1 //===- llvm/CodeGen/GlobalISel/Utils.cpp -------------------------*- C++ -*-==// 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 /// \file This file implements the utility functions used by the GlobalISel 9 /// pipeline. 10 //===----------------------------------------------------------------------===// 11 12 #include "llvm/CodeGen/GlobalISel/Utils.h" 13 #include "llvm/ADT/APFloat.h" 14 #include "llvm/ADT/APInt.h" 15 #include "llvm/CodeGen/CodeGenCommonISel.h" 16 #include "llvm/CodeGen/GlobalISel/GISelChangeObserver.h" 17 #include "llvm/CodeGen/GlobalISel/GISelKnownBits.h" 18 #include "llvm/CodeGen/GlobalISel/GenericMachineInstrs.h" 19 #include "llvm/CodeGen/GlobalISel/LostDebugLocObserver.h" 20 #include "llvm/CodeGen/GlobalISel/MIPatternMatch.h" 21 #include "llvm/CodeGen/MachineInstr.h" 22 #include "llvm/CodeGen/MachineInstrBuilder.h" 23 #include "llvm/CodeGen/MachineOptimizationRemarkEmitter.h" 24 #include "llvm/CodeGen/MachineRegisterInfo.h" 25 #include "llvm/CodeGen/MachineSizeOpts.h" 26 #include "llvm/CodeGen/RegisterBankInfo.h" 27 #include "llvm/CodeGen/StackProtector.h" 28 #include "llvm/CodeGen/TargetInstrInfo.h" 29 #include "llvm/CodeGen/TargetLowering.h" 30 #include "llvm/CodeGen/TargetPassConfig.h" 31 #include "llvm/CodeGen/TargetRegisterInfo.h" 32 #include "llvm/IR/Constants.h" 33 #include "llvm/Target/TargetMachine.h" 34 #include "llvm/Transforms/Utils/SizeOpts.h" 35 #include <numeric> 36 #include <optional> 37 38 #define DEBUG_TYPE "globalisel-utils" 39 40 using namespace llvm; 41 using namespace MIPatternMatch; 42 43 Register llvm::constrainRegToClass(MachineRegisterInfo &MRI, 44 const TargetInstrInfo &TII, 45 const RegisterBankInfo &RBI, Register Reg, 46 const TargetRegisterClass &RegClass) { 47 if (!RBI.constrainGenericRegister(Reg, RegClass, MRI)) 48 return MRI.createVirtualRegister(&RegClass); 49 50 return Reg; 51 } 52 53 Register llvm::constrainOperandRegClass( 54 const MachineFunction &MF, const TargetRegisterInfo &TRI, 55 MachineRegisterInfo &MRI, const TargetInstrInfo &TII, 56 const RegisterBankInfo &RBI, MachineInstr &InsertPt, 57 const TargetRegisterClass &RegClass, MachineOperand &RegMO) { 58 Register Reg = RegMO.getReg(); 59 // Assume physical registers are properly constrained. 60 assert(Reg.isVirtual() && "PhysReg not implemented"); 61 62 // Save the old register class to check whether 63 // the change notifications will be required. 64 // TODO: A better approach would be to pass 65 // the observers to constrainRegToClass(). 66 auto *OldRegClass = MRI.getRegClassOrNull(Reg); 67 Register ConstrainedReg = constrainRegToClass(MRI, TII, RBI, Reg, RegClass); 68 // If we created a new virtual register because the class is not compatible 69 // then create a copy between the new and the old register. 70 if (ConstrainedReg != Reg) { 71 MachineBasicBlock::iterator InsertIt(&InsertPt); 72 MachineBasicBlock &MBB = *InsertPt.getParent(); 73 // FIXME: The copy needs to have the classes constrained for its operands. 74 // Use operand's regbank to get the class for old register (Reg). 75 if (RegMO.isUse()) { 76 BuildMI(MBB, InsertIt, InsertPt.getDebugLoc(), 77 TII.get(TargetOpcode::COPY), ConstrainedReg) 78 .addReg(Reg); 79 } else { 80 assert(RegMO.isDef() && "Must be a definition"); 81 BuildMI(MBB, std::next(InsertIt), InsertPt.getDebugLoc(), 82 TII.get(TargetOpcode::COPY), Reg) 83 .addReg(ConstrainedReg); 84 } 85 if (GISelChangeObserver *Observer = MF.getObserver()) { 86 Observer->changingInstr(*RegMO.getParent()); 87 } 88 RegMO.setReg(ConstrainedReg); 89 if (GISelChangeObserver *Observer = MF.getObserver()) { 90 Observer->changedInstr(*RegMO.getParent()); 91 } 92 } else if (OldRegClass != MRI.getRegClassOrNull(Reg)) { 93 if (GISelChangeObserver *Observer = MF.getObserver()) { 94 if (!RegMO.isDef()) { 95 MachineInstr *RegDef = MRI.getVRegDef(Reg); 96 Observer->changedInstr(*RegDef); 97 } 98 Observer->changingAllUsesOfReg(MRI, Reg); 99 Observer->finishedChangingAllUsesOfReg(); 100 } 101 } 102 return ConstrainedReg; 103 } 104 105 Register llvm::constrainOperandRegClass( 106 const MachineFunction &MF, const TargetRegisterInfo &TRI, 107 MachineRegisterInfo &MRI, const TargetInstrInfo &TII, 108 const RegisterBankInfo &RBI, MachineInstr &InsertPt, const MCInstrDesc &II, 109 MachineOperand &RegMO, unsigned OpIdx) { 110 Register Reg = RegMO.getReg(); 111 // Assume physical registers are properly constrained. 112 assert(Reg.isVirtual() && "PhysReg not implemented"); 113 114 const TargetRegisterClass *OpRC = TII.getRegClass(II, OpIdx, &TRI, MF); 115 // Some of the target independent instructions, like COPY, may not impose any 116 // register class constraints on some of their operands: If it's a use, we can 117 // skip constraining as the instruction defining the register would constrain 118 // it. 119 120 if (OpRC) { 121 // Obtain the RC from incoming regbank if it is a proper sub-class. Operands 122 // can have multiple regbanks for a superclass that combine different 123 // register types (E.g., AMDGPU's VGPR and AGPR). The regbank ambiguity 124 // resolved by targets during regbankselect should not be overridden. 125 if (const auto *SubRC = TRI.getCommonSubClass( 126 OpRC, TRI.getConstrainedRegClassForOperand(RegMO, MRI))) 127 OpRC = SubRC; 128 129 OpRC = TRI.getAllocatableClass(OpRC); 130 } 131 132 if (!OpRC) { 133 assert((!isTargetSpecificOpcode(II.getOpcode()) || RegMO.isUse()) && 134 "Register class constraint is required unless either the " 135 "instruction is target independent or the operand is a use"); 136 // FIXME: Just bailing out like this here could be not enough, unless we 137 // expect the users of this function to do the right thing for PHIs and 138 // COPY: 139 // v1 = COPY v0 140 // v2 = COPY v1 141 // v1 here may end up not being constrained at all. Please notice that to 142 // reproduce the issue we likely need a destination pattern of a selection 143 // rule producing such extra copies, not just an input GMIR with them as 144 // every existing target using selectImpl handles copies before calling it 145 // and they never reach this function. 