 unsigned SOImmValV1 = 0, SOImmValV2 = 0; 879 880 if (ARM_AM::isSOImmTwoPartVal(ImmVal)) { // Expand into a movi + orr. 881 LO16 = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVi), DstReg); 882 HI16 = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::ORRri)) 883 .addReg(DstReg, RegState::Define | getDeadRegState(DstIsDead)) 884 .addReg(DstReg); 885 SOImmValV1 = ARM_AM::getSOImmTwoPartFirst(ImmVal); 886 SOImmValV2 = ARM_AM::getSOImmTwoPartSecond(ImmVal); 887 } else { // Expand into a mvn + sub. 888 LO16 = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MVNi), DstReg); 889 HI16 = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::SUBri)) 890 .addReg(DstReg, RegState::Define | getDeadRegState(DstIsDead)) 891 .addReg(DstReg); 892 SOImmValV1 = ARM_AM::getSOImmTwoPartFirst(-ImmVal); 893 SOImmValV2 = ARM_AM::getSOImmTwoPartSecond(-ImmVal); 894 SOImmValV1 = ~(-SOImmValV1); 895 } 896 897 unsigned MIFlags = MI.getFlags(); 898 LO16 = LO16.addImm(SOImmValV1); 899 HI16 = HI16.addImm(SOImmValV2); 900 LO16.cloneMemRefs(MI); 901 HI16.cloneMemRefs(MI); 902 LO16.setMIFlags(MIFlags); 903 HI16.setMIFlags(MIFlags); 904 LO16.addImm(Pred).addReg(PredReg).add(condCodeOp()); 905 HI16.addImm(Pred).addReg(PredReg).add(condCodeOp()); 906 if (isCC) 907 LO16.add(makeImplicit(MI.getOperand(1))); 908 TransferImpOps(MI, LO16, HI16); 909 MI.eraseFromParent(); 910 return; 911 } 912 913 unsigned LO16Opc = 0; 914 unsigned HI16Opc = 0; 915 unsigned MIFlags = MI.getFlags(); 916 if (Opcode == ARM::t2MOVi32imm || Opcode == ARM::t2MOVCCi32imm) { 917 LO16Opc = ARM::t2MOVi16; 918 HI16Opc = ARM::t2MOVTi16; 919 } else { 920 LO16Opc = ARM::MOVi16; 921 HI16Opc = ARM::MOVTi16; 922 } 923 924 LO16 = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(LO16Opc), DstReg); 925 HI16 = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(HI16Opc)) 926 .addReg(DstReg, RegState::Define | getDeadRegState(DstIsDead)) 927 .addReg(DstReg); 928 929 LO16.setMIFlags(MIFlags); 930 HI16.setMIFlags(MIFlags); 931 932 switch (MO.getType()) { 933 case MachineOperand::MO_Immediate: { 934 unsigned Imm = MO.getImm(); 935 unsigned Lo16 = Imm & 0xffff; 936 unsigned Hi16 = (Imm >> 16) & 0xffff; 937 LO16 = LO16.addImm(Lo16); 938 HI16 = HI16.addImm(Hi16); 939 break; 940 } 941 case MachineOperand::MO_ExternalSymbol: { 942 const char *ES = MO.getSymbolName(); 943 unsigned TF = MO.getTargetFlags(); 944 LO16 = LO16.addExternalSymbol(ES, TF | ARMII::MO_LO16); 945 HI16 = HI16.addExternalSymbol(ES, TF | ARMII::MO_HI16); 946 break; 947 } 948 default: { 949 const GlobalValue *GV = MO.getGlobal(); 950 unsigned TF = MO.getTargetFlags(); 951 LO16 = LO16.addGlobalAddress(GV, MO.getOffset(), TF | ARMII::MO_LO16); 952 HI16 = HI16.addGlobalAddress(GV, MO.getOffset(), TF | ARMII::MO_HI16); 953 break; 954 } 955 } 956 957 LO16.cloneMemRefs(MI); 958 HI16.cloneMemRefs(MI); 959 LO16.addImm(Pred).addReg(PredReg); 960 HI16.addImm(Pred).addReg(PredReg); 961 962 if (RequiresBundling) 963 finalizeBundle(MBB, LO16->getIterator(), MBBI->getIterator()); 964 965 if (isCC) 966 LO16.add(makeImplicit(MI.getOperand(1))); 967 TransferImpOps(MI, LO16, HI16); 968 MI.eraseFromParent(); 969 LLVM_DEBUG(dbgs() << "To: "; LO16.getInstr()->dump();); 970 LLVM_DEBUG(dbgs() << "And: "; HI16.getInstr()->dump();); 971 } 972 973 // The size of the area, accessed by that VLSTM/VLLDM 974 // S0-S31 + FPSCR + 8 more bytes (VPR + pad, or just pad) 975 static const int CMSE_FP_SAVE_SIZE = 136; 976 977 static void determineGPRegsToClear(const MachineInstr &MI, 978 const std::initializer_list<unsigned> &Regs, 979 SmallVectorImpl<unsigned> &ClearRegs) { 980 SmallVector<unsigned, 4> OpRegs; 981 for (const MachineOperand &Op : MI.operands()) { 982 if (!Op.isReg() || !Op.isUse()) 983 continue; 984 OpRegs.push_back(Op.getReg()); 985 } 986 llvm::sort(OpRegs); 987 988 std::set_difference(Regs.begin(), Regs.end(), OpRegs.begin(), OpRegs.end(), 989 std::back_inserter(ClearRegs)); 990 } 991 992 void ARMExpandPseudo::CMSEClearGPRegs( 993 MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI, 994 const DebugLoc &DL, const SmallVectorImpl<unsigned> &ClearRegs, 995 unsigned ClobberReg) { 996 997 if (STI->hasV8_1MMainlineOps()) { 998 // Clear the registers using the CLRM instruction. 999 MachineInstrBuilder CLRM = 1000 BuildMI(MBB, MBBI, DL, TII->get(ARM::t2CLRM)).add(predOps(ARMCC::AL)); 1001 for (unsigned R : ClearRegs) 1002 CLRM.addReg(R, RegState::Define); 1003 CLRM.addReg(ARM::APSR, RegState::Define); 1004 CLRM.addReg(ARM::CPSR, RegState::Define | RegState::Implicit); 1005 } else { 1006 // Clear the registers and flags by copying ClobberReg into them. 1007 // (Baseline can't do a high register clear in one instruction). 1008 for (unsigned Reg : ClearRegs) { 1009 if (Reg == ClobberReg) 1010 continue; 1011 BuildMI(MBB, MBBI, DL, TII->get(ARM::tMOVr), Reg) 1012 .addReg(ClobberReg) 1013 .add(predOps(ARMCC::AL)); 1014 } 1015 1016 BuildMI(MBB, MBBI, DL, TII->get(ARM::t2MSR_M)) 1017 .addImm(STI->hasDSP() ? 0xc00 : 0x800) 1018 .addReg(ClobberReg) 1019 .add(predOps(ARMCC::AL)); 1020 } 1021 } 1022 1023 // Find which FP registers need to be cleared. The parameter `ClearRegs` is 1024 // initialised with all elements set to true, and this function resets all the 1025 // bits, which correspond to register uses. Returns true if any floating point 1026 // register is defined, false otherwise. 1027 static bool determineFPRegsToClear(const MachineInstr &MI, 1028 BitVector &ClearRegs) { 1029 bool DefFP = false; 1030 for (const MachineOperand &Op : MI.operands()) { 1031 if (!Op.isReg()) 1032 continue; 1033 1034 unsigned Reg = Op.getReg(); 1035 if (Op.isDef()) { 1036 if ((Reg >= ARM::Q0 && Reg <= ARM::Q7) || 1037 (Reg >= ARM::D0 && Reg <= ARM::D15) || 1038 (Reg >= ARM::S0 && Reg <= ARM::S31)) 1039 DefFP = true; 1040 continue; 1041 } 1042 1043 if (Reg >= ARM::Q0 && Reg <= ARM::Q7) { 1044 int R = Reg - ARM::Q0; 1045 ClearRegs.reset(R * 4, (R + 1) * 4); 1046 } else if (Reg >= ARM::D0 && Reg <= ARM::D15) { 1047 int R = Reg - ARM::D0; 1048 ClearRegs.reset(R * 2, (R + 1) * 2); 1049 } else if (Reg >= ARM::S0 && Reg <= ARM::S31) { 1050 ClearRegs[Reg - ARM::S0] = false; 1051 } 1052 } 1053 return DefFP; 1054 } 1055 1056 MachineBasicBlock & 1057 ARMExpandPseudo::CMSEClearFPRegs(MachineBasicBlock &MBB, 1058 MachineBasicBlock::iterator MBBI) { 1059 BitVector ClearRegs(16, true); 1060 (void)determineFPRegsToClear(*MBBI, ClearRegs); 1061 1062 if (STI->hasV8_1MMainlineOps()) 1063 return CMSEClearFPRegsV81(MBB, MBBI, ClearRegs); 1064 else 1065 return CMSEClearFPRegsV8(MBB, MBBI, ClearRegs); 1066 } 1067 1068 // Clear the FP registers for v8.0-M, by copying over the content 1069 // of LR. Uses R12 as a scratch register. 1070 MachineBasicBlock & 1071 ARMExpandPseudo::CMSEClearFPRegsV8(MachineBasicBlock &MBB, 1072 MachineBasicBlock::iterator MBBI, 1073 const BitVector &ClearRegs) { 1074 if (!STI->hasFPRegs()) 1075 return MBB; 1076 1077 auto &RetI = *MBBI; 1078 const DebugLoc &DL = RetI.getDebugLoc(); 1079 1080 // If optimising for minimum size, clear FP registers unconditionally. 1081 // Otherwise, check the CONTROL.SFPA (Secure Floating-Point Active) bit and 1082 // don't clear them if they belong to the non-secure state. 1083 MachineBasicBlock *ClearBB, *DoneBB; 1084 if (STI->hasMinSize()) { 1085 ClearBB = DoneBB = &MBB; 1086 } else { 1087 MachineFunction *MF = MBB.getParent(); 1088 ClearBB = MF->CreateMachineBasicBlock(MBB.getBasicBlock()); 1089 DoneBB = MF->CreateMachineBasicBlock(MBB.getBasicBlock()); 1090 1091 MF->insert(++MBB.getIterator(), ClearBB); 1092 MF->insert(++ClearBB->getIterator(), DoneBB); 1093 1094 DoneBB->splice(DoneBB->end(), &MBB, MBBI, MBB.end()); 1095 DoneBB->transferSuccessors(&MBB); 1096 MBB.addSuccessor(ClearBB); 1097 MBB.addSuccessor(DoneBB); 1098 ClearBB->addSuccessor(DoneBB); 1099 1100 // At the new basic blocks we need to have live-in the registers, used 1101 // for the return value as well as LR, used to clear registers. 1102 for (const MachineOperand &Op : RetI.operands()) { 1103 if (!Op.isReg()) 1104 continue; 1105 Register Reg = Op.getReg(); 1106 if (Reg == ARM::NoRegister || Reg == ARM::LR) 1107 continue; 1108 assert(Register::isPhysicalRegister(Reg) && "Unallocated register"); 1109 ClearBB->addLiveIn(Reg); 1110 DoneBB->addLiveIn(Reg); 1111 } 1112 ClearBB->addLiveIn(ARM::LR); 1113 DoneBB->addLiveIn(ARM::LR); 1114 1115 // Read the CONTROL register. 1116 BuildMI(MBB, MBB.end(), DL, TII->get(ARM::t2MRS_M), ARM::R12) 1117 .addImm(20) 1118 .add(predOps(ARMCC::AL)); 1119 // Check bit 3 (SFPA). 1120 BuildMI(MBB, MBB.end(), DL, TII->get(ARM::t2TSTri)) 1121 .addReg(ARM::R12) 1122 .addImm(8) 1123 .add(predOps(ARMCC::AL)); 1124 // If SFPA is clear, jump over ClearBB to DoneBB. 