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/ADT/Optional.h" 16 #include "llvm/CodeGen/GlobalISel/GISelChangeObserver.h" 17 #include "llvm/CodeGen/GlobalISel/GISelKnownBits.h" 18 #include "llvm/CodeGen/GlobalISel/MIPatternMatch.h" 19 #include "llvm/CodeGen/GlobalISel/RegisterBankInfo.h" 20 #include "llvm/CodeGen/MachineInstr.h" 21 #include "llvm/CodeGen/MachineInstrBuilder.h" 22 #include "llvm/CodeGen/MachineOptimizationRemarkEmitter.h" 23 #include "llvm/CodeGen/MachineSizeOpts.h" 24 #include "llvm/CodeGen/MachineRegisterInfo.h" 25 #include "llvm/CodeGen/StackProtector.h" 26 #include "llvm/CodeGen/TargetInstrInfo.h" 27 #include "llvm/CodeGen/TargetLowering.h" 28 #include "llvm/CodeGen/TargetPassConfig.h" 29 #include "llvm/CodeGen/TargetRegisterInfo.h" 30 #include "llvm/IR/Constants.h" 31 #include "llvm/Target/TargetMachine.h" 32 33 #define DEBUG_TYPE "globalisel-utils" 34 35 using namespace llvm; 36 using namespace MIPatternMatch; 37 38 Register llvm::constrainRegToClass(MachineRegisterInfo &MRI, 39 const TargetInstrInfo &TII, 40 const RegisterBankInfo &RBI, Register Reg, 41 const TargetRegisterClass &RegClass) { 42 if (!RBI.constrainGenericRegister(Reg, RegClass, MRI)) 43 return MRI.createVirtualRegister(&RegClass); 44 45 return Reg; 46 } 47 48 Register llvm::constrainOperandRegClass( 49 const MachineFunction &MF, const TargetRegisterInfo &TRI, 50 MachineRegisterInfo &MRI, const TargetInstrInfo &TII, 51 const RegisterBankInfo &RBI, MachineInstr &InsertPt, 52 const TargetRegisterClass &RegClass, MachineOperand &RegMO) { 53 Register Reg = RegMO.getReg(); 54 // Assume physical registers are properly constrained. 55 assert(Register::isVirtualRegister(Reg) && "PhysReg not implemented"); 56 57 Register ConstrainedReg = constrainRegToClass(MRI, TII, RBI, Reg, RegClass); 58 // If we created a new virtual register because the class is not compatible 59 // then create a copy between the new and the old register. 60 if (ConstrainedReg != Reg) { 61 MachineBasicBlock::iterator InsertIt(&InsertPt); 62 MachineBasicBlock &MBB = *InsertPt.getParent(); 63 if (RegMO.isUse()) { 64 BuildMI(MBB, InsertIt, InsertPt.getDebugLoc(), 65 TII.get(TargetOpcode::COPY), ConstrainedReg) 66 .addReg(Reg); 67 } else { 68 assert(RegMO.isDef() && "Must be a definition"); 69 BuildMI(MBB, std::next(InsertIt), InsertPt.getDebugLoc(), 70 TII.get(TargetOpcode::COPY), Reg) 71 .addReg(ConstrainedReg); 72 } 73 if (GISelChangeObserver *Observer = MF.getObserver()) { 74 Observer->changingInstr(*RegMO.getParent()); 75 } 76 RegMO.setReg(ConstrainedReg); 77 if (GISelChangeObserver *Observer = MF.getObserver()) { 78 Observer->changedInstr(*RegMO.getParent()); 79 } 80 } else { 81 if (GISelChangeObserver *Observer = MF.getObserver()) { 82 if (!RegMO.isDef()) { 83 MachineInstr *RegDef = MRI.getVRegDef(Reg); 84 Observer->changedInstr(*RegDef); 85 } 86 Observer->changingAllUsesOfReg(MRI, Reg); 87 Observer->finishedChangingAllUsesOfReg(); 88 } 89 } 90 return ConstrainedReg; 91 } 92 93 Register llvm::constrainOperandRegClass( 94 const MachineFunction &MF, const TargetRegisterInfo &TRI, 95 MachineRegisterInfo &MRI, const TargetInstrInfo &TII, 96 const RegisterBankInfo &RBI, MachineInstr &InsertPt, const MCInstrDesc &II, 97 MachineOperand &RegMO, unsigned OpIdx) { 98 Register Reg = RegMO.getReg(); 99 // Assume physical registers are properly constrained. 100 assert(Register::isVirtualRegister(Reg) && "PhysReg not implemented"); 101 102 const TargetRegisterClass *RegClass = TII.getRegClass(II, OpIdx, &TRI, MF); 103 // Some of the target independent instructions, like COPY, may not impose any 104 // register class constraints on some of their operands: If it's a use, we can 105 // skip constraining as the instruction defining the register would constrain 106 // it. 107 108 // We can't constrain unallocatable register classes, because we can't create 109 // virtual registers for these classes, so we need to let targets handled this 110 // case. 111 if (RegClass && !RegClass->isAllocatable()) 112 RegClass = TRI.getConstrainedRegClassForOperand(RegMO, MRI); 113 114 if (!RegClass) { 115 assert((!isTargetSpecificOpcode(II.getOpcode()) || RegMO.isUse()) && 116 "Register class constraint is required unless either the " 117 "instruction is target independent or the operand is a use"); 118 // FIXME: Just bailing out like this here could be not enough, unless we 119 // expect the users of this function to do the right thing for PHIs and 120 // COPY: 121 // v1 = COPY v0 122 // v2 = COPY v1 123 // v1 here may end up not being constrained at all. Please notice that to 124 // reproduce the issue we likely need a destination pattern of a selection 125 // rule producing such extra copies, not just an input GMIR with them as 126 // every existing target using selectImpl handles copies before calling it 127 // and they never reach this function. 