146 return Reg; 147 } 148 return constrainOperandRegClass(MF, TRI, MRI, TII, RBI, InsertPt, *OpRC, 149 RegMO); 150 } 151 152 bool llvm::constrainSelectedInstRegOperands(MachineInstr &I, 153 const TargetInstrInfo &TII, 154 const TargetRegisterInfo &TRI, 155 const RegisterBankInfo &RBI) { 156 assert(!isPreISelGenericOpcode(I.getOpcode()) && 157 "A selected instruction is expected"); 158 MachineBasicBlock &MBB = *I.getParent(); 159 MachineFunction &MF = *MBB.getParent(); 160 MachineRegisterInfo &MRI = MF.getRegInfo(); 161 162 for (unsigned OpI = 0, OpE = I.getNumExplicitOperands(); OpI != OpE; ++OpI) { 163 MachineOperand &MO = I.getOperand(OpI); 164 165 // There's nothing to be done on non-register operands. 166 if (!MO.isReg()) 167 continue; 168 169 LLVM_DEBUG(dbgs() << "Converting operand: " << MO << '\n'); 170 assert(MO.isReg() && "Unsupported non-reg operand"); 171 172 Register Reg = MO.getReg(); 173 // Physical registers don't need to be constrained. 174 if (Reg.isPhysical()) 175 continue; 176 177 // Register operands with a value of 0 (e.g. predicate operands) don't need 178 // to be constrained. 179 if (Reg == 0) 180 continue; 181 182 // If the operand is a vreg, we should constrain its regclass, and only 183 // insert COPYs if that's impossible. 184 // constrainOperandRegClass does that for us. 185 constrainOperandRegClass(MF, TRI, MRI, TII, RBI, I, I.getDesc(), MO, OpI); 186 187 // Tie uses to defs as indicated in MCInstrDesc if this hasn't already been 188 // done. 189 if (MO.isUse()) { 190 int DefIdx = I.getDesc().getOperandConstraint(OpI, MCOI::TIED_TO); 191 if (DefIdx != -1 && !I.isRegTiedToUseOperand(DefIdx)) 192 I.tieOperands(DefIdx, OpI); 193 } 194 } 195 return true; 196 } 197 198 bool llvm::canReplaceReg(Register DstReg, Register SrcReg, 199 MachineRegisterInfo &MRI) { 200 // Give up if either DstReg or SrcReg is a physical register. 201 if (DstReg.isPhysical() || SrcReg.isPhysical()) 202 return false; 203 // Give up if the types don't match. 204 if (MRI.getType(DstReg) != MRI.getType(SrcReg)) 205 return false; 206 // Replace if either DstReg has no constraints or the register 207 // constraints match. 208 const auto &DstRBC = MRI.getRegClassOrRegBank(DstReg); 209 if (!DstRBC || DstRBC == MRI.getRegClassOrRegBank(SrcReg)) 210 return true; 211 212 // Otherwise match if the Src is already a regclass that is covered by the Dst 213 // RegBank. 214 return DstRBC.is<const RegisterBank *>() && MRI.getRegClassOrNull(SrcReg) && 215 DstRBC.get<const RegisterBank *>()->covers( 216 *MRI.getRegClassOrNull(SrcReg)); 217 } 218 219 bool llvm::isTriviallyDead(const MachineInstr &MI, 220 const MachineRegisterInfo &MRI) { 221 // FIXME: This logical is mostly duplicated with 222 // DeadMachineInstructionElim::isDead. Why is LOCAL_ESCAPE not considered in 223 // MachineInstr::isLabel? 224 225 // Don't delete frame allocation labels. 226 if (MI.getOpcode() == TargetOpcode::LOCAL_ESCAPE) 227 return false; 228 // LIFETIME markers should be preserved even if they seem dead. 229 if (MI.getOpcode() == TargetOpcode::LIFETIME_START || 230 MI.getOpcode() == TargetOpcode::LIFETIME_END) 231 return false; 232 233 // If we can move an instruction, we can remove it. Otherwise, it has 234 // a side-effect of some sort. 235 bool SawStore = false; 236 if (!MI.isSafeToMove(/*AA=*/nullptr, SawStore) && !MI.isPHI()) 237 return false; 238 239 // Instructions without side-effects are dead iff they only define dead vregs. 240 for (const auto &MO : MI.all_defs()) { 241 Register Reg = MO.getReg(); 242 if (Reg.isPhysical() || !MRI.use_nodbg_empty(Reg)) 243 return false; 244 } 245 return true; 246 } 247 248 static void reportGISelDiagnostic(DiagnosticSeverity Severity, 249 MachineFunction &MF, 250 const TargetPassConfig &TPC, 251 MachineOptimizationRemarkEmitter &MORE, 252 MachineOptimizationRemarkMissed &R) { 253 bool IsFatal = Severity == DS_Error && 254 TPC.isGlobalISelAbortEnabled(); 255 // Print the function name explicitly if we don't have a debug location (which 256 // makes the diagnostic less useful) or if we're going to emit a raw error. 257 if (!R.getLocation().isValid() || IsFatal) 258 R << (" (in function: " + MF.getName() + ")").str(); 259 260 if (IsFatal) 261 report_fatal_error(Twine(R.getMsg())); 262 else 263 MORE.emit(R); 264 } 265 266 void llvm::reportGISelWarning(MachineFunction &MF, const TargetPassConfig &TPC, 267 MachineOptimizationRemarkEmitter &MORE, 268 MachineOptimizationRemarkMissed &R) { 269 reportGISelDiagnostic(DS_Warning, MF, TPC, MORE, R); 270 } 271 272 void llvm::reportGISelFailure(MachineFunction &MF, const TargetPassConfig &TPC, 273 MachineOptimizationRemarkEmitter &MORE, 274 MachineOptimizationRemarkMissed &R) { 275 MF.getProperties().set(MachineFunctionProperties::Property::FailedISel); 276 reportGISelDiagnostic(DS_Error, MF, TPC, MORE, R); 277 } 278 279 void llvm::reportGISelFailure(MachineFunction &MF, const TargetPassConfig &TPC, 280 MachineOptimizationRemarkEmitter &MORE, 281 const char *PassName, StringRef Msg, 282 const MachineInstr &MI) { 283 MachineOptimizationRemarkMissed R(PassName, "GISelFailure: ", 284 MI.getDebugLoc(), MI.getParent()); 285 R << Msg; 286 // Printing MI is expensive; only do it if expensive remarks are enabled. 