1125 BuildMI(MBB, MBB.end(), DL, TII->get(ARM::tBcc)) 1126 .addMBB(DoneBB) 1127 .addImm(ARMCC::EQ) 1128 .addReg(ARM::CPSR, RegState::Kill); 1129 } 1130 1131 // Emit the clearing sequence 1132 for (unsigned D = 0; D < 8; D++) { 1133 // Attempt to clear as double 1134 if (ClearRegs[D * 2 + 0] && ClearRegs[D * 2 + 1]) { 1135 unsigned Reg = ARM::D0 + D; 1136 BuildMI(ClearBB, DL, TII->get(ARM::VMOVDRR), Reg) 1137 .addReg(ARM::LR) 1138 .addReg(ARM::LR) 1139 .add(predOps(ARMCC::AL)); 1140 } else { 1141 // Clear first part as single 1142 if (ClearRegs[D * 2 + 0]) { 1143 unsigned Reg = ARM::S0 + D * 2; 1144 BuildMI(ClearBB, DL, TII->get(ARM::VMOVSR), Reg) 1145 .addReg(ARM::LR) 1146 .add(predOps(ARMCC::AL)); 1147 } 1148 // Clear second part as single 1149 if (ClearRegs[D * 2 + 1]) { 1150 unsigned Reg = ARM::S0 + D * 2 + 1; 1151 BuildMI(ClearBB, DL, TII->get(ARM::VMOVSR), Reg) 1152 .addReg(ARM::LR) 1153 .add(predOps(ARMCC::AL)); 1154 } 1155 } 1156 } 1157 1158 // Clear FPSCR bits 0-4, 7, 28-31 1159 // The other bits are program global according to the AAPCS 1160 BuildMI(ClearBB, DL, TII->get(ARM::VMRS), ARM::R12) 1161 .add(predOps(ARMCC::AL)); 1162 BuildMI(ClearBB, DL, TII->get(ARM::t2BICri), ARM::R12) 1163 .addReg(ARM::R12) 1164 .addImm(0x0000009F) 1165 .add(predOps(ARMCC::AL)) 1166 .add(condCodeOp()); 1167 BuildMI(ClearBB, DL, TII->get(ARM::t2BICri), ARM::R12) 1168 .addReg(ARM::R12) 1169 .addImm(0xF0000000) 1170 .add(predOps(ARMCC::AL)) 1171 .add(condCodeOp()); 1172 BuildMI(ClearBB, DL, TII->get(ARM::VMSR)) 1173 .addReg(ARM::R12) 1174 .add(predOps(ARMCC::AL)); 1175 1176 return *DoneBB; 1177 } 1178 1179 MachineBasicBlock & 1180 ARMExpandPseudo::CMSEClearFPRegsV81(MachineBasicBlock &MBB, 1181 MachineBasicBlock::iterator MBBI, 1182 const BitVector &ClearRegs) { 1183 auto &RetI = *MBBI; 1184 1185 // Emit a sequence of VSCCLRM <sreglist> instructions, one instruction for 1186 // each contiguous sequence of S-registers. 1187 int Start = -1, End = -1; 1188 for (int S = 0, E = ClearRegs.size(); S != E; ++S) { 1189 if (ClearRegs[S] && S == End + 1) { 1190 End = S; // extend range 1191 continue; 1192 } 1193 // Emit current range. 1194 if (Start < End) { 1195 MachineInstrBuilder VSCCLRM = 1196 BuildMI(MBB, MBBI, RetI.getDebugLoc(), TII->get(ARM::VSCCLRMS)) 1197 .add(predOps(ARMCC::AL)); 1198 while (++Start <= End) 1199 VSCCLRM.addReg(ARM::S0 + Start, RegState::Define); 1200 VSCCLRM.addReg(ARM::VPR, RegState::Define); 1201 } 1202 Start = End = S; 1203 } 1204 // Emit last range. 1205 if (Start < End) { 1206 MachineInstrBuilder VSCCLRM = 1207 BuildMI(MBB, MBBI, RetI.getDebugLoc(), TII->get(ARM::VSCCLRMS)) 1208 .add(predOps(ARMCC::AL)); 1209 while (++Start <= End) 1210 VSCCLRM.addReg(ARM::S0 + Start, RegState::Define); 1211 VSCCLRM.addReg(ARM::VPR, RegState::Define); 1212 } 1213 1214 return MBB; 1215 } 1216 1217 void ARMExpandPseudo::CMSESaveClearFPRegs( 1218 MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI, DebugLoc &DL, 1219 const LivePhysRegs &LiveRegs, SmallVectorImpl<unsigned> &ScratchRegs) { 1220 if (STI->hasV8_1MMainlineOps()) 1221 CMSESaveClearFPRegsV81(MBB, MBBI, DL, LiveRegs); 1222 else 1223 CMSESaveClearFPRegsV8(MBB, MBBI, DL, LiveRegs, ScratchRegs); 1224 } 1225 1226 // Save and clear FP registers if present 1227 void ARMExpandPseudo::CMSESaveClearFPRegsV8( 1228 MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI, DebugLoc &DL, 1229 const LivePhysRegs &LiveRegs, SmallVectorImpl<unsigned> &ScratchRegs) { 1230 if (!STI->hasFPRegs()) 1231 return; 1232 1233 // Store an available register for FPSCR clearing 1234 assert(!ScratchRegs.empty()); 1235 unsigned SpareReg = ScratchRegs.front(); 1236 1237 // save space on stack for VLSTM 1238 BuildMI(MBB, MBBI, DL, TII->get(ARM::tSUBspi), ARM::SP) 1239 .addReg(ARM::SP) 1240 .addImm(CMSE_FP_SAVE_SIZE >> 2) 1241 .add(predOps(ARMCC::AL)); 1242 1243 // Use ScratchRegs to store the fp regs 1244 std::vector<std::tuple<unsigned, unsigned, unsigned>> ClearedFPRegs; 1245 std::vector<unsigned> NonclearedFPRegs; 1246 for (const MachineOperand &Op : MBBI->operands()) { 1247 if (Op.isReg() && Op.isUse()) { 1248 unsigned Reg = Op.getReg(); 1249 assert(!ARM::DPRRegClass.contains(Reg) || 1250 ARM::DPR_VFP2RegClass.contains(Reg)); 1251 assert(!ARM::QPRRegClass.contains(Reg)); 1252 if (ARM::DPR_VFP2RegClass.contains(Reg)) { 1253 if (ScratchRegs.size() >= 2) { 1254 unsigned SaveReg2 = ScratchRegs.pop_back_val(); 1255 unsigned SaveReg1 = ScratchRegs.pop_back_val(); 1256 ClearedFPRegs.emplace_back(Reg, SaveReg1, SaveReg2); 1257 1258 // Save the fp register to the normal registers 1259 BuildMI(MBB, MBBI, DL, TII->get(ARM::VMOVRRD)) 1260 .addReg(SaveReg1, RegState::Define) 1261 .addReg(SaveReg2, RegState::Define) 1262 .addReg(Reg) 1263 .add(predOps(ARMCC::AL)); 1264 } else { 1265 NonclearedFPRegs.push_back(Reg); 1266 } 1267 } else if (ARM::SPRRegClass.contains(Reg)) { 1268 if (ScratchRegs.size() >= 1) { 1269 unsigned SaveReg = ScratchRegs.pop_back_val(); 1270 ClearedFPRegs.emplace_back(Reg, SaveReg, 0); 1271 1272 // Save the fp register to the normal registers 1273 BuildMI(MBB, MBBI, DL, TII->get(ARM::VMOVRS), SaveReg) 1274 .addReg(Reg) 1275 .add(predOps(ARMCC::AL)); 1276 } else { 1277 NonclearedFPRegs.push_back(Reg); 1278 } 1279 } 1280 } 1281 } 1282 1283 bool passesFPReg = (!NonclearedFPRegs.empty() || !ClearedFPRegs.empty()); 1284 1285 // Lazy store all fp registers to the stack 1286 MachineInstrBuilder VLSTM = BuildMI(MBB, MBBI, DL, TII->get(ARM::VLSTM)) 1287 .addReg(ARM::SP) 1288 .add(predOps(ARMCC::AL)); 1289 for (auto R : {ARM::VPR, ARM::FPSCR, ARM::FPSCR_NZCV, ARM::Q0, ARM::Q1, 1290 ARM::Q2, ARM::Q3, ARM::Q4, ARM::Q5, ARM::Q6, ARM::Q7}) 1291 VLSTM.addReg(R, RegState::Implicit | 1292 (LiveRegs.contains(R) ? 0 : RegState::Undef)); 1293 1294 // Restore all arguments 1295 for (const auto &Regs : ClearedFPRegs) { 1296 unsigned Reg, SaveReg1, SaveReg2; 1297 std::tie(Reg, SaveReg1, SaveReg2) = Regs; 1298 if (ARM::DPR_VFP2RegClass.contains(Reg)) 1299 BuildMI(MBB, MBBI, DL, TII->get(ARM::VMOVDRR), Reg) 1300 .addReg(SaveReg1) 1301 .addReg(SaveReg2) 1302 .add(predOps(ARMCC::AL)); 1303 else if (ARM::SPRRegClass.contains(Reg)) 1304 BuildMI(MBB, MBBI, DL, TII->get(ARM::VMOVSR), Reg) 1305 .addReg(SaveReg1) 1306 .add(predOps(ARMCC::AL)); 1307 } 1308 1309 for (unsigned Reg : NonclearedFPRegs) { 1310 if (ARM::DPR_VFP2RegClass.contains(Reg)) { 1311 if (STI->isLittle()) { 1312 BuildMI(MBB, MBBI, DL, TII->get(ARM::VLDRD), Reg) 1313 .addReg(ARM::SP) 1314 .addImm((Reg - ARM::D0) * 2) 1315 .add(predOps(ARMCC::AL)); 1316 } else { 1317 // For big-endian targets we need to load the two subregisters of Reg 1318 // manually because VLDRD would load them in wrong order 1319 unsigned SReg0 = TRI->getSubReg(Reg, ARM::ssub_0); 1320 BuildMI(MBB, MBBI, DL, TII->get(ARM::VLDRS), SReg0) 1321 .addReg(ARM::SP) 1322 .addImm((Reg - ARM::D0) * 2) 1323 .add(predOps(ARMCC::AL)); 1324 BuildMI(MBB, MBBI, DL, TII->get(ARM::VLDRS), SReg0 + 1) 1325 .addReg(ARM::SP) 1326 .addImm((Reg - ARM::D0) * 2 + 1) 1327 .add(predOps(ARMCC::AL)); 1328 } 1329 } else if (ARM::SPRRegClass.contains(Reg)) { 1330 BuildMI(MBB, MBBI, DL, TII->get(ARM::VLDRS), Reg) 1331 .addReg(ARM::SP) 1332 .addImm(Reg - ARM::S0) 1333 .add(predOps(ARMCC::AL)); 1334 } 1335 } 1336 // restore FPSCR from stack and clear bits 0-4, 7, 28-31 1337 // The other bits are program global according to the AAPCS 1338 if (passesFPReg) { 1339 BuildMI(MBB, MBBI, DL, TII->get(ARM::t2LDRi8), SpareReg) 1340 .addReg(ARM::SP) 1341 .addImm(0x40) 1342 .add(predOps(ARMCC::AL)); 1343 BuildMI(MBB, MBBI, DL, TII->get(ARM::t2BICri), SpareReg) 1344 .addReg(SpareReg) 1345 .addImm(0x0000009F) 1346 .add(predOps(ARMCC::AL)) 1347 .add(condCodeOp()); 1348 BuildMI(MBB, MBBI, DL, TII->get(ARM::t2BICri), SpareReg) 1349 .addReg(SpareReg) 1350 .addImm(0xF0000000) 1351 .add(predOps(ARMCC::AL)) 1352 .add(condCodeOp()); 1353 BuildMI(MBB, MBBI, DL, TII->get(ARM::VMSR)) 1354 .addReg(SpareReg) 1355 .add(predOps(ARMCC::AL)); 1356 // The ldr must happen after a floating point instruction. To prevent the 1357 // post-ra scheduler to mess with the order, we create a bundle. 1358 finalizeBundle(MBB, VLSTM->getIterator(), MBBI->getIterator()); 1359 } 1360 } 1361 1362 void ARMExpandPseudo::CMSESaveClearFPRegsV81(MachineBasicBlock &MBB, 1363 MachineBasicBlock::iterator MBBI, 1364 DebugLoc &DL, 1365 const LivePhysRegs &LiveRegs) { 1366 BitVector ClearRegs(32, true); 1367 bool DefFP = determineFPRegsToClear(*MBBI, ClearRegs); 1368 1369 // If the instruction does not write to a FP register and no elements were 1370 // removed from the set, then no FP registers were used to pass 1371 // arguments/returns. 1372 if (!DefFP && ClearRegs.count() == ClearRegs.size()) { 1373 // save space on stack for VLSTM 1374 BuildMI(MBB, MBBI, DL, TII->get(ARM::tSUBspi), ARM::SP) 1375 .addReg(ARM::SP) 1376 .addImm(CMSE_FP_SAVE_SIZE >> 2) 1377 .add(predOps(ARMCC::AL)); 1378 1379 // Lazy store all FP registers to the stack 1380 MachineInstrBuilder VLSTM = BuildMI(MBB, MBBI, DL, TII->get(ARM::VLSTM)) 1381 .addReg(ARM::SP) 1382 .add(predOps(ARMCC::AL)); 1383 for (auto R : {ARM::VPR, ARM::FPSCR, ARM::FPSCR_NZCV, ARM::Q0, ARM::Q1, 1384 ARM::Q2, ARM::Q3, ARM::Q4, ARM::Q5, ARM::Q6, ARM::Q7}) 1385 VLSTM.addReg(R, RegState::Implicit | 1386 (LiveRegs.contains(R) ? 