1 //===-- PPCISelDAGToDAG.cpp - PPC --pattern matching inst selector --------===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 // This file defines a pattern matching instruction selector for PowerPC, 11 // converting from a legalized dag to a PPC dag. 12 // 13 //===----------------------------------------------------------------------===// 14 15 #include "PPC.h" 16 #include "MCTargetDesc/PPCPredicates.h" 17 #include "PPCMachineFunctionInfo.h" 18 #include "PPCTargetMachine.h" 19 #include "llvm/CodeGen/MachineFunction.h" 20 #include "llvm/CodeGen/MachineInstrBuilder.h" 21 #include "llvm/CodeGen/MachineRegisterInfo.h" 22 #include "llvm/CodeGen/SelectionDAG.h" 23 #include "llvm/CodeGen/SelectionDAGISel.h" 24 #include "llvm/IR/Constants.h" 25 #include "llvm/IR/Function.h" 26 #include "llvm/IR/GlobalAlias.h" 27 #include "llvm/IR/GlobalValue.h" 28 #include "llvm/IR/GlobalVariable.h" 29 #include "llvm/IR/Intrinsics.h" 30 #include "llvm/IR/Module.h" 31 #include "llvm/Support/CommandLine.h" 32 #include "llvm/Support/Debug.h" 33 #include "llvm/Support/ErrorHandling.h" 34 #include "llvm/Support/MathExtras.h" 35 #include "llvm/Support/raw_ostream.h" 36 #include "llvm/Target/TargetOptions.h" 37 using namespace llvm; 38 39 #define DEBUG_TYPE "ppc-codegen" 40 41 // FIXME: Remove this once the bug has been fixed! 42 cl::opt<bool> ANDIGlueBug("expose-ppc-andi-glue-bug", 43 cl::desc("expose the ANDI glue bug on PPC"), cl::Hidden); 44 45 namespace llvm { 46 void initializePPCDAGToDAGISelPass(PassRegistry&); 47 } 48 49 namespace { 50 //===--------------------------------------------------------------------===// 51 /// PPCDAGToDAGISel - PPC specific code to select PPC machine 52 /// instructions for SelectionDAG operations. 53 /// 54 class PPCDAGToDAGISel : public SelectionDAGISel { 55 const PPCTargetMachine &TM; 56 const PPCTargetLowering *PPCLowering; 57 const PPCSubtarget *PPCSubTarget; 58 unsigned GlobalBaseReg; 59 public: 60 explicit PPCDAGToDAGISel(PPCTargetMachine &tm) 61 : SelectionDAGISel(tm), TM(tm), 62 PPCLowering(TM.getTargetLowering()), 63 PPCSubTarget(TM.getSubtargetImpl()) { 64 initializePPCDAGToDAGISelPass(*PassRegistry::getPassRegistry()); 65 } 66 67 bool runOnMachineFunction(MachineFunction &MF) override { 68 // Make sure we re-emit a set of the global base reg if necessary 69 GlobalBaseReg = 0; 70 PPCLowering = TM.getTargetLowering(); 71 PPCSubTarget = TM.getSubtargetImpl(); 72 SelectionDAGISel::runOnMachineFunction(MF); 73 74 if (!PPCSubTarget->isSVR4ABI()) 75 InsertVRSaveCode(MF); 76 77 return true; 78 } 79 80 void PostprocessISelDAG() override; 81 82 /// getI32Imm - Return a target constant with the specified value, of type 83 /// i32. 84 inline SDValue getI32Imm(unsigned Imm) { 85 return CurDAG->getTargetConstant(Imm, MVT::i32); 86 } 87 88 /// getI64Imm - Return a target constant with the specified value, of type 89 /// i64. 90 inline SDValue getI64Imm(uint64_t Imm) { 91 return CurDAG->getTargetConstant(Imm, MVT::i64); 92 } 93 94 /// getSmallIPtrImm - Return a target constant of pointer type. 95 inline SDValue getSmallIPtrImm(unsigned Imm) { 96 return CurDAG->getTargetConstant(Imm, PPCLowering->getPointerTy()); 97 } 98 99 /// isRunOfOnes - Returns true iff Val consists of one contiguous run of 1s 100 /// with any number of 0s on either side. The 1s are allowed to wrap from 101 /// LSB to MSB, so 0x000FFF0, 0x0000FFFF, and 0xFF0000FF are all runs. 102 /// 0x0F0F0000 is not, since all 1s are not contiguous. 103 static bool isRunOfOnes(unsigned Val, unsigned &MB, unsigned &ME); 104 105 106 /// isRotateAndMask - Returns true if Mask and Shift can be folded into a 107 /// rotate and mask opcode and mask operation. 108 static bool isRotateAndMask(SDNode *N, unsigned Mask, bool isShiftMask, 109 unsigned &SH, unsigned &MB, unsigned &ME); 110 111 /// getGlobalBaseReg - insert code into the entry mbb to materialize the PIC 112 /// base register. Return the virtual register that holds this value. 113 SDNode *getGlobalBaseReg(); 114 115 // Select - Convert the specified operand from a target-independent to a 116 // target-specific node if it hasn't already been changed. 117 SDNode *Select(SDNode *N) override; 118 119 SDNode *SelectBitfieldInsert(SDNode *N); 120 121 /// SelectCC - Select a comparison of the specified values with the 122 /// specified condition code, returning the CR# of the expression. 123 SDValue SelectCC(SDValue LHS, SDValue RHS, ISD::CondCode CC, SDLoc dl); 124 125 /// SelectAddrImm - Returns true if the address N can be represented by 126 /// a base register plus a signed 16-bit displacement [r+imm]. 127 bool SelectAddrImm(SDValue N, SDValue &Disp, 128 SDValue &Base) { 129 return PPCLowering->SelectAddressRegImm(N, Disp, Base, *CurDAG, false); 130 } 131 132 /// SelectAddrImmOffs - Return true if the operand is valid for a preinc 133 /// immediate field. Note that the operand at this point is already the 134 /// result of a prior SelectAddressRegImm call. 135 bool SelectAddrImmOffs(SDValue N, SDValue &Out) const { 136 if (N.getOpcode() == ISD::TargetConstant || 137 N.getOpcode() == ISD::TargetGlobalAddress) { 138 Out = N; 139 return true; 140 } 141 142 return false; 143 } 144 145 /// SelectAddrIdx - Given the specified addressed, check to see if it can be 146 /// represented as an indexed [r+r] operation. Returns false if it can 147 /// be represented by [r+imm], which are preferred. 148 bool SelectAddrIdx(SDValue N, SDValue &Base, SDValue &Index) { 149 return PPCLowering->SelectAddressRegReg(N, Base, Index, *CurDAG); 150 } 151 152 /// SelectAddrIdxOnly - Given the specified addressed, force it to be 153 /// represented as an indexed [r+r] operation. 154 bool SelectAddrIdxOnly(SDValue N, SDValue &Base, SDValue &Index) { 155 return PPCLowering->SelectAddressRegRegOnly(N, Base, Index, *CurDAG); 156 } 157 158 /// SelectAddrImmX4 - Returns true if the address N can be represented by 159 /// a base register plus a signed 16-bit displacement that is a multiple of 4. 160 /// Suitable for use by STD and friends. 161 bool SelectAddrImmX4(SDValue N, SDValue &Disp, SDValue &Base) { 162 return PPCLowering->SelectAddressRegImm(N, Disp, Base, *CurDAG, true); 163 } 164 165 // Select an address into a single register. 166 bool SelectAddr(SDValue N, SDValue &Base) { 167 Base = N; 168 return true; 169 } 170 171 /// SelectInlineAsmMemoryOperand - Implement addressing mode selection for 172 /// inline asm expressions. It is always correct to compute the value into 173 /// a register. The case of adding a (possibly relocatable) constant to a 174 /// register can be improved, but it is wrong to substitute Reg+Reg for 175 /// Reg in an asm, because the load or store opcode would have to change. 176 bool SelectInlineAsmMemoryOperand(const SDValue &Op, 177 char ConstraintCode, 178 std::vector<SDValue> &OutOps) override { 179 OutOps.push_back(Op); 180 return false; 181 } 182 183 void InsertVRSaveCode(MachineFunction &MF); 184 185 const char *getPassName() const override { 186 return "PowerPC DAG->DAG Pattern Instruction Selection"; 187 } 188 189 // Include the pieces autogenerated from the target description. 190 #include "PPCGenDAGISel.inc" 191 192 private: 193 SDNode *SelectSETCC(SDNode *N); 194 195 void PeepholePPC64(); 196 void PeepholeCROps(); 197 198 bool AllUsersSelectZero(SDNode *N); 199 void SwapAllSelectUsers(SDNode *N); 200 }; 201 } 202 203 /// InsertVRSaveCode - Once the entire function has been instruction selected, 204 /// all virtual registers are created and all machine instructions are built, 205 /// check to see if we need to save/restore VRSAVE. If so, do it. 206 void PPCDAGToDAGISel::InsertVRSaveCode(MachineFunction &Fn) { 207 // Check to see if this function uses vector registers, which means we have to 208 // save and restore the VRSAVE register and update it with the regs we use. 209 // 210 // In this case, there will be virtual registers of vector type created 211 // by the scheduler. Detect them now. 212 bool HasVectorVReg = false; 213 for (unsigned i = 0, e = RegInfo->getNumVirtRegs(); i != e; ++i) { 214 unsigned Reg = TargetRegisterInfo::index2VirtReg(i); 215 if (RegInfo->getRegClass(Reg) == &PPC::VRRCRegClass) { 216 HasVectorVReg = true; 217 break; 218 } 219 } 220 if (!HasVectorVReg) return; // nothing to do. 221 222 // If we have a vector register, we want to emit code into the entry and exit 223 // blocks to save and restore the VRSAVE register. We do this here (instead 224 // of marking all vector instructions as clobbering VRSAVE) for two reasons: 225 // 226 // 1. This (trivially) reduces the load on the register allocator, by not 227 // having to represent the live range of the VRSAVE register. 228 // 2. This (more significantly) allows us to create a temporary virtual 229 // register to hold the saved VRSAVE value, allowing this temporary to be 230 // register allocated, instead of forcing it to be spilled to the stack. 231 232 // Create two vregs - one to hold the VRSAVE register that is live-in to the 233 // function and one for the value after having bits or'd into it. 234 unsigned InVRSAVE = RegInfo->createVirtualRegister(&PPC::GPRCRegClass); 235 unsigned UpdatedVRSAVE = RegInfo->createVirtualRegister(&PPC::GPRCRegClass); 236 237 const TargetInstrInfo &TII = *TM.getInstrInfo(); 238 MachineBasicBlock &EntryBB = *Fn.begin(); 239 DebugLoc dl; 240 // Emit the following code into the entry block: 241 // InVRSAVE = MFVRSAVE 242 // UpdatedVRSAVE = UPDATE_VRSAVE InVRSAVE 243 // MTVRSAVE UpdatedVRSAVE 244 MachineBasicBlock::iterator IP = EntryBB.begin(); // Insert Point 245 BuildMI(EntryBB, IP, dl, TII.get(PPC::MFVRSAVE), InVRSAVE); 246 BuildMI(EntryBB, IP, dl, TII.get(PPC::UPDATE_VRSAVE), 247 UpdatedVRSAVE).addReg(InVRSAVE); 248 BuildMI(EntryBB, IP, dl, TII.get(PPC::MTVRSAVE)).addReg(UpdatedVRSAVE); 249 250 // Find all return blocks, outputting a restore in each epilog. 