1 //===- HexagonHardwareLoops.cpp - Identify and generate hardware loops ----===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // This pass identifies loops where we can generate the Hexagon hardware 10 // loop instruction. The hardware loop can perform loop branches with a 11 // zero-cycle overhead. 12 // 13 // The pattern that defines the induction variable can changed depending on 14 // prior optimizations. For example, the IndVarSimplify phase run by 'opt' 15 // normalizes induction variables, and the Loop Strength Reduction pass 16 // run by 'llc' may also make changes to the induction variable. 17 // The pattern detected by this phase is due to running Strength Reduction. 18 // 19 // Criteria for hardware loops: 20 // - Countable loops (w/ ind. var for a trip count) 21 // - Assumes loops are normalized by IndVarSimplify 22 // - Try inner-most loops first 23 // - No function calls in loops. 24 // 25 //===----------------------------------------------------------------------===// 26 27 #include "HexagonInstrInfo.h" 28 #include "HexagonSubtarget.h" 29 #include "llvm/ADT/ArrayRef.h" 30 #include "llvm/ADT/STLExtras.h" 31 #include "llvm/ADT/SmallSet.h" 32 #include "llvm/ADT/SmallVector.h" 33 #include "llvm/ADT/Statistic.h" 34 #include "llvm/ADT/StringRef.h" 35 #include "llvm/CodeGen/MachineBasicBlock.h" 36 #include "llvm/CodeGen/MachineDominators.h" 37 #include "llvm/CodeGen/MachineFunction.h" 38 #include "llvm/CodeGen/MachineFunctionPass.h" 39 #include "llvm/CodeGen/MachineInstr.h" 40 #include "llvm/CodeGen/MachineInstrBuilder.h" 41 #include "llvm/CodeGen/MachineLoopInfo.h" 42 #include "llvm/CodeGen/MachineOperand.h" 43 #include "llvm/CodeGen/MachineRegisterInfo.h" 44 #include "llvm/CodeGen/TargetRegisterInfo.h" 45 #include "llvm/IR/Constants.h" 46 #include "llvm/IR/DebugLoc.h" 47 #include "llvm/InitializePasses.h" 48 #include "llvm/Pass.h" 49 #include "llvm/Support/CommandLine.h" 50 #include "llvm/Support/Debug.h" 51 #include "llvm/Support/ErrorHandling.h" 52 #include "llvm/Support/MathExtras.h" 53 #include "llvm/Support/raw_ostream.h" 54 #include <cassert> 55 #include <cstdint> 56 #include <cstdlib> 57 #include <iterator> 58 #include <map> 59 #include <set> 60 #include <string> 61 #include <utility> 62 #include <vector> 63 64 using namespace llvm; 65 66 #define DEBUG_TYPE "hwloops" 67 68 #ifndef NDEBUG 69 static cl::opt<int> HWLoopLimit("hexagon-max-hwloop", cl::Hidden, cl::init(-1)); 70 71 // Option to create preheader only for a specific function. 72 static cl::opt<std::string> PHFn("hexagon-hwloop-phfn", cl::Hidden, 73 cl::init("")); 74 #endif 75 76 // Option to create a preheader if one doesn't exist. 77 static cl::opt<bool> HWCreatePreheader("hexagon-hwloop-preheader", 78 cl::Hidden, cl::init(true), 79 cl::desc("Add a preheader to a hardware loop if one doesn't exist")); 80 81 // Turn it off by default. If a preheader block is not created here, the 82 // software pipeliner may be unable to find a block suitable to serve as 83 // a preheader. In that case SWP will not run. 84 static cl::opt<bool> SpecPreheader("hwloop-spec-preheader", cl::Hidden, 85 cl::desc("Allow speculation of preheader " 86 "instructions")); 87 88 STATISTIC(NumHWLoops, "Number of loops converted to hardware loops"); 89 90 namespace llvm { 91 92 FunctionPass *createHexagonHardwareLoops(); 93 void initializeHexagonHardwareLoopsPass(PassRegistry&); 94 95 } // end namespace llvm 96 97 namespace { 98 99 class CountValue; 100 101 struct HexagonHardwareLoops : public MachineFunctionPass { 102 MachineLoopInfo *MLI; 