0 : RegState::Undef)); 1387 } else { 1388 // Push all the callee-saved registers (s16-s31). 1389 MachineInstrBuilder VPUSH = 1390 BuildMI(MBB, MBBI, DL, TII->get(ARM::VSTMSDB_UPD), ARM::SP) 1391 .addReg(ARM::SP) 1392 .add(predOps(ARMCC::AL)); 1393 for (int Reg = ARM::S16; Reg <= ARM::S31; ++Reg) 1394 VPUSH.addReg(Reg); 1395 1396 // Clear FP registers with a VSCCLRM. 1397 (void)CMSEClearFPRegsV81(MBB, MBBI, ClearRegs); 1398 1399 // Save floating-point context. 1400 BuildMI(MBB, MBBI, DL, TII->get(ARM::VSTR_FPCXTS_pre), ARM::SP) 1401 .addReg(ARM::SP) 1402 .addImm(-8) 1403 .add(predOps(ARMCC::AL)); 1404 } 1405 } 1406 1407 // Restore FP registers if present 1408 void ARMExpandPseudo::CMSERestoreFPRegs( 1409 MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI, DebugLoc &DL, 1410 SmallVectorImpl<unsigned> &AvailableRegs) { 1411 if (STI->hasV8_1MMainlineOps()) 1412 CMSERestoreFPRegsV81(MBB, MBBI, DL, AvailableRegs); 1413 else 1414 CMSERestoreFPRegsV8(MBB, MBBI, DL, AvailableRegs); 1415 } 1416 1417 void ARMExpandPseudo::CMSERestoreFPRegsV8( 1418 MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI, DebugLoc &DL, 1419 SmallVectorImpl<unsigned> &AvailableRegs) { 1420 if (!STI->hasFPRegs()) 1421 return; 1422 1423 // Use AvailableRegs to store the fp regs 1424 std::vector<std::tuple<unsigned, unsigned, unsigned>> ClearedFPRegs; 1425 std::vector<unsigned> NonclearedFPRegs; 1426 for (const MachineOperand &Op : MBBI->operands()) { 1427 if (Op.isReg() && Op.isDef()) { 1428 unsigned Reg = Op.getReg(); 1429 assert(!ARM::DPRRegClass.contains(Reg) || 1430 ARM::DPR_VFP2RegClass.contains(Reg)); 1431 assert(!ARM::QPRRegClass.contains(Reg)); 1432 if (ARM::DPR_VFP2RegClass.contains(Reg)) { 1433 if (AvailableRegs.size() >= 2) { 1434 unsigned SaveReg2 = AvailableRegs.pop_back_val(); 1435 unsigned SaveReg1 = AvailableRegs.pop_back_val(); 1436 ClearedFPRegs.emplace_back(Reg, SaveReg1, SaveReg2); 1437 1438 // Save the fp register to the normal registers 1439 BuildMI(MBB, MBBI, DL, TII->get(ARM::VMOVRRD)) 1440 .addReg(SaveReg1, RegState::Define) 1441 .addReg(SaveReg2, RegState::Define) 1442 .addReg(Reg) 1443 .add(predOps(ARMCC::AL)); 1444 } else { 1445 NonclearedFPRegs.push_back(Reg); 1446 } 1447 } else if (ARM::SPRRegClass.contains(Reg)) { 1448 if (AvailableRegs.size() >= 1) { 1449 unsigned SaveReg = AvailableRegs.pop_back_val(); 1450 ClearedFPRegs.emplace_back(Reg, SaveReg, 0); 1451 1452 // Save the fp register to the normal registers 1453 BuildMI(MBB, MBBI, DL, TII->get(ARM::VMOVRS), SaveReg) 1454 .addReg(Reg) 1455 .add(predOps(ARMCC::AL)); 1456 } else { 1457 NonclearedFPRegs.push_back(Reg); 1458 } 1459 } 1460 } 1461 } 1462 1463 // Push FP regs that cannot be restored via normal registers on the stack 1464 for (unsigned Reg : NonclearedFPRegs) { 1465 if (ARM::DPR_VFP2RegClass.contains(Reg)) 1466 BuildMI(MBB, MBBI, DL, TII->get(ARM::VSTRD), Reg) 1467 .addReg(ARM::SP) 1468 .addImm((Reg - ARM::D0) * 2) 1469 .add(predOps(ARMCC::AL)); 1470 else if (ARM::SPRRegClass.contains(Reg)) 1471 BuildMI(MBB, MBBI, DL, TII->get(ARM::VSTRS), Reg) 1472 .addReg(ARM::SP) 1473 .addImm(Reg - ARM::S0) 1474 .add(predOps(ARMCC::AL)); 1475 } 1476 1477 // Lazy load fp regs from stack 1478 BuildMI(MBB, MBBI, DL, TII->get(ARM::VLLDM)) 1479 .addReg(ARM::SP) 1480 .add(predOps(ARMCC::AL)); 1481 1482 // Restore all FP registers via normal registers 1483 for (const auto &Regs : ClearedFPRegs) { 1484 unsigned Reg, SaveReg1, SaveReg2; 1485 std::tie(Reg, SaveReg1, SaveReg2) = Regs; 1486 if (ARM::DPR_VFP2RegClass.contains(Reg)) 1487 BuildMI(MBB, MBBI, DL, TII->get(ARM::VMOVDRR), Reg) 1488 .addReg(SaveReg1) 1489 .addReg(SaveReg2) 1490 .add(predOps(ARMCC::AL)); 1491 else if (ARM::SPRRegClass.contains(Reg)) 1492 BuildMI(MBB, MBBI, DL, TII->get(ARM::VMOVSR), Reg) 1493 .addReg(SaveReg1) 1494 .add(predOps(ARMCC::AL)); 1495 } 1496 1497 // Pop the stack space 1498 BuildMI(MBB, MBBI, DL, TII->get(ARM::tADDspi), ARM::SP) 1499 .addReg(ARM::SP) 1500 .addImm(CMSE_FP_SAVE_SIZE >> 2) 1501 .add(predOps(ARMCC::AL)); 1502 } 1503 1504 static bool definesOrUsesFPReg(const MachineInstr &MI) { 1505 for (const MachineOperand &Op : MI.operands()) { 1506 if (!Op.isReg()) 1507 continue; 1508 unsigned Reg = Op.getReg(); 1509 if ((Reg >= ARM::Q0 && Reg <= ARM::Q7) || 1510 (Reg >= ARM::D0 && Reg <= ARM::D15) || 1511 (Reg >= ARM::S0 && Reg <= ARM::S31)) 1512 return true; 1513 } 1514 return false; 1515 } 1516 1517 void ARMExpandPseudo::CMSERestoreFPRegsV81( 1518 MachineBasicBlock &MBB, MachineBasicBlock::iterator MBBI, DebugLoc &DL, 1519 SmallVectorImpl<unsigned> &AvailableRegs) { 1520 if (!definesOrUsesFPReg(*MBBI)) { 1521 // Load FP registers from stack. 1522 BuildMI(MBB, MBBI, DL, TII->get(ARM::VLLDM)) 1523 .addReg(ARM::SP) 1524 .add(predOps(ARMCC::AL)); 1525 1526 // Pop the stack space 1527 BuildMI(MBB, MBBI, DL, TII->get(ARM::tADDspi), ARM::SP) 1528 .addReg(ARM::SP) 1529 .addImm(CMSE_FP_SAVE_SIZE >> 2) 1530 .add(predOps(ARMCC::AL)); 1531 } else { 1532 // Restore the floating point context. 1533 BuildMI(MBB, MBBI, MBBI->getDebugLoc(), TII->get(ARM::VLDR_FPCXTS_post), 1534 ARM::SP) 1535 .addReg(ARM::SP) 1536 .addImm(8) 1537 .add(predOps(ARMCC::AL)); 1538 1539 // Pop all the callee-saved registers (s16-s31). 1540 MachineInstrBuilder VPOP = 1541 BuildMI(MBB, MBBI, DL, TII->get(ARM::VLDMSIA_UPD), ARM::SP) 1542 .addReg(ARM::SP) 1543 .add(predOps(ARMCC::AL)); 1544 for (int Reg = ARM::S16; Reg <= ARM::S31; ++Reg) 1545 VPOP.addReg(Reg, RegState::Define); 1546 } 1547 } 1548 1549 /// Expand a CMP_SWAP pseudo-inst to an ldrex/strex loop as simply as 1550 /// possible. This only gets used at -O0 so we don't care about efficiency of 1551 /// the generated code. 1552 bool ARMExpandPseudo::ExpandCMP_SWAP(MachineBasicBlock &MBB, 1553 MachineBasicBlock::iterator MBBI, 1554 unsigned LdrexOp, unsigned StrexOp, 1555 unsigned UxtOp, 1556 MachineBasicBlock::iterator &NextMBBI) { 1557 bool IsThumb = STI->isThumb(); 1558 MachineInstr &MI = *MBBI; 1559 DebugLoc DL = MI.getDebugLoc(); 1560 const MachineOperand &Dest = MI.getOperand(0); 1561 Register TempReg = MI.getOperand(1).getReg(); 1562 // Duplicating undef operands into 2 instructions does not guarantee the same 1563 // value on both; However undef should be replaced by xzr anyway. 1564 assert(!MI.getOperand(2).isUndef() && "cannot handle undef"); 1565 Register AddrReg = MI.getOperand(2).getReg(); 1566 Register DesiredReg = MI.getOperand(3).getReg(); 1567 Register NewReg = MI.getOperand(4).getReg(); 1568 1569 if (IsThumb) { 1570 assert(STI->hasV8MBaselineOps() && 1571 "CMP_SWAP not expected to be custom expanded for Thumb1"); 1572 assert((UxtOp == 0 || UxtOp == ARM::tUXTB || UxtOp == ARM::tUXTH) && 1573 "ARMv8-M.baseline does not have t2UXTB/t2UXTH"); 1574 assert(ARM::tGPRRegClass.contains(DesiredReg) && 1575 "DesiredReg used for UXT op must be tGPR"); 1576 } 1577 1578 MachineFunction *MF = MBB.getParent(); 1579 auto LoadCmpBB = MF->CreateMachineBasicBlock(MBB.getBasicBlock()); 1580 auto StoreBB = MF->CreateMachineBasicBlock(MBB.getBasicBlock()); 1581 auto DoneBB = MF->CreateMachineBasicBlock(MBB.getBasicBlock()); 1582 1583 MF->insert(++MBB.getIterator(), LoadCmpBB); 1584 MF->insert(++LoadCmpBB->getIterator(), StoreBB); 1585 MF->insert(++StoreBB->getIterator(), DoneBB); 1586 1587 if (UxtOp) { 1588 MachineInstrBuilder MIB = 1589 BuildMI(MBB, MBBI, DL, TII->get(UxtOp), DesiredReg) 1590 .addReg(DesiredReg, RegState::Kill); 1591 if (!IsThumb) 1592 MIB.addImm(0); 1593 MIB.add(predOps(ARMCC::AL)); 1594 } 1595 1596 // .Lloadcmp: 1597 // ldrex rDest, [rAddr] 1598 // cmp rDest, rDesired 1599 // bne .Ldone 1600 1601 MachineInstrBuilder MIB; 1602 MIB = BuildMI(LoadCmpBB, DL, TII->get(LdrexOp), Dest.getReg()); 1603 MIB.addReg(AddrReg); 1604 if (LdrexOp == ARM::t2LDREX) 1605 MIB.addImm(0); // a 32-bit Thumb ldrex (only) allows an offset. 1606 MIB.add(predOps(ARMCC::AL)); 1607 1608 unsigned CMPrr = IsThumb ? ARM::tCMPhir : ARM::CMPrr; 1609 BuildMI(LoadCmpBB, DL, TII->get(CMPrr)) 1610 .addReg(Dest.getReg(), getKillRegState(Dest.isDead())) 1611 .addReg(DesiredReg) 1612 .add(predOps(ARMCC::AL)); 1613 unsigned Bcc = IsThumb ? ARM::tBcc : ARM::Bcc; 1614 BuildMI(LoadCmpBB, DL, TII->get(Bcc)) 1615 .addMBB(DoneBB) 1616 .addImm(ARMCC::NE) 1617 .addReg(ARM::CPSR, RegState::Kill); 1618 LoadCmpBB->addSuccessor(DoneBB); 1619 LoadCmpBB->addSuccessor(StoreBB); 1620 1621 // .Lstore: 1622 // strex rTempReg, rNew, [rAddr] 1623 // cmp rTempReg, #0 1624 // bne .Lloadcmp 1625 MIB = BuildMI(StoreBB, DL, TII->get(StrexOp), TempReg) 1626 .addReg(NewReg) 1627 .addReg(AddrReg); 1628 if (StrexOp == ARM::t2STREX) 1629 MIB.addImm(0); // a 32-bit Thumb strex (only) allows an offset. 1630 MIB.add(predOps(ARMCC::AL)); 1631 1632 unsigned CMPri = IsThumb ? ARM::t2CMPri : ARM::CMPri; 1633 BuildMI(StoreBB, DL, TII->get(CMPri)) 1634 .addReg(TempReg, RegState::Kill) 1635 .addImm(0) 1636 .add(predOps(ARMCC::AL)); 1637 BuildMI(StoreBB, DL, TII->get(Bcc)) 1638 .addMBB(LoadCmpBB) 1639 .addImm(ARMCC::NE) 1640 .addReg(ARM::CPSR, RegState::Kill); 1641 StoreBB->addSuccessor(LoadCmpBB); 1642 StoreBB->addSuccessor(DoneBB); 1643 1644 DoneBB->splice(DoneBB->end(), &MBB, MI, MBB.end()); 1645 DoneBB->transferSuccessors(&MBB); 1646 1647 MBB.addSuccessor(LoadCmpBB); 1648 1649 NextMBBI = MBB.end(); 1650 MI.eraseFromParent(); 1651 1652 // Recompute livein lists. 