251 for (MachineFunction::iterator BB = Fn.begin(), E = Fn.end(); BB != E; ++BB) { 252 if (!BB->empty() && BB->back().isReturn()) { 253 IP = BB->end(); --IP; 254 255 // Skip over all terminator instructions, which are part of the return 256 // sequence. 257 MachineBasicBlock::iterator I2 = IP; 258 while (I2 != BB->begin() && (--I2)->isTerminator()) 259 IP = I2; 260 261 // Emit: MTVRSAVE InVRSave 262 BuildMI(*BB, IP, dl, TII.get(PPC::MTVRSAVE)).addReg(InVRSAVE); 263 } 264 } 265 } 266 267 268 /// getGlobalBaseReg - Output the instructions required to put the 269 /// base address to use for accessing globals into a register. 270 /// 271 SDNode *PPCDAGToDAGISel::getGlobalBaseReg() { 272 if (!GlobalBaseReg) { 273 const TargetInstrInfo &TII = *TM.getInstrInfo(); 274 // Insert the set of GlobalBaseReg into the first MBB of the function 275 MachineBasicBlock &FirstMBB = MF->front(); 276 MachineBasicBlock::iterator MBBI = FirstMBB.begin(); 277 const Module *M = MF->getFunction()->getParent(); 278 DebugLoc dl; 279 280 if (PPCLowering->getPointerTy() == MVT::i32) { 281 if (PPCSubTarget->isTargetELF()) { 282 GlobalBaseReg = PPC::R30; 283 if (M->getPICLevel() == PICLevel::Small) { 284 BuildMI(FirstMBB, MBBI, dl, TII.get(PPC::MoveGOTtoLR)); 285 BuildMI(FirstMBB, MBBI, dl, TII.get(PPC::MFLR), GlobalBaseReg); 286 } else { 287 BuildMI(FirstMBB, MBBI, dl, TII.get(PPC::MovePCtoLR)); 288 BuildMI(FirstMBB, MBBI, dl, TII.get(PPC::MFLR), GlobalBaseReg); 289 unsigned TempReg = RegInfo->createVirtualRegister(&PPC::GPRCRegClass); 290 BuildMI(FirstMBB, MBBI, dl, 291 TII.get(PPC::UpdateGBR)).addReg(GlobalBaseReg) 292 .addReg(TempReg, RegState::Define).addReg(GlobalBaseReg); 293 MF->getInfo<PPCFunctionInfo>()->setUsesPICBase(true); 294 } 295 } else { 296 GlobalBaseReg = 297 RegInfo->createVirtualRegister(&PPC::GPRC_NOR0RegClass); 298 BuildMI(FirstMBB, MBBI, dl, TII.get(PPC::MovePCtoLR)); 299 BuildMI(FirstMBB, MBBI, dl, TII.get(PPC::MFLR), GlobalBaseReg); 300 } 301 } else { 302 GlobalBaseReg = RegInfo->createVirtualRegister(&PPC::G8RC_NOX0RegClass); 303 BuildMI(FirstMBB, MBBI, dl, TII.get(PPC::MovePCtoLR8)); 304 BuildMI(FirstMBB, MBBI, dl, TII.get(PPC::MFLR8), GlobalBaseReg); 305 } 306 } 307 return CurDAG->getRegister(GlobalBaseReg, 308 PPCLowering->getPointerTy()).getNode(); 309 } 310 311 /// isIntS16Immediate - This method tests to see if the node is either a 32-bit 312 /// or 64-bit immediate, and if the value can be accurately represented as a 313 /// sign extension from a 16-bit value. If so, this returns true and the 314 /// immediate. 315 static bool isIntS16Immediate(SDNode *N, short &Imm) { 316 if (N->getOpcode() != ISD::Constant) 317 return false; 318 319 Imm = (short)cast<ConstantSDNode>(N)->getZExtValue(); 320 if (N->getValueType(0) == MVT::i32) 321 return Imm == (int32_t)cast<ConstantSDNode>(N)->getZExtValue(); 322 else 323 return Imm == (int64_t)cast<ConstantSDNode>(N)->getZExtValue(); 324 } 325 326 static bool isIntS16Immediate(SDValue Op, short &Imm) { 327 return isIntS16Immediate(Op.getNode(), Imm); 328 } 329 330 331 /// isInt32Immediate - This method tests to see if the node is a 32-bit constant 332 /// operand. If so Imm will receive the 32-bit value. 333 static bool isInt32Immediate(SDNode *N, unsigned &Imm) { 334 if (N->getOpcode() == ISD::Constant && N->getValueType(0) == MVT::i32) { 335 Imm = cast<ConstantSDNode>(N)->getZExtValue(); 336 return true; 337 } 338 return false; 339 } 340 341 /// isInt64Immediate - This method tests to see if the node is a 64-bit constant 342 /// operand. If so Imm will receive the 64-bit value. 343 static bool isInt64Immediate(SDNode *N, uint64_t &Imm) { 344 if (N->getOpcode() == ISD::Constant && N->getValueType(0) == MVT::i64) { 345 Imm = cast<ConstantSDNode>(N)->getZExtValue(); 346 return true; 347 } 348 return false; 349 } 350 351 // isInt32Immediate - This method tests to see if a constant operand. 352 // If so Imm will receive the 32 bit value. 353 static bool isInt32Immediate(SDValue N, unsigned &Imm) { 354 return isInt32Immediate(N.getNode(), Imm); 355 } 356 357 358 // isOpcWithIntImmediate - This method tests to see if the node is a specific 359 // opcode and that it has a immediate integer right operand. 360 // If so Imm will receive the 32 bit value. 361 static bool isOpcWithIntImmediate(SDNode *N, unsigned Opc, unsigned& Imm) { 362 return N->getOpcode() == Opc 363 && isInt32Immediate(N->getOperand(1).getNode(), Imm); 364 } 365 366 bool PPCDAGToDAGISel::isRunOfOnes(unsigned Val, unsigned &MB, unsigned &ME) { 367 if (!Val) 368 return false; 369 370 if (isShiftedMask_32(Val)) { 371 // look for the first non-zero bit 372 MB = countLeadingZeros(Val); 373 // look for the first zero bit after the run of ones 374 ME = countLeadingZeros((Val - 1) ^ Val); 375 return true; 376 } else { 377 Val = ~Val; // invert mask 378 if (isShiftedMask_32(Val)) { 379 // effectively look for the first zero bit 380 ME = countLeadingZeros(Val) - 1; 381 // effectively look for the first one bit after the run of zeros 382 MB = countLeadingZeros((Val - 1) ^ Val) + 1; 383 return true; 384 } 385 } 386 // no run present 387 return false; 388 } 389 390 bool PPCDAGToDAGISel::isRotateAndMask(SDNode *N, unsigned Mask, 391 bool isShiftMask, unsigned &SH, 392 unsigned &MB, unsigned &ME) { 393 // Don't even go down this path for i64, since different logic will be 394 // necessary for rldicl/rldicr/rldimi. 395 if (N->getValueType(0) != MVT::i32) 396 return false; 397 398 unsigned Shift = 32; 399 unsigned Indeterminant = ~0; // bit mask marking indeterminant results 400 unsigned Opcode = N->getOpcode(); 401 if (N->getNumOperands() != 2 || 402 !isInt32Immediate(N->getOperand(1).getNode(), Shift) || (Shift > 31)) 403 return false; 404 405 if (Opcode == ISD::SHL) { 406 // apply shift left to mask if it comes first 407 if (isShiftMask) Mask = Mask << Shift; 408 // determine which bits are made indeterminant by shift 409 Indeterminant = ~(0xFFFFFFFFu << Shift); 410 } else if (Opcode == ISD::SRL) { 411 // apply shift right to mask if it comes first 412 if (isShiftMask) Mask = Mask >> Shift; 413 // determine which bits are made indeterminant by shift 414 Indeterminant = ~(0xFFFFFFFFu >> Shift); 415 // adjust for the left rotate 416 Shift = 32 - Shift; 417 } else if (Opcode == ISD::ROTL) { 418 Indeterminant = 0; 419 } else { 420 return false; 421 } 422 423 // if the mask doesn't intersect any Indeterminant bits 424 if (Mask && !(Mask & Indeterminant)) { 425 SH = Shift & 31; 426 // make sure the mask is still a mask (wrap arounds may not be) 427 return isRunOfOnes(Mask, MB, ME); 428 } 429 return false; 430 } 431 432 /// SelectBitfieldInsert - turn an or of two masked values into 433 /// the rotate left word immediate then mask insert (rlwimi) instruction. 434 SDNode *PPCDAGToDAGISel::SelectBitfieldInsert(SDNode *N) { 435 SDValue Op0 = N->getOperand(0); 436 SDValue Op1 = N->getOperand(1); 437 SDLoc dl(N); 438 439 APInt LKZ, LKO, RKZ, RKO; 440 CurDAG->computeKnownBits(Op0, LKZ, LKO); 441 CurDAG->computeKnownBits(Op1, RKZ, RKO); 442 443 unsigned TargetMask = LKZ.getZExtValue(); 444 unsigned InsertMask = RKZ.getZExtValue(); 445 446 if ((TargetMask | InsertMask) == 0xFFFFFFFF) { 447 unsigned Op0Opc = Op0.getOpcode(); 448 unsigned Op1Opc = Op1.getOpcode(); 449 unsigned Value, SH = 0; 450 TargetMask = ~TargetMask; 451 InsertMask = ~InsertMask; 452 453 // If the LHS has a foldable shift and the RHS does not, then swap it to the 454 // RHS so that we can fold the shift into the insert. 455 if (Op0Opc == ISD::AND && Op1Opc == ISD::AND) { 456 if (Op0.getOperand(0).getOpcode() == ISD::SHL || 457 Op0.getOperand(0).getOpcode() == ISD::SRL) { 458 if (Op1.getOperand(0).getOpcode() != ISD::SHL && 459 Op1.getOperand(0).getOpcode() != ISD::SRL) { 460 std::swap(Op0, Op1); 461 std::swap(Op0Opc, Op1Opc); 462 std::swap(TargetMask, InsertMask); 463 } 464 } 465 } else if (Op0Opc == ISD::SHL || Op0Opc == ISD::SRL) { 466 if (Op1Opc == ISD::AND && Op1.getOperand(0).getOpcode() != ISD::SHL && 467 Op1.getOperand(0).getOpcode() != ISD::SRL) { 468 std::swap(Op0, Op1); 469 std::swap(Op0Opc, Op1Opc); 470 std::swap(TargetMask, InsertMask); 471 } 472 } 473 474 unsigned MB, ME; 475 if (isRunOfOnes(InsertMask, MB, ME)) { 476 SDValue Tmp1, Tmp2; 477 478 if ((Op1Opc == ISD::SHL || Op1Opc == ISD::SRL) && 479 isInt32Immediate(Op1.getOperand(1), Value)) { 480 Op1 = Op1.getOperand(0); 481 SH = (Op1Opc == ISD::SHL) ? Value : 32 - Value; 482 } 483 if (Op1Opc == ISD::AND) { 484 // The AND mask might not be a constant, and we need to make sure that 485 // if we're going to fold the masking with the insert, all bits not 486 // know to be zero in the mask are known to be one. 487 APInt MKZ, MKO; 488 CurDAG->computeKnownBits(Op1.getOperand(1), MKZ, MKO); 489 bool CanFoldMask = InsertMask == MKO.getZExtValue(); 490 491 unsigned SHOpc = Op1.getOperand(0).getOpcode(); 492 if ((SHOpc == ISD::SHL || SHOpc == ISD::SRL) && CanFoldMask && 493 isInt32Immediate(Op1.getOperand(0).getOperand(1), Value)) { 494 // Note that Value must be in range here (less than 32) because 495 // otherwise there would not be any bits set in InsertMask. 496 Op1 = Op1.getOperand(0).getOperand(0); 497 SH = (SHOpc == ISD::SHL) ? Value : 32 - Value; 498 } 499 } 500 501 SH &= 31; 502 SDValue Ops[] = { Op0, Op1, getI32Imm(SH), getI32Imm(MB), 503 getI32Imm(ME) }; 504 return CurDAG->getMachineNode(PPC::RLWIMI, dl, MVT::i32, Ops); 505 } 506 } 507 return nullptr; 508 } 509 510 /// SelectCC - Select a comparison of the specified values with the specified 511 /// condition code, returning the CR# of the expression. 