103 MachineRegisterInfo *MRI; 104 MachineDominatorTree *MDT; 105 const HexagonInstrInfo *TII; 106 const HexagonRegisterInfo *TRI; 107 #ifndef NDEBUG 108 static int Counter; 109 #endif 110 111 public: 112 static char ID; 113 114 HexagonHardwareLoops() : MachineFunctionPass(ID) {} 115 116 bool runOnMachineFunction(MachineFunction &MF) override; 117 118 StringRef getPassName() const override { return "Hexagon Hardware Loops"; } 119 120 void getAnalysisUsage(AnalysisUsage &AU) const override { 121 AU.addRequired<MachineDominatorTree>(); 122 AU.addRequired<MachineLoopInfo>(); 123 MachineFunctionPass::getAnalysisUsage(AU); 124 } 125 126 private: 127 using LoopFeederMap = std::map<Register, MachineInstr *>; 128 129 /// Kinds of comparisons in the compare instructions. 130 struct Comparison { 131 enum Kind { 132 EQ = 0x01, 133 NE = 0x02, 134 L = 0x04, 135 G = 0x08, 136 U = 0x40, 137 LTs = L, 138 LEs = L | EQ, 139 GTs = G, 140 GEs = G | EQ, 141 LTu = L | U, 142 LEu = L | EQ | U, 143 GTu = G | U, 144 GEu = G | EQ | U 145 }; 146 147 static Kind getSwappedComparison(Kind Cmp) { 148 assert ((!((Cmp & L) && (Cmp & G))) && "Malformed comparison operator"); 149 if ((Cmp & L) || (Cmp & G)) 150 return (Kind)(Cmp ^ (L|G)); 151 return Cmp; 152 } 153 154 static Kind getNegatedComparison(Kind Cmp) { 155 if ((Cmp & L) || (Cmp & G)) 156 return (Kind)((Cmp ^ (L | G)) ^ EQ); 157 if ((Cmp & NE) || (Cmp & EQ)) 158 return (Kind)(Cmp ^ (EQ | NE)); 159 return (Kind)0; 160 } 161 162 static bool isSigned(Kind Cmp) { 163 return (Cmp & (L | G) && !(Cmp & U)); 164 } 165 166 static bool isUnsigned(Kind Cmp) { 167 return (Cmp & U); 168 } 169 }; 170 171 /// Find the register that contains the loop controlling 172 /// induction variable. 173 /// If successful, it will return true and set the \p Reg, \p IVBump 174 /// and \p IVOp arguments. Otherwise it will return false. 175 /// The returned induction register is the register R that follows the 176 /// following induction pattern: 177 /// loop: 178 /// R = phi ..., [ R.next, LatchBlock ] 179 /// R.next = R + #bump 180 /// if (R.next < #N) goto loop 181 /// IVBump is the immediate value added to R, and IVOp is the instruction 182 /// "R.next = R + #bump". 183 bool findInductionRegister(MachineLoop *L, Register &Reg, 184 int64_t &IVBump, MachineInstr *&IVOp) const; 185 186 /// Return the comparison kind for the specified opcode. 187 Comparison::Kind getComparisonKind(unsigned CondOpc, 188 MachineOperand *InitialValue, 189 const MachineOperand *Endvalue, 190 int64_t IVBump) const; 191 192 /// Analyze the statements in a loop to determine if the loop 193 /// has a computable trip count and, if so, return a value that represents 194 /// the trip count expression. 195 CountValue *getLoopTripCount(MachineLoop *L, 196 SmallVectorImpl<MachineInstr *> &OldInsts); 197 198 /// Return the expression that represents the number of times 199 /// a loop iterates. The function takes the operands that represent the 200 /// loop start value, loop end value, and induction value. Based upon 201 /// these operands, the function attempts to compute the trip count. 202 /// If the trip count is not directly available (as an immediate value, 203 /// or a register), the function will attempt to insert computation of it 204 /// to the loop's preheader. 