128 return Reg; 129 } 130 return constrainOperandRegClass(MF, TRI, MRI, TII, RBI, InsertPt, *RegClass, 131 RegMO); 132 } 133 134 bool llvm::constrainSelectedInstRegOperands(MachineInstr &I, 135 const TargetInstrInfo &TII, 136 const TargetRegisterInfo &TRI, 137 const RegisterBankInfo &RBI) { 138 assert(!isPreISelGenericOpcode(I.getOpcode()) && 139 "A selected instruction is expected"); 140 MachineBasicBlock &MBB = *I.getParent(); 141 MachineFunction &MF = *MBB.getParent(); 142 MachineRegisterInfo &MRI = MF.getRegInfo(); 143 144 for (unsigned OpI = 0, OpE = I.getNumExplicitOperands(); OpI != OpE; ++OpI) { 145 MachineOperand &MO = I.getOperand(OpI); 146 147 // There's nothing to be done on non-register operands. 148 if (!MO.isReg()) 149 continue; 150 151 LLVM_DEBUG(dbgs() << "Converting operand: " << MO << '\n'); 152 assert(MO.isReg() && "Unsupported non-reg operand"); 153 154 Register Reg = MO.getReg(); 155 // Physical registers don't need to be constrained. 156 if (Register::isPhysicalRegister(Reg)) 157 continue; 158 159 // Register operands with a value of 0 (e.g. predicate operands) don't need 160 // to be constrained. 161 if (Reg == 0) 162 continue; 163 164 // If the operand is a vreg, we should constrain its regclass, and only 165 // insert COPYs if that's impossible. 166 // constrainOperandRegClass does that for us. 167 constrainOperandRegClass(MF, TRI, MRI, TII, RBI, I, I.getDesc(), MO, OpI); 168 169 // Tie uses to defs as indicated in MCInstrDesc if this hasn't already been 170 // done. 171 if (MO.isUse()) { 172 int DefIdx = I.getDesc().getOperandConstraint(OpI, MCOI::TIED_TO); 173 if (DefIdx != -1 && !I.isRegTiedToUseOperand(DefIdx)) 174 I.tieOperands(DefIdx, OpI); 175 } 176 } 177 return true; 178 } 179 180 bool llvm::canReplaceReg(Register DstReg, Register SrcReg, 181 MachineRegisterInfo &MRI) { 182 // Give up if either DstReg or SrcReg is a physical register. 183 if (DstReg.isPhysical() || SrcReg.isPhysical()) 184 return false; 185 // Give up if the types don't match. 186 if (MRI.getType(DstReg) != MRI.getType(SrcReg)) 187 return false; 188 // Replace if either DstReg has no constraints or the register 189 // constraints match. 190 return !MRI.getRegClassOrRegBank(DstReg) || 191 MRI.getRegClassOrRegBank(DstReg) == MRI.getRegClassOrRegBank(SrcReg); 192 } 193 194 bool llvm::isTriviallyDead(const MachineInstr &MI, 195 const MachineRegisterInfo &MRI) { 196 // FIXME: This logical is mostly duplicated with 197 // DeadMachineInstructionElim::isDead. Why is LOCAL_ESCAPE not considered in 198 // MachineInstr::isLabel? 199 200 // Don't delete frame allocation labels. 201 if (MI.getOpcode() == TargetOpcode::LOCAL_ESCAPE) 202 return false; 203 // LIFETIME markers should be preserved even if they seem dead. 204 if (MI.getOpcode() == TargetOpcode::LIFETIME_START || 205 MI.getOpcode() == TargetOpcode::LIFETIME_END) 206 return false; 207 208 // If we can move an instruction, we can remove it. Otherwise, it has 209 // a side-effect of some sort. 210 bool SawStore = false; 211 if (!MI.isSafeToMove(/*AA=*/nullptr, SawStore) && !MI.isPHI()) 212 return false; 213 214 // Instructions without side-effects are dead iff they only define dead vregs. 215 for (auto &MO : MI.operands()) { 216 if (!MO.isReg() || !MO.isDef()) 217 continue; 218 219 Register Reg = MO.getReg(); 220 if (Register::isPhysicalRegister(Reg) || !MRI.use_nodbg_empty(Reg)) 221 return false; 222 } 223 return true; 224 } 225 226 static void reportGISelDiagnostic(DiagnosticSeverity Severity, 227 MachineFunction &MF, 228 const TargetPassConfig &TPC, 229 MachineOptimizationRemarkEmitter &MORE, 230 MachineOptimizationRemarkMissed &R) { 231 bool IsFatal = Severity == DS_Error && 232 TPC.isGlobalISelAbortEnabled(); 233 // Print the function name explicitly if we don't have a debug location (which 234 // makes the diagnostic less useful) or if we're going to emit a raw error. 