512 SDValue PPCDAGToDAGISel::SelectCC(SDValue LHS, SDValue RHS, 513 ISD::CondCode CC, SDLoc dl) { 514 // Always select the LHS. 515 unsigned Opc; 516 517 if (LHS.getValueType() == MVT::i32) { 518 unsigned Imm; 519 if (CC == ISD::SETEQ || CC == ISD::SETNE) { 520 if (isInt32Immediate(RHS, Imm)) { 521 // SETEQ/SETNE comparison with 16-bit immediate, fold it. 522 if (isUInt<16>(Imm)) 523 return SDValue(CurDAG->getMachineNode(PPC::CMPLWI, dl, MVT::i32, LHS, 524 getI32Imm(Imm & 0xFFFF)), 0); 525 // If this is a 16-bit signed immediate, fold it. 526 if (isInt<16>((int)Imm)) 527 return SDValue(CurDAG->getMachineNode(PPC::CMPWI, dl, MVT::i32, LHS, 528 getI32Imm(Imm & 0xFFFF)), 0); 529 530 // For non-equality comparisons, the default code would materialize the 531 // constant, then compare against it, like this: 532 // lis r2, 4660 533 // ori r2, r2, 22136 534 // cmpw cr0, r3, r2 535 // Since we are just comparing for equality, we can emit this instead: 536 // xoris r0,r3,0x1234 537 // cmplwi cr0,r0,0x5678 538 // beq cr0,L6 539 SDValue Xor(CurDAG->getMachineNode(PPC::XORIS, dl, MVT::i32, LHS, 540 getI32Imm(Imm >> 16)), 0); 541 return SDValue(CurDAG->getMachineNode(PPC::CMPLWI, dl, MVT::i32, Xor, 542 getI32Imm(Imm & 0xFFFF)), 0); 543 } 544 Opc = PPC::CMPLW; 545 } else if (ISD::isUnsignedIntSetCC(CC)) { 546 if (isInt32Immediate(RHS, Imm) && isUInt<16>(Imm)) 547 return SDValue(CurDAG->getMachineNode(PPC::CMPLWI, dl, MVT::i32, LHS, 548 getI32Imm(Imm & 0xFFFF)), 0); 549 Opc = PPC::CMPLW; 550 } else { 551 short SImm; 552 if (isIntS16Immediate(RHS, SImm)) 553 return SDValue(CurDAG->getMachineNode(PPC::CMPWI, dl, MVT::i32, LHS, 554 getI32Imm((int)SImm & 0xFFFF)), 555 0); 556 Opc = PPC::CMPW; 557 } 558 } else if (LHS.getValueType() == MVT::i64) { 559 uint64_t Imm; 560 if (CC == ISD::SETEQ || CC == ISD::SETNE) { 561 if (isInt64Immediate(RHS.getNode(), Imm)) { 562 // SETEQ/SETNE comparison with 16-bit immediate, fold it. 563 if (isUInt<16>(Imm)) 564 return SDValue(CurDAG->getMachineNode(PPC::CMPLDI, dl, MVT::i64, LHS, 565 getI32Imm(Imm & 0xFFFF)), 0); 566 // If this is a 16-bit signed immediate, fold it. 567 if (isInt<16>(Imm)) 568 return SDValue(CurDAG->getMachineNode(PPC::CMPDI, dl, MVT::i64, LHS, 569 getI32Imm(Imm & 0xFFFF)), 0); 570 571 // For non-equality comparisons, the default code would materialize the 572 // constant, then compare against it, like this: 573 // lis r2, 4660 574 // ori r2, r2, 22136 575 // cmpd cr0, r3, r2 576 // Since we are just comparing for equality, we can emit this instead: 577 // xoris r0,r3,0x1234 578 // cmpldi cr0,r0,0x5678 579 // beq cr0,L6 580 if (isUInt<32>(Imm)) { 581 SDValue Xor(CurDAG->getMachineNode(PPC::XORIS8, dl, MVT::i64, LHS, 582 getI64Imm(Imm >> 16)), 0); 583 return SDValue(CurDAG->getMachineNode(PPC::CMPLDI, dl, MVT::i64, Xor, 584 getI64Imm(Imm & 0xFFFF)), 0); 585 } 586 } 587 Opc = PPC::CMPLD; 588 } else if (ISD::isUnsignedIntSetCC(CC)) { 589 if (isInt64Immediate(RHS.getNode(), Imm) && isUInt<16>(Imm)) 590 return SDValue(CurDAG->getMachineNode(PPC::CMPLDI, dl, MVT::i64, LHS, 591 getI64Imm(Imm & 0xFFFF)), 0); 592 Opc = PPC::CMPLD; 593 } else { 594 short SImm; 595 if (isIntS16Immediate(RHS, SImm)) 596 return SDValue(CurDAG->getMachineNode(PPC::CMPDI, dl, MVT::i64, LHS, 597 getI64Imm(SImm & 0xFFFF)), 598 0); 599 Opc = PPC::CMPD; 600 } 601 } else if (LHS.getValueType() == MVT::f32) { 602 Opc = PPC::FCMPUS; 603 } else { 604 assert(LHS.getValueType() == MVT::f64 && "Unknown vt!"); 605 Opc = PPCSubTarget->hasVSX() ? PPC::XSCMPUDP : PPC::FCMPUD; 606 } 607 return SDValue(CurDAG->getMachineNode(Opc, dl, MVT::i32, LHS, RHS), 0); 608 } 609 610 static PPC::Predicate getPredicateForSetCC(ISD::CondCode CC) { 611 switch (CC) { 612 case ISD::SETUEQ: 613 case ISD::SETONE: 614 case ISD::SETOLE: 615 case ISD::SETOGE: 616 llvm_unreachable("Should be lowered by legalize!"); 617 default: llvm_unreachable("Unknown condition!"); 618 case ISD::SETOEQ: 619 case ISD::SETEQ: return PPC::PRED_EQ; 620 case ISD::SETUNE: 621 case ISD::SETNE: return PPC::PRED_NE; 622 case ISD::SETOLT: 623 case ISD::SETLT: return PPC::PRED_LT; 624 case ISD::SETULE: 625 case ISD::SETLE: return PPC::PRED_LE; 626 case ISD::SETOGT: 627 case ISD::SETGT: return PPC::PRED_GT; 628 case ISD::SETUGE: 629 case ISD::SETGE: return PPC::PRED_GE; 630 case ISD::SETO: return PPC::PRED_NU; 631 case ISD::SETUO: return PPC::PRED_UN; 632 // These two are invalid for floating point. Assume we have int. 633 case ISD::SETULT: return PPC::PRED_LT; 634 case ISD::SETUGT: return PPC::PRED_GT; 635 } 636 } 637 638 /// getCRIdxForSetCC - Return the index of the condition register field 639 /// associated with the SetCC condition, and whether or not the field is 640 /// treated as inverted. That is, lt = 0; ge = 0 inverted. 641 static unsigned getCRIdxForSetCC(ISD::CondCode CC, bool &Invert) { 642 Invert = false; 643 switch (CC) { 644 default: llvm_unreachable("Unknown condition!"); 645 case ISD::SETOLT: 646 case ISD::SETLT: return 0; // Bit #0 = SETOLT 647 case ISD::SETOGT: 648 case ISD::SETGT: return 1; // Bit #1 = SETOGT 649 case ISD::SETOEQ: 650 case ISD::SETEQ: return 2; // Bit #2 = SETOEQ 651 case ISD::SETUO: return 3; // Bit #3 = SETUO 652 case ISD::SETUGE: 653 case ISD::SETGE: Invert = true; return 0; // !Bit #0 = SETUGE 654 case ISD::SETULE: 655 case ISD::SETLE: Invert = true; return 1; // !Bit #1 = SETULE 656 case ISD::SETUNE: 657 case ISD::SETNE: Invert = true; return 2; // !Bit #2 = SETUNE 658 case ISD::SETO: Invert = true; return 3; // !Bit #3 = SETO 659 case ISD::SETUEQ: 660 case ISD::SETOGE: 661 case ISD::SETOLE: 662 case ISD::SETONE: 663 llvm_unreachable("Invalid branch code: should be expanded by legalize"); 664 // These are invalid for floating point. Assume integer. 665 case ISD::SETULT: return 0; 666 case ISD::SETUGT: return 1; 667 } 668 } 669 670 // getVCmpInst: return the vector compare instruction for the specified 671 // vector type and condition code. Since this is for altivec specific code, 672 // only support the altivec types (v16i8, v8i16, v4i32, and v4f32). 673 static unsigned int getVCmpInst(MVT VecVT, ISD::CondCode CC, 674 bool HasVSX, bool &Swap, bool &Negate) { 675 Swap = false; 676 Negate = false; 677 678 if (VecVT.isFloatingPoint()) { 679 /* Handle some cases by swapping input operands. */ 680 switch (CC) { 681 case ISD::SETLE: CC = ISD::SETGE; Swap = true; break; 682 case ISD::SETLT: CC = ISD::SETGT; Swap = true; break; 683 case ISD::SETOLE: CC = ISD::SETOGE; Swap = true; break; 684 case ISD::SETOLT: CC = ISD::SETOGT; Swap = true; break; 685 case ISD::SETUGE: CC = ISD::SETULE; Swap = true; break; 686 case ISD::SETUGT: CC = ISD::SETULT; Swap = true; break; 687 default: break; 688 } 689 /* Handle some cases by negating the result. */ 690 switch (CC) { 691 case ISD::SETNE: CC = ISD::SETEQ; Negate = true; break; 692 case ISD::SETUNE: CC = ISD::SETOEQ; Negate = true; break; 693 case ISD::SETULE: CC = ISD::SETOGT; Negate = true; break; 694 case ISD::SETULT: CC = ISD::SETOGE; Negate = true; break; 695 default: break; 696 } 697 /* We have instructions implementing the remaining cases. */ 698 switch (CC) { 699 case ISD::SETEQ: 700 case ISD::SETOEQ: 701 if (VecVT == MVT::v4f32) 702 return HasVSX ? PPC::XVCMPEQSP : PPC::VCMPEQFP; 703 else if (VecVT == MVT::v2f64) 704 return PPC::XVCMPEQDP; 705 break; 706 case ISD::SETGT: 707 case ISD::SETOGT: 708 if (VecVT == MVT::v4f32) 709 return HasVSX ? PPC::XVCMPGTSP : PPC::VCMPGTFP; 710 else if (VecVT == MVT::v2f64) 711 return PPC::XVCMPGTDP; 712 break; 713 case ISD::SETGE: 714 case ISD::SETOGE: 715 if (VecVT == MVT::v4f32) 716 return HasVSX ? PPC::XVCMPGESP : PPC::VCMPGEFP; 717 else if (VecVT == MVT::v2f64) 718 return PPC::XVCMPGEDP; 719 break; 720 default: 721 break; 722 } 723 llvm_unreachable("Invalid floating-point vector compare condition"); 724 } else { 725 /* Handle some cases by swapping input operands. */ 726 switch (CC) { 727 case ISD::SETGE: CC = ISD::SETLE; Swap = true; break; 728 case ISD::SETLT: CC = ISD::SETGT; Swap = true; break; 729 case ISD::SETUGE: CC = ISD::SETULE; Swap = true; break; 730 case ISD::SETULT: CC = ISD::SETUGT; Swap = true; break; 731 default: break; 732 } 733 /* Handle some cases by negating the result. */ 734 switch (CC) { 735 case ISD::SETNE: CC = ISD::SETEQ; Negate = true; break; 736 case ISD::SETUNE: CC = ISD::SETUEQ; Negate = true; break; 737 case ISD::SETLE: CC = ISD::SETGT; Negate = true; break; 738 case ISD::SETULE: CC = ISD::SETUGT; Negate = true; break; 739 default: break; 740 } 741 /* We have instructions implementing the remaining cases. */ 742 switch (CC) { 743 case ISD::SETEQ: 744 case ISD::SETUEQ: 745 if (VecVT == MVT::v16i8) 746 return PPC::VCMPEQUB; 747 else if (VecVT == MVT::v8i16) 748 return PPC::VCMPEQUH; 749 else if (VecVT == MVT::v4i32) 750 return PPC::VCMPEQUW; 751 break; 752 case ISD::SETGT: 753 if (VecVT == MVT::v16i8) 754 return PPC::VCMPGTSB; 755 else if (VecVT == MVT::v8i16) 756 return PPC::VCMPGTSH; 757 else if (VecVT == MVT::v4i32) 758 return PPC::VCMPGTSW; 759 break; 760 case ISD::SETUGT: 761 if (VecVT == MVT::v16i8) 762 return PPC::VCMPGTUB; 763 else if (VecVT == MVT::v8i16) 764 return PPC::VCMPGTUH; 765 else if (VecVT == MVT::v4i32) 766 return PPC::VCMPGTUW; 767 break; 768 default: 769 break; 770 } 771 llvm_unreachable("Invalid integer vector compare condition"); 772 } 773 } 774 775 SDNode *PPCDAGToDAGISel::SelectSETCC(SDNode *N) { 776 SDLoc dl(N); 777 unsigned Imm; 778 ISD::CondCode CC = cast<CondCodeSDNode>(N->getOperand(2))->get(); 779 EVT PtrVT = CurDAG->getTargetLoweringInfo().getPointerTy(); 780 bool isPPC64 = (PtrVT == MVT::i64); 781 782 if (!PPCSubTarget->useCRBits() && 783 isInt32Immediate(N->getOperand(1), Imm)) { 784 // We can codegen setcc op, imm very efficiently compared to a brcond. 