1653 LivePhysRegs LiveRegs; 1654 computeAndAddLiveIns(LiveRegs, *DoneBB); 1655 computeAndAddLiveIns(LiveRegs, *StoreBB); 1656 computeAndAddLiveIns(LiveRegs, *LoadCmpBB); 1657 // Do an extra pass around the loop to get loop carried registers right. 1658 StoreBB->clearLiveIns(); 1659 computeAndAddLiveIns(LiveRegs, *StoreBB); 1660 LoadCmpBB->clearLiveIns(); 1661 computeAndAddLiveIns(LiveRegs, *LoadCmpBB); 1662 1663 return true; 1664 } 1665 1666 /// ARM's ldrexd/strexd take a consecutive register pair (represented as a 1667 /// single GPRPair register), Thumb's take two separate registers so we need to 1668 /// extract the subregs from the pair. 1669 static void addExclusiveRegPair(MachineInstrBuilder &MIB, MachineOperand &Reg, 1670 unsigned Flags, bool IsThumb, 1671 const TargetRegisterInfo *TRI) { 1672 if (IsThumb) { 1673 Register RegLo = TRI->getSubReg(Reg.getReg(), ARM::gsub_0); 1674 Register RegHi = TRI->getSubReg(Reg.getReg(), ARM::gsub_1); 1675 MIB.addReg(RegLo, Flags); 1676 MIB.addReg(RegHi, Flags); 1677 } else 1678 MIB.addReg(Reg.getReg(), Flags); 1679 } 1680 1681 /// Expand a 64-bit CMP_SWAP to an ldrexd/strexd loop. 1682 bool ARMExpandPseudo::ExpandCMP_SWAP_64(MachineBasicBlock &MBB, 1683 MachineBasicBlock::iterator MBBI, 1684 MachineBasicBlock::iterator &NextMBBI) { 1685 bool IsThumb = STI->isThumb(); 1686 MachineInstr &MI = *MBBI; 1687 DebugLoc DL = MI.getDebugLoc(); 1688 MachineOperand &Dest = MI.getOperand(0); 1689 Register TempReg = MI.getOperand(1).getReg(); 1690 // Duplicating undef operands into 2 instructions does not guarantee the same 1691 // value on both; However undef should be replaced by xzr anyway. 1692 assert(!MI.getOperand(2).isUndef() && "cannot handle undef"); 1693 Register AddrReg = MI.getOperand(2).getReg(); 1694 Register DesiredReg = MI.getOperand(3).getReg(); 1695 MachineOperand New = MI.getOperand(4); 1696 New.setIsKill(false); 1697 1698 Register DestLo = TRI->getSubReg(Dest.getReg(), ARM::gsub_0); 1699 Register DestHi = TRI->getSubReg(Dest.getReg(), ARM::gsub_1); 1700 Register DesiredLo = TRI->getSubReg(DesiredReg, ARM::gsub_0); 1701 Register DesiredHi = TRI->getSubReg(DesiredReg, ARM::gsub_1); 1702 1703 MachineFunction *MF = MBB.getParent(); 1704 auto LoadCmpBB = MF->CreateMachineBasicBlock(MBB.getBasicBlock()); 1705 auto StoreBB = MF->CreateMachineBasicBlock(MBB.getBasicBlock()); 1706 auto DoneBB = MF->CreateMachineBasicBlock(MBB.getBasicBlock()); 1707 1708 MF->insert(++MBB.getIterator(), LoadCmpBB); 1709 MF->insert(++LoadCmpBB->getIterator(), StoreBB); 1710 MF->insert(++StoreBB->getIterator(), DoneBB); 1711 1712 // .Lloadcmp: 1713 // ldrexd rDestLo, rDestHi, [rAddr] 1714 // cmp rDestLo, rDesiredLo 1715 // sbcs dead rTempReg, rDestHi, rDesiredHi 1716 // bne .Ldone 1717 unsigned LDREXD = IsThumb ? ARM::t2LDREXD : ARM::LDREXD; 1718 MachineInstrBuilder MIB; 1719 MIB = BuildMI(LoadCmpBB, DL, TII->get(LDREXD)); 1720 addExclusiveRegPair(MIB, Dest, RegState::Define, IsThumb, TRI); 1721 MIB.addReg(AddrReg).add(predOps(ARMCC::AL)); 1722 1723 unsigned CMPrr = IsThumb ? ARM::tCMPhir : ARM::CMPrr; 1724 BuildMI(LoadCmpBB, DL, TII->get(CMPrr)) 1725 .addReg(DestLo, getKillRegState(Dest.isDead())) 1726 .addReg(DesiredLo) 1727 .add(predOps(ARMCC::AL)); 1728 1729 BuildMI(LoadCmpBB, DL, TII->get(CMPrr)) 1730 .addReg(DestHi, getKillRegState(Dest.isDead())) 1731 .addReg(DesiredHi) 1732 .addImm(ARMCC::EQ).addReg(ARM::CPSR, RegState::Kill); 1733 1734 unsigned Bcc = IsThumb ? ARM::tBcc : ARM::Bcc; 1735 BuildMI(LoadCmpBB, DL, TII->get(Bcc)) 1736 .addMBB(DoneBB) 1737 .addImm(ARMCC::NE) 1738 .addReg(ARM::CPSR, RegState::Kill); 1739 LoadCmpBB->addSuccessor(DoneBB); 1740 LoadCmpBB->addSuccessor(StoreBB); 1741 1742 // .Lstore: 1743 // strexd rTempReg, rNewLo, rNewHi, [rAddr] 1744 // cmp rTempReg, #0 1745 // bne .Lloadcmp 1746 unsigned STREXD = IsThumb ? ARM::t2STREXD : ARM::STREXD; 1747 MIB = BuildMI(StoreBB, DL, TII->get(STREXD), TempReg); 1748 unsigned Flags = getKillRegState(New.isDead()); 1749 addExclusiveRegPair(MIB, New, Flags, IsThumb, TRI); 1750 MIB.addReg(AddrReg).add(predOps(ARMCC::AL)); 1751 1752 unsigned CMPri = IsThumb ? ARM::t2CMPri : ARM::CMPri; 1753 BuildMI(StoreBB, DL, TII->get(CMPri)) 1754 .addReg(TempReg, RegState::Kill) 1755 .addImm(0) 1756 .add(predOps(ARMCC::AL)); 1757 BuildMI(StoreBB, DL, TII->get(Bcc)) 1758 .addMBB(LoadCmpBB) 1759 .addImm(ARMCC::NE) 1760 .addReg(ARM::CPSR, RegState::Kill); 1761 StoreBB->addSuccessor(LoadCmpBB); 1762 StoreBB->addSuccessor(DoneBB); 1763 1764 DoneBB->splice(DoneBB->end(), &MBB, MI, MBB.end()); 1765 DoneBB->transferSuccessors(&MBB); 1766 1767 MBB.addSuccessor(LoadCmpBB); 1768 1769 NextMBBI = MBB.end(); 1770 MI.eraseFromParent(); 1771 1772 // Recompute livein lists. 1773 LivePhysRegs LiveRegs; 1774 computeAndAddLiveIns(LiveRegs, *DoneBB); 1775 computeAndAddLiveIns(LiveRegs, *StoreBB); 1776 computeAndAddLiveIns(LiveRegs, *LoadCmpBB); 1777 // Do an extra pass around the loop to get loop carried registers right. 1778 StoreBB->clearLiveIns(); 1779 computeAndAddLiveIns(LiveRegs, *StoreBB); 1780 LoadCmpBB->clearLiveIns(); 1781 computeAndAddLiveIns(LiveRegs, *LoadCmpBB); 1782 1783 return true; 1784 } 1785 1786 static void CMSEPushCalleeSaves(const TargetInstrInfo &TII, 1787 MachineBasicBlock &MBB, 1788 MachineBasicBlock::iterator MBBI, int JumpReg, 1789 const LivePhysRegs &LiveRegs, bool Thumb1Only) { 1790 const DebugLoc &DL = MBBI->getDebugLoc(); 1791 if (Thumb1Only) { // push Lo and Hi regs separately 1792 MachineInstrBuilder PushMIB = 1793 BuildMI(MBB, MBBI, DL, TII.get(ARM::tPUSH)).add(predOps(ARMCC::AL)); 1794 for (int Reg = ARM::R4; Reg < ARM::R8; ++Reg) { 1795 PushMIB.addReg( 1796 Reg, Reg == JumpReg || LiveRegs.contains(Reg) ? 0 : RegState::Undef); 1797 } 1798 1799 // Thumb1 can only tPUSH low regs, so we copy the high regs to the low 1800 // regs that we just saved and push the low regs again, taking care to 1801 // not clobber JumpReg. If JumpReg is one of the low registers, push first 1802 // the values of r9-r11, and then r8. That would leave them ordered in 1803 // memory, and allow us to later pop them with a single instructions. 1804 // FIXME: Could also use any of r0-r3 that are free (including in the 1805 // first PUSH above). 1806 for (int LoReg = ARM::R7, HiReg = ARM::R11; LoReg >= ARM::R4; --LoReg) { 1807 if (JumpReg == LoReg) 1808 continue; 1809 BuildMI(MBB, MBBI, DL, TII.get(ARM::tMOVr), LoReg) 1810 .addReg(HiReg, LiveRegs.contains(HiReg) ? 0 : RegState::Undef) 1811 .add(predOps(ARMCC::AL)); 1812 --HiReg; 1813 } 1814 MachineInstrBuilder PushMIB2 = 1815 BuildMI(MBB, MBBI, DL, TII.get(ARM::tPUSH)).add(predOps(ARMCC::AL)); 1816 for (int Reg = ARM::R4; Reg < ARM::R8; ++Reg) { 1817 if (Reg == JumpReg) 1818 continue; 1819 PushMIB2.addReg(Reg, RegState::Kill); 1820 } 1821 1822 // If we couldn't use a low register for temporary storage (because it was 1823 // the JumpReg), use r4 or r5, whichever is not JumpReg. It has already been 1824 // saved. 1825 if (JumpReg >= ARM::R4 && JumpReg <= ARM::R7) { 1826 int LoReg = JumpReg == ARM::R4 ? ARM::R5 : ARM::R4; 1827 BuildMI(MBB, MBBI, DL, TII.get(ARM::tMOVr), LoReg) 1828 .addReg(ARM::R8, LiveRegs.contains(ARM::R8) ? 0 : RegState::Undef) 1829 .add(predOps(ARMCC::AL)); 1830 BuildMI(MBB, MBBI, DL, TII.get(ARM::tPUSH)) 1831 .add(predOps(ARMCC::AL)) 1832 .addReg(LoReg, RegState::Kill); 1833 } 1834 } else { // push Lo and Hi registers with a single instruction 1835 MachineInstrBuilder PushMIB = 1836 BuildMI(MBB, MBBI, DL, TII.get(ARM::t2STMDB_UPD), ARM::SP) 1837 .addReg(ARM::SP) 1838 .add(predOps(ARMCC::AL)); 1839 for (int Reg = ARM::R4; Reg < ARM::R12; ++Reg) { 1840 PushMIB.addReg( 1841 Reg, Reg == JumpReg || LiveRegs.contains(Reg) ? 