205 CountValue *computeCount(MachineLoop *Loop, const MachineOperand *Start, 206 const MachineOperand *End, Register IVReg, 207 int64_t IVBump, Comparison::Kind Cmp) const; 208 209 /// Return true if the instruction is not valid within a hardware 210 /// loop. 211 bool isInvalidLoopOperation(const MachineInstr *MI, 212 bool IsInnerHWLoop) const; 213 214 /// Return true if the loop contains an instruction that inhibits 215 /// using the hardware loop. 216 bool containsInvalidInstruction(MachineLoop *L, bool IsInnerHWLoop) const; 217 218 /// Given a loop, check if we can convert it to a hardware loop. 219 /// If so, then perform the conversion and return true. 220 bool convertToHardwareLoop(MachineLoop *L, bool &L0used, bool &L1used); 221 222 /// Return true if the instruction is now dead. 223 bool isDead(const MachineInstr *MI, 224 SmallVectorImpl<MachineInstr *> &DeadPhis) const; 225 226 /// Remove the instruction if it is now dead. 227 void removeIfDead(MachineInstr *MI); 228 229 /// Make sure that the "bump" instruction executes before the 230 /// compare. We need that for the IV fixup, so that the compare 231 /// instruction would not use a bumped value that has not yet been 232 /// defined. If the instructions are out of order, try to reorder them. 233 bool orderBumpCompare(MachineInstr *BumpI, MachineInstr *CmpI); 234 235 /// Return true if MO and MI pair is visited only once. If visited 236 /// more than once, this indicates there is recursion. In such a case, 237 /// return false. 238 bool isLoopFeeder(MachineLoop *L, MachineBasicBlock *A, MachineInstr *MI, 239 const MachineOperand *MO, 240 LoopFeederMap &LoopFeederPhi) const; 241 242 /// Return true if the Phi may generate a value that may underflow, 243 /// or may wrap. 244 bool phiMayWrapOrUnderflow(MachineInstr *Phi, const MachineOperand *EndVal, 245 MachineBasicBlock *MBB, MachineLoop *L, 246 LoopFeederMap &LoopFeederPhi) const; 247 248 /// Return true if the induction variable may underflow an unsigned 249 /// value in the first iteration. 250 bool loopCountMayWrapOrUnderFlow(const MachineOperand *InitVal, 251 const MachineOperand *EndVal, 252 MachineBasicBlock *MBB, MachineLoop *L, 253 LoopFeederMap &LoopFeederPhi) const; 254 255 /// Check if the given operand has a compile-time known constant 256 /// value. Return true if yes, and false otherwise. When returning true, set 257 /// Val to the corresponding constant value. 258 bool checkForImmediate(const MachineOperand &MO, int64_t &Val) const; 259 260 /// Check if the operand has a compile-time known constant value. 261 bool isImmediate(const MachineOperand &MO) const { 262 int64_t V; 263 return checkForImmediate(MO, V); 264 } 265 266 /// Return the immediate for the specified operand. 267 int64_t getImmediate(const MachineOperand &MO) const { 268 int64_t V; 269 if (!checkForImmediate(MO, V)) 270 llvm_unreachable("Invalid operand"); 271 return V; 272 } 273 274 /// Reset the given machine operand to now refer to a new immediate 275 /// value. Assumes that the operand was already referencing an immediate 276 /// value, either directly, or via a register. 277 void setImmediate(MachineOperand &MO, int64_t Val); 278 279 /// Fix the data flow of the induction variable. 280 /// The desired flow is: phi ---> bump -+-> comparison-in-latch. 281 /// | 282 /// +-> back to phi 283 /// where "bump" is the increment of the induction variable: 284 /// iv = iv + #const. 285 /// Due to some prior code transformations, the actual flow may look 286 /// like this: 287 /// phi -+-> bump ---> back to phi 288 /// | 289 /// +-> comparison-in-latch (against upper_bound-bump), 290 /// i.e. the comparison that controls the loop execution may be using 291 /// the value of the induction variable from before the increment. 