287 if (TPC.isGlobalISelAbortEnabled() || MORE.allowExtraAnalysis(PassName)) 288 R << ": " << ore::MNV("Inst", MI); 289 reportGISelFailure(MF, TPC, MORE, R); 290 } 291 292 std::optional<APInt> llvm::getIConstantVRegVal(Register VReg, 293 const MachineRegisterInfo &MRI, 294 bool LookThroughInstrs) { 295 std::optional<ValueAndVReg> ValAndVReg = 296 getIConstantVRegValWithLookThrough(VReg, MRI, LookThroughInstrs); 297 assert((!ValAndVReg || ValAndVReg->VReg == VReg) && 298 "Value found while looking through instrs"); 299 if (!ValAndVReg) 300 return std::nullopt; 301 return ValAndVReg->Value; 302 } 303 304 std::optional<int64_t> 305 llvm::getIConstantVRegSExtVal(Register VReg, const MachineRegisterInfo &MRI, 306 bool LookThroughInstrs) { 307 std::optional<APInt> Val = getIConstantVRegVal(VReg, MRI, LookThroughInstrs); 308 if (Val && Val->getBitWidth() <= 64) 309 return Val->getSExtValue(); 310 return std::nullopt; 311 } 312 313 namespace { 314 315 typedef std::function<bool(const MachineInstr *)> IsOpcodeFn; 316 typedef std::function<std::optional<APInt>(const MachineInstr *MI)> GetAPCstFn; 317 318 std::optional<ValueAndVReg> getConstantVRegValWithLookThrough( 319 Register VReg, const MachineRegisterInfo &MRI, IsOpcodeFn IsConstantOpcode, 320 GetAPCstFn getAPCstValue, bool LookThroughInstrs = true, 321 bool LookThroughAnyExt = false) { 322 SmallVector<std::pair<unsigned, unsigned>, 4> SeenOpcodes; 323 MachineInstr *MI; 324 325 while ((MI = MRI.getVRegDef(VReg)) && !IsConstantOpcode(MI) && 326 LookThroughInstrs) { 327 switch (MI->getOpcode()) { 328 case TargetOpcode::G_ANYEXT: 329 if (!LookThroughAnyExt) 330 return std::nullopt; 331 [[fallthrough]]; 332 case TargetOpcode::G_TRUNC: 333 case TargetOpcode::G_SEXT: 334 case TargetOpcode::G_ZEXT: 335 SeenOpcodes.push_back(std::make_pair( 336 MI->getOpcode(), 337 MRI.getType(MI->getOperand(0).getReg()).getSizeInBits())); 338 VReg = MI->getOperand(1).getReg(); 339 break; 340 case TargetOpcode::COPY: 341 VReg = MI->getOperand(1).getReg(); 342 if (VReg.isPhysical()) 343 return std::nullopt; 344 break; 345 case TargetOpcode::G_INTTOPTR: 346 VReg = MI->getOperand(1).getReg(); 347 break; 348 default: 349 return std::nullopt; 350 } 351 } 352 if (!MI || !IsConstantOpcode(MI)) 353 return std::nullopt; 354 355 std::optional<APInt> MaybeVal = getAPCstValue(MI); 356 if (!MaybeVal) 357 return std::nullopt; 358 APInt &Val = *MaybeVal; 359 while (!SeenOpcodes.empty()) { 360 std::pair<unsigned, unsigned> OpcodeAndSize = SeenOpcodes.pop_back_val(); 361 switch (OpcodeAndSize.first) { 362 case TargetOpcode::G_TRUNC: 363 Val = Val.trunc(OpcodeAndSize.second); 364 break; 365 case TargetOpcode::G_ANYEXT: 366 case TargetOpcode::G_SEXT: 367 Val = Val.sext(OpcodeAndSize.second); 368 break; 369 case TargetOpcode::G_ZEXT: 370 Val = Val.zext(OpcodeAndSize.second); 371 break; 372 } 373 } 374 375 return ValueAndVReg{Val, VReg}; 376 } 377 378 bool isIConstant(const MachineInstr *MI) { 379 if (!MI) 380 return false; 381 return MI->getOpcode() == TargetOpcode::G_CONSTANT; 382 } 383 384 bool isFConstant(const MachineInstr *MI) { 385 if (!MI) 386 return false; 387 return MI->getOpcode() == TargetOpcode::G_FCONSTANT; 388 } 389 390 bool isAnyConstant(const MachineInstr *MI) { 391 if (!MI) 392 return false; 393 unsigned Opc = MI->getOpcode(); 394 return Opc == TargetOpcode::G_CONSTANT || Opc == TargetOpcode::G_FCONSTANT; 395 } 396 397 std::optional<APInt> getCImmAsAPInt(const MachineInstr *MI) { 398 const MachineOperand &CstVal = MI->getOperand(1); 399 if (CstVal.isCImm()) 400 return CstVal.getCImm()->getValue(); 401 return std::nullopt; 402 } 403 404 std::optional<APInt> getCImmOrFPImmAsAPInt(const MachineInstr *MI) { 405 const MachineOperand &CstVal = MI->getOperand(1); 406 if (CstVal.isCImm()) 407 return CstVal.getCImm()->getValue(); 408 if (CstVal.isFPImm()) 409 return CstVal.getFPImm()->getValueAPF().bitcastToAPInt(); 410 return std::nullopt; 411 } 412 413 } // end anonymous namespace 414 415 std::optional<ValueAndVReg> llvm::getIConstantVRegValWithLookThrough( 416 Register VReg, const MachineRegisterInfo &MRI, bool LookThroughInstrs) { 417 return getConstantVRegValWithLookThrough(VReg, MRI, isIConstant, 418 getCImmAsAPInt, LookThroughInstrs); 419 } 420 421 std::optional<ValueAndVReg> llvm::getAnyConstantVRegValWithLookThrough( 422 Register VReg, const MachineRegisterInfo &MRI, bool LookThroughInstrs, 423 bool LookThroughAnyExt) { 424 return getConstantVRegValWithLookThrough( 425 VReg, MRI, isAnyConstant, getCImmOrFPImmAsAPInt, LookThroughInstrs, 426 LookThroughAnyExt); 427 } 428 429 std::optional<FPValueAndVReg> llvm::getFConstantVRegValWithLookThrough( 430 Register VReg, const MachineRegisterInfo &MRI, bool LookThroughInstrs) { 431 auto Reg = getConstantVRegValWithLookThrough( 432 VReg, MRI, isFConstant, getCImmOrFPImmAsAPInt, LookThroughInstrs); 433 if (!Reg) 434 return std::nullopt; 435 return FPValueAndVReg{getConstantFPVRegVal(Reg->VReg, MRI)->getValueAPF(), 436 Reg->VReg}; 437 } 438 439 const ConstantFP * 440 llvm::getConstantFPVRegVal(Register VReg, const MachineRegisterInfo &MRI) { 441 MachineInstr *MI = MRI.getVRegDef(VReg); 442 if (TargetOpcode::G_FCONSTANT != MI->getOpcode()) 443 return nullptr; 444 return MI->getOperand(1).getFPImm(); 445 } 446 447 std::optional<DefinitionAndSourceRegister> 448 llvm::getDefSrcRegIgnoringCopies(Register Reg, const MachineRegisterInfo &MRI) { 449 Register DefSrcReg = Reg; 450 auto *DefMI = MRI.getVRegDef(Reg); 451 auto DstTy = MRI.getType(DefMI->getOperand(0).getReg()); 452 if (!DstTy.isValid()) 453 return std::nullopt; 454 unsigned Opc = DefMI->getOpcode(); 455 while (Opc == TargetOpcode::COPY || isPreISelGenericOptimizationHint(Opc)) { 456 Register SrcReg = DefMI->getOperand(1).getReg(); 457 auto SrcTy = MRI.getType(SrcReg); 458 if (!SrcTy.isValid()) 459 break; 460 DefMI = MRI.getVRegDef(SrcReg); 461 DefSrcReg = SrcReg; 462 Opc = DefMI->getOpcode(); 463 } 464 return DefinitionAndSourceRegister{DefMI, DefSrcReg}; 465 } 466 467 MachineInstr *llvm::getDefIgnoringCopies(Register Reg, 468 const MachineRegisterInfo &MRI) { 469 std::optional<DefinitionAndSourceRegister> DefSrcReg = 470 getDefSrcRegIgnoringCopies(Reg, MRI); 471 return DefSrcReg ? DefSrcReg->MI : nullptr; 472 } 473 474 Register llvm::getSrcRegIgnoringCopies(Register Reg, 475 const MachineRegisterInfo &MRI) { 476 std::optional<DefinitionAndSourceRegister> DefSrcReg = 477 getDefSrcRegIgnoringCopies(Reg, MRI); 478 return DefSrcReg ? DefSrcReg->Reg : Register(); 479 } 480 481 MachineInstr *llvm::getOpcodeDef(unsigned Opcode, Register Reg, 482 const MachineRegisterInfo &MRI) { 483 MachineInstr *DefMI = getDefIgnoringCopies(Reg, MRI); 484 return DefMI && DefMI->getOpcode() == Opcode ? DefMI : nullptr; 485 } 486 487 APFloat llvm::getAPFloatFromSize(double Val, unsigned Size) { 488 if (Size == 32) 489 return