785 // Check for those cases here. 786 // setcc op, 0 787 if (Imm == 0) { 788 SDValue Op = N->getOperand(0); 789 switch (CC) { 790 default: break; 791 case ISD::SETEQ: { 792 Op = SDValue(CurDAG->getMachineNode(PPC::CNTLZW, dl, MVT::i32, Op), 0); 793 SDValue Ops[] = { Op, getI32Imm(27), getI32Imm(5), getI32Imm(31) }; 794 return CurDAG->SelectNodeTo(N, PPC::RLWINM, MVT::i32, Ops); 795 } 796 case ISD::SETNE: { 797 if (isPPC64) break; 798 SDValue AD = 799 SDValue(CurDAG->getMachineNode(PPC::ADDIC, dl, MVT::i32, MVT::Glue, 800 Op, getI32Imm(~0U)), 0); 801 return CurDAG->SelectNodeTo(N, PPC::SUBFE, MVT::i32, AD, Op, 802 AD.getValue(1)); 803 } 804 case ISD::SETLT: { 805 SDValue Ops[] = { Op, getI32Imm(1), getI32Imm(31), getI32Imm(31) }; 806 return CurDAG->SelectNodeTo(N, PPC::RLWINM, MVT::i32, Ops); 807 } 808 case ISD::SETGT: { 809 SDValue T = 810 SDValue(CurDAG->getMachineNode(PPC::NEG, dl, MVT::i32, Op), 0); 811 T = SDValue(CurDAG->getMachineNode(PPC::ANDC, dl, MVT::i32, T, Op), 0); 812 SDValue Ops[] = { T, getI32Imm(1), getI32Imm(31), getI32Imm(31) }; 813 return CurDAG->SelectNodeTo(N, PPC::RLWINM, MVT::i32, Ops); 814 } 815 } 816 } else if (Imm == ~0U) { // setcc op, -1 817 SDValue Op = N->getOperand(0); 818 switch (CC) { 819 default: break; 820 case ISD::SETEQ: 821 if (isPPC64) break; 822 Op = SDValue(CurDAG->getMachineNode(PPC::ADDIC, dl, MVT::i32, MVT::Glue, 823 Op, getI32Imm(1)), 0); 824 return CurDAG->SelectNodeTo(N, PPC::ADDZE, MVT::i32, 825 SDValue(CurDAG->getMachineNode(PPC::LI, dl, 826 MVT::i32, 827 getI32Imm(0)), 0), 828 Op.getValue(1)); 829 case ISD::SETNE: { 830 if (isPPC64) break; 831 Op = SDValue(CurDAG->getMachineNode(PPC::NOR, dl, MVT::i32, Op, Op), 0); 832 SDNode *AD = CurDAG->getMachineNode(PPC::ADDIC, dl, MVT::i32, MVT::Glue, 833 Op, getI32Imm(~0U)); 834 return CurDAG->SelectNodeTo(N, PPC::SUBFE, MVT::i32, SDValue(AD, 0), 835 Op, SDValue(AD, 1)); 836 } 837 case ISD::SETLT: { 838 SDValue AD = SDValue(CurDAG->getMachineNode(PPC::ADDI, dl, MVT::i32, Op, 839 getI32Imm(1)), 0); 840 SDValue AN = SDValue(CurDAG->getMachineNode(PPC::AND, dl, MVT::i32, AD, 841 Op), 0); 842 SDValue Ops[] = { AN, getI32Imm(1), getI32Imm(31), getI32Imm(31) }; 843 return CurDAG->SelectNodeTo(N, PPC::RLWINM, MVT::i32, Ops); 844 } 845 case ISD::SETGT: { 846 SDValue Ops[] = { Op, getI32Imm(1), getI32Imm(31), getI32Imm(31) }; 847 Op = SDValue(CurDAG->getMachineNode(PPC::RLWINM, dl, MVT::i32, Ops), 848 0); 849 return CurDAG->SelectNodeTo(N, PPC::XORI, MVT::i32, Op, 850 getI32Imm(1)); 851 } 852 } 853 } 854 } 855 856 SDValue LHS = N->getOperand(0); 857 SDValue RHS = N->getOperand(1); 858 859 // Altivec Vector compare instructions do not set any CR register by default and 860 // vector compare operations return the same type as the operands. 861 if (LHS.getValueType().isVector()) { 862 EVT VecVT = LHS.getValueType(); 863 bool Swap, Negate; 864 unsigned int VCmpInst = getVCmpInst(VecVT.getSimpleVT(), CC, 865 PPCSubTarget->hasVSX(), Swap, Negate); 866 if (Swap) 867 std::swap(LHS, RHS); 868 869 if (Negate) { 870 SDValue VCmp(CurDAG->getMachineNode(VCmpInst, dl, VecVT, LHS, RHS), 0); 871 return CurDAG->SelectNodeTo(N, PPCSubTarget->hasVSX() ? PPC::XXLNOR : 872 PPC::VNOR, 873 VecVT, VCmp, VCmp); 874 } 875 876 return CurDAG->SelectNodeTo(N, VCmpInst, VecVT, LHS, RHS); 877 } 878 879 if (PPCSubTarget->useCRBits()) 880 return nullptr; 881 882 bool Inv; 883 unsigned Idx = getCRIdxForSetCC(CC, Inv); 884 SDValue CCReg = SelectCC(LHS, RHS, CC, dl); 885 SDValue IntCR; 886 887 // Force the ccreg into CR7. 888 SDValue CR7Reg = CurDAG->getRegister(PPC::CR7, MVT::i32); 889 890 SDValue InFlag(nullptr, 0); // Null incoming flag value. 891 CCReg = CurDAG->getCopyToReg(CurDAG->getEntryNode(), dl, CR7Reg, CCReg, 892 InFlag).getValue(1); 893 894 IntCR = SDValue(CurDAG->getMachineNode(PPC::MFOCRF, dl, MVT::i32, CR7Reg, 895 CCReg), 0); 896 897 SDValue Ops[] = { IntCR, getI32Imm((32-(3-Idx)) & 31), 898 getI32Imm(31), getI32Imm(31) }; 899 if (!Inv) 900 return CurDAG->SelectNodeTo(N, PPC::RLWINM, MVT::i32, Ops); 901 902 // Get the specified bit. 903 SDValue Tmp = 904 SDValue(CurDAG->getMachineNode(PPC::RLWINM, dl, MVT::i32, Ops), 0); 905 return CurDAG->SelectNodeTo(N, PPC::XORI, MVT::i32, Tmp, getI32Imm(1)); 906 } 907 908 909 // Select - Convert the specified operand from a target-independent to a 910 // target-specific node if it hasn't already been changed. 911 SDNode *PPCDAGToDAGISel::Select(SDNode *N) { 912 SDLoc dl(N); 913 if (N->isMachineOpcode()) { 914 N->setNodeId(-1); 915 return nullptr; // Already selected. 916 } 917 918 switch (N->getOpcode()) { 919 default: break; 920 921 case ISD::Constant: { 922 if (N->getValueType(0) == MVT::i64) { 923 // Get 64 bit value. 924 int64_t Imm = cast<ConstantSDNode>(N)->getZExtValue(); 925 // Assume no remaining bits. 926 unsigned Remainder = 0; 927 // Assume no shift required. 928 unsigned Shift = 0; 929 930 // If it can't be represented as a 32 bit value. 931 if (!isInt<32>(Imm)) { 932 Shift = countTrailingZeros<uint64_t>(Imm); 933 int64_t ImmSh = static_cast<uint64_t>(Imm) >> Shift; 934 935 // If the shifted value fits 32 bits. 936 if (isInt<32>(ImmSh)) { 937 // Go with the shifted value. 938 Imm = ImmSh; 939 } else { 940 // Still stuck with a 64 bit value. 941 Remainder = Imm; 942 Shift = 32; 943 Imm >>= 32; 944 } 945 } 946 947 // Intermediate operand. 948 SDNode *Result; 949 950 // Handle first 32 bits. 951 unsigned Lo = Imm & 0xFFFF; 952 unsigned Hi = (Imm >> 16) & 0xFFFF; 953 954 // Simple value. 955 if (isInt<16>(Imm)) { 956 // Just the Lo bits. 957 Result = CurDAG->getMachineNode(PPC::LI8, dl, MVT::i64, getI32Imm(Lo)); 958 } else if (Lo) { 959 // Handle the Hi bits. 960 unsigned OpC = Hi ? PPC::LIS8 : PPC::LI8; 961 Result = CurDAG->getMachineNode(OpC, dl, MVT::i64, getI32Imm(Hi)); 962 // And Lo bits. 963 Result = CurDAG->getMachineNode(PPC::ORI8, dl, MVT::i64, 964 SDValue(Result, 0), getI32Imm(Lo)); 965 } else { 966 // Just the Hi bits. 967 Result = CurDAG->getMachineNode(PPC::LIS8, dl, MVT::i64, getI32Imm(Hi)); 968 } 969 970 // If no shift, we're done. 971 if (!Shift) return Result; 972 973 // Shift for next step if the upper 32-bits were not zero. 974 if (Imm) { 975 Result = CurDAG->getMachineNode(PPC::RLDICR, dl, MVT::i64, 976 SDValue(Result, 0), 977 getI32Imm(Shift), 978 getI32Imm(63 - Shift)); 979 } 980 981 // Add in the last bits as required. 982 if ((Hi = (Remainder >> 16) & 0xFFFF)) { 983 Result = CurDAG->getMachineNode(PPC::ORIS8, dl, MVT::i64, 984 SDValue(Result, 0), getI32Imm(Hi)); 985 } 986 if ((Lo = Remainder & 0xFFFF)) { 987 Result = CurDAG->getMachineNode(PPC::ORI8, dl, MVT::i64, 988 SDValue(Result, 0), getI32Imm(Lo)); 989 } 990 991 return Result; 992 } 993 break; 994 } 995 996 case ISD::SETCC: { 997 SDNode *SN = SelectSETCC(N); 998 if (SN) 999 return SN; 1000 break; 1001 } 1002 case PPCISD::GlobalBaseReg: 1003 return getGlobalBaseReg(); 1004 1005 case ISD::FrameIndex: { 1006 int FI = cast<FrameIndexSDNode>(N)->getIndex(); 1007 SDValue TFI = CurDAG->getTargetFrameIndex(FI, N->getValueType(0)); 1008 unsigned Opc = N->getValueType(0) == MVT::i32 ? PPC::ADDI : PPC::ADDI8; 1009 if (N->hasOneUse()) 1010 return CurDAG->SelectNodeTo(N, Opc, N->getValueType(0), TFI, 1011 getSmallIPtrImm(0)); 1012 return CurDAG->getMachineNode(Opc, dl, N->getValueType(0), TFI, 1013 getSmallIPtrImm(0)); 1014 } 1015 1016 case PPCISD::MFOCRF: { 1017 SDValue InFlag = N->getOperand(1); 1018 return CurDAG->getMachineNode(PPC::MFOCRF, dl, MVT::i32, 1019 N->getOperand(0), InFlag); 1020 } 1021 1022 case ISD::SDIV: { 1023 // FIXME: since this depends on the setting of the carry flag from the srawi 1024 // we should really be making notes about that for the scheduler. 1025 // FIXME: It sure would be nice if we could cheaply recognize the 1026 // srl/add/sra pattern the dag combiner will generate for this as 1027 // sra/addze rather than having to handle sdiv ourselves. oh well. 1028 unsigned Imm; 1029 if (isInt32Immediate(N->getOperand(1), Imm)) { 1030 SDValue N0 = N->getOperand(0); 1031 if ((signed)Imm > 0 && isPowerOf2_32(Imm)) { 1032 SDNode *Op = 1033 CurDAG->getMachineNode(PPC::SRAWI, dl, MVT::i32, MVT::Glue, 1034 N0, getI32Imm(Log2_32(Imm))); 1035 return CurDAG->SelectNodeTo(N, PPC::ADDZE, MVT::i32, 1036 SDValue(Op, 0), SDValue(Op, 1)); 1037 } else if ((signed)Imm < 0 && isPowerOf2_32(-Imm)) { 1038 SDNode *Op = 1039 CurDAG->getMachineNode(PPC::SRAWI, dl, MVT::i32, MVT::Glue, 1040 N0, getI32Imm(Log2_32(-Imm))); 1041 SDValue PT = 1042 SDValue(CurDAG->getMachineNode(PPC::ADDZE, dl, MVT::i32, 1043 SDValue(Op, 0), SDValue(Op, 1)), 1044 0); 1045 return CurDAG->SelectNodeTo(N, PPC::NEG, MVT::i32, PT); 1046 } 1047 } 1048 1049 // Other cases are autogenerated. 1050 break; 1051 } 1052 1053 case ISD::LOAD: { 1054 // Handle preincrement loads. 1055 LoadSDNode *LD = cast<LoadSDNode>(N); 1056 EVT LoadedVT = LD->getMemoryVT(); 1057 1058 // Normal loads are handled by code generated from the .td file. 1059 if (LD->getAddressingMode() != ISD::PRE_INC) 1060 break; 1061 1062 SDValue Offset = LD->getOffset(); 1063 if (Offset.getOpcode() == ISD::TargetConstant || 1064 Offset.getOpcode() == ISD::TargetGlobalAddress) { 1065 1066 unsigned Opcode; 1067 bool isSExt = LD->getExtensionType() == ISD::SEXTLOAD; 1068 if (LD->getValueType(0) != MVT::i64) { 1069 // Handle PPC32 integer and normal FP loads. 1070 assert((!isSExt || LoadedVT == MVT::i16) && "Invalid sext update load"); 1071 switch (LoadedVT.getSimpleVT().SimpleTy) { 1072 default: llvm_unreachable("Invalid PPC load type!"); 1073 case MVT::f64: Opcode = PPC::LFDU; break; 1074 case MVT::f32: Opcode = PPC::LFSU; break; 1075 case MVT::i32: Opcode = PPC::LWZU; break; 1076 case MVT::i16: Opcode = isSExt ? PPC::LHAU : PPC::LHZU; break; 1077 case MVT::i1: 1078 case MVT::i8: Opcode = PPC::LBZU; break; 1079 } 1080 } else { 1081 assert(LD->getValueType(0) == MVT::i64 && "Unknown load result type!"); 1082 assert((!isSExt || LoadedVT == MVT::i16) && "Invalid sext update load"); 1083 switch (LoadedVT.getSimpleVT().SimpleTy) { 1084 default: llvm_unreachable("Invalid PPC load type!"); 1085 case MVT::i64: Opcode = PPC::LDU; break; 1086 case MVT::i32: Opcode = PPC::LWZU8; break; 1087 case MVT::i16: Opcode = isSExt ? PPC::LHAU8 : PPC::LHZU8; break; 1088 case MVT::i1: 1089 case MVT::i8: Opcode = PPC::LBZU8; break; 1090 } 1091 } 1092 1093 SDValue Chain = LD->getChain(); 1094 SDValue Base = LD->getBasePtr(); 1095 SDValue Ops[] = { Offset, Base, Chain }; 1096 return CurDAG->getMachineNode(Opcode, dl, LD->getValueType(0), 1097 PPCLowering->getPointerTy(), 1098 MVT::Other, Ops); 1099 } else { 1100 unsigned Opcode; 1101 bool isSExt = LD->getExtensionType() == ISD::SEXTLOAD; 1102 if (LD->getValueType(0) != MVT::i64) { 1103 // Handle PPC32 integer and normal FP loads. 