235 if (!R.getLocation().isValid() || IsFatal) 236 R << (" (in function: " + MF.getName() + ")").str(); 237 238 if (IsFatal) 239 report_fatal_error(R.getMsg()); 240 else 241 MORE.emit(R); 242 } 243 244 void llvm::reportGISelWarning(MachineFunction &MF, const TargetPassConfig &TPC, 245 MachineOptimizationRemarkEmitter &MORE, 246 MachineOptimizationRemarkMissed &R) { 247 reportGISelDiagnostic(DS_Warning, MF, TPC, MORE, R); 248 } 249 250 void llvm::reportGISelFailure(MachineFunction &MF, const TargetPassConfig &TPC, 251 MachineOptimizationRemarkEmitter &MORE, 252 MachineOptimizationRemarkMissed &R) { 253 MF.getProperties().set(MachineFunctionProperties::Property::FailedISel); 254 reportGISelDiagnostic(DS_Error, MF, TPC, MORE, R); 255 } 256 257 void llvm::reportGISelFailure(MachineFunction &MF, const TargetPassConfig &TPC, 258 MachineOptimizationRemarkEmitter &MORE, 259 const char *PassName, StringRef Msg, 260 const MachineInstr &MI) { 261 MachineOptimizationRemarkMissed R(PassName, "GISelFailure: ", 262 MI.getDebugLoc(), MI.getParent()); 263 R << Msg; 264 // Printing MI is expensive; only do it if expensive remarks are enabled. 265 if (TPC.isGlobalISelAbortEnabled() || MORE.allowExtraAnalysis(PassName)) 266 R << ": " << ore::MNV("Inst", MI); 267 reportGISelFailure(MF, TPC, MORE, R); 268 } 269 270 Optional<APInt> llvm::getIConstantVRegVal(Register VReg, 271 const MachineRegisterInfo &MRI) { 272 Optional<ValueAndVReg> ValAndVReg = getIConstantVRegValWithLookThrough( 273 VReg, MRI, /*LookThroughInstrs*/ false); 274 assert((!ValAndVReg || ValAndVReg->VReg == VReg) && 275 "Value found while looking through instrs"); 276 if (!ValAndVReg) 277 return None; 278 return ValAndVReg->Value; 279 } 280 281 Optional<int64_t> 282 llvm::getIConstantVRegSExtVal(Register VReg, const MachineRegisterInfo &MRI) { 283 Optional<APInt> Val = getIConstantVRegVal(VReg, MRI); 284 if (Val && Val->getBitWidth() <= 64) 285 return Val->getSExtValue(); 286 return None; 287 } 288 289 namespace { 290 291 typedef std::function<bool(const MachineInstr *)> IsOpcodeFn; 292 typedef std::function<Optional<APInt>(const MachineInstr *MI)> GetAPCstFn; 293 294 Optional<ValueAndVReg> getConstantVRegValWithLookThrough( 295 Register VReg, const MachineRegisterInfo &MRI, IsOpcodeFn IsConstantOpcode, 296 GetAPCstFn getAPCstValue, bool LookThroughInstrs = true, 297 bool LookThroughAnyExt = false) { 298 SmallVector<std::pair<unsigned, unsigned>, 4> SeenOpcodes; 299 MachineInstr *MI; 300 301 while ((MI = MRI.getVRegDef(VReg)) && !IsConstantOpcode(MI) && 302 LookThroughInstrs) { 303 switch (MI->getOpcode()) { 304 case TargetOpcode::G_ANYEXT: 305 if (!LookThroughAnyExt) 306 return None; 307 LLVM_FALLTHROUGH; 308 case TargetOpcode::G_TRUNC: 309 case TargetOpcode::G_SEXT: 310 case TargetOpcode::G_ZEXT: 311 SeenOpcodes.push_back(std::make_pair( 312 MI->getOpcode(), 313 MRI.getType(MI->getOperand(0).getReg()).getSizeInBits())); 314 VReg = MI->getOperand(1).getReg(); 315 break; 316 case TargetOpcode::COPY: 317 VReg = MI->getOperand(1).getReg(); 318 if (Register::isPhysicalRegister(VReg)) 319 return None; 320 break; 321 case TargetOpcode::G_INTTOPTR: 322 VReg = MI->getOperand(1).getReg(); 323 break; 324 default: 325 return None; 326 } 327 } 328 if (!MI || !IsConstantOpcode(MI)) 329 return None; 330 331 Optional<APInt> MaybeVal = getAPCstValue(MI); 332 if (!MaybeVal) 333 return None; 334 APInt &Val = *MaybeVal; 335 while (!SeenOpcodes.empty()) { 336 std::pair<unsigned, unsigned> OpcodeAndSize = SeenOpcodes.pop_back_val(); 337 switch (OpcodeAndSize.first) { 338 case TargetOpcode::G_TRUNC: 339 Val = Val.trunc(OpcodeAndSize.second); 340 break; 341 case TargetOpcode::G_ANYEXT: 342 case TargetOpcode::G_SEXT: 343 Val = Val.sext(OpcodeAndSize.second); 344 break; 345 case TargetOpcode::G_ZEXT: 346 Val = Val.zext(OpcodeAndSize.second); 347 break; 348 } 349 } 350 351 return ValueAndVReg{Val, VReg}; 352 } 353 354 bool isIConstant(const MachineInstr *MI) { 355 if (!MI) 356 return false; 357 return MI->getOpcode() == TargetOpcode::G_CONSTANT; 358 } 359 360 bool isFConstant(const MachineInstr *MI) { 361 if (!MI) 362 return false; 363 return MI->getOpcode() == TargetOpcode::G_FCONSTANT; 364 } 365 366 bool isAnyConstant(const MachineInstr *MI) { 367 if (!MI) 368 return false; 369 unsigned Opc = MI->getOpcode(); 370 return Opc == TargetOpcode::G_CONSTANT || Opc == TargetOpcode::G_FCONSTANT; 371 } 372 373 Optional<APInt> getCImmAsAPInt(const