292 /// 293 /// Return true if the loop's flow is the desired one (i.e. it's 294 /// either been fixed, or no fixing was necessary). 295 /// Otherwise, return false. This can happen if the induction variable 296 /// couldn't be identified, or if the value in the latch's comparison 297 /// cannot be adjusted to reflect the post-bump value. 298 bool fixupInductionVariable(MachineLoop *L); 299 300 /// Given a loop, if it does not have a preheader, create one. 301 /// Return the block that is the preheader. 302 MachineBasicBlock *createPreheaderForLoop(MachineLoop *L); 303 }; 304 305 char HexagonHardwareLoops::ID = 0; 306 #ifndef NDEBUG 307 int HexagonHardwareLoops::Counter = 0; 308 #endif 309 310 /// Abstraction for a trip count of a loop. A smaller version 311 /// of the MachineOperand class without the concerns of changing the 312 /// operand representation. 313 class CountValue { 314 public: 315 enum CountValueType { 316 CV_Register, 317 CV_Immediate 318 }; 319 320 private: 321 CountValueType Kind; 322 union Values { 323 Values() : R{Register(), 0} {} 324 Values(const Values&) = default; 325 struct { 326 Register Reg; 327 unsigned Sub; 328 } R; 329 unsigned ImmVal; 330 } Contents; 331 332 public: 333 explicit CountValue(CountValueType t, Register v, unsigned u = 0) { 334 Kind = t; 335 if (Kind == CV_Register) { 336 Contents.R.Reg = v; 337 Contents.R.Sub = u; 338 } else { 339 Contents.ImmVal = v; 340 } 341 } 342 343 bool isReg() const { return Kind == CV_Register; } 344 bool isImm() const { return Kind == CV_Immediate; } 345 346 Register getReg() const { 347 assert(isReg() && "Wrong CountValue accessor"); 348 return Contents.R.Reg; 349 } 350 351 unsigned getSubReg() const { 352 assert(isReg() && "Wrong CountValue accessor"); 353 return Contents.R.Sub; 354 } 355 356 unsigned getImm() const { 357 assert(isImm() && "Wrong CountValue accessor"); 358 return Contents.ImmVal; 359 } 360 361 void print(raw_ostream &OS, const TargetRegisterInfo *TRI = nullptr) const { 362 if (isReg()) { OS << printReg(Contents.R.Reg, TRI, Contents.R.Sub); } 363 if (isImm()) { OS << Contents.ImmVal; } 364 } 365 }; 366 367 } // end anonymous namespace 368 369 INITIALIZE_PASS_BEGIN(HexagonHardwareLoops, "hwloops", 370 "Hexagon Hardware Loops", false, false) 371 INITIALIZE_PASS_DEPENDENCY(MachineDominatorTree) 372 INITIALIZE_PASS_DEPENDENCY(MachineLoopInfo) 373 INITIALIZE_PASS_END(HexagonHardwareLoops, "hwloops", 374 "Hexagon Hardware Loops", false, false) 375 376 FunctionPass *llvm::createHexagonHardwareLoops() { 377 return new HexagonHardwareLoops(); 378 } 379 380 bool HexagonHardwareLoops::runOnMachineFunction(MachineFunction &MF) { 381 LLVM_DEBUG(dbgs() << "********* Hexagon Hardware Loops *********\n"); 382 if (skipFunction(MF.getFunction())) 383 return false; 384 385 bool Changed = false; 386 387 MLI = &getAnalysis<MachineLoopInfo>(); 388 MRI = &MF.getRegInfo(); 389 MDT = &getAnalysis<MachineDominatorTree>(); 390 const HexagonSubtarget &HST = MF.getSubtarget<HexagonSubtarget>(); 391 TII = HST.getInstrInfo(); 392 TRI = HST.getRegisterInfo(); 393 394 for (auto &L : *MLI) 395 if (L->isOutermost()) { 396 bool L0Used = false; 397 bool L1Used = false; 398 Changed |= convertToHardwareLoop(L, L0Used, L1Used); 399 } 400 401 return Changed; 402 } 403 404 bool HexagonHardwareLoops::findInductionRegister(MachineLoop *L, 405 Register &Reg, 406 int64_t &IVBump, 407 MachineInstr *&IVOp 408 ) const { 409 MachineBasicBlock *Header = L->getHeader(); 410 MachineBasicBlock *Preheader = MLI->findLoopPreheader(L, SpecPreheader); 411 MachineBasicBlock *Latch = L->getLoopLatch(); 412 MachineBasicBlock *ExitingBlock = L->findLoopControlBlock(); 413 if (!Header || !Preheader || !Latch || !ExitingBlock) 414 return false; 415 416 // This pair represents an induction register together with an immediate 417 // value that will be added to it in each loop iteration. 