1104 assert((!isSExt || LoadedVT == MVT::i16) && "Invalid sext update load"); 1105 switch (LoadedVT.getSimpleVT().SimpleTy) { 1106 default: llvm_unreachable("Invalid PPC load type!"); 1107 case MVT::f64: Opcode = PPC::LFDUX; break; 1108 case MVT::f32: Opcode = PPC::LFSUX; break; 1109 case MVT::i32: Opcode = PPC::LWZUX; break; 1110 case MVT::i16: Opcode = isSExt ? PPC::LHAUX : PPC::LHZUX; break; 1111 case MVT::i1: 1112 case MVT::i8: Opcode = PPC::LBZUX; break; 1113 } 1114 } else { 1115 assert(LD->getValueType(0) == MVT::i64 && "Unknown load result type!"); 1116 assert((!isSExt || LoadedVT == MVT::i16 || LoadedVT == MVT::i32) && 1117 "Invalid sext update load"); 1118 switch (LoadedVT.getSimpleVT().SimpleTy) { 1119 default: llvm_unreachable("Invalid PPC load type!"); 1120 case MVT::i64: Opcode = PPC::LDUX; break; 1121 case MVT::i32: Opcode = isSExt ? PPC::LWAUX : PPC::LWZUX8; break; 1122 case MVT::i16: Opcode = isSExt ? PPC::LHAUX8 : PPC::LHZUX8; break; 1123 case MVT::i1: 1124 case MVT::i8: Opcode = PPC::LBZUX8; break; 1125 } 1126 } 1127 1128 SDValue Chain = LD->getChain(); 1129 SDValue Base = LD->getBasePtr(); 1130 SDValue Ops[] = { Base, Offset, Chain }; 1131 return CurDAG->getMachineNode(Opcode, dl, LD->getValueType(0), 1132 PPCLowering->getPointerTy(), 1133 MVT::Other, Ops); 1134 } 1135 } 1136 1137 case ISD::AND: { 1138 unsigned Imm, Imm2, SH, MB, ME; 1139 uint64_t Imm64; 1140 1141 // If this is an and of a value rotated between 0 and 31 bits and then and'd 1142 // with a mask, emit rlwinm 1143 if (isInt32Immediate(N->getOperand(1), Imm) && 1144 isRotateAndMask(N->getOperand(0).getNode(), Imm, false, SH, MB, ME)) { 1145 SDValue Val = N->getOperand(0).getOperand(0); 1146 SDValue Ops[] = { Val, getI32Imm(SH), getI32Imm(MB), getI32Imm(ME) }; 1147 return CurDAG->SelectNodeTo(N, PPC::RLWINM, MVT::i32, Ops); 1148 } 1149 // If this is just a masked value where the input is not handled above, and 1150 // is not a rotate-left (handled by a pattern in the .td file), emit rlwinm 1151 if (isInt32Immediate(N->getOperand(1), Imm) && 1152 isRunOfOnes(Imm, MB, ME) && 1153 N->getOperand(0).getOpcode() != ISD::ROTL) { 1154 SDValue Val = N->getOperand(0); 1155 SDValue Ops[] = { Val, getI32Imm(0), getI32Imm(MB), getI32Imm(ME) }; 1156 return CurDAG->SelectNodeTo(N, PPC::RLWINM, MVT::i32, Ops); 1157 } 1158 // If this is a 64-bit zero-extension mask, emit rldicl. 1159 if (isInt64Immediate(N->getOperand(1).getNode(), Imm64) && 1160 isMask_64(Imm64)) { 1161 SDValue Val = N->getOperand(0); 1162 MB = 64 - CountTrailingOnes_64(Imm64); 1163 SH = 0; 1164 1165 // If the operand is a logical right shift, we can fold it into this 1166 // instruction: rldicl(rldicl(x, 64-n, n), 0, mb) -> rldicl(x, 64-n, mb) 1167 // for n <= mb. The right shift is really a left rotate followed by a 1168 // mask, and this mask is a more-restrictive sub-mask of the mask implied 1169 // by the shift. 1170 if (Val.getOpcode() == ISD::SRL && 1171 isInt32Immediate(Val.getOperand(1).getNode(), Imm) && Imm <= MB) { 1172 assert(Imm < 64 && "Illegal shift amount"); 1173 Val = Val.getOperand(0); 1174 SH = 64 - Imm; 1175 } 1176 1177 SDValue Ops[] = { Val, getI32Imm(SH), getI32Imm(MB) }; 1178 return CurDAG->SelectNodeTo(N, PPC::RLDICL, MVT::i64, Ops); 1179 } 1180 // AND X, 0 -> 0, not "rlwinm 32". 1181 if (isInt32Immediate(N->getOperand(1), Imm) && (Imm == 0)) { 1182 ReplaceUses(SDValue(N, 0), N->getOperand(1)); 1183 return nullptr; 1184 } 1185 // ISD::OR doesn't get all the bitfield insertion fun. 1186 // (and (or x, c1), c2) where isRunOfOnes(~(c1^c2)) is a bitfield insert 1187 if (isInt32Immediate(N->getOperand(1), Imm) && 1188 N->getOperand(0).getOpcode() == ISD::OR && 1189 isInt32Immediate(N->getOperand(0).getOperand(1), Imm2)) { 1190 unsigned MB, ME; 1191 Imm = ~(Imm^Imm2); 1192 if (isRunOfOnes(Imm, MB, ME)) { 1193 SDValue Ops[] = { N->getOperand(0).getOperand(0), 1194 N->getOperand(0).getOperand(1), 1195 getI32Imm(0), getI32Imm(MB),getI32Imm(ME) }; 1196 return CurDAG->getMachineNode(PPC::RLWIMI, dl, MVT::i32, Ops); 1197 } 1198 } 1199 1200 // Other cases are autogenerated. 1201 break; 1202 } 1203 case ISD::OR: 1204 if (N->getValueType(0) == MVT::i32) 1205 if (SDNode *I = SelectBitfieldInsert(N)) 1206 return I; 1207 1208 // Other cases are autogenerated. 1209 break; 1210 case ISD::SHL: { 1211 unsigned Imm, SH, MB, ME; 1212 if (isOpcWithIntImmediate(N->getOperand(0).getNode(), ISD::AND, Imm) && 1213 isRotateAndMask(N, Imm, true, SH, MB, ME)) { 1214 SDValue Ops[] = { N->getOperand(0).getOperand(0), 1215 getI32Imm(SH), getI32Imm(MB), getI32Imm(ME) }; 1216 return CurDAG->SelectNodeTo(N, PPC::RLWINM, MVT::i32, Ops); 1217 } 1218 1219 // Other cases are autogenerated. 1220 break; 1221 } 1222 case ISD::SRL: { 1223 unsigned Imm, SH, MB, ME; 1224 if (isOpcWithIntImmediate(N->getOperand(0).getNode(), ISD::AND, Imm) && 1225 isRotateAndMask(N, Imm, true, SH, MB, ME)) { 1226 SDValue Ops[] = { N->getOperand(0).getOperand(0), 1227 getI32Imm(SH), getI32Imm(MB), getI32Imm(ME) }; 1228 return CurDAG->SelectNodeTo(N, PPC::RLWINM, MVT::i32, Ops); 1229 } 1230 1231 // Other cases are autogenerated. 1232 break; 1233 } 1234 // FIXME: Remove this once the ANDI glue bug is fixed: 1235 case PPCISD::ANDIo_1_EQ_BIT: 1236 case PPCISD::ANDIo_1_GT_BIT: { 1237 if (!ANDIGlueBug) 1238 break; 1239 1240 EVT InVT = N->getOperand(0).getValueType(); 1241 assert((InVT == MVT::i64 || InVT == MVT::i32) && 1242 "Invalid input type for ANDIo_1_EQ_BIT"); 1243 1244 unsigned Opcode = (InVT == MVT::i64) ? PPC::ANDIo8 : PPC::ANDIo; 1245 SDValue AndI(CurDAG->getMachineNode(Opcode, dl, InVT, MVT::Glue, 1246 N->getOperand(0), 1247 CurDAG->getTargetConstant(1, InVT)), 0); 1248 SDValue CR0Reg = CurDAG->getRegister(PPC::CR0, MVT::i32); 1249 SDValue SRIdxVal = 1250 CurDAG->getTargetConstant(N->getOpcode() == PPCISD::ANDIo_1_EQ_BIT ? 1251 PPC::sub_eq : PPC::sub_gt, MVT::i32); 1252 1253 return CurDAG->SelectNodeTo(N, TargetOpcode::EXTRACT_SUBREG, MVT::i1, 1254 CR0Reg, SRIdxVal, 1255 SDValue(AndI.getNode(), 1) /* glue */); 1256 } 1257 case ISD::SELECT_CC: { 1258 ISD::CondCode CC = cast<CondCodeSDNode>(N->getOperand(4))->get(); 1259 EVT PtrVT = CurDAG->getTargetLoweringInfo().getPointerTy(); 1260 bool isPPC64 = (PtrVT == MVT::i64); 1261 1262 // If this is a select of i1 operands, we'll pattern match it. 1263 if (PPCSubTarget->useCRBits() && 1264 N->getOperand(0).getValueType() == MVT::i1) 1265 break; 1266 1267 // Handle the setcc cases here. select_cc lhs, 0, 1, 0, cc 1268 if (!isPPC64) 1269 if (ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N->getOperand(1))) 1270 if (ConstantSDNode *N2C = dyn_cast<ConstantSDNode>(N->getOperand(2))) 1271 if (ConstantSDNode *N3C = dyn_cast<ConstantSDNode>(N->getOperand(3))) 1272 if (N1C->isNullValue() && N3C->isNullValue() && 1273 N2C->getZExtValue() == 1ULL && CC == ISD::SETNE && 1274 // FIXME: Implement this optzn for PPC64. 1275 N->getValueType(0) == MVT::i32) { 1276 SDNode *Tmp = 1277 CurDAG->getMachineNode(PPC::ADDIC, dl, MVT::i32, MVT::Glue, 1278 N->getOperand(0), getI32Imm(~0U)); 1279 return CurDAG->SelectNodeTo(N, PPC::SUBFE, MVT::i32, 1280 SDValue(Tmp, 0), N->getOperand(0), 1281 SDValue(Tmp, 1)); 1282 } 1283 1284 SDValue CCReg = SelectCC(N->getOperand(0), N->getOperand(1), CC, dl); 1285 1286 if (N->getValueType(0) == MVT::i1) { 1287 // An i1 select is: (c & t) | (!c & f). 1288 bool Inv; 1289 unsigned Idx = getCRIdxForSetCC(CC, Inv); 1290 1291 unsigned SRI; 1292 switch (Idx) { 1293 default: llvm_unreachable("Invalid CC index"); 1294 case 0: SRI = PPC::sub_lt; break; 1295 case 1: SRI = PPC::sub_gt; break; 1296 case 2: SRI = PPC::sub_eq; break; 1297 case 3: SRI = PPC::sub_un; break; 1298 } 1299 1300 SDValue CCBit = CurDAG->getTargetExtractSubreg(SRI, dl, MVT::i1, CCReg); 1301 1302 SDValue NotCCBit(CurDAG->getMachineNode(PPC::CRNOR, dl, MVT::i1, 1303 CCBit, CCBit), 0); 1304 SDValue C = Inv ? NotCCBit : CCBit, 1305 NotC = Inv ? CCBit : NotCCBit; 1306 1307 SDValue CAndT(CurDAG->getMachineNode(PPC::CRAND, dl, MVT::i1, 1308 C, N->getOperand(2)), 0); 1309 SDValue NotCAndF(CurDAG->getMachineNode(PPC::CRAND, dl, MVT::i1, 1310 NotC, N->getOperand(3)), 0); 1311 1312 return CurDAG->SelectNodeTo(N, PPC::CROR, MVT::i1, CAndT, NotCAndF); 1313 } 1314 1315 unsigned BROpc = getPredicateForSetCC(CC); 1316 1317 unsigned SelectCCOp; 1318 if (N->getValueType(0) == MVT::i32) 1319 SelectCCOp = PPC::SELECT_CC_I4; 1320 else if (N->getValueType(0) == MVT::i64) 1321 SelectCCOp = PPC::SELECT_CC_I8; 1322 else if (N->getValueType(0) == MVT::f32) 1323 SelectCCOp = PPC::SELECT_CC_F4; 1324 else if (N->getValueType(0) == MVT::f64) 1325 SelectCCOp = PPC::SELECT_CC_F8; 1326 else 1327 SelectCCOp = PPC::SELECT_CC_VRRC; 1328 1329 SDValue Ops[] = { CCReg, N->getOperand(2), N->getOperand(3), 1330 getI32Imm(BROpc) }; 1331 return CurDAG->SelectNodeTo(N, SelectCCOp, N->getValueType(0), Ops); 1332 } 1333 case ISD::VSELECT: 1334 if (PPCSubTarget->hasVSX()) { 1335 SDValue Ops[] = { N->getOperand(2), N->getOperand(1), N->getOperand(0) }; 1336 return CurDAG->SelectNodeTo(N, PPC::XXSEL, N->getValueType(0), Ops); 1337 } 1338 1339 break; 1340 case ISD::VECTOR_SHUFFLE: 1341 if (PPCSubTarget->hasVSX() && (N->getValueType(0) == MVT::v2f64 || 1342 N->getValueType(0) == MVT::v2i64)) { 1343 ShuffleVectorSDNode *SVN = cast<ShuffleVectorSDNode>(N); 1344 1345 SDValue Op1 = N->getOperand(SVN->getMaskElt(0) < 2 ? 0 : 1), 1346 Op2 = N->getOperand(SVN->getMaskElt(1) < 2 ? 