APFloat(float(Val)); 490 if (Size == 64) 491 return APFloat(Val); 492 if (Size != 16) 493 llvm_unreachable("Unsupported FPConstant size"); 494 bool Ignored; 495 APFloat APF(Val); 496 APF.convert(APFloat::IEEEhalf(), APFloat::rmNearestTiesToEven, &Ignored); 497 return APF; 498 } 499 500 std::optional<APInt> llvm::ConstantFoldBinOp(unsigned Opcode, 501 const Register Op1, 502 const Register Op2, 503 const MachineRegisterInfo &MRI) { 504 auto MaybeOp2Cst = getAnyConstantVRegValWithLookThrough(Op2, MRI, false); 505 if (!MaybeOp2Cst) 506 return std::nullopt; 507 508 auto MaybeOp1Cst = getAnyConstantVRegValWithLookThrough(Op1, MRI, false); 509 if (!MaybeOp1Cst) 510 return std::nullopt; 511 512 const APInt &C1 = MaybeOp1Cst->Value; 513 const APInt &C2 = MaybeOp2Cst->Value; 514 switch (Opcode) { 515 default: 516 break; 517 case TargetOpcode::G_ADD: 518 case TargetOpcode::G_PTR_ADD: 519 return C1 + C2; 520 case TargetOpcode::G_AND: 521 return C1 & C2; 522 case TargetOpcode::G_ASHR: 523 return C1.ashr(C2); 524 case TargetOpcode::G_LSHR: 525 return C1.lshr(C2); 526 case TargetOpcode::G_MUL: 527 return C1 * C2; 528 case TargetOpcode::G_OR: 529 return C1 | C2; 530 case TargetOpcode::G_SHL: 531 return C1 << C2; 532 case TargetOpcode::G_SUB: 533 return C1 - C2; 534 case TargetOpcode::G_XOR: 535 return C1 ^ C2; 536 case TargetOpcode::G_UDIV: 537 if (!C2.getBoolValue()) 538 break; 539 return C1.udiv(C2); 540 case TargetOpcode::G_SDIV: 541 if (!C2.getBoolValue()) 542 break; 543 return C1.sdiv(C2); 544 case TargetOpcode::G_UREM: 545 if (!C2.getBoolValue()) 546 break; 547 return C1.urem(C2); 548 case TargetOpcode::G_SREM: 549 if (!C2.getBoolValue()) 550 break; 551 return C1.srem(C2); 552 case TargetOpcode::G_SMIN: 553 return APIntOps::smin(C1, C2); 554 case TargetOpcode::G_SMAX: 555 return APIntOps::smax(C1, C2); 556 case TargetOpcode::G_UMIN: 557 return APIntOps::umin(C1, C2); 558 case TargetOpcode::G_UMAX: 559 return APIntOps::umax(C1, C2); 560 } 561 562 return std::nullopt; 563 } 564 565 std::optional<APFloat> 566 llvm::ConstantFoldFPBinOp(unsigned Opcode, const Register Op1, 567 const Register Op2, const MachineRegisterInfo &MRI) { 568 const ConstantFP *Op2Cst = getConstantFPVRegVal(Op2, MRI); 569 if (!Op2Cst) 570 return std::nullopt; 571 572 const ConstantFP *Op1Cst = getConstantFPVRegVal(Op1, MRI); 573 if (!Op1Cst) 574 return std::nullopt; 575 576 APFloat C1 = Op1Cst->getValueAPF(); 577 const APFloat &C2 = Op2Cst->getValueAPF(); 578 switch (Opcode) { 579 case TargetOpcode::G_FADD: 580 C1.add(C2, APFloat::rmNearestTiesToEven); 581 return C1; 582 case TargetOpcode::G_FSUB: 583 C1.subtract(C2, APFloat::rmNearestTiesToEven); 584 return C1; 585 case TargetOpcode::G_FMUL: 586 C1.multiply(C2, APFloat::rmNearestTiesToEven); 587 return C1; 588 case TargetOpcode::G_FDIV: 589 C1.divide(C2, APFloat::rmNearestTiesToEven); 590 return C1; 591 case TargetOpcode::G_FREM: 592 C1.mod(C2); 593 return C1; 594 case TargetOpcode::G_FCOPYSIGN: 595 C1.copySign(C2); 596 return C1; 597 case TargetOpcode::G_FMINNUM: 598 return minnum(C1, C2); 599 case TargetOpcode::G_FMAXNUM: 600 return maxnum(C1, C2); 601 case TargetOpcode::G_FMINIMUM: 602 return minimum(C1, C2); 603 case TargetOpcode::G_FMAXIMUM: 604 return maximum(C1, C2); 605 case TargetOpcode::G_FMINNUM_IEEE: 606 case TargetOpcode::G_FMAXNUM_IEEE: 607 // FIXME: These operations were unfortunately named. fminnum/fmaxnum do not 608 // follow the IEEE behavior for signaling nans and follow libm's fmin/fmax, 609 // and currently there isn't a nice wrapper in APFloat for the version with 610 // correct snan handling. 611 break; 612 default: 613 break; 614 } 615 616 return std::nullopt; 617 } 618 619 SmallVector<APInt> 620 llvm::ConstantFoldVectorBinop(unsigned Opcode, const Register Op1, 621 const Register Op2, 622 const MachineRegisterInfo &MRI) { 623 auto *SrcVec2 = getOpcodeDef<GBuildVector>(Op2, MRI); 624 if (!SrcVec2) 625 return SmallVector<APInt>(); 626 627 auto *SrcVec1 = getOpcodeDef<GBuildVector>(Op1, MRI); 628 if (!SrcVec1) 629 return SmallVector<APInt>(); 630 631 SmallVector<APInt> FoldedElements; 632 for (unsigned Idx = 0, E = SrcVec1->getNumSources(); Idx < E; ++Idx) { 633 auto MaybeCst = ConstantFoldBinOp(Opcode, SrcVec1->getSourceReg(Idx), 634 SrcVec2->getSourceReg(Idx), MRI); 635 if (!MaybeCst) 636 return SmallVector<APInt>(); 637 FoldedElements.push_back(*MaybeCst); 638 } 639 return FoldedElements; 640 } 641 642 bool llvm::isKnownNeverNaN(Register Val, const MachineRegisterInfo &MRI, 643 bool SNaN) { 644 const MachineInstr *DefMI = MRI.getVRegDef(Val); 645 if (!DefMI) 646 return false; 647 648 const TargetMachine& TM = DefMI->getMF()->getTarget(); 649 if (DefMI->getFlag(MachineInstr::FmNoNans) || TM.Options.NoNaNsFPMath) 650 return true; 651 652 // If the value is a constant, we can obviously see if it is a NaN or not. 653 if (const ConstantFP *FPVal = getConstantFPVRegVal(Val, MRI)) { 654 return !FPVal->getValueAPF().isNaN() || 655 (SNaN && !FPVal->getValueAPF().isSignaling()); 656 } 657 658 if (DefMI->getOpcode() == TargetOpcode::G_BUILD_VECTOR) { 659 for (const auto &Op : DefMI->uses()) 660 if (!isKnownNeverNaN(Op.getReg(), MRI, SNaN)) 661 return false; 662 return true; 663 } 664 665 switch (DefMI->getOpcode()) { 666 default: 667 break; 668 case TargetOpcode::G_FADD: 669 case TargetOpcode::G_FSUB: 670 case TargetOpcode::G_FMUL: 671 case TargetOpcode::G_FDIV: 672 case TargetOpcode::G_FREM: 673 case TargetOpcode::G_FSIN: 674 case TargetOpcode::G_FCOS: 675 case TargetOpcode::G_FMA: 676 case TargetOpcode::G_FMAD: 677 if (SNaN) 678 return true; 679 680 // TODO: Need isKnownNeverInfinity 681 return false; 682 case TargetOpcode::G_FMINNUM_IEEE: 683 case TargetOpcode::G_FMAXNUM_IEEE: { 684 if (SNaN) 685 return true; 686 // This can return a NaN if either operand is an sNaN, or if both operands 687 // are NaN. 688 return (isKnownNeverNaN(DefMI->getOperand(1).getReg(), MRI) && 689 isKnownNeverSNaN(DefMI->getOperand(2).getReg(), MRI)) || 690 (isKnownNeverSNaN(DefMI->getOperand(1).getReg(), MRI) && 691 isKnownNeverNaN(DefMI->getOperand(2).getReg(), MRI)); 692 } 693 case TargetOpcode::G_FMINNUM: 694 case TargetOpcode::G_FMAXNUM: { 695 // Only one needs to be known not-nan, since it will be returned if the 696 // other ends up being one. 697 return isKnownNeverNaN(DefMI->getOperand(1).getReg(), MRI, SNaN) || 698 isKnownNeverNaN(DefMI->getOperand(2).getReg(), MRI, SNaN); 699 } 700 } 701 702 if (SNaN) { 703 // FP operations quiet. For now, just handle the ones inserted during 704 // legalization. 