418 using RegisterBump = std::pair<Register, int64_t>; 419 420 // Mapping: R.next -> (R, bump), where R, R.next and bump are derived 421 // from an induction operation 422 // R.next = R + bump 423 // where bump is an immediate value. 424 using InductionMap = std::map<Register, RegisterBump>; 425 426 InductionMap IndMap; 427 428 using instr_iterator = MachineBasicBlock::instr_iterator; 429 430 for (instr_iterator I = Header->instr_begin(), E = Header->instr_end(); 431 I != E && I->isPHI(); ++I) { 432 MachineInstr *Phi = &*I; 433 434 // Have a PHI instruction. Get the operand that corresponds to the 435 // latch block, and see if is a result of an addition of form "reg+imm", 436 // where the "reg" is defined by the PHI node we are looking at. 437 for (unsigned i = 1, n = Phi->getNumOperands(); i < n; i += 2) { 438 if (Phi->getOperand(i+1).getMBB() != Latch) 439 continue; 440 441 Register PhiOpReg = Phi->getOperand(i).getReg(); 442 MachineInstr *DI = MRI->getVRegDef(PhiOpReg); 443 444 if (DI->getDesc().isAdd()) { 445 // If the register operand to the add is the PHI we're looking at, this 446 // meets the induction pattern. 447 Register IndReg = DI->getOperand(1).getReg(); 448 MachineOperand &Opnd2 = DI->getOperand(2); 449 int64_t V; 450 if (MRI->getVRegDef(IndReg) == Phi && checkForImmediate(Opnd2, V)) { 451 Register UpdReg = DI->getOperand(0).getReg(); 452 IndMap.insert(std::make_pair(UpdReg, std::make_pair(IndReg, V))); 453 } 454 } 455 } // for (i) 456 } // for (instr) 457 458 SmallVector<MachineOperand,2> Cond; 459 MachineBasicBlock *TB = nullptr, *FB = nullptr; 460 bool NotAnalyzed = TII->analyzeBranch(*ExitingBlock, TB, FB, Cond, false); 461 if (NotAnalyzed) 462 return false; 463 464 Register PredR; 465 unsigned PredPos, PredRegFlags; 466 if (!TII->getPredReg(Cond, PredR, PredPos, PredRegFlags)) 467 return false; 468 469 MachineInstr *PredI = MRI->getVRegDef(PredR); 470 if (!PredI->isCompare()) 471 return false; 472 473 Register CmpReg1, CmpReg2; 474 int64_t CmpImm = 0, CmpMask = 0; 475 bool CmpAnalyzed = 476 TII->analyzeCompare(*PredI, CmpReg1, CmpReg2, CmpMask, CmpImm); 477 // Fail if the compare was not analyzed, or it's not comparing a register 478 // with an immediate value. Not checking the mask here, since we handle 479 // the individual compare opcodes (including A4_cmpb*) later on. 480 if (!CmpAnalyzed) 481 return false; 482 483 // Exactly one of the input registers to the comparison should be among 484 // the induction registers. 485 InductionMap::iterator IndMapEnd = IndMap.end(); 486 InductionMap::iterator F = IndMapEnd; 487 if (CmpReg1 != 0) { 488 InductionMap::iterator F1 = IndMap.find(CmpReg1); 489 if (F1 != IndMapEnd) 490 F = F1; 491 } 492 if (CmpReg2 != 0) { 493 InductionMap::iterator F2 = IndMap.find(CmpReg2); 494 if (F2 != IndMapEnd) { 495 if (F != IndMapEnd) 496 return false; 497 F = F2; 498 } 499 } 500 if (F == IndMapEnd) 501 return false; 502 503 Reg = F->second.first; 504 IVBump = F->second.second; 505 IVOp = MRI->getVRegDef(F->first); 506 return true; 507 } 508 509 // Return the comparison kind for the specified opcode. 