0 : RegState::Undef); 1842 } 1843 } 1844 } 1845 1846 static void CMSEPopCalleeSaves(const TargetInstrInfo &TII, 1847 MachineBasicBlock &MBB, 1848 MachineBasicBlock::iterator MBBI, int JumpReg, 1849 bool Thumb1Only) { 1850 const DebugLoc &DL = MBBI->getDebugLoc(); 1851 if (Thumb1Only) { 1852 MachineInstrBuilder PopMIB = 1853 BuildMI(MBB, MBBI, DL, TII.get(ARM::tPOP)).add(predOps(ARMCC::AL)); 1854 for (int R = 0; R < 4; ++R) { 1855 PopMIB.addReg(ARM::R4 + R, RegState::Define); 1856 BuildMI(MBB, MBBI, DL, TII.get(ARM::tMOVr), ARM::R8 + R) 1857 .addReg(ARM::R4 + R, RegState::Kill) 1858 .add(predOps(ARMCC::AL)); 1859 } 1860 MachineInstrBuilder PopMIB2 = 1861 BuildMI(MBB, MBBI, DL, TII.get(ARM::tPOP)).add(predOps(ARMCC::AL)); 1862 for (int R = 0; R < 4; ++R) 1863 PopMIB2.addReg(ARM::R4 + R, RegState::Define); 1864 } else { // pop Lo and Hi registers with a single instruction 1865 MachineInstrBuilder PopMIB = 1866 BuildMI(MBB, MBBI, DL, TII.get(ARM::t2LDMIA_UPD), ARM::SP) 1867 .addReg(ARM::SP) 1868 .add(predOps(ARMCC::AL)); 1869 for (int Reg = ARM::R4; Reg < ARM::R12; ++Reg) 1870 PopMIB.addReg(Reg, RegState::Define); 1871 } 1872 } 1873 1874 bool ARMExpandPseudo::ExpandMI(MachineBasicBlock &MBB, 1875 MachineBasicBlock::iterator MBBI, 1876 MachineBasicBlock::iterator &NextMBBI) { 1877 MachineInstr &MI = *MBBI; 1878 unsigned Opcode = MI.getOpcode(); 1879 switch (Opcode) { 1880 default: 1881 return false; 1882 1883 case ARM::VBSPd: 1884 case ARM::VBSPq: { 1885 Register DstReg = MI.getOperand(0).getReg(); 1886 if (DstReg == MI.getOperand(3).getReg()) { 1887 // Expand to VBIT 1888 unsigned NewOpc = Opcode == ARM::VBSPd ? ARM::VBITd : ARM::VBITq; 1889 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(NewOpc)) 1890 .add(MI.getOperand(0)) 1891 .add(MI.getOperand(3)) 1892 .add(MI.getOperand(2)) 1893 .add(MI.getOperand(1)) 1894 .addImm(MI.getOperand(4).getImm()) 1895 .add(MI.getOperand(5)); 1896 } else if (DstReg == MI.getOperand(2).getReg()) { 1897 // Expand to VBIF 1898 unsigned NewOpc = Opcode == ARM::VBSPd ? ARM::VBIFd : ARM::VBIFq; 1899 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(NewOpc)) 1900 .add(MI.getOperand(0)) 1901 .add(MI.getOperand(2)) 1902 .add(MI.getOperand(3)) 1903 .add(MI.getOperand(1)) 1904 .addImm(MI.getOperand(4).getImm()) 1905 .add(MI.getOperand(5)); 1906 } else { 1907 // Expand to VBSL 1908 unsigned NewOpc = Opcode == ARM::VBSPd ? ARM::VBSLd : ARM::VBSLq; 1909 if (DstReg == MI.getOperand(1).getReg()) { 1910 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(NewOpc)) 1911 .add(MI.getOperand(0)) 1912 .add(MI.getOperand(1)) 1913 .add(MI.getOperand(2)) 1914 .add(MI.getOperand(3)) 1915 .addImm(MI.getOperand(4).getImm()) 1916 .add(MI.getOperand(5)); 1917 } else { 1918 // Use move to satisfy constraints 1919 unsigned MoveOpc = Opcode == ARM::VBSPd ? ARM::VORRd : ARM::VORRq; 1920 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(MoveOpc)) 1921 .addReg(DstReg, 1922 RegState::Define | 1923 getRenamableRegState(MI.getOperand(0).isRenamable())) 1924 .add(MI.getOperand(1)) 1925 .add(MI.getOperand(1)) 1926 .addImm(MI.getOperand(4).getImm()) 1927 .add(MI.getOperand(5)); 1928 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(NewOpc)) 1929 .add(MI.getOperand(0)) 1930 .addReg(DstReg, 1931 RegState::Kill | 1932 getRenamableRegState(MI.getOperand(0).isRenamable())) 1933 .add(MI.getOperand(2)) 1934 .add(MI.getOperand(3)) 1935 .addImm(MI.getOperand(4).getImm()) 1936 .add(MI.getOperand(5)); 1937 } 1938 } 1939 MI.eraseFromParent(); 1940 return true; 1941 } 1942 1943 case ARM::TCRETURNdi: 1944 case ARM::TCRETURNri: { 1945 MachineBasicBlock::iterator MBBI = MBB.getLastNonDebugInstr(); 1946 assert(MBBI->isReturn() && 1947 "Can only insert epilog into returning blocks"); 1948 unsigned RetOpcode = MBBI->getOpcode(); 1949 DebugLoc dl = MBBI->getDebugLoc(); 1950 const ARMBaseInstrInfo &TII = *static_cast<const ARMBaseInstrInfo *>( 1951 MBB.getParent()->getSubtarget().getInstrInfo()); 1952 1953 // Tail call return: adjust the stack pointer and jump to callee. 1954 MBBI = MBB.getLastNonDebugInstr(); 1955 MachineOperand &JumpTarget = MBBI->getOperand(0); 1956 1957 // Jump to label or value in register. 1958 if (RetOpcode == ARM::TCRETURNdi) { 1959 unsigned TCOpcode = 1960 STI->isThumb() 1961 ? (STI->isTargetMachO() ? ARM::tTAILJMPd : ARM::tTAILJMPdND) 1962 : ARM::TAILJMPd; 1963 MachineInstrBuilder MIB = BuildMI(MBB, MBBI, dl, TII.get(TCOpcode)); 1964 if (JumpTarget.isGlobal()) 1965 MIB.addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset(), 1966 JumpTarget.getTargetFlags()); 1967 else { 1968 assert(JumpTarget.isSymbol()); 1969 MIB.addExternalSymbol(JumpTarget.getSymbolName(), 1970 JumpTarget.getTargetFlags()); 1971 } 1972 1973 // Add the default predicate in Thumb mode. 1974 if (STI->isThumb()) 1975 MIB.add(predOps(ARMCC::AL)); 1976 } else if (RetOpcode == ARM::TCRETURNri) { 1977 unsigned Opcode = 1978 STI->isThumb() ? ARM::tTAILJMPr 1979 : (STI->hasV4TOps() ? ARM::TAILJMPr : ARM::TAILJMPr4); 1980 BuildMI(MBB, MBBI, dl, 1981 TII.get(Opcode)) 1982 .addReg(JumpTarget.getReg(), RegState::Kill); 1983 } 1984 1985 auto NewMI = std::prev(MBBI); 1986 for (unsigned i = 1, e = MBBI->getNumOperands(); i != e; ++i) 1987 NewMI->addOperand(MBBI->getOperand(i)); 1988 1989 1990 // Update call site info and delete the pseudo instruction TCRETURN. 1991 if (MI.isCandidateForCallSiteEntry()) 1992 MI.getMF()->moveCallSiteInfo(&MI, &*NewMI); 1993 MBB.erase(MBBI); 1994 1995 MBBI = NewMI; 1996 return true; 1997 } 1998 case ARM::tBXNS_RET: { 1999 MachineBasicBlock &AfterBB = CMSEClearFPRegs(MBB, MBBI); 2000 2001 if (STI->hasV8_1MMainlineOps()) { 2002 // Restore the non-secure floating point context. 2003 BuildMI(MBB, MBBI, MBBI->getDebugLoc(), 2004 TII->get(ARM::VLDR_FPCXTNS_post), ARM::SP) 2005 .addReg(ARM::SP) 2006 .addImm(4) 2007 .add(predOps(ARMCC::AL)); 2008 } 2009 2010 // Clear all GPR that are not a use of the return instruction. 2011 assert(llvm::all_of(MBBI->operands(), [](const MachineOperand &Op) { 2012 return !Op.isReg() || Op.getReg() != ARM::R12; 2013 })); 2014 SmallVector<unsigned, 5> ClearRegs; 2015 determineGPRegsToClear( 2016 *MBBI, {ARM::R0, ARM::R1, ARM::R2, ARM::R3, ARM::R12}, ClearRegs); 2017 CMSEClearGPRegs(AfterBB, AfterBB.end(), MBBI->getDebugLoc(), ClearRegs, 2018 ARM::LR); 2019 2020 MachineInstrBuilder NewMI = 2021 BuildMI(AfterBB, AfterBB.end(), MBBI->getDebugLoc(), 2022 TII->get(ARM::tBXNS)) 2023 .addReg(ARM::LR) 2024 .add(predOps(ARMCC::AL)); 2025 for (const MachineOperand &Op : MI.operands()) 2026 NewMI->addOperand(Op); 2027 MI.eraseFromParent(); 2028 return true; 2029 } 2030 case ARM::tBLXNS_CALL: { 2031 DebugLoc DL = MBBI->getDebugLoc(); 2032 unsigned JumpReg = MBBI->getOperand(0).getReg(); 2033 2034 // Figure out which registers are live at the point immediately before the 2035 // call. When we indiscriminately push a set of registers, the live 2036 // registers are added as ordinary use operands, whereas dead registers 2037 // are "undef". 2038 LivePhysRegs LiveRegs(*TRI); 2039 LiveRegs.addLiveOuts(MBB); 2040 for (const MachineInstr &MI : make_range(MBB.rbegin(), MBBI.getReverse())) 2041 LiveRegs.stepBackward(MI); 2042 LiveRegs.stepBackward(*MBBI); 2043 2044 CMSEPushCalleeSaves(*TII, MBB, MBBI, JumpReg, LiveRegs, 2045 AFI->isThumb1OnlyFunction()); 2046 2047 SmallVector<unsigned, 16> ClearRegs; 2048 determineGPRegsToClear(*MBBI, 2049 {ARM::R0, ARM::R1, ARM::R2, ARM::R3, ARM::R4, 2050 ARM::R5, ARM::R6, ARM::R7, ARM::R8, ARM::R9, 2051 ARM::R10, ARM::R11, ARM::R12}, 2052 ClearRegs); 2053 auto OriginalClearRegs = ClearRegs; 2054 2055 // Get the first cleared register as a scratch (to use later with tBIC). 2056 // We need to use the first so we can ensure it is a low register. 2057 unsigned ScratchReg = ClearRegs.front(); 2058 2059 // Clear LSB of JumpReg 2060 if (AFI->isThumb2Function()) { 2061 BuildMI(MBB, MBBI, DL, TII->get(ARM::t2BICri), JumpReg) 2062 .addReg(JumpReg) 2063 .addImm(1) 2064 .add(predOps(ARMCC::AL)) 2065 .add(condCodeOp()); 2066 } else { 2067 // We need to use an extra register to cope with 8M Baseline, 2068 // since we have saved all of the registers we are ok to trash a non 2069 // argument register here. 2070 BuildMI(MBB, MBBI, DL, TII->get(ARM::tMOVi8), ScratchReg) 2071 .add(condCodeOp()) 2072 .addImm(1) 2073 .add(predOps(ARMCC::AL)); 2074 BuildMI(MBB, MBBI, DL, TII->get(ARM::tBIC), JumpReg) 2075 .addReg(ARM::CPSR, RegState::Define) 2076 .addReg(JumpReg) 2077 .addReg(ScratchReg) 2078 .add(predOps(ARMCC::AL)); 2079 } 2080 2081 CMSESaveClearFPRegs(MBB, MBBI, DL, LiveRegs, 2082 ClearRegs); // save+clear FP regs with ClearRegs 2083 CMSEClearGPRegs(MBB, MBBI, DL, ClearRegs, JumpReg); 2084 2085 const MachineInstrBuilder NewCall = 2086 BuildMI(MBB, MBBI, DL, TII->get(ARM::tBLXNSr)) 2087 .add(predOps(ARMCC::AL)) 2088 .addReg(JumpReg, RegState::Kill); 2089 2090 for (int I = 1, E = MI.getNumOperands(); I != E; ++I) 2091 NewCall->addOperand(MI.getOperand(I)); 2092 if (MI.isCandidateForCallSiteEntry()) 2093 MI.getMF()->moveCallSiteInfo(&MI, NewCall.getInstr()); 2094 2095 CMSERestoreFPRegs(MBB, MBBI, DL, OriginalClearRegs); // restore FP registers 2096 2097 CMSEPopCalleeSaves(*TII, MBB, MBBI, JumpReg, AFI->isThumb1OnlyFunction()); 2098 2099 MI.eraseFromParent(); 2100 return true; 2101 } 2102 case ARM::VMOVHcc: 2103 case ARM::VMOVScc: 2104 case ARM::VMOVDcc: { 2105 unsigned newOpc = Opcode != ARM::VMOVDcc ? ARM::VMOVS : ARM::VMOVD; 2106 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(newOpc), 2107 MI.getOperand(1).getReg()) 2108 .add(MI.getOperand(2)) 2109 .addImm(MI.getOperand(3).getImm()) // 'pred' 2110 .add(MI.getOperand(4)) 2111 .add(makeImplicit(MI.getOperand(1))); 2112 2113 MI.eraseFromParent(); 2114 return true; 2115 } 2116 case ARM::t2MOVCCr: 2117 case ARM::MOVCCr: { 2118 unsigned Opc = AFI->isThumbFunction() ? ARM::t2MOVr : ARM::MOVr; 2119 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(Opc), 2120 MI.getOperand(1).getReg()) 2121 .add(MI.getOperand(2)) 2122 .addImm(MI.getOperand(3).getImm()) // 