705 switch (DefMI->getOpcode()) { 706 case TargetOpcode::G_FPEXT: 707 case TargetOpcode::G_FPTRUNC: 708 case TargetOpcode::G_FCANONICALIZE: 709 return true; 710 default: 711 return false; 712 } 713 } 714 715 return false; 716 } 717 718 Align llvm::inferAlignFromPtrInfo(MachineFunction &MF, 719 const MachinePointerInfo &MPO) { 720 auto PSV = dyn_cast_if_present<const PseudoSourceValue *>(MPO.V); 721 if (auto FSPV = dyn_cast_or_null<FixedStackPseudoSourceValue>(PSV)) { 722 MachineFrameInfo &MFI = MF.getFrameInfo(); 723 return commonAlignment(MFI.getObjectAlign(FSPV->getFrameIndex()), 724 MPO.Offset); 725 } 726 727 if (const Value *V = dyn_cast_if_present<const Value *>(MPO.V)) { 728 const Module *M = MF.getFunction().getParent(); 729 return V->getPointerAlignment(M->getDataLayout()); 730 } 731 732 return Align(1); 733 } 734 735 Register llvm::getFunctionLiveInPhysReg(MachineFunction &MF, 736 const TargetInstrInfo &TII, 737 MCRegister PhysReg, 738 const TargetRegisterClass &RC, 739 const DebugLoc &DL, LLT RegTy) { 740 MachineBasicBlock &EntryMBB = MF.front(); 741 MachineRegisterInfo &MRI = MF.getRegInfo(); 742 Register LiveIn = MRI.getLiveInVirtReg(PhysReg); 743 if (LiveIn) { 744 MachineInstr *Def = MRI.getVRegDef(LiveIn); 745 if (Def) { 746 // FIXME: Should the verifier check this is in the entry block? 747 assert(Def->getParent() == &EntryMBB && "live-in copy not in entry block"); 748 return LiveIn; 749 } 750 751 // It's possible the incoming argument register and copy was added during 752 // lowering, but later deleted due to being/becoming dead. If this happens, 753 // re-insert the copy. 754 } else { 755 // The live in register was not present, so add it. 756 LiveIn = MF.addLiveIn(PhysReg, &RC); 757 if (RegTy.isValid()) 758 MRI.setType(LiveIn, RegTy); 759 } 760 761 BuildMI(EntryMBB, EntryMBB.begin(), DL, TII.get(TargetOpcode::COPY), LiveIn) 762 .addReg(PhysReg); 763 if (!EntryMBB.isLiveIn(PhysReg)) 764 EntryMBB.addLiveIn(PhysReg); 765 return LiveIn; 766 } 767 768 std::optional<APInt> llvm::ConstantFoldExtOp(unsigned Opcode, 769 const Register Op1, uint64_t Imm, 770 const MachineRegisterInfo &MRI) { 771 auto MaybeOp1Cst = getIConstantVRegVal(Op1, MRI); 772 if (MaybeOp1Cst) { 773 switch (Opcode) { 774 default: 775 break; 776 case TargetOpcode::G_SEXT_INREG: { 777 LLT Ty = MRI.getType(Op1); 778 return MaybeOp1Cst->trunc(Imm).sext(Ty.getScalarSizeInBits()); 779 } 780 } 781 } 782 return std::nullopt; 783 } 784 785 std::optional<APInt> llvm::ConstantFoldCastOp(unsigned Opcode, LLT DstTy, 786 const Register Op0, 787 const MachineRegisterInfo &MRI) { 788 std::optional<APInt> Val = getIConstantVRegVal(Op0, MRI); 789 if (!Val) 790 return Val; 791 792 const unsigned DstSize = DstTy.getScalarSizeInBits(); 793 794 switch (Opcode) { 795 case TargetOpcode::G_SEXT: 796 return Val->sext(DstSize); 797 case TargetOpcode::G_ZEXT: 798 case TargetOpcode::G_ANYEXT: 799 // TODO: DAG considers target preference when constant folding any_extend. 800 return Val->zext(DstSize); 801 default: 802 break; 803 } 804 805 llvm_unreachable("unexpected cast opcode to constant fold"); 806 } 807 808 std::optional<APFloat> 809 llvm::ConstantFoldIntToFloat(unsigned Opcode, LLT DstTy, Register Src, 810 const MachineRegisterInfo &MRI) { 811 assert(Opcode == TargetOpcode::G_SITOFP || Opcode == TargetOpcode::G_UITOFP); 812 if (auto MaybeSrcVal = getIConstantVRegVal(Src, MRI)) { 813 APFloat DstVal(getFltSemanticForLLT(DstTy)); 814 DstVal.convertFromAPInt(*MaybeSrcVal, Opcode == TargetOpcode::G_SITOFP, 815 APFloat::rmNearestTiesToEven); 816 return DstVal; 817 } 818 return std::nullopt; 819 } 820 821 std::optional<SmallVector<unsigned>> 822 llvm::ConstantFoldCTLZ(Register Src, const MachineRegisterInfo &MRI) { 823 LLT Ty = MRI.getType(Src); 824 SmallVector<unsigned> FoldedCTLZs; 825 auto tryFoldScalar = [&](Register R) -> std::optional<unsigned> { 826 auto MaybeCst = getIConstantVRegVal(R, MRI); 827 if (!MaybeCst) 828 return std::nullopt; 829 return MaybeCst->countl_zero(); 830 }; 831 if (Ty.isVector()) { 832 // Try to constant fold each element. 833 auto *BV = getOpcodeDef<GBuildVector>(Src, MRI); 834 if (!BV) 835 return std::nullopt; 836 for (unsigned SrcIdx = 0; SrcIdx < BV->getNumSources(); ++SrcIdx) { 837 if (auto MaybeFold = tryFoldScalar(BV->getSourceReg(SrcIdx))) { 838 FoldedCTLZs.emplace_back(*MaybeFold); 839 continue; 840 } 841 return std::nullopt; 842 } 843 return FoldedCTLZs; 844 } 845 if (auto MaybeCst = tryFoldScalar(Src)) { 846 FoldedCTLZs.emplace_back(*MaybeCst); 847 return FoldedCTLZs; 848 } 849 return std::nullopt; 850 } 851 852 bool llvm::isKnownToBeAPowerOfTwo(Register Reg, const MachineRegisterInfo &MRI, 853 GISelKnownBits *KB) { 854 std::optional<DefinitionAndSourceRegister> DefSrcReg = 855 getDefSrcRegIgnoringCopies(Reg, MRI); 856 if (!DefSrcReg) 857 return false; 858 859 const MachineInstr &MI = *DefSrcReg->MI; 860 const LLT Ty = MRI.getType(Reg); 861 862 switch (MI.getOpcode()) { 863 case TargetOpcode::G_CONSTANT: { 864 unsigned BitWidth = Ty.getScalarSizeInBits(); 865 const ConstantInt *CI = MI.getOperand(1).getCImm(); 866 return CI->getValue().zextOrTrunc(BitWidth).isPowerOf2(); 867 } 868 case TargetOpcode::G_SHL: { 869 // A left-shift of a constant one will have exactly one bit set because 870 // shifting the bit off the end is undefined. 871 872 // TODO: Constant splat 873 if (auto ConstLHS = getIConstantVRegVal(MI.getOperand(1).getReg(), MRI)) { 874 if (*ConstLHS == 1) 875 return true; 876 } 877 878 break; 879 } 880 case TargetOpcode::G_LSHR: { 881 if (auto ConstLHS = getIConstantVRegVal(MI.getOperand(1).getReg(), MRI)) { 882 if (ConstLHS->isSignMask()) 883 return true; 884 } 885 886 break; 887 } 888 case TargetOpcode::G_BUILD_VECTOR: { 889 // TODO: Probably should have a recursion depth guard since you could have 890 // bitcasted vector elements. 