MachineInstr *MI) { 374 const MachineOperand &CstVal = MI->getOperand(1); 375 if (CstVal.isCImm()) 376 return CstVal.getCImm()->getValue(); 377 return None; 378 } 379 380 Optional<APInt> getCImmOrFPImmAsAPInt(const MachineInstr *MI) { 381 const MachineOperand &CstVal = MI->getOperand(1); 382 if (CstVal.isCImm()) 383 return CstVal.getCImm()->getValue(); 384 if (CstVal.isFPImm()) 385 return CstVal.getFPImm()->getValueAPF().bitcastToAPInt(); 386 return None; 387 } 388 389 } // end anonymous namespace 390 391 Optional<ValueAndVReg> llvm::getIConstantVRegValWithLookThrough( 392 Register VReg, const MachineRegisterInfo &MRI, bool LookThroughInstrs) { 393 return getConstantVRegValWithLookThrough(VReg, MRI, isIConstant, 394 getCImmAsAPInt, LookThroughInstrs); 395 } 396 397 Optional<ValueAndVReg> llvm::getAnyConstantVRegValWithLookThrough( 398 Register VReg, const MachineRegisterInfo &MRI, bool LookThroughInstrs, 399 bool LookThroughAnyExt) { 400 return getConstantVRegValWithLookThrough( 401 VReg, MRI, isAnyConstant, getCImmOrFPImmAsAPInt, LookThroughInstrs, 402 LookThroughAnyExt); 403 } 404 405 Optional<FPValueAndVReg> llvm::getFConstantVRegValWithLookThrough( 406 Register VReg, const MachineRegisterInfo &MRI, bool LookThroughInstrs) { 407 auto Reg = getConstantVRegValWithLookThrough( 408 VReg, MRI, isFConstant, getCImmOrFPImmAsAPInt, LookThroughInstrs); 409 if (!Reg) 410 return None; 411 return FPValueAndVReg{getConstantFPVRegVal(Reg->VReg, MRI)->getValueAPF(), 412 Reg->VReg}; 413 } 414 415 const ConstantFP * 416 llvm::getConstantFPVRegVal(Register VReg, const MachineRegisterInfo &MRI) { 417 MachineInstr *MI = MRI.getVRegDef(VReg); 418 if (TargetOpcode::G_FCONSTANT != MI->getOpcode()) 419 return nullptr; 420 return MI->getOperand(1).getFPImm(); 421 } 422 423 Optional<DefinitionAndSourceRegister> 424 llvm::getDefSrcRegIgnoringCopies(Register Reg, const MachineRegisterInfo &MRI) { 425 Register DefSrcReg = Reg; 426 auto *DefMI = MRI.getVRegDef(Reg); 427 auto DstTy = MRI.getType(DefMI->getOperand(0).getReg()); 428 if (!DstTy.isValid()) 429 return None; 430 unsigned Opc = DefMI->getOpcode(); 431 while (Opc == TargetOpcode::COPY || isPreISelGenericOptimizationHint(Opc)) { 432 Register SrcReg = DefMI->getOperand(1).getReg(); 433 auto SrcTy = MRI.getType(SrcReg); 434 if (!SrcTy.isValid()) 435 break; 436 DefMI = MRI.getVRegDef(SrcReg); 437 DefSrcReg = SrcReg; 438 Opc = DefMI->getOpcode(); 439 } 440 return DefinitionAndSourceRegister{DefMI, DefSrcReg}; 441 } 442 443 MachineInstr *llvm::getDefIgnoringCopies(Register Reg, 444 const MachineRegisterInfo &MRI) { 445 Optional<DefinitionAndSourceRegister> DefSrcReg = 446 getDefSrcRegIgnoringCopies(Reg, MRI); 447 return DefSrcReg ? DefSrcReg->MI : nullptr; 448 } 449 450 Register llvm::getSrcRegIgnoringCopies(Register Reg, 451 const MachineRegisterInfo &MRI) { 452 Optional<DefinitionAndSourceRegister> DefSrcReg = 453 getDefSrcRegIgnoringCopies(Reg, MRI); 454 return DefSrcReg ? DefSrcReg->Reg : Register(); 455 } 456 457 MachineInstr *llvm::getOpcodeDef(unsigned Opcode, Register Reg, 458 const MachineRegisterInfo &MRI) { 459 MachineInstr *DefMI = getDefIgnoringCopies(Reg, MRI); 460 return DefMI && DefMI->getOpcode() == Opcode ? DefMI : nullptr; 461 } 462 463 APFloat llvm::getAPFloatFromSize(double Val, unsigned Size) { 464 if (Size == 32) 465 return APFloat(float(Val)); 466 if (Size == 64) 467 return APFloat(Val); 468 if (Size != 16) 469 llvm_unreachable("Unsupported FPConstant size"); 470 bool Ignored; 471 APFloat APF(Val); 472 APF.convert(APFloat::IEEEhalf(), APFloat::rmNearestTiesToEven, &Ignored); 473 return APF; 474 } 475 476 Optional<APInt> llvm::ConstantFoldBinOp(unsigned Opcode, const Register Op1, 477 const Register Op2, 478 const MachineRegisterInfo &MRI) { 479 auto MaybeOp2Cst = getAnyConstantVRegValWithLookThrough(Op2, MRI, false); 480 if (!MaybeOp2Cst) 481 return None; 482 483 auto MaybeOp1Cst = getAnyConstantVRegValWithLookThrough(Op1, MRI, false); 484 if (!MaybeOp1Cst) 485 return None; 486 487 const APInt &C1 = MaybeOp1Cst->Value; 488 const APInt &C2 = MaybeOp2Cst->Value; 489 switch (Opcode) { 490 default: 491 break; 492 case TargetOpcode::G_ADD: 493 return C1 + C2; 494 case TargetOpcode::G_AND: 495 return C1 & C2; 496 case TargetOpcode::G_ASHR: 497 return C1.ashr(C2); 498 case TargetOpcode::G_LSHR: 499 return C1.lshr(C2); 500 case TargetOpcode::G_MUL: 501 return C1 * C2; 502 case