0 : 1); 1347 unsigned DM[2]; 1348 1349 for (int i = 0; i < 2; ++i) 1350 if (SVN->getMaskElt(i) <= 0 || SVN->getMaskElt(i) == 2) 1351 DM[i] = 0; 1352 else 1353 DM[i] = 1; 1354 1355 SDValue DMV = CurDAG->getTargetConstant(DM[1] | (DM[0] << 1), MVT::i32); 1356 1357 if (Op1 == Op2 && DM[0] == 0 && DM[1] == 0 && 1358 Op1.getOpcode() == ISD::SCALAR_TO_VECTOR && 1359 isa<LoadSDNode>(Op1.getOperand(0))) { 1360 LoadSDNode *LD = cast<LoadSDNode>(Op1.getOperand(0)); 1361 SDValue Base, Offset; 1362 1363 if (LD->isUnindexed() && 1364 SelectAddrIdxOnly(LD->getBasePtr(), Base, Offset)) { 1365 SDValue Chain = LD->getChain(); 1366 SDValue Ops[] = { Base, Offset, Chain }; 1367 return CurDAG->SelectNodeTo(N, PPC::LXVDSX, 1368 N->getValueType(0), Ops); 1369 } 1370 } 1371 1372 SDValue Ops[] = { Op1, Op2, DMV }; 1373 return CurDAG->SelectNodeTo(N, PPC::XXPERMDI, N->getValueType(0), Ops); 1374 } 1375 1376 break; 1377 case PPCISD::BDNZ: 1378 case PPCISD::BDZ: { 1379 bool IsPPC64 = PPCSubTarget->isPPC64(); 1380 SDValue Ops[] = { N->getOperand(1), N->getOperand(0) }; 1381 return CurDAG->SelectNodeTo(N, N->getOpcode() == PPCISD::BDNZ ? 1382 (IsPPC64 ? PPC::BDNZ8 : PPC::BDNZ) : 1383 (IsPPC64 ? PPC::BDZ8 : PPC::BDZ), 1384 MVT::Other, Ops); 1385 } 1386 case PPCISD::COND_BRANCH: { 1387 // Op #0 is the Chain. 1388 // Op #1 is the PPC::PRED_* number. 1389 // Op #2 is the CR# 1390 // Op #3 is the Dest MBB 1391 // Op #4 is the Flag. 1392 // Prevent PPC::PRED_* from being selected into LI. 1393 SDValue Pred = 1394 getI32Imm(cast<ConstantSDNode>(N->getOperand(1))->getZExtValue()); 1395 SDValue Ops[] = { Pred, N->getOperand(2), N->getOperand(3), 1396 N->getOperand(0), N->getOperand(4) }; 1397 return CurDAG->SelectNodeTo(N, PPC::BCC, MVT::Other, Ops); 1398 } 1399 case ISD::BR_CC: { 1400 ISD::CondCode CC = cast<CondCodeSDNode>(N->getOperand(1))->get(); 1401 unsigned PCC = getPredicateForSetCC(CC); 1402 1403 if (N->getOperand(2).getValueType() == MVT::i1) { 1404 unsigned Opc; 1405 bool Swap; 1406 switch (PCC) { 1407 default: llvm_unreachable("Unexpected Boolean-operand predicate"); 1408 case PPC::PRED_LT: Opc = PPC::CRANDC; Swap = true; break; 1409 case PPC::PRED_LE: Opc = PPC::CRORC; Swap = true; break; 1410 case PPC::PRED_EQ: Opc = PPC::CREQV; Swap = false; break; 1411 case PPC::PRED_GE: Opc = PPC::CRORC; Swap = false; break; 1412 case PPC::PRED_GT: Opc = PPC::CRANDC; Swap = false; break; 1413 case PPC::PRED_NE: Opc = PPC::CRXOR; Swap = false; break; 1414 } 1415 1416 SDValue BitComp(CurDAG->getMachineNode(Opc, dl, MVT::i1, 1417 N->getOperand(Swap ? 3 : 2), 1418 N->getOperand(Swap ? 2 : 3)), 0); 1419 return CurDAG->SelectNodeTo(N, PPC::BC, MVT::Other, 1420 BitComp, N->getOperand(4), N->getOperand(0)); 1421 } 1422 1423 SDValue CondCode = SelectCC(N->getOperand(2), N->getOperand(3), CC, dl); 1424 SDValue Ops[] = { getI32Imm(PCC), CondCode, 1425 N->getOperand(4), N->getOperand(0) }; 1426 return CurDAG->SelectNodeTo(N, PPC::BCC, MVT::Other, Ops); 1427 } 1428 case ISD::BRIND: { 1429 // FIXME: Should custom lower this. 1430 SDValue Chain = N->getOperand(0); 1431 SDValue Target = N->getOperand(1); 1432 unsigned Opc = Target.getValueType() == MVT::i32 ? PPC::MTCTR : PPC::MTCTR8; 1433 unsigned Reg = Target.getValueType() == MVT::i32 ? PPC::BCTR : PPC::BCTR8; 1434 Chain = SDValue(CurDAG->getMachineNode(Opc, dl, MVT::Glue, Target, 1435 Chain), 0); 1436 return CurDAG->SelectNodeTo(N, Reg, MVT::Other, Chain); 1437 } 1438 case PPCISD::TOC_ENTRY: { 1439 assert ((PPCSubTarget->isPPC64() || PPCSubTarget->isSVR4ABI()) && 1440 "Only supported for 64-bit ABI and 32-bit SVR4"); 1441 if (PPCSubTarget->isSVR4ABI() && !PPCSubTarget->isPPC64()) { 1442 SDValue GA = N->getOperand(0); 1443 return CurDAG->getMachineNode(PPC::LWZtoc, dl, MVT::i32, GA, 1444 N->getOperand(1)); 1445 } 1446 1447 // For medium and large code model, we generate two instructions as 1448 // described below. Otherwise we allow SelectCodeCommon to handle this, 1449 // selecting one of LDtoc, LDtocJTI, and LDtocCPT. 1450 CodeModel::Model CModel = TM.getCodeModel(); 1451 if (CModel != CodeModel::Medium && CModel != CodeModel::Large) 1452 break; 1453 1454 // The first source operand is a TargetGlobalAddress or a TargetJumpTable. 1455 // If it is an externally defined symbol, a symbol with common linkage, 1456 // a non-local function address, or a jump table address, or if we are 1457 // generating code for large code model, we generate: 1458 // LDtocL(<ga:@sym>, ADDIStocHA(%X2, <ga:@sym>)) 1459 // Otherwise we generate: 1460 // ADDItocL(ADDIStocHA(%X2, <ga:@sym>), <ga:@sym>) 1461 SDValue GA = N->getOperand(0); 1462 SDValue TOCbase = N->getOperand(1); 1463 SDNode *Tmp = CurDAG->getMachineNode(PPC::ADDIStocHA, dl, MVT::i64, 1464 TOCbase, GA); 1465 1466 if (isa<JumpTableSDNode>(GA) || CModel == CodeModel::Large) 1467 return CurDAG->getMachineNode(PPC::LDtocL, dl, MVT::i64, GA, 1468 SDValue(Tmp, 0)); 1469 1470 if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(GA)) { 1471 const GlobalValue *GValue = G->getGlobal(); 1472 if ((GValue->getType()->getElementType()->isFunctionTy() && 1473 (GValue->isDeclaration() || GValue->isWeakForLinker())) || 1474 GValue->isDeclaration() || GValue->hasCommonLinkage() || 1475 GValue->hasAvailableExternallyLinkage()) 1476 return CurDAG->getMachineNode(PPC::LDtocL, dl, MVT::i64, GA, 1477 SDValue(Tmp, 0)); 1478 } 1479 1480 return CurDAG->getMachineNode(PPC::ADDItocL, dl, MVT::i64, 1481 SDValue(Tmp, 0), GA); 1482 } 1483 case PPCISD::PPC32_PICGOT: { 1484 // Generate a PIC-safe GOT reference. 1485 assert(!PPCSubTarget->isPPC64() && PPCSubTarget->isSVR4ABI() && 1486 "PPCISD::PPC32_PICGOT is only supported for 32-bit SVR4"); 1487 return CurDAG->SelectNodeTo(N, PPC::PPC32PICGOT, PPCLowering->getPointerTy(), MVT::i32); 1488 } 1489 case PPCISD::VADD_SPLAT: { 1490 // This expands into one of three sequences, depending on whether 1491 // the first operand is odd or even, positive or negative. 1492 assert(isa<ConstantSDNode>(N->getOperand(0)) && 1493 isa<ConstantSDNode>(N->getOperand(1)) && 1494 "Invalid operand on VADD_SPLAT!"); 1495 1496 int Elt = N->getConstantOperandVal(0); 1497 int EltSize = N->getConstantOperandVal(1); 1498 unsigned Opc1, Opc2, Opc3; 1499 EVT VT; 1500 1501 if (EltSize == 1) { 1502 Opc1 = PPC::VSPLTISB; 1503 Opc2 = PPC::VADDUBM; 1504 Opc3 = PPC::VSUBUBM; 1505 VT = MVT::v16i8; 1506 } else if (EltSize == 2) { 1507 Opc1 = PPC::VSPLTISH; 1508 Opc2 = PPC::VADDUHM; 1509 Opc3 = PPC::VSUBUHM; 1510 VT = MVT::v8i16; 1511 } else { 1512 assert(EltSize == 4 && "Invalid element size on VADD_SPLAT!"); 1513 Opc1 = PPC::VSPLTISW; 1514 Opc2 = PPC::VADDUWM; 1515 Opc3 = PPC::VSUBUWM; 1516 VT = MVT::v4i32; 1517 } 1518 1519 if ((Elt & 1) == 0) { 1520 // Elt is even, in the range [-32,-18] + [16,30]. 1521 // 1522 // Convert: VADD_SPLAT elt, size 1523 // Into: tmp = VSPLTIS[BHW] elt 1524 // VADDU[BHW]M tmp, tmp 1525 // Where: [BHW] = B for size = 1, H for size = 2, W for size = 4 1526 SDValue EltVal = getI32Imm(Elt >> 1); 1527 SDNode *Tmp = CurDAG->getMachineNode(Opc1, dl, VT, EltVal); 1528 SDValue TmpVal = SDValue(Tmp, 0); 1529 return CurDAG->getMachineNode(Opc2, dl, VT, TmpVal, TmpVal); 1530 1531 } else if (Elt > 0) { 1532 // Elt is odd and positive, in the range [17,31]. 1533 // 1534 // Convert: VADD_SPLAT elt, size 1535 // Into: tmp1 = VSPLTIS[BHW] elt-16 1536 // tmp2 = VSPLTIS[BHW] -16 1537 // VSUBU[BHW]M tmp1, tmp2 1538 SDValue EltVal = getI32Imm(Elt - 16); 1539 SDNode *Tmp1 = CurDAG->getMachineNode(Opc1, dl, VT, EltVal); 1540 EltVal = getI32Imm(-16); 1541 SDNode *Tmp2 = CurDAG->getMachineNode(Opc1, dl, VT, EltVal); 1542 return CurDAG->getMachineNode(Opc3, dl, VT, SDValue(Tmp1, 0), 1543 SDValue(Tmp2, 0)); 1544 1545 } else { 1546 // Elt is odd and negative, in the range [-31,-17]. 1547 // 1548 // Convert: VADD_SPLAT elt, size 1549 // Into: tmp1 = VSPLTIS[BHW] elt+16 1550 // tmp2 = VSPLTIS[BHW] -16 1551 // VADDU[BHW]M tmp1, tmp2 1552 SDValue EltVal = getI32Imm(Elt + 16); 1553 SDNode *Tmp1 = CurDAG->getMachineNode(Opc1, dl, VT, EltVal); 1554 EltVal = getI32Imm(-16); 1555 SDNode *Tmp2 = CurDAG->getMachineNode(Opc1, dl, VT, EltVal); 1556 return CurDAG->getMachineNode(Opc2, dl, VT, SDValue(Tmp1, 0), 1557 SDValue(Tmp2, 0)); 1558 } 1559 } 1560 } 1561 1562 return SelectCode(N); 1563 } 1564 1565 /// PostprocessISelDAG - Perform some late peephole optimizations 1566 /// on the DAG representation. 1567 void PPCDAGToDAGISel::PostprocessISelDAG() { 1568 1569 // Skip peepholes at -O0. 1570 if (TM.getOptLevel() == CodeGenOpt::None) 1571 return; 1572 1573 PeepholePPC64(); 1574 PeepholeCROps(); 1575 } 1576 1577 // Check if all users of this node will become isel where the second operand 1578 // is the constant zero. If this is so, and if we can negate the condition, 1579 // then we can flip the true and false operands. This will allow the zero to 1580 // be folded with the isel so that we don't need to materialize a register 1581 // containing zero. 1582 bool PPCDAGToDAGISel::AllUsersSelectZero(SDNode *N) { 1583 // If we're not using isel, then this does not matter. 