'pred' 2123 .add(MI.getOperand(4)) 2124 .add(condCodeOp()) // 's' bit 2125 .add(makeImplicit(MI.getOperand(1))); 2126 2127 MI.eraseFromParent(); 2128 return true; 2129 } 2130 case ARM::MOVCCsi: { 2131 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVsi), 2132 (MI.getOperand(1).getReg())) 2133 .add(MI.getOperand(2)) 2134 .addImm(MI.getOperand(3).getImm()) 2135 .addImm(MI.getOperand(4).getImm()) // 'pred' 2136 .add(MI.getOperand(5)) 2137 .add(condCodeOp()) // 's' bit 2138 .add(makeImplicit(MI.getOperand(1))); 2139 2140 MI.eraseFromParent(); 2141 return true; 2142 } 2143 case ARM::MOVCCsr: { 2144 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVsr), 2145 (MI.getOperand(1).getReg())) 2146 .add(MI.getOperand(2)) 2147 .add(MI.getOperand(3)) 2148 .addImm(MI.getOperand(4).getImm()) 2149 .addImm(MI.getOperand(5).getImm()) // 'pred' 2150 .add(MI.getOperand(6)) 2151 .add(condCodeOp()) // 's' bit 2152 .add(makeImplicit(MI.getOperand(1))); 2153 2154 MI.eraseFromParent(); 2155 return true; 2156 } 2157 case ARM::t2MOVCCi16: 2158 case ARM::MOVCCi16: { 2159 unsigned NewOpc = AFI->isThumbFunction() ? ARM::t2MOVi16 : ARM::MOVi16; 2160 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(NewOpc), 2161 MI.getOperand(1).getReg()) 2162 .addImm(MI.getOperand(2).getImm()) 2163 .addImm(MI.getOperand(3).getImm()) // 'pred' 2164 .add(MI.getOperand(4)) 2165 .add(makeImplicit(MI.getOperand(1))); 2166 MI.eraseFromParent(); 2167 return true; 2168 } 2169 case ARM::t2MOVCCi: 2170 case ARM::MOVCCi: { 2171 unsigned Opc = AFI->isThumbFunction() ? ARM::t2MOVi : ARM::MOVi; 2172 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(Opc), 2173 MI.getOperand(1).getReg()) 2174 .addImm(MI.getOperand(2).getImm()) 2175 .addImm(MI.getOperand(3).getImm()) // 'pred' 2176 .add(MI.getOperand(4)) 2177 .add(condCodeOp()) // 's' bit 2178 .add(makeImplicit(MI.getOperand(1))); 2179 2180 MI.eraseFromParent(); 2181 return true; 2182 } 2183 case ARM::t2MVNCCi: 2184 case ARM::MVNCCi: { 2185 unsigned Opc = AFI->isThumbFunction() ? ARM::t2MVNi : ARM::MVNi; 2186 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(Opc), 2187 MI.getOperand(1).getReg()) 2188 .addImm(MI.getOperand(2).getImm()) 2189 .addImm(MI.getOperand(3).getImm()) // 'pred' 2190 .add(MI.getOperand(4)) 2191 .add(condCodeOp()) // 's' bit 2192 .add(makeImplicit(MI.getOperand(1))); 2193 2194 MI.eraseFromParent(); 2195 return true; 2196 } 2197 case ARM::t2MOVCClsl: 2198 case ARM::t2MOVCClsr: 2199 case ARM::t2MOVCCasr: 2200 case ARM::t2MOVCCror: { 2201 unsigned NewOpc; 2202 switch (Opcode) { 2203 case ARM::t2MOVCClsl: NewOpc = ARM::t2LSLri; break; 2204 case ARM::t2MOVCClsr: NewOpc = ARM::t2LSRri; break; 2205 case ARM::t2MOVCCasr: NewOpc = ARM::t2ASRri; break; 2206 case ARM::t2MOVCCror: NewOpc = ARM::t2RORri; break; 2207 default: llvm_unreachable("unexpeced conditional move"); 2208 } 2209 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(NewOpc), 2210 MI.getOperand(1).getReg()) 2211 .add(MI.getOperand(2)) 2212 .addImm(MI.getOperand(3).getImm()) 2213 .addImm(MI.getOperand(4).getImm()) // 'pred' 2214 .add(MI.getOperand(5)) 2215 .add(condCodeOp()) // 's' bit 2216 .add(makeImplicit(MI.getOperand(1))); 2217 MI.eraseFromParent(); 2218 return true; 2219 } 2220 case ARM::Int_eh_sjlj_dispatchsetup: { 2221 MachineFunction &MF = *MI.getParent()->getParent(); 2222 const ARMBaseInstrInfo *AII = 2223 static_cast<const ARMBaseInstrInfo*>(TII); 2224 const ARMBaseRegisterInfo &RI = AII->getRegisterInfo(); 2225 // For functions using a base pointer, we rematerialize it (via the frame 2226 // pointer) here since eh.sjlj.setjmp and eh.sjlj.longjmp don't do it 2227 // for us. Otherwise, expand to nothing. 2228 if (RI.hasBasePointer(MF)) { 2229 int32_t NumBytes = AFI->getFramePtrSpillOffset(); 2230 Register FramePtr = RI.getFrameRegister(MF); 2231 assert(MF.getSubtarget().getFrameLowering()->hasFP(MF) && 2232 "base pointer without frame pointer?"); 2233 2234 if (AFI->isThumb2Function()) { 2235 emitT2RegPlusImmediate(MBB, MBBI, MI.getDebugLoc(), ARM::R6, 2236 FramePtr, -NumBytes, ARMCC::AL, 0, *TII); 2237 } else if (AFI->isThumbFunction()) { 2238 emitThumbRegPlusImmediate(MBB, MBBI, MI.getDebugLoc(), ARM::R6, 2239 FramePtr, -NumBytes, *TII, RI); 2240 } else { 2241 emitARMRegPlusImmediate(MBB, MBBI, MI.getDebugLoc(), ARM::R6, 2242 FramePtr, -NumBytes, ARMCC::AL, 0, 2243 *TII); 2244 } 2245 // If there's dynamic realignment, adjust for it. 2246 if (RI.hasStackRealignment(MF)) { 2247 MachineFrameInfo &MFI = MF.getFrameInfo(); 2248 Align MaxAlign = MFI.getMaxAlign(); 2249 assert (!AFI->isThumb1OnlyFunction()); 2250 // Emit bic r6, r6, MaxAlign 2251 assert(MaxAlign <= Align(256) && 2252 "The BIC instruction cannot encode " 2253 "immediates larger than 256 with all lower " 2254 "bits set."); 2255 unsigned bicOpc = AFI->isThumbFunction() ? 2256 ARM::t2BICri : ARM::BICri; 2257 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(bicOpc), ARM::R6) 2258 .addReg(ARM::R6, RegState::Kill) 2259 .addImm(MaxAlign.value() - 1) 2260 .add(predOps(ARMCC::AL)) 2261 .add(condCodeOp()); 2262 } 2263 } 2264 MI.eraseFromParent(); 2265 return true; 2266 } 2267 2268 case ARM::MOVsrl_flag: 2269 case ARM::MOVsra_flag: { 2270 // These are just fancy MOVs instructions. 2271 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVsi), 2272 MI.getOperand(0).getReg()) 2273 .add(MI.getOperand(1)) 2274 .addImm(ARM_AM::getSORegOpc( 2275 (Opcode == ARM::MOVsrl_flag ? ARM_AM::lsr : ARM_AM::asr), 1)) 2276 .add(predOps(ARMCC::AL)) 2277 .addReg(ARM::CPSR, RegState::Define); 2278 MI.eraseFromParent(); 2279 return true; 2280 } 2281 case ARM::RRX: { 2282 // This encodes as "MOVs Rd, Rm, rrx 2283 MachineInstrBuilder MIB = 2284 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::MOVsi), 2285 MI.getOperand(0).getReg()) 2286 .add(MI.getOperand(1)) 2287 .addImm(ARM_AM::getSORegOpc(ARM_AM::rrx, 0)) 2288 .add(predOps(ARMCC::AL)) 2289 .add(condCodeOp()); 2290 TransferImpOps(MI, MIB, MIB); 2291 MI.eraseFromParent(); 2292 return true; 2293 } 2294 case ARM::tTPsoft: 2295 case ARM::TPsoft: { 2296 const bool Thumb = Opcode == ARM::tTPsoft; 2297 2298 MachineInstrBuilder MIB; 2299 MachineFunction *MF = MBB.getParent(); 2300 if (STI->genLongCalls()) { 2301 MachineConstantPool *MCP = MF->getConstantPool(); 2302 unsigned PCLabelID = AFI->createPICLabelUId(); 2303 MachineConstantPoolValue *CPV = 2304 ARMConstantPoolSymbol::Create(MF->getFunction().getContext(), 2305 "__aeabi_read_tp", PCLabelID, 0); 2306 Register Reg = MI.getOperand(0).getReg(); 2307 MIB = 2308 BuildMI(MBB, MBBI, MI.getDebugLoc(), 2309 TII->get(Thumb ? ARM::tLDRpci : ARM::LDRi12), Reg) 2310 .addConstantPoolIndex(MCP->getConstantPoolIndex(CPV, Align(4))); 2311 if (!Thumb) 2312 MIB.addImm(0); 2313 MIB.add(predOps(ARMCC::AL)); 2314 2315 MIB = 2316 BuildMI(MBB, MBBI, MI.getDebugLoc(), 2317 TII->get(Thumb ? gettBLXrOpcode(*MF) : getBLXOpcode(*MF))); 2318 if (Thumb) 2319 MIB.add(predOps(ARMCC::AL)); 2320 MIB.addReg(Reg, RegState::Kill); 2321 } else { 2322 MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(), 2323 TII->get(Thumb ? ARM::tBL : ARM::BL)); 2324 if (Thumb) 2325 MIB.add(predOps(ARMCC::AL)); 2326 MIB.addExternalSymbol("__aeabi_read_tp", 0); 2327 } 2328 2329 MIB.cloneMemRefs(MI); 2330 TransferImpOps(MI, MIB, MIB); 2331 // Update the call site info. 2332 if (MI.isCandidateForCallSiteEntry()) 2333 MF->moveCallSiteInfo(&MI, &*MIB); 2334 MI.eraseFromParent(); 2335 return true; 2336 } 2337 case ARM::tLDRpci_pic: 2338 case ARM::t2LDRpci_pic: { 2339 unsigned NewLdOpc = (Opcode == ARM::tLDRpci_pic) 2340 ? ARM::tLDRpci : ARM::t2LDRpci; 2341 Register DstReg = MI.getOperand(0).getReg(); 2342 bool DstIsDead = MI.getOperand(0).isDead(); 2343 MachineInstrBuilder MIB1 = 2344 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(NewLdOpc), DstReg) 2345 .add(MI.getOperand(1)) 2346 .add(predOps(ARMCC::AL)); 2347 MIB1.cloneMemRefs(MI); 2348 MachineInstrBuilder MIB2 = 2349 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::tPICADD)) 2350 .addReg(DstReg, RegState::Define | getDeadRegState(DstIsDead)) 2351 .addReg(DstReg) 2352 .add(MI.getOperand(2)); 2353 TransferImpOps(MI, MIB1, MIB2); 2354 MI.eraseFromParent(); 2355 return true; 2356 } 2357 2358 case ARM::LDRLIT_ga_abs: 2359 case ARM::LDRLIT_ga_pcrel: 2360 case ARM::LDRLIT_ga_pcrel_ldr: 2361 case ARM::tLDRLIT_ga_abs: 2362 case ARM::tLDRLIT_ga_pcrel: { 2363 Register DstReg = MI.getOperand(0).getReg(); 2364 bool DstIsDead = MI.getOperand(0).isDead(); 2365 const MachineOperand &MO1 = MI.getOperand(1); 2366 auto Flags = MO1.getTargetFlags(); 2367 const GlobalValue *GV = MO1.getGlobal(); 2368 bool IsARM = 2369 Opcode != ARM::tLDRLIT_ga_pcrel && Opcode != ARM::tLDRLIT_ga_abs; 2370 bool IsPIC = 2371 Opcode != ARM::LDRLIT_ga_abs && Opcode != ARM::tLDRLIT_ga_abs; 2372 unsigned LDRLITOpc = IsARM ? ARM::LDRi12 : ARM::tLDRpci; 2373 unsigned PICAddOpc = 2374 IsARM 2375 ? (Opcode == ARM::LDRLIT_ga_pcrel_ldr ? ARM::PICLDR : ARM::PICADD) 2376 : ARM::tPICADD; 2377 2378 // We need a new const-pool entry to load from. 2379 MachineConstantPool *MCP = MBB.getParent()->getConstantPool(); 2380 unsigned ARMPCLabelIndex = 0; 2381 MachineConstantPoolValue *CPV; 2382 2383 if (IsPIC) { 2384 unsigned PCAdj = IsARM ? 