TargetOpcode::G_OR: 503 return C1 | C2; 504 case TargetOpcode::G_SHL: 505 return C1 << C2; 506 case TargetOpcode::G_SUB: 507 return C1 - C2; 508 case TargetOpcode::G_XOR: 509 return C1 ^ C2; 510 case TargetOpcode::G_UDIV: 511 if (!C2.getBoolValue()) 512 break; 513 return C1.udiv(C2); 514 case TargetOpcode::G_SDIV: 515 if (!C2.getBoolValue()) 516 break; 517 return C1.sdiv(C2); 518 case TargetOpcode::G_UREM: 519 if (!C2.getBoolValue()) 520 break; 521 return C1.urem(C2); 522 case TargetOpcode::G_SREM: 523 if (!C2.getBoolValue()) 524 break; 525 return C1.srem(C2); 526 } 527 528 return None; 529 } 530 531 Optional<APFloat> llvm::ConstantFoldFPBinOp(unsigned Opcode, const Register Op1, 532 const Register Op2, 533 const MachineRegisterInfo &MRI) { 534 const ConstantFP *Op2Cst = getConstantFPVRegVal(Op2, MRI); 535 if (!Op2Cst) 536 return None; 537 538 const ConstantFP *Op1Cst = getConstantFPVRegVal(Op1, MRI); 539 if (!Op1Cst) 540 return None; 541 542 APFloat C1 = Op1Cst->getValueAPF(); 543 const APFloat &C2 = Op2Cst->getValueAPF(); 544 switch (Opcode) { 545 case TargetOpcode::G_FADD: 546 C1.add(C2, APFloat::rmNearestTiesToEven); 547 return C1; 548 case TargetOpcode::G_FSUB: 549 C1.subtract(C2, APFloat::rmNearestTiesToEven); 550 return C1; 551 case TargetOpcode::G_FMUL: 552 C1.multiply(C2, APFloat::rmNearestTiesToEven); 553 return C1; 554 case TargetOpcode::G_FDIV: 555 C1.divide(C2, APFloat::rmNearestTiesToEven); 556 return C1; 557 case TargetOpcode::G_FREM: 558 C1.mod(C2); 559 return C1; 560 case TargetOpcode::G_FCOPYSIGN: 561 C1.copySign(C2); 562 return C1; 563 case TargetOpcode::G_FMINNUM: 564 return minnum(C1, C2); 565 case TargetOpcode::G_FMAXNUM: 566 return maxnum(C1, C2); 567 case TargetOpcode::G_FMINIMUM: 568 return minimum(C1, C2); 569 case TargetOpcode::G_FMAXIMUM: 570 return maximum(C1, C2); 571 case TargetOpcode::G_FMINNUM_IEEE: 572 case TargetOpcode::G_FMAXNUM_IEEE: 573 // FIXME: These operations were unfortunately named. fminnum/fmaxnum do not 574 // follow the IEEE behavior for signaling nans and follow libm's fmin/fmax, 575 // and currently there isn't a nice wrapper in APFloat for the version with 576 // correct snan handling. 577 break; 578 default: 579 break; 580 } 581 582 return None; 583 } 584 585 bool llvm::isKnownNeverNaN(Register Val, const MachineRegisterInfo &MRI, 586 bool SNaN) { 587 const MachineInstr *DefMI = MRI.getVRegDef(Val); 588 if (!DefMI) 589 return false; 590 591 const TargetMachine& TM = DefMI->getMF()->getTarget(); 592 if (DefMI->getFlag(MachineInstr::FmNoNans) || TM.Options.NoNaNsFPMath) 593 return true; 594 595 // If the value is a constant, we can obviously see if it is a NaN or not. 596 if (const ConstantFP *FPVal = getConstantFPVRegVal(Val, MRI)) { 597 return !FPVal->getValueAPF().isNaN() || 598 (SNaN && !FPVal->getValueAPF().isSignaling()); 599 } 600 601 if (DefMI->getOpcode() == TargetOpcode::G_BUILD_VECTOR) { 602 for (const auto &Op : DefMI->uses()) 603 if (!isKnownNeverNaN(Op.getReg(), MRI, SNaN)) 604 return false; 605 return true; 606 } 607 608 switch (DefMI->getOpcode()) { 609 default: 610 break; 611 case TargetOpcode::G_FMINNUM_IEEE: 612 case TargetOpcode::G_FMAXNUM_IEEE: { 613 if (SNaN) 614 return true; 615 // This can return a NaN if either operand is an sNaN, or if both operands 616 // are NaN. 617 return (isKnownNeverNaN(DefMI->getOperand(1).getReg(), MRI) && 618 isKnownNeverSNaN(DefMI->getOperand(2).getReg(), MRI)) || 619 (isKnownNeverSNaN(DefMI->getOperand(1).getReg(), MRI) && 620 isKnownNeverNaN(DefMI->getOperand(2).getReg(), MRI)); 621 } 622 case TargetOpcode::G_FMINNUM: 623 case TargetOpcode::G_FMAXNUM: { 624 // Only one needs to be known not-nan, since it will be returned if the 625 // other ends up being one. 626 return isKnownNeverNaN(DefMI->getOperand(1).getReg(), MRI, SNaN) || 627 isKnownNeverNaN(DefMI->getOperand(2).getReg(), MRI, SNaN); 628 } 629 } 630 631 if (SNaN) { 632 // FP operations quiet. For now, just handle the ones inserted during 633 // legalization. 