1584 if (!PPCSubTarget->hasISEL()) 1585 return false; 1586 1587 for (SDNode::use_iterator UI = N->use_begin(), UE = N->use_end(); 1588 UI != UE; ++UI) { 1589 SDNode *User = *UI; 1590 if (!User->isMachineOpcode()) 1591 return false; 1592 if (User->getMachineOpcode() != PPC::SELECT_I4 && 1593 User->getMachineOpcode() != PPC::SELECT_I8) 1594 return false; 1595 1596 SDNode *Op2 = User->getOperand(2).getNode(); 1597 if (!Op2->isMachineOpcode()) 1598 return false; 1599 1600 if (Op2->getMachineOpcode() != PPC::LI && 1601 Op2->getMachineOpcode() != PPC::LI8) 1602 return false; 1603 1604 ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op2->getOperand(0)); 1605 if (!C) 1606 return false; 1607 1608 if (!C->isNullValue()) 1609 return false; 1610 } 1611 1612 return true; 1613 } 1614 1615 void PPCDAGToDAGISel::SwapAllSelectUsers(SDNode *N) { 1616 SmallVector<SDNode *, 4> ToReplace; 1617 for (SDNode::use_iterator UI = N->use_begin(), UE = N->use_end(); 1618 UI != UE; ++UI) { 1619 SDNode *User = *UI; 1620 assert((User->getMachineOpcode() == PPC::SELECT_I4 || 1621 User->getMachineOpcode() == PPC::SELECT_I8) && 1622 "Must have all select users"); 1623 ToReplace.push_back(User); 1624 } 1625 1626 for (SmallVector<SDNode *, 4>::iterator UI = ToReplace.begin(), 1627 UE = ToReplace.end(); UI != UE; ++UI) { 1628 SDNode *User = *UI; 1629 SDNode *ResNode = 1630 CurDAG->getMachineNode(User->getMachineOpcode(), SDLoc(User), 1631 User->getValueType(0), User->getOperand(0), 1632 User->getOperand(2), 1633 User->getOperand(1)); 1634 1635 DEBUG(dbgs() << "CR Peephole replacing:\nOld: "); 1636 DEBUG(User->dump(CurDAG)); 1637 DEBUG(dbgs() << "\nNew: "); 1638 DEBUG(ResNode->dump(CurDAG)); 1639 DEBUG(dbgs() << "\n"); 1640 1641 ReplaceUses(User, ResNode); 1642 } 1643 } 1644 1645 void PPCDAGToDAGISel::PeepholeCROps() { 1646 bool IsModified; 1647 do { 1648 IsModified = false; 1649 for (SelectionDAG::allnodes_iterator I = CurDAG->allnodes_begin(), 1650 E = CurDAG->allnodes_end(); I != E; ++I) { 1651 MachineSDNode *MachineNode = dyn_cast<MachineSDNode>(I); 1652 if (!MachineNode || MachineNode->use_empty()) 1653 continue; 1654 SDNode *ResNode = MachineNode; 1655 1656 bool Op1Set = false, Op1Unset = false, 1657 Op1Not = false, 1658 Op2Set = false, Op2Unset = false, 1659 Op2Not = false; 1660 1661 unsigned Opcode = MachineNode->getMachineOpcode(); 1662 switch (Opcode) { 1663 default: break; 1664 case PPC::CRAND: 1665 case PPC::CRNAND: 1666 case PPC::CROR: 1667 case PPC::CRXOR: 1668 case PPC::CRNOR: 1669 case PPC::CREQV: 1670 case PPC::CRANDC: 1671 case PPC::CRORC: { 1672 SDValue Op = MachineNode->getOperand(1); 1673 if (Op.isMachineOpcode()) { 1674 if (Op.getMachineOpcode() == PPC::CRSET) 1675 Op2Set = true; 1676 else if (Op.getMachineOpcode() == PPC::CRUNSET) 1677 Op2Unset = true; 1678 else if (Op.getMachineOpcode() == PPC::CRNOR && 1679 Op.getOperand(0) == Op.getOperand(1)) 1680 Op2Not = true; 1681 } 1682 } // fallthrough 1683 case PPC::BC: 1684 case PPC::BCn: 1685 case PPC::SELECT_I4: 1686 case PPC::SELECT_I8: 1687 case PPC::SELECT_F4: 1688 case PPC::SELECT_F8: 1689 case PPC::SELECT_VRRC: { 1690 SDValue Op = MachineNode->getOperand(0); 1691 if (Op.isMachineOpcode()) { 1692 if (Op.getMachineOpcode() == PPC::CRSET) 1693 Op1Set = true; 1694 else if (Op.getMachineOpcode() == PPC::CRUNSET) 1695 Op1Unset = true; 1696 else if (Op.getMachineOpcode() == PPC::CRNOR && 1697 Op.getOperand(0) == Op.getOperand(1)) 1698 Op1Not = true; 1699 } 1700 } 1701 break; 1702 } 1703 1704 bool SelectSwap = false; 1705 switch (Opcode) { 1706 default: break; 1707 case PPC::CRAND: 1708 if (MachineNode->getOperand(0) == MachineNode->getOperand(1)) 1709 // x & x = x 1710 ResNode = MachineNode->getOperand(0).getNode(); 1711 else if (Op1Set) 1712 // 1 & y = y 1713 ResNode = MachineNode->getOperand(1).getNode(); 1714 else if (Op2Set) 1715 // x & 1 = x 1716 ResNode = MachineNode->getOperand(0).getNode(); 1717 else if (Op1Unset || Op2Unset) 1718 // x & 0 = 0 & y = 0 1719 ResNode = CurDAG->getMachineNode(PPC::CRUNSET, SDLoc(MachineNode), 1720 MVT::i1); 1721 else if (Op1Not) 1722 // ~x & y = andc(y, x) 1723 ResNode = CurDAG->getMachineNode(PPC::CRANDC, SDLoc(MachineNode), 1724 MVT::i1, MachineNode->getOperand(1), 1725 MachineNode->getOperand(0). 1726 getOperand(0)); 1727 else if (Op2Not) 1728 // x & ~y = andc(x, y) 1729 ResNode = CurDAG->getMachineNode(PPC::CRANDC, SDLoc(MachineNode), 1730 MVT::i1, MachineNode->getOperand(0), 1731 MachineNode->getOperand(1). 1732 getOperand(0)); 1733 else if (AllUsersSelectZero(MachineNode)) 1734 ResNode = CurDAG->getMachineNode(PPC::CRNAND, SDLoc(MachineNode), 1735 MVT::i1, MachineNode->getOperand(0), 1736 MachineNode->getOperand(1)), 1737 SelectSwap = true; 1738 break; 1739 case PPC::CRNAND: 1740 if (MachineNode->getOperand(0) == MachineNode->getOperand(1)) 1741 // nand(x, x) -> nor(x, x) 1742 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode), 1743 MVT::i1, MachineNode->getOperand(0), 1744 MachineNode->getOperand(0)); 1745 else if (Op1Set) 1746 // nand(1, y) -> nor(y, y) 1747 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode), 1748 MVT::i1, MachineNode->getOperand(1), 1749 MachineNode->getOperand(1)); 1750 else if (Op2Set) 1751 // nand(x, 1) -> nor(x, x) 1752 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode), 1753 MVT::i1, MachineNode->getOperand(0), 1754 MachineNode->getOperand(0)); 1755 else if (Op1Unset || Op2Unset) 1756 // nand(x, 0) = nand(0, y) = 1 1757 ResNode = CurDAG->getMachineNode(PPC::CRSET, SDLoc(MachineNode), 1758 MVT::i1); 1759 else if (Op1Not) 1760 // nand(~x, y) = ~(~x & y) = x | ~y = orc(x, y) 1761 ResNode = CurDAG->getMachineNode(PPC::CRORC, SDLoc(MachineNode), 1762 MVT::i1, MachineNode->getOperand(0). 1763 getOperand(0), 1764 MachineNode->getOperand(1)); 1765 else if (Op2Not) 1766 // nand(x, ~y) = ~x | y = orc(y, x) 1767 ResNode = CurDAG->getMachineNode(PPC::CRORC, SDLoc(MachineNode), 1768 MVT::i1, MachineNode->getOperand(1). 1769 getOperand(0), 1770 MachineNode->getOperand(0)); 1771 else if (AllUsersSelectZero(MachineNode)) 1772 ResNode = CurDAG->getMachineNode(PPC::CRAND, SDLoc(MachineNode), 1773 MVT::i1, MachineNode->getOperand(0), 1774 MachineNode->getOperand(1)), 1775 SelectSwap = true; 1776 break; 1777 case PPC::CROR: 1778 if (MachineNode->getOperand(0) == MachineNode->getOperand(1)) 1779 // x | x = x 1780 ResNode = MachineNode->getOperand(0).getNode(); 1781 else if (Op1Set || Op2Set) 1782 // x | 1 = 1 | y = 1 1783 ResNode = CurDAG->getMachineNode(PPC::CRSET, SDLoc(MachineNode), 1784 MVT::i1); 1785 else if (Op1Unset) 1786 // 0 | y = y 1787 ResNode = MachineNode->getOperand(1).getNode(); 1788 else if (Op2Unset) 1789 // x | 0 = x 1790 ResNode = MachineNode->getOperand(0).getNode(); 1791 else if (Op1Not) 1792 // ~x | y = orc(y, x) 1793 ResNode = CurDAG->getMachineNode(PPC::CRORC, SDLoc(MachineNode), 1794 MVT::i1, MachineNode->getOperand(1), 1795 MachineNode->getOperand(0). 1796 getOperand(0)); 1797 else if (Op2Not) 1798 // x | ~y = orc(x, y) 1799 ResNode = CurDAG->getMachineNode(PPC::CRORC, SDLoc(MachineNode), 1800 MVT::i1, MachineNode->getOperand(0), 1801 MachineNode->getOperand(1). 1802 getOperand(0)); 1803 else if (AllUsersSelectZero(MachineNode)) 1804 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode), 1805 MVT::i1, MachineNode->getOperand(0), 1806 MachineNode->getOperand(1)), 1807 SelectSwap = true; 1808 break; 1809 case PPC::CRXOR: 1810 if (MachineNode->getOperand(0) == MachineNode->getOperand(1)) 1811 // xor(x, x) = 0 1812 ResNode = CurDAG->getMachineNode(PPC::CRUNSET, SDLoc(MachineNode), 1813 MVT::i1); 1814 else if (Op1Set) 1815 // xor(1, y) -> nor(y, y) 1816 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode), 1817 MVT::i1, MachineNode->getOperand(1), 1818 MachineNode->getOperand(1)); 1819 else if (Op2Set) 1820 // xor(x, 1) -> nor(x, x) 1821 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode), 1822 MVT::i1, MachineNode->getOperand(0), 1823 MachineNode->getOperand(0)); 1824 else if (Op1Unset) 1825 // xor(0, y) = y 1826 ResNode = MachineNode->getOperand(1).getNode(); 1827 else if (Op2Unset) 1828 // xor(x, 0) = x 1829 ResNode = MachineNode->getOperand(0).getNode(); 1830 else if (Op1Not) 1831 // xor(~x, y) = eqv(x, y) 1832 ResNode = CurDAG->getMachineNode(PPC::CREQV, SDLoc(MachineNode), 1833 MVT::i1, MachineNode->getOperand(0). 1834 getOperand(0), 1835 MachineNode->getOperand(1)); 1836 else if (Op2Not) 1837 // xor(x, ~y) = eqv(x, y) 1838 ResNode = CurDAG->getMachineNode(PPC::CREQV, SDLoc(MachineNode), 1839 MVT::i1, MachineNode->getOperand(0), 1840 MachineNode->getOperand(1). 1841 getOperand(0)); 1842 else if (AllUsersSelectZero(MachineNode)) 1843 ResNode = CurDAG->getMachineNode(PPC::CREQV, SDLoc(MachineNode), 1844 MVT::i1, MachineNode->getOperand(0), 1845 MachineNode->getOperand(1)), 1846 SelectSwap = true; 1847 break; 1848 case PPC::CRNOR: 1849 if (Op1Set || Op2Set) 1850 // nor(1, y) -> 0 1851 ResNode = CurDAG->getMachineNode(PPC::CRUNSET, SDLoc(MachineNode), 1852 MVT::i1); 1853 else if (Op1Unset) 1854 // nor(0, y) = ~y -> nor(y, y) 1855 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode), 1856 MVT::i1, MachineNode->getOperand(1), 1857 MachineNode->getOperand(1)); 1858 else if (Op2Unset) 1859 // nor(x, 0) = ~x 1860 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode), 1861 MVT::i1, MachineNode->getOperand(0), 1862 MachineNode->getOperand(0)); 1863 else if (Op1Not) 1864 // nor(~x, y) = andc(x, y) 1865 ResNode = CurDAG->getMachineNode(PPC::CRANDC, SDLoc(MachineNode), 1866 MVT::i1, MachineNode->getOperand(0). 1867 getOperand(0), 1868 MachineNode->getOperand(1)); 1869 else if (Op2Not) 1870 // nor(x, ~y) = andc(y, x) 1871 ResNode = CurDAG->getMachineNode(PPC::CRANDC, SDLoc(MachineNode), 1872 MVT::i1, MachineNode->getOperand(1). 1873 getOperand(0), 1874 MachineNode->getOperand(0)); 1875 else if (AllUsersSelectZero(MachineNode)) 1876 ResNode = CurDAG->getMachineNode(PPC::CROR, SDLoc(MachineNode), 1877 MVT::i1, MachineNode->getOperand(0), 1878 MachineNode->getOperand(1)), 1879 SelectSwap = true; 1880 break; 1881 case PPC::CREQV: 1882 if (MachineNode->getOperand(0) == MachineNode->getOperand(1)) 1883 // eqv(x, x) = 1 1884 ResNode = CurDAG->getMachineNode(PPC::CRSET, SDLoc(MachineNode), 1885 MVT::i1); 1886 else if (Op1Set) 1887 // eqv(1, y) = y 1888 ResNode = MachineNode->getOperand(1).getNode(); 1889 else if (Op2Set) 1890 // eqv(x, 1) = x 1891 ResNode = MachineNode->getOperand(0).getNode(); 1892 else if (Op1Unset) 1893 // eqv(0, y) = ~y -> nor(y, y) 1894 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode), 1895 MVT::i1, MachineNode->getOperand(1), 1896 MachineNode->getOperand(1)); 1897 else if (Op2Unset) 1898 // eqv(x, 0) = ~x 1899 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode), 1900 MVT::i1, MachineNode->getOperand(0), 1901 MachineNode->getOperand(0)); 1902 else if (Op1Not) 1903 // eqv(~x, y) = xor(x, y) 1904 ResNode = CurDAG->getMachineNode(PPC::CRXOR, SDLoc(MachineNode), 1905 MVT::i1, MachineNode->getOperand(0). 1906 getOperand(0), 1907 MachineNode->getOperand(1)); 1908 else if (Op2Not) 1909 // eqv(x, ~y) = xor(x, y) 1910 ResNode = CurDAG->getMachineNode(PPC::CRXOR, SDLoc(MachineNode), 1911 MVT::i1, MachineNode->getOperand(0), 1912 MachineNode->getOperand(1). 