891 for (const MachineOperand &MO : llvm::drop_begin(MI.operands())) 892 if (!isKnownToBeAPowerOfTwo(MO.getReg(), MRI, KB)) 893 return false; 894 895 return true; 896 } 897 case TargetOpcode::G_BUILD_VECTOR_TRUNC: { 898 // Only handle constants since we would need to know if number of leading 899 // zeros is greater than the truncation amount. 900 const unsigned BitWidth = Ty.getScalarSizeInBits(); 901 for (const MachineOperand &MO : llvm::drop_begin(MI.operands())) { 902 auto Const = getIConstantVRegVal(MO.getReg(), MRI); 903 if (!Const || !Const->zextOrTrunc(BitWidth).isPowerOf2()) 904 return false; 905 } 906 907 return true; 908 } 909 default: 910 break; 911 } 912 913 if (!KB) 914 return false; 915 916 // More could be done here, though the above checks are enough 917 // to handle some common cases. 918 919 // Fall back to computeKnownBits to catch other known cases. 920 KnownBits Known = KB->getKnownBits(Reg); 921 return (Known.countMaxPopulation() == 1) && (Known.countMinPopulation() == 1); 922 } 923 924 void llvm::getSelectionDAGFallbackAnalysisUsage(AnalysisUsage &AU) { 925 AU.addPreserved<StackProtector>(); 926 } 927 928 LLT llvm::getLCMType(LLT OrigTy, LLT TargetTy) { 929 const unsigned OrigSize = OrigTy.getSizeInBits(); 930 const unsigned TargetSize = TargetTy.getSizeInBits(); 931 932 if (OrigSize == TargetSize) 933 return OrigTy; 934 935 if (OrigTy.isVector()) { 936 const LLT OrigElt = OrigTy.getElementType(); 937 938 if (TargetTy.isVector()) { 939 const LLT TargetElt = TargetTy.getElementType(); 940 941 if (OrigElt.getSizeInBits() == TargetElt.getSizeInBits()) { 942 int GCDElts = 943 std::gcd(OrigTy.getNumElements(), TargetTy.getNumElements()); 944 // Prefer the original element type. 945 ElementCount Mul = OrigTy.getElementCount() * TargetTy.getNumElements(); 946 return LLT::vector(Mul.divideCoefficientBy(GCDElts), 947 OrigTy.getElementType()); 948 } 949 } else { 950 if (OrigElt.getSizeInBits() == TargetSize) 951 return OrigTy; 952 } 953 954 unsigned LCMSize = std::lcm(OrigSize, TargetSize); 955 return LLT::fixed_vector(LCMSize / OrigElt.getSizeInBits(), OrigElt); 956 } 957 958 if (TargetTy.isVector()) { 959 unsigned LCMSize = std::lcm(OrigSize, TargetSize); 960 return LLT::fixed_vector(LCMSize / OrigSize, OrigTy); 961 } 962 963 unsigned LCMSize = std::lcm(OrigSize, TargetSize); 964 965 // Preserve pointer types. 966 if (LCMSize == OrigSize) 967 return OrigTy; 968 if (LCMSize == TargetSize) 969 return TargetTy; 970 971 return LLT::scalar(LCMSize); 972 } 973 974 LLT llvm::getCoverTy(LLT OrigTy, LLT TargetTy) { 975 if (!OrigTy.isVector() || !TargetTy.isVector() || OrigTy == TargetTy || 976 (OrigTy.getScalarSizeInBits() != TargetTy.getScalarSizeInBits())) 977 return getLCMType(OrigTy, TargetTy); 978 979 unsigned OrigTyNumElts = OrigTy.getNumElements(); 980 unsigned TargetTyNumElts = TargetTy.getNumElements(); 981 if (OrigTyNumElts % TargetTyNumElts == 0) 982 return OrigTy; 983 984 unsigned NumElts = alignTo(OrigTyNumElts, TargetTyNumElts); 985 return LLT::scalarOrVector(ElementCount::getFixed(NumElts), 986 OrigTy.getElementType()); 987 } 988 989 LLT llvm::getGCDType(LLT OrigTy, LLT TargetTy) { 990 const unsigned OrigSize = OrigTy.getSizeInBits(); 991 const unsigned TargetSize = TargetTy.getSizeInBits(); 992 993 if (OrigSize == TargetSize) 994 return OrigTy; 995 996 if (OrigTy.isVector()) { 997 LLT OrigElt = OrigTy.getElementType(); 998 if (TargetTy.isVector()) { 999 LLT TargetElt = TargetTy.getElementType(); 1000 if (OrigElt.getSizeInBits() == TargetElt.getSizeInBits()) { 1001 int GCD = std::gcd(OrigTy.getNumElements(), TargetTy.getNumElements()); 1002 return LLT::scalarOrVector(ElementCount::getFixed(GCD), OrigElt); 1003 } 1004 } else { 1005 // If the source is a vector of pointers, return a pointer element. 1006 if (OrigElt.getSizeInBits() == TargetSize) 1007 return OrigElt; 1008 } 1009 1010 unsigned GCD = std::gcd(OrigSize, TargetSize); 1011 if (GCD == OrigElt.getSizeInBits()) 1012 return OrigElt; 1013 1014 // If we can't produce the original element type, we have to use a smaller 1015 // scalar. 1016 if (GCD < OrigElt.getSizeInBits()) 1017 return LLT::scalar(GCD); 1018 return LLT::fixed_vector(GCD / OrigElt.getSizeInBits(), OrigElt); 1019 } 1020 1021 if (TargetTy.isVector()) { 1022 // Try to preserve the original element type. 1023 LLT TargetElt = TargetTy.getElementType(); 1024 if (TargetElt.getSizeInBits() == OrigSize) 1025 return OrigTy; 1026 } 1027 1028 unsigned GCD = std::gcd(OrigSize, TargetSize); 1029 return LLT::scalar(GCD); 1030 } 1031 1032 std::optional<int> llvm::getSplatIndex(MachineInstr &MI) { 1033 assert(MI.getOpcode() == TargetOpcode::G_SHUFFLE_VECTOR && 1034 "Only G_SHUFFLE_VECTOR can have a splat index!"); 1035 ArrayRef<int> Mask = MI.getOperand(3).getShuffleMask(); 1036 auto FirstDefinedIdx = find_if(Mask, [](int Elt) { return Elt >= 0; }); 1037 1038 // If all elements are undefined, this shuffle can be considered a splat. 1039 // Return 0 for better potential for callers to simplify. 1040 if (FirstDefinedIdx == Mask.end()) 1041 return 0; 1042 1043 // Make sure all remaining elements are either undef or the same 1044 // as the first non-undef value. 1045 int SplatValue = *FirstDefinedIdx; 1046 if (any_of(make_range(std::next(FirstDefinedIdx), Mask.end()), 1047 [&SplatValue](int Elt) { return Elt >= 0 && Elt != SplatValue; })) 1048 return std::nullopt; 1049 1050 return SplatValue; 1051 } 1052 1053 static bool isBuildVectorOp(unsigned Opcode) { 1054 return Opcode == TargetOpcode::G_BUILD_VECTOR || 1055 Opcode == TargetOpcode::G_BUILD_VECTOR_TRUNC; 1056 } 1057 1058 namespace { 1059 1060 std::optional<ValueAndVReg> getAnyConstantSplat(Register VReg, 1061 const MachineRegisterInfo &MRI, 1062 bool AllowUndef) { 1063 MachineInstr *MI = getDefIgnoringCopies(VReg, MRI); 1064 if (!MI) 1065 return std::nullopt; 1066 1067 bool isConcatVectorsOp = MI->getOpcode() == TargetOpcode::G_CONCAT_VECTORS; 1068 if (!isBuildVectorOp(MI->getOpcode()) && !isConcatVectorsOp) 1069 return std::nullopt; 1070 1071 std::optional<ValueAndVReg> SplatValAndReg; 1072 for (MachineOperand &Op : MI->uses()) { 1073 Register Element = Op.getReg(); 1074 // If we have a G_CONCAT_VECTOR, we recursively look into the 1075 // vectors that we're concatenating to see if they're splats. 