634 switch (DefMI->getOpcode()) { 635 case TargetOpcode::G_FPEXT: 636 case TargetOpcode::G_FPTRUNC: 637 case TargetOpcode::G_FCANONICALIZE: 638 return true; 639 default: 640 return false; 641 } 642 } 643 644 return false; 645 } 646 647 Align llvm::inferAlignFromPtrInfo(MachineFunction &MF, 648 const MachinePointerInfo &MPO) { 649 auto PSV = MPO.V.dyn_cast<const PseudoSourceValue *>(); 650 if (auto FSPV = dyn_cast_or_null<FixedStackPseudoSourceValue>(PSV)) { 651 MachineFrameInfo &MFI = MF.getFrameInfo(); 652 return commonAlignment(MFI.getObjectAlign(FSPV->getFrameIndex()), 653 MPO.Offset); 654 } 655 656 if (const Value *V = MPO.V.dyn_cast<const Value *>()) { 657 const Module *M = MF.getFunction().getParent(); 658 return V->getPointerAlignment(M->getDataLayout()); 659 } 660 661 return Align(1); 662 } 663 664 Register llvm::getFunctionLiveInPhysReg(MachineFunction &MF, 665 const TargetInstrInfo &TII, 666 MCRegister PhysReg, 667 const TargetRegisterClass &RC, 668 LLT RegTy) { 669 DebugLoc DL; // FIXME: Is no location the right choice? 670 MachineBasicBlock &EntryMBB = MF.front(); 671 MachineRegisterInfo &MRI = MF.getRegInfo(); 672 Register LiveIn = MRI.getLiveInVirtReg(PhysReg); 673 if (LiveIn) { 674 MachineInstr *Def = MRI.getVRegDef(LiveIn); 675 if (Def) { 676 // FIXME: Should the verifier check this is in the entry block? 677 assert(Def->getParent() == &EntryMBB && "live-in copy not in entry block"); 678 return LiveIn; 679 } 680 681 // It's possible the incoming argument register and copy was added during 682 // lowering, but later deleted due to being/becoming dead. If this happens, 683 // re-insert the copy. 684 } else { 685 // The live in register was not present, so add it. 686 LiveIn = MF.addLiveIn(PhysReg, &RC); 687 if (RegTy.isValid()) 688 MRI.setType(LiveIn, RegTy); 689 } 690 691 BuildMI(EntryMBB, EntryMBB.begin(), DL, TII.get(TargetOpcode::COPY), LiveIn) 692 .addReg(PhysReg); 693 if (!EntryMBB.isLiveIn(PhysReg)) 694 EntryMBB.addLiveIn(PhysReg); 695 return LiveIn; 696 } 697 698 Optional<APInt> llvm::ConstantFoldExtOp(unsigned Opcode, const Register Op1, 699 uint64_t Imm, 700 const MachineRegisterInfo &MRI) { 701 auto MaybeOp1Cst = getIConstantVRegVal(Op1, MRI); 702 if (MaybeOp1Cst) { 703 switch (Opcode) { 704 default: 705 break; 706 case TargetOpcode::G_SEXT_INREG: { 707 LLT Ty = MRI.getType(Op1); 708 return MaybeOp1Cst->trunc(Imm).sext(Ty.getScalarSizeInBits()); 709 } 710 } 711 } 712 return None; 713 } 714 715 Optional<APFloat> llvm::ConstantFoldIntToFloat(unsigned Opcode, LLT DstTy, 716 Register Src, 717 const MachineRegisterInfo &MRI) { 718 assert(Opcode == TargetOpcode::G_SITOFP || Opcode == TargetOpcode::G_UITOFP); 719 if (auto MaybeSrcVal = getIConstantVRegVal(Src, MRI)) { 720 APFloat DstVal(getFltSemanticForLLT(DstTy)); 721 DstVal.convertFromAPInt(*MaybeSrcVal, Opcode == TargetOpcode::G_SITOFP, 722 APFloat::rmNearestTiesToEven); 723 return DstVal; 724 } 725 return None; 726 } 727 728 bool llvm::isKnownToBeAPowerOfTwo(Register Reg, const MachineRegisterInfo &MRI, 729 GISelKnownBits *KB) { 730 Optional<DefinitionAndSourceRegister> DefSrcReg = 731 getDefSrcRegIgnoringCopies(Reg, MRI); 732 if (!DefSrcReg) 733 return false; 734 735 const MachineInstr &MI = *DefSrcReg->MI; 736 const LLT Ty = MRI.getType(Reg); 737 738 switch (MI.getOpcode()) { 739 case TargetOpcode::G_CONSTANT: { 740 unsigned BitWidth = Ty.getScalarSizeInBits(); 741 const ConstantInt *CI = MI.getOperand(1).getCImm(); 742 return CI->getValue().zextOrTrunc(BitWidth).isPowerOf2(); 743 } 744 case TargetOpcode::G_SHL: { 745 // A left-shift of a constant one will have exactly one bit set because 746 // shifting the bit off the end is undefined. 747 748 // TODO: Constant splat 749 if (auto ConstLHS = getIConstantVRegVal(MI.getOperand(1).getReg(), MRI)) { 750 if (*ConstLHS == 1) 751 return true; 752 } 753 754 break; 755 } 756 case TargetOpcode::G_LSHR: { 757 if (auto ConstLHS = getIConstantVRegVal(MI.getOperand(1).getReg(), MRI)) { 758 if (ConstLHS->isSignMask()) 759 return true; 760 } 761 762 break; 763 } 764 case TargetOpcode::G_BUILD_VECTOR: { 765 // TODO: Probably should have a recursion depth guard since you could have 766 // bitcasted vector elements. 767 for (unsigned I = 1, E = MI.getNumOperands(); I != E; ++I) { 768 if (!isKnownToBeAPowerOfTwo(MI.getOperand(I).getReg(), MRI, KB)) 769 return false; 770 } 771 772 return true; 773 } 774 case TargetOpcode::G_BUILD_VECTOR_TRUNC: { 775 // Only handle constants since we would need to know if number of leading 776 // zeros is greater than the truncation amount. 777 const unsigned BitWidth = Ty.getScalarSizeInBits(); 778 for (unsigned I = 1, E = MI.getNumOperands(); I != E; ++I) { 779 auto Const = getIConstantVRegVal(MI.getOperand(I).getReg(), MRI); 780 if (!Const || !Const->zextOrTrunc(BitWidth).isPowerOf2()) 781 return false; 782 } 783 784 return true; 785 } 786 default: 787 break; 788 } 789 790 if (!KB) 791 return false; 792 793 // More could be done here, though the above checks are enough 794 // to handle some common cases. 