8 : 4; 2385 auto Modifier = (Flags & ARMII::MO_GOT) 2386 ? ARMCP::GOT_PREL 2387 : ARMCP::no_modifier; 2388 ARMPCLabelIndex = AFI->createPICLabelUId(); 2389 CPV = ARMConstantPoolConstant::Create( 2390 GV, ARMPCLabelIndex, ARMCP::CPValue, PCAdj, Modifier, 2391 /*AddCurrentAddr*/ Modifier == ARMCP::GOT_PREL); 2392 } else 2393 CPV = ARMConstantPoolConstant::Create(GV, ARMCP::no_modifier); 2394 2395 MachineInstrBuilder MIB = 2396 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(LDRLITOpc), DstReg) 2397 .addConstantPoolIndex(MCP->getConstantPoolIndex(CPV, Align(4))); 2398 if (IsARM) 2399 MIB.addImm(0); 2400 MIB.add(predOps(ARMCC::AL)); 2401 2402 if (IsPIC) { 2403 MachineInstrBuilder MIB = 2404 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(PICAddOpc)) 2405 .addReg(DstReg, RegState::Define | getDeadRegState(DstIsDead)) 2406 .addReg(DstReg) 2407 .addImm(ARMPCLabelIndex); 2408 2409 if (IsARM) 2410 MIB.add(predOps(ARMCC::AL)); 2411 } 2412 2413 MI.eraseFromParent(); 2414 return true; 2415 } 2416 case ARM::MOV_ga_pcrel: 2417 case ARM::MOV_ga_pcrel_ldr: 2418 case ARM::t2MOV_ga_pcrel: { 2419 // Expand into movw + movw. Also "add pc" / ldr [pc] in PIC mode. 2420 unsigned LabelId = AFI->createPICLabelUId(); 2421 Register DstReg = MI.getOperand(0).getReg(); 2422 bool DstIsDead = MI.getOperand(0).isDead(); 2423 const MachineOperand &MO1 = MI.getOperand(1); 2424 const GlobalValue *GV = MO1.getGlobal(); 2425 unsigned TF = MO1.getTargetFlags(); 2426 bool isARM = Opcode != ARM::t2MOV_ga_pcrel; 2427 unsigned LO16Opc = isARM ? ARM::MOVi16_ga_pcrel : ARM::t2MOVi16_ga_pcrel; 2428 unsigned HI16Opc = isARM ? ARM::MOVTi16_ga_pcrel :ARM::t2MOVTi16_ga_pcrel; 2429 unsigned LO16TF = TF | ARMII::MO_LO16; 2430 unsigned HI16TF = TF | ARMII::MO_HI16; 2431 unsigned PICAddOpc = isARM 2432 ? (Opcode == ARM::MOV_ga_pcrel_ldr ? ARM::PICLDR : ARM::PICADD) 2433 : ARM::tPICADD; 2434 MachineInstrBuilder MIB1 = BuildMI(MBB, MBBI, MI.getDebugLoc(), 2435 TII->get(LO16Opc), DstReg) 2436 .addGlobalAddress(GV, MO1.getOffset(), TF | LO16TF) 2437 .addImm(LabelId); 2438 2439 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(HI16Opc), DstReg) 2440 .addReg(DstReg) 2441 .addGlobalAddress(GV, MO1.getOffset(), TF | HI16TF) 2442 .addImm(LabelId); 2443 2444 MachineInstrBuilder MIB3 = BuildMI(MBB, MBBI, MI.getDebugLoc(), 2445 TII->get(PICAddOpc)) 2446 .addReg(DstReg, RegState::Define | getDeadRegState(DstIsDead)) 2447 .addReg(DstReg).addImm(LabelId); 2448 if (isARM) { 2449 MIB3.add(predOps(ARMCC::AL)); 2450 if (Opcode == ARM::MOV_ga_pcrel_ldr) 2451 MIB3.cloneMemRefs(MI); 2452 } 2453 TransferImpOps(MI, MIB1, MIB3); 2454 MI.eraseFromParent(); 2455 return true; 2456 } 2457 2458 case ARM::MOVi32imm: 2459 case ARM::MOVCCi32imm: 2460 case ARM::t2MOVi32imm: 2461 case ARM::t2MOVCCi32imm: 2462 ExpandMOV32BitImm(MBB, MBBI); 2463 return true; 2464 2465 case ARM::SUBS_PC_LR: { 2466 MachineInstrBuilder MIB = 2467 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::SUBri), ARM::PC) 2468 .addReg(ARM::LR) 2469 .add(MI.getOperand(0)) 2470 .add(MI.getOperand(1)) 2471 .add(MI.getOperand(2)) 2472 .addReg(ARM::CPSR, RegState::Undef); 2473 TransferImpOps(MI, MIB, MIB); 2474 MI.eraseFromParent(); 2475 return true; 2476 } 2477 case ARM::VLDMQIA: { 2478 unsigned NewOpc = ARM::VLDMDIA; 2479 MachineInstrBuilder MIB = 2480 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(NewOpc)); 2481 unsigned OpIdx = 0; 2482 2483 // Grab the Q register destination. 2484 bool DstIsDead = MI.getOperand(OpIdx).isDead(); 2485 Register DstReg = MI.getOperand(OpIdx++).getReg(); 2486 2487 // Copy the source register. 2488 MIB.add(MI.getOperand(OpIdx++)); 2489 2490 // Copy the predicate operands. 2491 MIB.add(MI.getOperand(OpIdx++)); 2492 MIB.add(MI.getOperand(OpIdx++)); 2493 2494 // Add the destination operands (D subregs). 2495 Register D0 = TRI->getSubReg(DstReg, ARM::dsub_0); 2496 Register D1 = TRI->getSubReg(DstReg, ARM::dsub_1); 2497 MIB.addReg(D0, RegState::Define | getDeadRegState(DstIsDead)) 2498 .addReg(D1, RegState::Define | getDeadRegState(DstIsDead)); 2499 2500 // Add an implicit def for the super-register. 2501 MIB.addReg(DstReg, RegState::ImplicitDefine | getDeadRegState(DstIsDead)); 2502 TransferImpOps(MI, MIB, MIB); 2503 MIB.cloneMemRefs(MI); 2504 MI.eraseFromParent(); 2505 return true; 2506 } 2507 2508 case ARM::VSTMQIA: { 2509 unsigned NewOpc = ARM::VSTMDIA; 2510 MachineInstrBuilder MIB = 2511 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(NewOpc)); 2512 unsigned OpIdx = 0; 2513 2514 // Grab the Q register source. 2515 bool SrcIsKill = MI.getOperand(OpIdx).isKill(); 2516 Register SrcReg = MI.getOperand(OpIdx++).getReg(); 2517 2518 // Copy the destination register. 2519 MachineOperand Dst(MI.getOperand(OpIdx++)); 2520 MIB.add(Dst); 2521 2522 // Copy the predicate operands. 2523 MIB.add(MI.getOperand(OpIdx++)); 2524 MIB.add(MI.getOperand(OpIdx++)); 2525 2526 // Add the source operands (D subregs). 2527 Register D0 = TRI->getSubReg(SrcReg, ARM::dsub_0); 2528 Register D1 = TRI->getSubReg(SrcReg, ARM::dsub_1); 2529 MIB.addReg(D0, SrcIsKill ? RegState::Kill : 0) 2530 .addReg(D1, SrcIsKill ? RegState::Kill : 0); 2531 2532 if (SrcIsKill) // Add an implicit kill for the Q register. 2533 MIB->addRegisterKilled(SrcReg, TRI, true); 2534 2535 TransferImpOps(MI, MIB, MIB); 2536 MIB.cloneMemRefs(MI); 2537 MI.eraseFromParent(); 2538 return true; 2539 } 2540 2541 case ARM::VLD2q8Pseudo: 2542 case ARM::VLD2q16Pseudo: 2543 case ARM::VLD2q32Pseudo: 2544 case ARM::VLD2q8PseudoWB_fixed: 2545 case ARM::VLD2q16PseudoWB_fixed: 2546 case ARM::VLD2q32PseudoWB_fixed: 2547 case ARM::VLD2q8PseudoWB_register: 2548 case ARM::VLD2q16PseudoWB_register: 2549 case ARM::VLD2q32PseudoWB_register: 2550 case ARM::VLD3d8Pseudo: 2551 case ARM::VLD3d16Pseudo: 2552 case ARM::VLD3d32Pseudo: 2553 case ARM::VLD1d8TPseudo: 2554 case ARM::VLD1d16TPseudo: 2555 case ARM::VLD1d32TPseudo: 2556 case ARM::VLD1d64TPseudo: 2557 case ARM::VLD1d64TPseudoWB_fixed: 2558 case ARM::VLD1d64TPseudoWB_register: 2559 case ARM::VLD3d8Pseudo_UPD: 2560 case ARM::VLD3d16Pseudo_UPD: 2561 case ARM::VLD3d32Pseudo_UPD: 2562 case ARM::VLD3q8Pseudo_UPD: 2563 case ARM::VLD3q16Pseudo_UPD: 2564 case ARM::VLD3q32Pseudo_UPD: 2565 case ARM::VLD3q8oddPseudo: 2566 case ARM::VLD3q16oddPseudo: 2567 case ARM::VLD3q32oddPseudo: 2568 case ARM::VLD3q8oddPseudo_UPD: 2569 case ARM::VLD3q16oddPseudo_UPD: 2570 case ARM::VLD3q32oddPseudo_UPD: 2571 case ARM::VLD4d8Pseudo: 2572 case ARM::VLD4d16Pseudo: 2573 case ARM::VLD4d32Pseudo: 2574 case ARM::VLD1d8QPseudo: 2575 case ARM::VLD1d16QPseudo: 2576 case ARM::VLD1d32QPseudo: 2577 case ARM::VLD1d64QPseudo: 2578 case ARM::VLD1d64QPseudoWB_fixed: 2579 case ARM::VLD1d64QPseudoWB_register: 2580 case ARM::VLD1q8HighQPseudo: 2581 case ARM::VLD1q8LowQPseudo_UPD: 2582 case ARM::VLD1q8HighTPseudo: 2583 case ARM::VLD1q8LowTPseudo_UPD: 2584 case ARM::VLD1q16HighQPseudo: 2585 case ARM::VLD1q16LowQPseudo_UPD: 2586 case ARM::VLD1q16HighTPseudo: 2587 case ARM::VLD1q16LowTPseudo_UPD: 2588 case ARM::VLD1q32HighQPseudo: 2589 case ARM::VLD1q32LowQPseudo_UPD: 2590 case ARM::VLD1q32HighTPseudo: 2591 case ARM::VLD1q32LowTPseudo_UPD: 2592 case ARM::VLD1q64HighQPseudo: 2593 case ARM::VLD1q64LowQPseudo_UPD: 2594 case ARM::VLD1q64HighTPseudo: 2595 case ARM::VLD1q64LowTPseudo_UPD: 2596 case ARM::VLD4d8Pseudo_UPD: 2597 case ARM::VLD4d16Pseudo_UPD: 2598 case ARM::VLD4d32Pseudo_UPD: 2599 case ARM::VLD4q8Pseudo_UPD: 2600 case ARM::VLD4q16Pseudo_UPD: 2601 case ARM::VLD4q32Pseudo_UPD: 2602 case ARM::VLD4q8oddPseudo: 2603 case ARM::VLD4q16oddPseudo: 2604 case ARM::VLD4q32oddPseudo: 2605 case ARM::VLD4q8oddPseudo_UPD: 2606 case ARM::VLD4q16oddPseudo_UPD: 2607 case ARM::VLD4q32oddPseudo_UPD: 2608 case ARM::VLD3DUPd8Pseudo: 2609 case ARM::VLD3DUPd16Pseudo: 2610 case ARM::VLD3DUPd32Pseudo: 2611 case ARM::VLD3DUPd8Pseudo_UPD: 2612 case ARM::VLD3DUPd16Pseudo_UPD: 2613 case ARM::VLD3DUPd32Pseudo_UPD: 2614 case ARM::VLD4DUPd8Pseudo: 2615 case ARM::VLD4DUPd16Pseudo: 2616 case ARM::VLD4DUPd32Pseudo: 2617 case ARM::VLD4DUPd8Pseudo_UPD: 2618 case ARM::VLD4DUPd16Pseudo_UPD: 2619 case ARM::VLD4DUPd32Pseudo_UPD: 2620 case ARM::VLD2DUPq8EvenPseudo: 2621 case ARM::VLD2DUPq8OddPseudo: 2622 case ARM::VLD2DUPq16EvenPseudo: 2623 case ARM::VLD2DUPq16OddPseudo: 2624 case ARM::VLD2DUPq32EvenPseudo: 2625 case ARM::VLD2DUPq32OddPseudo: 2626 case ARM::VLD3DUPq8EvenPseudo: 2627 case ARM::VLD3DUPq8OddPseudo: 2628 case ARM::VLD3DUPq16EvenPseudo: 2629 case ARM::VLD3DUPq16OddPseudo: 2630 case ARM::VLD3DUPq32EvenPseudo: 2631 case ARM::VLD3DUPq32OddPseudo: 2632 case ARM::VLD4DUPq8EvenPseudo: 2633 case ARM::VLD4DUPq8OddPseudo: 2634 case ARM::VLD4DUPq16EvenPseudo: 2635 case ARM::VLD4DUPq16OddPseudo: 2636 case ARM::VLD4DUPq32EvenPseudo: 2637 case ARM::VLD4DUPq32OddPseudo: 2638 ExpandVLD(MBBI); 2639 return true; 2640 2641 