1076 auto ElementValAndReg = 1077 isConcatVectorsOp 1078 ? getAnyConstantSplat(Element, MRI, AllowUndef) 1079 : getAnyConstantVRegValWithLookThrough(Element, MRI, true, true); 1080 1081 // If AllowUndef, treat undef as value that will result in a constant splat. 1082 if (!ElementValAndReg) { 1083 if (AllowUndef && isa<GImplicitDef>(MRI.getVRegDef(Element))) 1084 continue; 1085 return std::nullopt; 1086 } 1087 1088 // Record splat value 1089 if (!SplatValAndReg) 1090 SplatValAndReg = ElementValAndReg; 1091 1092 // Different constant than the one already recorded, not a constant splat. 1093 if (SplatValAndReg->Value != ElementValAndReg->Value) 1094 return std::nullopt; 1095 } 1096 1097 return SplatValAndReg; 1098 } 1099 1100 } // end anonymous namespace 1101 1102 bool llvm::isBuildVectorConstantSplat(const Register Reg, 1103 const MachineRegisterInfo &MRI, 1104 int64_t SplatValue, bool AllowUndef) { 1105 if (auto SplatValAndReg = getAnyConstantSplat(Reg, MRI, AllowUndef)) 1106 return mi_match(SplatValAndReg->VReg, MRI, m_SpecificICst(SplatValue)); 1107 return false; 1108 } 1109 1110 bool llvm::isBuildVectorConstantSplat(const MachineInstr &MI, 1111 const MachineRegisterInfo &MRI, 1112 int64_t SplatValue, bool AllowUndef) { 1113 return isBuildVectorConstantSplat(MI.getOperand(0).getReg(), MRI, SplatValue, 1114 AllowUndef); 1115 } 1116 1117 std::optional<APInt> 1118 llvm::getIConstantSplatVal(const Register Reg, const MachineRegisterInfo &MRI) { 1119 if (auto SplatValAndReg = 1120 getAnyConstantSplat(Reg, MRI, /* AllowUndef */ false)) { 1121 std::optional<ValueAndVReg> ValAndVReg = 1122 getIConstantVRegValWithLookThrough(SplatValAndReg->VReg, MRI); 1123 return ValAndVReg->Value; 1124 } 1125 1126 return std::nullopt; 1127 } 1128 1129 std::optional<APInt> 1130 llvm::getIConstantSplatVal(const MachineInstr &MI, 1131 const MachineRegisterInfo &MRI) { 1132 return getIConstantSplatVal(MI.getOperand(0).getReg(), MRI); 1133 } 1134 1135 std::optional<int64_t> 1136 llvm::getIConstantSplatSExtVal(const Register Reg, 1137 const MachineRegisterInfo &MRI) { 1138 if (auto SplatValAndReg = 1139 getAnyConstantSplat(Reg, MRI, /* AllowUndef */ false)) 1140 return getIConstantVRegSExtVal(SplatValAndReg->VReg, MRI); 1141 return std::nullopt; 1142 } 1143 1144 std::optional<int64_t> 1145 llvm::getIConstantSplatSExtVal(const MachineInstr &MI, 1146 const MachineRegisterInfo &MRI) { 1147 return getIConstantSplatSExtVal(MI.getOperand(0).getReg(), MRI); 1148 } 1149 1150 std::optional<FPValueAndVReg> 1151 llvm::getFConstantSplat(Register VReg, const MachineRegisterInfo &MRI, 1152 bool AllowUndef) { 1153 if (auto SplatValAndReg = getAnyConstantSplat(VReg, MRI, AllowUndef)) 1154 return getFConstantVRegValWithLookThrough(SplatValAndReg->VReg, MRI); 1155 return std::nullopt; 1156 } 1157 1158 bool llvm::isBuildVectorAllZeros(const MachineInstr &MI, 1159 const MachineRegisterInfo &MRI, 1160 bool AllowUndef) { 1161 return isBuildVectorConstantSplat(MI, MRI, 0, AllowUndef); 1162 } 1163 1164 bool llvm::isBuildVectorAllOnes(const MachineInstr &MI, 1165 const MachineRegisterInfo &MRI, 1166 bool AllowUndef) { 1167 return isBuildVectorConstantSplat(MI, MRI, -1, AllowUndef); 1168 } 1169 1170 std::optional<RegOrConstant> 1171 llvm::getVectorSplat(const MachineInstr &MI, const MachineRegisterInfo &MRI) { 1172 unsigned Opc = MI.getOpcode(); 1173 if (!isBuildVectorOp(Opc)) 1174 return std::nullopt; 1175 if (auto Splat = getIConstantSplatSExtVal(MI, MRI)) 1176 return RegOrConstant(*Splat); 1177 auto Reg = MI.getOperand(1).getReg(); 1178 if (any_of(drop_begin(MI.operands(), 2), 1179 [&Reg](const MachineOperand &Op) { return Op.getReg() != Reg; })) 1180 return std::nullopt; 1181 return RegOrConstant(Reg); 1182 } 1183 1184 static bool isConstantScalar(const MachineInstr &MI, 1185 const MachineRegisterInfo &MRI, 1186 bool AllowFP = true, 1187 bool AllowOpaqueConstants = true) { 1188 switch (MI.getOpcode()) { 1189 case TargetOpcode::G_CONSTANT: 1190 case TargetOpcode::G_IMPLICIT_DEF: 1191 return true; 1192 case TargetOpcode::G_FCONSTANT: 1193 return AllowFP; 1194 case TargetOpcode::G_GLOBAL_VALUE: 1195 case TargetOpcode::G_FRAME_INDEX: 1196 case TargetOpcode::G_BLOCK_ADDR: 1197 case TargetOpcode::G_JUMP_TABLE: 1198 return AllowOpaqueConstants; 1199 default: 1200 return false; 1201 } 1202 } 1203 1204 bool llvm::isConstantOrConstantVector(MachineInstr &MI, 1205 const MachineRegisterInfo &MRI) { 1206 Register Def = MI.getOperand(0).getReg(); 1207 if (auto C = getIConstantVRegValWithLookThrough(Def, MRI)) 1208 return true; 1209 GBuildVector *BV = dyn_cast<GBuildVector>(&MI); 1210 if (!BV) 1211 return false; 1212 for (unsigned SrcIdx = 0; SrcIdx < BV->getNumSources(); ++SrcIdx) { 1213 if (getIConstantVRegValWithLookThrough(BV->getSourceReg(SrcIdx), MRI) || 1214 getOpcodeDef<GImplicitDef>(BV->getSourceReg(SrcIdx), MRI)) 1215 continue; 1216 return false; 1217 } 1218 return true; 1219 } 1220 1221 bool llvm::isConstantOrConstantVector(const MachineInstr &MI, 1222 const MachineRegisterInfo &MRI, 1223 bool AllowFP, bool AllowOpaqueConstants) { 1224 if (isConstantScalar(MI, MRI, AllowFP, AllowOpaqueConstants)) 1225 return true; 1226 1227 if (!isBuildVectorOp(MI.getOpcode())) 1228 return false; 1229 1230 const unsigned NumOps = MI.getNumOperands(); 1231 for (unsigned I = 1; I != NumOps; ++I) { 1232 const MachineInstr *ElementDef = MRI.getVRegDef(MI.getOperand(I).getReg()); 1233 if (!isConstantScalar(*ElementDef, MRI, AllowFP, AllowOpaqueConstants)) 1234 return false; 1235 } 1236 1237 return true; 1238 } 1239 1240 std::optional<APInt> 1241 