1913 getOperand(0)); 1914 else if (AllUsersSelectZero(MachineNode)) 1915 ResNode = CurDAG->getMachineNode(PPC::CRXOR, SDLoc(MachineNode), 1916 MVT::i1, MachineNode->getOperand(0), 1917 MachineNode->getOperand(1)), 1918 SelectSwap = true; 1919 break; 1920 case PPC::CRANDC: 1921 if (MachineNode->getOperand(0) == MachineNode->getOperand(1)) 1922 // andc(x, x) = 0 1923 ResNode = CurDAG->getMachineNode(PPC::CRUNSET, SDLoc(MachineNode), 1924 MVT::i1); 1925 else if (Op1Set) 1926 // andc(1, y) = ~y 1927 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode), 1928 MVT::i1, MachineNode->getOperand(1), 1929 MachineNode->getOperand(1)); 1930 else if (Op1Unset || Op2Set) 1931 // andc(0, y) = andc(x, 1) = 0 1932 ResNode = CurDAG->getMachineNode(PPC::CRUNSET, SDLoc(MachineNode), 1933 MVT::i1); 1934 else if (Op2Unset) 1935 // andc(x, 0) = x 1936 ResNode = MachineNode->getOperand(0).getNode(); 1937 else if (Op1Not) 1938 // andc(~x, y) = ~(x | y) = nor(x, y) 1939 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode), 1940 MVT::i1, MachineNode->getOperand(0). 1941 getOperand(0), 1942 MachineNode->getOperand(1)); 1943 else if (Op2Not) 1944 // andc(x, ~y) = x & y 1945 ResNode = CurDAG->getMachineNode(PPC::CRAND, SDLoc(MachineNode), 1946 MVT::i1, MachineNode->getOperand(0), 1947 MachineNode->getOperand(1). 1948 getOperand(0)); 1949 else if (AllUsersSelectZero(MachineNode)) 1950 ResNode = CurDAG->getMachineNode(PPC::CRORC, SDLoc(MachineNode), 1951 MVT::i1, MachineNode->getOperand(1), 1952 MachineNode->getOperand(0)), 1953 SelectSwap = true; 1954 break; 1955 case PPC::CRORC: 1956 if (MachineNode->getOperand(0) == MachineNode->getOperand(1)) 1957 // orc(x, x) = 1 1958 ResNode = CurDAG->getMachineNode(PPC::CRSET, SDLoc(MachineNode), 1959 MVT::i1); 1960 else if (Op1Set || Op2Unset) 1961 // orc(1, y) = orc(x, 0) = 1 1962 ResNode = CurDAG->getMachineNode(PPC::CRSET, SDLoc(MachineNode), 1963 MVT::i1); 1964 else if (Op2Set) 1965 // orc(x, 1) = x 1966 ResNode = MachineNode->getOperand(0).getNode(); 1967 else if (Op1Unset) 1968 // orc(0, y) = ~y 1969 ResNode = CurDAG->getMachineNode(PPC::CRNOR, SDLoc(MachineNode), 1970 MVT::i1, MachineNode->getOperand(1), 1971 MachineNode->getOperand(1)); 1972 else if (Op1Not) 1973 // orc(~x, y) = ~(x & y) = nand(x, y) 1974 ResNode = CurDAG->getMachineNode(PPC::CRNAND, SDLoc(MachineNode), 1975 MVT::i1, MachineNode->getOperand(0). 1976 getOperand(0), 1977 MachineNode->getOperand(1)); 1978 else if (Op2Not) 1979 // orc(x, ~y) = x | y 1980 ResNode = CurDAG->getMachineNode(PPC::CROR, SDLoc(MachineNode), 1981 MVT::i1, MachineNode->getOperand(0), 1982 MachineNode->getOperand(1). 1983 getOperand(0)); 1984 else if (AllUsersSelectZero(MachineNode)) 1985 ResNode = CurDAG->getMachineNode(PPC::CRANDC, SDLoc(MachineNode), 1986 MVT::i1, MachineNode->getOperand(1), 1987 MachineNode->getOperand(0)), 1988 SelectSwap = true; 1989 break; 1990 case PPC::SELECT_I4: 1991 case PPC::SELECT_I8: 1992 case PPC::SELECT_F4: 1993 case PPC::SELECT_F8: 1994 case PPC::SELECT_VRRC: 1995 if (Op1Set) 1996 ResNode = MachineNode->getOperand(1).getNode(); 1997 else if (Op1Unset) 1998 ResNode = MachineNode->getOperand(2).getNode(); 1999 else if (Op1Not) 2000 ResNode = CurDAG->getMachineNode(MachineNode->getMachineOpcode(), 2001 SDLoc(MachineNode), 2002 MachineNode->getValueType(0), 2003 MachineNode->getOperand(0). 2004 getOperand(0), 2005 MachineNode->getOperand(2), 2006 MachineNode->getOperand(1)); 2007 break; 2008 case PPC::BC: 2009 case PPC::BCn: 2010 if (Op1Not) 2011 ResNode = CurDAG->getMachineNode(Opcode == PPC::BC ? PPC::BCn : 2012 PPC::BC, 2013 SDLoc(MachineNode), 2014 MVT::Other, 2015 MachineNode->getOperand(0). 2016 getOperand(0), 2017 MachineNode->getOperand(1), 2018 MachineNode->getOperand(2)); 2019 // FIXME: Handle Op1Set, Op1Unset here too. 2020 break; 2021 } 2022 2023 // If we're inverting this node because it is used only by selects that 2024 // we'd like to swap, then swap the selects before the node replacement. 2025 if (SelectSwap) 2026 SwapAllSelectUsers(MachineNode); 2027 2028 if (ResNode != MachineNode) { 2029 DEBUG(dbgs() << "CR Peephole replacing:\nOld: "); 2030 DEBUG(MachineNode->dump(CurDAG)); 2031 DEBUG(dbgs() << "\nNew: "); 2032 DEBUG(ResNode->dump(CurDAG)); 2033 DEBUG(dbgs() << "\n"); 2034 2035 ReplaceUses(MachineNode, ResNode); 2036 IsModified = true; 2037 } 2038 } 2039 if (IsModified) 2040 CurDAG->RemoveDeadNodes(); 2041 } while (IsModified); 2042 } 2043 2044 void PPCDAGToDAGISel::PeepholePPC64() { 2045 // These optimizations are currently supported only for 64-bit SVR4. 2046 if (PPCSubTarget->isDarwin() || !PPCSubTarget->isPPC64()) 2047 return; 2048 2049 SelectionDAG::allnodes_iterator Position(CurDAG->getRoot().getNode()); 2050 ++Position; 2051 2052 while (Position != CurDAG->allnodes_begin()) { 2053 SDNode *N = --Position; 2054 // Skip dead nodes and any non-machine opcodes. 2055 if (N->use_empty() || !N->isMachineOpcode()) 2056 continue; 2057 2058 unsigned FirstOp; 2059 unsigned StorageOpcode = N->getMachineOpcode(); 2060 2061 switch (StorageOpcode) { 2062 default: continue; 2063 2064 case PPC::LBZ: 2065 case PPC::LBZ8: 2066 case PPC::LD: 2067 case PPC::LFD: 2068 case PPC::LFS: 2069 case PPC::LHA: 2070 case PPC::LHA8: 2071 case PPC::LHZ: 2072 case PPC::LHZ8: 2073 case PPC::LWA: 2074 case PPC::LWZ: 2075 case PPC::LWZ8: 2076 FirstOp = 0; 2077 break; 2078 2079 case PPC::STB: 2080 case PPC::STB8: 2081 case PPC::STD: 2082 case PPC::STFD: 2083 case PPC::STFS: 2084 case PPC::STH: 2085 case PPC::STH8: 2086 case PPC::STW: 2087 case PPC::STW8: 2088 FirstOp = 1; 2089 break; 2090 } 2091 2092 // If this is a load or store with a zero offset, we may be able to 2093 // fold an add-immediate into the memory operation. 2094 if (!isa<ConstantSDNode>(N->getOperand(FirstOp)) || 2095 N->getConstantOperandVal(FirstOp) != 0) 2096 continue; 2097 2098 SDValue Base = N->getOperand(FirstOp + 1); 2099 if (!Base.isMachineOpcode()) 2100 continue; 2101 2102 unsigned Flags = 0; 2103 bool ReplaceFlags = true; 2104 2105 // When the feeding operation is an add-immediate of some sort, 2106 // determine whether we need to add relocation information to the 2107 // target flags on the immediate operand when we fold it into the 2108 // load instruction. 2109 // 2110 // For something like ADDItocL, the relocation information is 2111 // inferred from the opcode; when we process it in the AsmPrinter, 2112 // we add the necessary relocation there. A load, though, can receive 2113 // relocation from various flavors of ADDIxxx, so we need to carry 2114 // the relocation information in the target flags. 2115 switch (Base.getMachineOpcode()) { 2116 default: continue; 2117 2118 case PPC::ADDI8: 2119 case PPC::ADDI: 2120 // In some cases (such as TLS) the relocation information 2121 // is already in place on the operand, so copying the operand 2122 // is sufficient. 2123 ReplaceFlags = false; 2124 // For these cases, the immediate may not be divisible by 4, in 2125 // which case the fold is illegal for DS-form instructions. (The 2126 // other cases provide aligned addresses and are always safe.) 2127 if ((StorageOpcode == PPC::LWA || 2128 StorageOpcode == PPC::LD || 2129 StorageOpcode == PPC::STD) && 2130 (!isa<ConstantSDNode>(Base.getOperand(1)) || 2131 Base.getConstantOperandVal(1) % 4 != 0)) 2132 continue; 2133 break; 2134 case PPC::ADDIdtprelL: 2135 Flags = PPCII::MO_DTPREL_LO; 2136 break; 2137 case PPC::ADDItlsldL: 2138 Flags = PPCII::MO_TLSLD_LO; 2139 break; 2140 case PPC::ADDItocL: 2141 Flags = PPCII::MO_TOC_LO; 2142 break; 2143 } 2144 2145 // We found an opportunity. Reverse the operands from the add 2146 // immediate and substitute them into the load or store. If 2147 // needed, update the target flags for the immediate operand to 2148 // reflect the necessary relocation information. 2149 DEBUG(dbgs() << "Folding add-immediate into mem-op:\nBase: "); 2150 DEBUG(Base->dump(CurDAG)); 2151 DEBUG(dbgs() << "\nN: "); 2152 DEBUG(N->dump(CurDAG)); 2153 DEBUG(dbgs() << "\n"); 2154 2155 SDValue ImmOpnd = Base.getOperand(1); 2156 2157 // If the relocation information isn't already present on the 2158 // immediate operand, add it now. 2159 if (ReplaceFlags) { 2160 if (GlobalAddressSDNode *GA = dyn_cast<GlobalAddressSDNode>(ImmOpnd)) { 2161 SDLoc dl(GA); 2162 const GlobalValue *GV = GA->getGlobal(); 2163 // We can't perform this optimization for data whose alignment 2164 // is insufficient for the instruction encoding. 2165 if (GV->getAlignment() < 4 && 2166 (StorageOpcode == PPC::LD || StorageOpcode == PPC::STD || 2167 StorageOpcode == PPC::LWA)) { 2168 DEBUG(dbgs() << "Rejected this candidate for alignment.\n\n"); 2169 continue; 2170 } 2171 ImmOpnd = CurDAG->getTargetGlobalAddress(GV, dl, MVT::i64, 0, Flags); 2172 } else if (ConstantPoolSDNode *CP = 2173 dyn_cast<ConstantPoolSDNode>(ImmOpnd)) { 2174 const Constant *C = CP->getConstVal(); 2175 ImmOpnd = CurDAG->getTargetConstantPool(C, MVT::i64, 2176 CP->getAlignment(), 2177 0, Flags); 2178 } 2179 } 2180 2181 if (FirstOp == 1) // Store 2182 (void)CurDAG->UpdateNodeOperands(N, N->getOperand(0), ImmOpnd, 2183 Base.getOperand(0), N->getOperand(3)); 2184 else // Load 2185 (void)CurDAG->UpdateNodeOperands(N, ImmOpnd, Base.getOperand(0), 2186 N->getOperand(2)); 2187 2188 // The add-immediate may now be dead, in which case remove it. 2189 if (Base.getNode()->use_empty()) 2190 CurDAG->RemoveDeadNode(Base.getNode()); 2191 } 2192 } 2193 2194 2195 /// createPPCISelDag - This pass converts a legalized DAG into a 2196 /// PowerPC-specific DAG, ready for instruction scheduling. 2197 /// 2198 FunctionPass *llvm::createPPCISelDag(PPCTargetMachine &TM) { 2199 return new PPCDAGToDAGISel(TM); 2200 } 2201 2202 static void initializePassOnce(PassRegistry &Registry) { 2203 const char *Name = "PowerPC DAG->DAG Pattern Instruction Selection"; 2204 PassInfo *PI = new PassInfo(Name, "ppc-codegen", &SelectionDAGISel::ID, 2205 nullptr, false, false); 2206 Registry.registerPass(*PI, true); 2207 } 2208 2209 void llvm::initializePPCDAGToDAGISelPass(PassRegistry &Registry) { 2210 CALL_ONCE_INITIALIZATION(initializePassOnce); 2211 } 2212 2213