case ARM::VST2q8Pseudo: 2642 case ARM::VST2q16Pseudo: 2643 case ARM::VST2q32Pseudo: 2644 case ARM::VST2q8PseudoWB_fixed: 2645 case ARM::VST2q16PseudoWB_fixed: 2646 case ARM::VST2q32PseudoWB_fixed: 2647 case ARM::VST2q8PseudoWB_register: 2648 case ARM::VST2q16PseudoWB_register: 2649 case ARM::VST2q32PseudoWB_register: 2650 case ARM::VST3d8Pseudo: 2651 case ARM::VST3d16Pseudo: 2652 case ARM::VST3d32Pseudo: 2653 case ARM::VST1d8TPseudo: 2654 case ARM::VST1d16TPseudo: 2655 case ARM::VST1d32TPseudo: 2656 case ARM::VST1d64TPseudo: 2657 case ARM::VST3d8Pseudo_UPD: 2658 case ARM::VST3d16Pseudo_UPD: 2659 case ARM::VST3d32Pseudo_UPD: 2660 case ARM::VST1d64TPseudoWB_fixed: 2661 case ARM::VST1d64TPseudoWB_register: 2662 case ARM::VST3q8Pseudo_UPD: 2663 case ARM::VST3q16Pseudo_UPD: 2664 case ARM::VST3q32Pseudo_UPD: 2665 case ARM::VST3q8oddPseudo: 2666 case ARM::VST3q16oddPseudo: 2667 case ARM::VST3q32oddPseudo: 2668 case ARM::VST3q8oddPseudo_UPD: 2669 case ARM::VST3q16oddPseudo_UPD: 2670 case ARM::VST3q32oddPseudo_UPD: 2671 case ARM::VST4d8Pseudo: 2672 case ARM::VST4d16Pseudo: 2673 case ARM::VST4d32Pseudo: 2674 case ARM::VST1d8QPseudo: 2675 case ARM::VST1d16QPseudo: 2676 case ARM::VST1d32QPseudo: 2677 case ARM::VST1d64QPseudo: 2678 case ARM::VST4d8Pseudo_UPD: 2679 case ARM::VST4d16Pseudo_UPD: 2680 case ARM::VST4d32Pseudo_UPD: 2681 case ARM::VST1d64QPseudoWB_fixed: 2682 case ARM::VST1d64QPseudoWB_register: 2683 case ARM::VST1q8HighQPseudo: 2684 case ARM::VST1q8LowQPseudo_UPD: 2685 case ARM::VST1q8HighTPseudo: 2686 case ARM::VST1q8LowTPseudo_UPD: 2687 case ARM::VST1q16HighQPseudo: 2688 case ARM::VST1q16LowQPseudo_UPD: 2689 case ARM::VST1q16HighTPseudo: 2690 case ARM::VST1q16LowTPseudo_UPD: 2691 case ARM::VST1q32HighQPseudo: 2692 case ARM::VST1q32LowQPseudo_UPD: 2693 case ARM::VST1q32HighTPseudo: 2694 case ARM::VST1q32LowTPseudo_UPD: 2695 case ARM::VST1q64HighQPseudo: 2696 case ARM::VST1q64LowQPseudo_UPD: 2697 case ARM::VST1q64HighTPseudo: 2698 case ARM::VST1q64LowTPseudo_UPD: 2699 case ARM::VST4q8Pseudo_UPD: 2700 case ARM::VST4q16Pseudo_UPD: 2701 case ARM::VST4q32Pseudo_UPD: 2702 case ARM::VST4q8oddPseudo: 2703 case ARM::VST4q16oddPseudo: 2704 case ARM::VST4q32oddPseudo: 2705 case ARM::VST4q8oddPseudo_UPD: 2706 case ARM::VST4q16oddPseudo_UPD: 2707 case ARM::VST4q32oddPseudo_UPD: 2708 ExpandVST(MBBI); 2709 return true; 2710 2711 case ARM::VLD1LNq8Pseudo: 2712 case ARM::VLD1LNq16Pseudo: 2713 case ARM::VLD1LNq32Pseudo: 2714 case ARM::VLD1LNq8Pseudo_UPD: 2715 case ARM::VLD1LNq16Pseudo_UPD: 2716 case ARM::VLD1LNq32Pseudo_UPD: 2717 case ARM::VLD2LNd8Pseudo: 2718 case ARM::VLD2LNd16Pseudo: 2719 case ARM::VLD2LNd32Pseudo: 2720 case ARM::VLD2LNq16Pseudo: 2721 case ARM::VLD2LNq32Pseudo: 2722 case ARM::VLD2LNd8Pseudo_UPD: 2723 case ARM::VLD2LNd16Pseudo_UPD: 2724 case ARM::VLD2LNd32Pseudo_UPD: 2725 case ARM::VLD2LNq16Pseudo_UPD: 2726 case ARM::VLD2LNq32Pseudo_UPD: 2727 case ARM::VLD3LNd8Pseudo: 2728 case ARM::VLD3LNd16Pseudo: 2729 case ARM::VLD3LNd32Pseudo: 2730 case ARM::VLD3LNq16Pseudo: 2731 case ARM::VLD3LNq32Pseudo: 2732 case ARM::VLD3LNd8Pseudo_UPD: 2733 case ARM::VLD3LNd16Pseudo_UPD: 2734 case ARM::VLD3LNd32Pseudo_UPD: 2735 case ARM::VLD3LNq16Pseudo_UPD: 2736 case ARM::VLD3LNq32Pseudo_UPD: 2737 case ARM::VLD4LNd8Pseudo: 2738 case ARM::VLD4LNd16Pseudo: 2739 case ARM::VLD4LNd32Pseudo: 2740 case ARM::VLD4LNq16Pseudo: 2741 case ARM::VLD4LNq32Pseudo: 2742 case ARM::VLD4LNd8Pseudo_UPD: 2743 case ARM::VLD4LNd16Pseudo_UPD: 2744 case ARM::VLD4LNd32Pseudo_UPD: 2745 case ARM::VLD4LNq16Pseudo_UPD: 2746 case ARM::VLD4LNq32Pseudo_UPD: 2747 case ARM::VST1LNq8Pseudo: 2748 case ARM::VST1LNq16Pseudo: 2749 case ARM::VST1LNq32Pseudo: 2750 case ARM::VST1LNq8Pseudo_UPD: 2751 case ARM::VST1LNq16Pseudo_UPD: 2752 case ARM::VST1LNq32Pseudo_UPD: 2753 case ARM::VST2LNd8Pseudo: 2754 case ARM::VST2LNd16Pseudo: 2755 case ARM::VST2LNd32Pseudo: 2756 case ARM::VST2LNq16Pseudo: 2757 case ARM::VST2LNq32Pseudo: 2758 case ARM::VST2LNd8Pseudo_UPD: 2759 case ARM::VST2LNd16Pseudo_UPD: 2760 case ARM::VST2LNd32Pseudo_UPD: 2761 case ARM::VST2LNq16Pseudo_UPD: 2762 case ARM::VST2LNq32Pseudo_UPD: 2763 case ARM::VST3LNd8Pseudo: 2764 case ARM::VST3LNd16Pseudo: 2765 case ARM::VST3LNd32Pseudo: 2766 case ARM::VST3LNq16Pseudo: 2767 case ARM::VST3LNq32Pseudo: 2768 case ARM::VST3LNd8Pseudo_UPD: 2769 case ARM::VST3LNd16Pseudo_UPD: 2770 case ARM::VST3LNd32Pseudo_UPD: 2771 case ARM::VST3LNq16Pseudo_UPD: 2772 case ARM::VST3LNq32Pseudo_UPD: 2773 case ARM::VST4LNd8Pseudo: 2774 case ARM::VST4LNd16Pseudo: 2775 case ARM::VST4LNd32Pseudo: 2776 case ARM::VST4LNq16Pseudo: 2777 case ARM::VST4LNq32Pseudo: 2778 case ARM::VST4LNd8Pseudo_UPD: 2779 case ARM::VST4LNd16Pseudo_UPD: 2780 case ARM::VST4LNd32Pseudo_UPD: 2781 case ARM::VST4LNq16Pseudo_UPD: 2782 case ARM::VST4LNq32Pseudo_UPD: 2783 ExpandLaneOp(MBBI); 2784 return true; 2785 2786 case ARM::VTBL3Pseudo: ExpandVTBL(MBBI, ARM::VTBL3, false); return true; 2787 case ARM::VTBL4Pseudo: ExpandVTBL(MBBI, ARM::VTBL4, false); return true; 2788 case ARM::VTBX3Pseudo: ExpandVTBL(MBBI, ARM::VTBX3, true); return true; 2789 case ARM::VTBX4Pseudo: ExpandVTBL(MBBI, ARM::VTBX4, true); return true; 2790 2791 case ARM::tCMP_SWAP_8: 2792 assert(STI->isThumb()); 2793 return ExpandCMP_SWAP(MBB, MBBI, ARM::t2LDREXB, ARM::t2STREXB, ARM::tUXTB, 2794 NextMBBI); 2795 case ARM::tCMP_SWAP_16: 2796 assert(STI->isThumb()); 2797 return ExpandCMP_SWAP(MBB, MBBI, ARM::t2LDREXH, ARM::t2STREXH, ARM::tUXTH, 2798 NextMBBI); 2799 2800 case ARM::CMP_SWAP_8: 2801 assert(!STI->isThumb()); 2802 return ExpandCMP_SWAP(MBB, MBBI, ARM::LDREXB, ARM::STREXB, ARM::UXTB, 2803 NextMBBI); 2804 case ARM::CMP_SWAP_16: 2805 assert(!STI->isThumb()); 2806 return ExpandCMP_SWAP(MBB, MBBI, ARM::LDREXH, ARM::STREXH, ARM::UXTH, 2807 NextMBBI); 2808 case ARM::CMP_SWAP_32: 2809 if (STI->isThumb()) 2810 return ExpandCMP_SWAP(MBB, MBBI, ARM::t2LDREX, ARM::t2STREX, 0, 2811 NextMBBI); 2812 else 2813 return ExpandCMP_SWAP(MBB, MBBI, ARM::LDREX, ARM::STREX, 0, NextMBBI); 2814 2815 case ARM::CMP_SWAP_64: 2816 return ExpandCMP_SWAP_64(MBB, MBBI, NextMBBI); 2817 2818 case ARM::tBL_PUSHLR: 2819 case ARM::BL_PUSHLR: { 2820 const bool Thumb = Opcode == ARM::tBL_PUSHLR; 2821 Register Reg = MI.getOperand(0).getReg(); 2822 assert(Reg == ARM::LR && "expect LR register!"); 2823 MachineInstrBuilder MIB; 2824 if (Thumb) { 2825 // push {lr} 2826 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::tPUSH)) 2827 .add(predOps(ARMCC::AL)) 2828 .addReg(Reg); 2829 2830 // bl __gnu_mcount_nc 2831 MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::tBL)); 2832 } else { 2833 // stmdb sp!, {lr} 2834 BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::STMDB_UPD)) 2835 .addReg(ARM::SP, RegState::Define) 2836 .addReg(ARM::SP) 2837 .add(predOps(ARMCC::AL)) 2838 .addReg(Reg); 2839 2840 // bl __gnu_mcount_nc 2841 MIB = BuildMI(MBB, MBBI, MI.getDebugLoc(), TII->get(ARM::BL)); 2842 } 2843 MIB.cloneMemRefs(MI); 2844 for (unsigned i = 1; i < MI.getNumOperands(); ++i) MIB.add(MI.getOperand(i)); 2845 MI.eraseFromParent(); 2846 return true; 2847 } 2848 case ARM::LOADDUAL: 2849 case ARM::STOREDUAL: { 2850 Register PairReg = MI.getOperand(0).getReg(); 2851 2852 MachineInstrBuilder MIB = 2853 BuildMI(MBB, MBBI, MI.getDebugLoc(), 2854 TII->get(Opcode == ARM::LOADDUAL ? ARM::LDRD : ARM::STRD)) 2855 .addReg(TRI->getSubReg(PairReg, ARM::gsub_0), 2856 Opcode == ARM::LOADDUAL ? RegState::Define : 0) 2857 .addReg(TRI->getSubReg(PairReg, ARM::gsub_1), 2858 Opcode == ARM::LOADDUAL ? RegState::Define : 0); 2859 for (unsigned i = 1; i < MI.getNumOperands(); i++) 2860 MIB.add(MI.getOperand(i)); 2861 MIB.add(predOps(ARMCC::AL)); 2862 MIB.cloneMemRefs(MI); 2863 MI.eraseFromParent(); 2864 return true; 2865 } 2866 } 2867 } 2868 2869 bool ARMExpandPseudo::ExpandMBB(MachineBasicBlock &MBB) { 2870 bool Modified = false; 2871 2872 MachineBasicBlock::iterator MBBI = MBB.begin(), E = MBB.end(); 2873 while (MBBI != E) { 2874 MachineBasicBlock::iterator NMBBI = std::next(MBBI); 2875 Modified |= ExpandMI(MBB, MBBI, NMBBI); 2876 MBBI = NMBBI; 2877 } 2878 2879 return Modified; 2880 } 2881 2882 bool ARMExpandPseudo::runOnMachineFunction(MachineFunction &MF) { 2883 STI = &static_cast<const ARMSubtarget &>(MF.getSubtarget()); 2884 TII = STI->getInstrInfo(); 2885 TRI = STI->getRegisterInfo(); 2886 AFI = MF.getInfo<ARMFunctionInfo>(); 2887 2888 LLVM_DEBUG(dbgs() << "********** ARM EXPAND PSEUDO INSTRUCTIONS **********\n" 2889 << "********** Function: " << MF.getName() << '\n'); 2890 2891 bool Modified = false; 2892 for (MachineBasicBlock &MBB : MF) 2893 Modified |= ExpandMBB(MBB); 2894 if (VerifyARMPseudo) 2895 MF.verify(this, "After expanding ARM pseudo instructions."); 2896 2897 LLVM_DEBUG(dbgs() << "***************************************************\n"); 2898 return Modified; 2899 } 2900 2901 /// createARMExpandPseudoPass - returns an instance of the pseudo instruction 2902 /// expansion pass. 2903 FunctionPass *llvm::createARMExpandPseudoPass() { 2904 return new ARMExpandPseudo(); 2905 } 2906