llvm::isConstantOrConstantSplatVector(MachineInstr &MI, 1242 const MachineRegisterInfo &MRI) { 1243 Register Def = MI.getOperand(0).getReg(); 1244 if (auto C = getIConstantVRegValWithLookThrough(Def, MRI)) 1245 return C->Value; 1246 auto MaybeCst = getIConstantSplatSExtVal(MI, MRI); 1247 if (!MaybeCst) 1248 return std::nullopt; 1249 const unsigned ScalarSize = MRI.getType(Def).getScalarSizeInBits(); 1250 return APInt(ScalarSize, *MaybeCst, true); 1251 } 1252 1253 bool llvm::isNullOrNullSplat(const MachineInstr &MI, 1254 const MachineRegisterInfo &MRI, bool AllowUndefs) { 1255 switch (MI.getOpcode()) { 1256 case TargetOpcode::G_IMPLICIT_DEF: 1257 return AllowUndefs; 1258 case TargetOpcode::G_CONSTANT: 1259 return MI.getOperand(1).getCImm()->isNullValue(); 1260 case TargetOpcode::G_FCONSTANT: { 1261 const ConstantFP *FPImm = MI.getOperand(1).getFPImm(); 1262 return FPImm->isZero() && !FPImm->isNegative(); 1263 } 1264 default: 1265 if (!AllowUndefs) // TODO: isBuildVectorAllZeros assumes undef is OK already 1266 return false; 1267 return isBuildVectorAllZeros(MI, MRI); 1268 } 1269 } 1270 1271 bool llvm::isAllOnesOrAllOnesSplat(const MachineInstr &MI, 1272 const MachineRegisterInfo &MRI, 1273 bool AllowUndefs) { 1274 switch (MI.getOpcode()) { 1275 case TargetOpcode::G_IMPLICIT_DEF: 1276 return AllowUndefs; 1277 case TargetOpcode::G_CONSTANT: 1278 return MI.getOperand(1).getCImm()->isAllOnesValue(); 1279 default: 1280 if (!AllowUndefs) // TODO: isBuildVectorAllOnes assumes undef is OK already 1281 return false; 1282 return isBuildVectorAllOnes(MI, MRI); 1283 } 1284 } 1285 1286 bool llvm::matchUnaryPredicate( 1287 const MachineRegisterInfo &MRI, Register Reg, 1288 std::function<bool(const Constant *ConstVal)> Match, bool AllowUndefs) { 1289 1290 const MachineInstr *Def = getDefIgnoringCopies(Reg, MRI); 1291 if (AllowUndefs && Def->getOpcode() == TargetOpcode::G_IMPLICIT_DEF) 1292 return Match(nullptr); 1293 1294 // TODO: Also handle fconstant 1295 if (Def->getOpcode() == TargetOpcode::G_CONSTANT) 1296 return Match(Def->getOperand(1).getCImm()); 1297 1298 if (Def->getOpcode() != TargetOpcode::G_BUILD_VECTOR) 1299 return false; 1300 1301 for (unsigned I = 1, E = Def->getNumOperands(); I != E; ++I) { 1302 Register SrcElt = Def->getOperand(I).getReg(); 1303 const MachineInstr *SrcDef = getDefIgnoringCopies(SrcElt, MRI); 1304 if (AllowUndefs && SrcDef->getOpcode() == TargetOpcode::G_IMPLICIT_DEF) { 1305 if (!Match(nullptr)) 1306 return false; 1307 continue; 1308 } 1309 1310 if (SrcDef->getOpcode() != TargetOpcode::G_CONSTANT || 1311 !Match(SrcDef->getOperand(1).getCImm())) 1312 return false; 1313 } 1314 1315 return true; 1316 } 1317 1318 bool llvm::isConstTrueVal(const TargetLowering &TLI, int64_t Val, bool IsVector, 1319 bool IsFP) { 1320 switch (TLI.getBooleanContents(IsVector, IsFP)) { 1321 case TargetLowering::UndefinedBooleanContent: 1322 return Val & 0x1; 1323 case TargetLowering::ZeroOrOneBooleanContent: 1324 return Val == 1; 1325 case TargetLowering::ZeroOrNegativeOneBooleanContent: 1326 return Val == -1; 1327 } 1328 llvm_unreachable("Invalid boolean contents"); 1329 } 1330 1331 bool llvm::isConstFalseVal(const TargetLowering &TLI, int64_t Val, 1332 bool IsVector, bool IsFP) { 1333 switch (TLI.getBooleanContents(IsVector, IsFP)) { 1334 case TargetLowering::UndefinedBooleanContent: 1335 return ~Val & 0x1; 1336 case TargetLowering::ZeroOrOneBooleanContent: 1337 case TargetLowering::ZeroOrNegativeOneBooleanContent: 1338 return Val == 0; 1339 } 1340 llvm_unreachable("Invalid boolean contents"); 1341 } 1342 1343 int64_t llvm::getICmpTrueVal(const TargetLowering &TLI, bool IsVector, 1344 bool IsFP) { 1345 switch (TLI.getBooleanContents(IsVector, IsFP)) { 1346 case TargetLowering::UndefinedBooleanContent: 1347 case TargetLowering::ZeroOrOneBooleanContent: 1348 return 1; 1349 case TargetLowering::ZeroOrNegativeOneBooleanContent: 1350 return -1; 1351 } 1352 llvm_unreachable("Invalid boolean contents"); 1353 } 1354 1355 bool llvm::shouldOptForSize(const MachineBasicBlock &MBB, 1356 ProfileSummaryInfo *PSI, BlockFrequencyInfo *BFI) { 1357 const auto &F = MBB.getParent()->getFunction(); 1358 return F.hasOptSize() || F.hasMinSize() || 1359 llvm::shouldOptimizeForSize(MBB.getBasicBlock(), PSI, BFI); 1360 } 1361 1362 void llvm::saveUsesAndErase(MachineInstr &MI, MachineRegisterInfo &MRI, 1363 LostDebugLocObserver *LocObserver, 1364 SmallInstListTy &DeadInstChain) { 1365 for (MachineOperand &Op : MI.uses()) { 1366 if (Op.isReg() && Op.getReg().isVirtual()) 1367 DeadInstChain.insert(MRI.getVRegDef(Op.getReg())); 1368 } 1369 LLVM_DEBUG(dbgs() << MI << "Is dead; erasing.\n"); 1370 DeadInstChain.remove(&MI); 1371 MI.eraseFromParent(); 1372 if (LocObserver) 1373 LocObserver->checkpoint(false); 1374 } 1375 1376 void llvm::eraseInstrs(ArrayRef<MachineInstr *> DeadInstrs, 1377 MachineRegisterInfo &MRI, 1378 LostDebugLocObserver *LocObserver) { 1379 SmallInstListTy DeadInstChain; 1380 for (MachineInstr *MI : DeadInstrs) 1381 saveUsesAndErase(*MI, MRI, LocObserver, DeadInstChain); 1382 1383 while (!DeadInstChain.empty()) { 1384 MachineInstr *Inst = DeadInstChain.pop_back_val(); 1385 if (!isTriviallyDead(*Inst, MRI)) 1386 continue; 1387 saveUsesAndErase(*Inst, MRI, LocObserver, DeadInstChain); 1388 } 1389 } 1390 1391 void llvm::eraseInstr(MachineInstr &MI, MachineRegisterInfo &MRI, 1392 LostDebugLocObserver *LocObserver) { 1393 return eraseInstrs({&MI}, MRI, LocObserver); 1394 } 1395 1396 void llvm::salvageDebugInfo(const MachineRegisterInfo &MRI, MachineInstr &MI) { 1397 for (auto &Def : MI.defs()) { 1398 assert(Def.isReg() && "Must be a reg"); 1399 1400 SmallVector<MachineOperand *, 16> DbgUsers; 1401 for (auto &MOUse : MRI.use_operands(Def.getReg())) { 1402 MachineInstr *DbgValue = MOUse.getParent(); 1403 // Ignore partially formed DBG_VALUEs. 1404 if (DbgValue->isNonListDebugValue() && DbgValue->getNumOperands() == 4) { 1405 DbgUsers.push_back(&MOUse); 1406 } 1407 } 1408 1409 if (!DbgUsers.empty()) { 1410 salvageDebugInfoForDbgValue(MRI, MI, DbgUsers); 1411 } 1412 } 1413 } 1414