795 796 // Fall back to computeKnownBits to catch other known cases. 797 KnownBits Known = KB->getKnownBits(Reg); 798 return (Known.countMaxPopulation() == 1) && (Known.countMinPopulation() == 1); 799 } 800 801 void llvm::getSelectionDAGFallbackAnalysisUsage(AnalysisUsage &AU) { 802 AU.addPreserved<StackProtector>(); 803 } 804 805 static unsigned getLCMSize(unsigned OrigSize, unsigned TargetSize) { 806 unsigned Mul = OrigSize * TargetSize; 807 unsigned GCDSize = greatestCommonDivisor(OrigSize, TargetSize); 808 return Mul / GCDSize; 809 } 810 811 LLT llvm::getLCMType(LLT OrigTy, LLT TargetTy) { 812 const unsigned OrigSize = OrigTy.getSizeInBits(); 813 const unsigned TargetSize = TargetTy.getSizeInBits(); 814 815 if (OrigSize == TargetSize) 816 return OrigTy; 817 818 if (OrigTy.isVector()) { 819 const LLT OrigElt = OrigTy.getElementType(); 820 821 if (TargetTy.isVector()) { 822 const LLT TargetElt = TargetTy.getElementType(); 823 824 if (OrigElt.getSizeInBits() == TargetElt.getSizeInBits()) { 825 int GCDElts = greatestCommonDivisor(OrigTy.getNumElements(), 826 TargetTy.getNumElements()); 827 // Prefer the original element type. 828 ElementCount Mul = OrigTy.getElementCount() * TargetTy.getNumElements(); 829 return LLT::vector(Mul.divideCoefficientBy(GCDElts), 830 OrigTy.getElementType()); 831 } 832 } else { 833 if (OrigElt.getSizeInBits() == TargetSize) 834 return OrigTy; 835 } 836 837 unsigned LCMSize = getLCMSize(OrigSize, TargetSize); 838 return LLT::fixed_vector(LCMSize / OrigElt.getSizeInBits(), OrigElt); 839 } 840 841 if (TargetTy.isVector()) { 842 unsigned LCMSize = getLCMSize(OrigSize, TargetSize); 843 return LLT::fixed_vector(LCMSize / OrigSize, OrigTy); 844 } 845 846 unsigned LCMSize = getLCMSize(OrigSize, TargetSize); 847 848 // Preserve pointer types. 849 if (LCMSize == OrigSize) 850 return OrigTy; 851 if (LCMSize == TargetSize) 852 return TargetTy; 853 854 return LLT::scalar(LCMSize); 855 } 856 857 LLT llvm::getGCDType(LLT OrigTy, LLT TargetTy) { 858 const unsigned OrigSize = OrigTy.getSizeInBits(); 859 const unsigned TargetSize = TargetTy.getSizeInBits(); 860 861 if (OrigSize == TargetSize) 862 return OrigTy; 863 864 if (OrigTy.isVector()) { 865 LLT OrigElt = OrigTy.getElementType(); 866 if (TargetTy.isVector()) { 867 LLT TargetElt = TargetTy.getElementType(); 868 if (OrigElt.getSizeInBits() == TargetElt.getSizeInBits()) { 869 int GCD = greatestCommonDivisor(OrigTy.getNumElements(), 870 TargetTy.getNumElements()); 871 return LLT::scalarOrVector(ElementCount::getFixed(GCD), OrigElt); 872 } 873 } else { 874 // If the source is a vector of pointers, return a pointer element. 875 if (OrigElt.getSizeInBits() == TargetSize) 876 return OrigElt; 877 } 878 879 unsigned GCD = greatestCommonDivisor(OrigSize, TargetSize); 880 if (GCD == OrigElt.getSizeInBits()) 881 return OrigElt; 882 883 // If we can't produce the original element type, we have to use a smaller 884 // scalar. 885 if (GCD < OrigElt.getSizeInBits()) 886 return LLT::scalar(GCD); 887 return LLT::fixed_vector(GCD / OrigElt.getSizeInBits(), OrigElt); 888 } 889 890 if (TargetTy.isVector()) { 891 // Try to preserve the original element type. 892 LLT TargetElt = TargetTy.getElementType(); 893 if (TargetElt.getSizeInBits() == OrigSize) 894 return OrigTy; 895 } 896 897 unsigned GCD = greatestCommonDivisor(OrigSize, TargetSize); 898 return LLT::scalar(GCD); 899 } 900 901 Optional<int> llvm::getSplatIndex(MachineInstr &MI) { 902 assert(MI.getOpcode() == TargetOpcode::G_SHUFFLE_VECTOR && 903 "Only G_SHUFFLE_VECTOR can have a splat index!"); 904 ArrayRef<int> Mask = MI.getOperand(3).getShuffleMask(); 905 auto FirstDefinedIdx = find_if(Mask, [](int Elt) { return Elt >= 0; }); 906 907 // If all elements are undefined, this shuffle can be considered a splat. 908 // Return 0 for better potential for callers to simplify. 909 if (FirstDefinedIdx == Mask.end()) 910 return 0; 911 912 // Make sure all remaining elements are either undef or the same 913 // as the first non-undef value. 914 int SplatValue = *FirstDefinedIdx; 915 if (any_of(make_range(std::next(FirstDefinedIdx), Mask.end()), 916 [&SplatValue](int Elt) { return Elt >= 0 && Elt != SplatValue; })) 917 return None; 918 919 return SplatValue; 920 } 921 922 static bool isBuildVectorOp(unsigned Opcode) { 923 return Opcode == TargetOpcode::G_BUILD_VECTOR || 924 Opcode == TargetOpcode::G_BUILD_VECTOR_TRUNC; 925 } 926 927 // TODO: Handle mixed undef elements. 928 static bool isBuildVectorConstantSplat(const MachineInstr &MI, 929 const MachineRegisterInfo &MRI, 930 int64_t SplatValue) { 931 if (!isBuildVectorOp(MI.getOpcode())) 932 return false; 933 934 const unsigned NumOps = MI.getNumOperands(); 935 for (unsigned I = 1; I != NumOps; ++I) { 936 Register Element = MI.getOperand(I).getReg(); 937 if (!mi_match(Element, MRI, m_SpecificICst(SplatValue))) 938 return false; 939 } 940 941 return true; 942 } 943 944 Optional<int64_t> 945 llvm::getBuildVectorConstantSplat(const MachineInstr &MI, 946 const MachineRegisterInfo &MRI) { 947 if (!isBuildVectorOp(MI.getOpcode())) 948 return None; 949 950 const unsigned NumOps = MI.getNumOperands(); 951 Optional<int64_t> Scalar; 952 for (unsigned I = 1; I != NumOps; ++I) { 953 Register Element = MI.getOperand(I).getReg(); 954 int64_t ElementValue; 955 if (!mi_match(Element, MRI, m_ICst(ElementValue))) 956 return None; 957 if (!Scalar) 958 Scalar = ElementValue; 959 else if (*Scalar != ElementValue) 960 return None; 961 } 962 963 return Scalar; 964 } 965 966 bool llvm::isBuildVectorAllZeros(const MachineInstr &MI, 967 const MachineRegisterInfo &MRI) { 968 return isBuildVectorConstantSplat(MI, MRI, 0); 969 } 970 971 bool llvm::isBuildVectorAllOnes(const MachineInstr &MI, 972 const MachineRegisterInfo &MRI) { 973 return isBuildVectorConstantSplat(MI, MRI, -1); 974 } 975 976 Optional<RegOrConstant> llvm::getVectorSplat(const MachineInstr &MI, 977 const MachineRegisterInfo &MRI) { 978 unsigned Opc = MI.getOpcode(); 979 if (!isBuildVectorOp(Opc)) 980 return None; 981 if (auto Splat = getBuildVectorConstantSplat(MI, MRI)) 982 return RegOrConstant(*Splat); 983 auto Reg = MI.getOperand(1).getReg(); 984 if (any_of(make_range(MI.operands_begin() + 2, MI.operands_end()), 985 [&Reg](const MachineOperand &Op) { return Op.getReg() != Reg; })) 986 return None; 987 return RegOrConstant(Reg); 988 } 989 990 bool llvm::matchUnaryPredicate( 991 const MachineRegisterInfo &MRI, Register Reg, 992 std::function<bool(const Constant *ConstVal)> Match, bool AllowUndefs) { 993 994 const MachineInstr *Def = getDefIgnoringCopies(Reg, MRI); 995 if (AllowUndefs && Def->getOpcode() == TargetOpcode::G_IMPLICIT_DEF) 996 return Match(nullptr); 997 998 // TODO: Also handle fconstant 999 if (Def->getOpcode() == TargetOpcode::G_CONSTANT) 1000 return Match(Def->getOperand(1).getCImm()); 1001 1002 if (Def->getOpcode() != TargetOpcode::G_BUILD_VECTOR) 1003 return false; 1004 1005 for (unsigned I = 1, E = Def->getNumOperands(); I != E; ++I) { 1006 Register SrcElt = Def->getOperand(I).getReg(); 1007 const MachineInstr *SrcDef = getDefIgnoringCopies(SrcElt, MRI); 1008 if (AllowUndefs && SrcDef->getOpcode() == TargetOpcode::G_IMPLICIT_DEF) { 1009 if (!Match(nullptr)) 1010 return false; 1011 continue; 1012 } 1013 1014 if (SrcDef->getOpcode() != TargetOpcode::G_CONSTANT || 1015 !Match(SrcDef->getOperand(1).getCImm())) 1016 return false; 1017 } 1018 1019 return true; 1020 } 1021 1022 bool llvm::isConstTrueVal(const TargetLowering &TLI, int64_t Val, bool IsVector, 1023 bool IsFP) { 1024 switch (TLI.getBooleanContents(IsVector, IsFP)) { 1025 case TargetLowering::UndefinedBooleanContent: 1026 return Val & 0x1; 1027 case TargetLowering::ZeroOrOneBooleanContent: 1028 return Val == 1; 1029 case TargetLowering::ZeroOrNegativeOneBooleanContent: 1030 return Val == -1; 1031 } 1032 llvm_unreachable("Invalid boolean contents"); 1033 } 1034 1035 int64_t llvm::getICmpTrueVal(const TargetLowering &TLI, bool IsVector, 1036 bool IsFP) { 1037 switch (TLI.getBooleanContents(IsVector, IsFP)) { 1038 case TargetLowering::UndefinedBooleanContent: 1039 case TargetLowering::ZeroOrOneBooleanContent: 1040 return 1; 1041 case TargetLowering::ZeroOrNegativeOneBooleanContent: 1042 return -1; 1043 } 1044 llvm_unreachable("Invalid boolean contents"); 1045 } 1046 1047 bool llvm::shouldOptForSize(const MachineBasicBlock &MBB, 1048 ProfileSummaryInfo *PSI, BlockFrequencyInfo *BFI) { 1049 const auto &F = MBB.getParent()->getFunction(); 1050 return F.hasOptSize() || F.hasMinSize() || 1051 llvm::shouldOptimizeForSize(MBB.getBasicBlock(), PSI, BFI); 1052 } 1053