510 HexagonHardwareLoops::Comparison::Kind 511 HexagonHardwareLoops::getComparisonKind(unsigned CondOpc, 512 MachineOperand *InitialValue, 513 const MachineOperand *EndValue, 514 int64_t IVBump) const { 515 Comparison::Kind Cmp = (Comparison::Kind)0; 516 switch (CondOpc) { 517 case Hexagon::C2_cmpeq: 518 case Hexagon::C2_cmpeqi: 519 case Hexagon::C2_cmpeqp: 520 Cmp = Comparison::EQ; 521 break; 522 case Hexagon::C4_cmpneq: 523 case Hexagon::C4_cmpneqi: 524 Cmp = Comparison::NE; 525 break; 526 case Hexagon::C2_cmplt: 527 Cmp = Comparison::LTs; 528 break; 529 case Hexagon::C2_cmpltu: 530 Cmp = Comparison::LTu; 531 break; 532 case Hexagon::C4_cmplte: 533 case Hexagon::C4_cmpltei: 534 Cmp = Comparison::LEs; 535 break; 536 case Hexagon::C4_cmplteu: 537 case Hexagon::C4_cmplteui: 538 Cmp = Comparison::LEu; 539 break; 540 case Hexagon::C2_cmpgt: 541 case Hexagon::C2_cmpgti: 542 case Hexagon::C2_cmpgtp: 543 Cmp = Comparison::GTs; 544 break; 545 case Hexagon::C2_cmpgtu: 546 case Hexagon::C2_cmpgtui: 547 case Hexagon::C2_cmpgtup: 548 Cmp = Comparison::GTu; 549 break; 550 case Hexagon::C2_cmpgei: 551 Cmp = Comparison::GEs; 552 break; 553 case Hexagon::C2_cmpgeui: 554 Cmp = Comparison::GEs; 555 break; 556 default: 557 return (Comparison::Kind)0; 558 } 559 return Cmp; 560 } 561 562 /// Analyze the statements in a loop to determine if the loop has 563 /// a computable trip count and, if so, return a value that represents 564 /// the trip count expression. 565 /// 566 /// This function iterates over the phi nodes in the loop to check for 567 /// induction variable patterns that are used in the calculation for 568 /// the number of time the loop is executed. 569 CountValue *HexagonHardwareLoops::getLoopTripCount(MachineLoop *L, 570 SmallVectorImpl<MachineInstr *> &OldInsts) { 571 MachineBasicBlock *TopMBB = L->getTopBlock(); 572 MachineBasicBlock::pred_iterator PI = TopMBB->pred_begin(); 573 assert(PI != TopMBB->pred_end() && 574 "Loop must have more than one incoming edge!"); 575 MachineBasicBlock *Backedge = *PI++; 576 if (PI == TopMBB->pred_end()) // dead loop? 577 return nullptr; 578 MachineBasicBlock *Incoming = *PI++; 579 if (PI != TopMBB->pred_end()) // multiple backedges? 580 return nullptr; 581 582 // Make sure there is one incoming and one backedge and determine which 583 // is which. 584 if (L->contains(Incoming)) { 585 if (L->contains(Backedge)) 586 return nullptr; 587 std::swap(Incoming, Backedge); 588 } else if (!L->contains(Backedge)) 589 return nullptr; 590 591 // Look for the cmp instruction to determine if we can get a useful trip 592 // count. The trip count can be either a register or an immediate. The 593 // location of the value depends upon the type (reg or imm). 594 MachineBasicBlock *ExitingBlock = L->findLoopControlBlock(); 595 if (!ExitingBlock) 596 return nullptr; 597 598 Register IVReg = 0; 599 int64_t IVBump = 0; 600 MachineInstr *IVOp; 601 bool FoundIV = findInductionRegister(L, IVReg, IVBump, IVOp); 602 if (!FoundIV) 603 return nullptr; 604 605 MachineBasicBlock *Preheader = MLI->findLoopPreheader(L, SpecPreheader); 606 607 MachineOperand *InitialValue = nullptr; 608 MachineInstr *IV_Phi = MRI->getVRegDef(IVReg); 609 MachineBasicBlock *Latch = L->getLoopLatch(); 610 for (unsigned i = 1, n = IV_Phi->getNumOperands(); i < n; i += 2) { 611 MachineBasicBlock *MBB = IV_Phi->getOperand(i+1).getMBB(); 612 if (MBB == Preheader) 613 InitialValue = &IV_Phi->getOperand(i); 614 else if (MBB == Latch) 615 IVReg = IV_Phi->getOperand(i).getReg(); // Want IV reg after bump. 616 } 617 if (!InitialValue) 618 return nullptr; 619 620 SmallVector<MachineOperand,2> Cond; 621 MachineBasicBlock *TB = nullptr, *FB = nullptr; 622 bool NotAnalyzed = TII->analyzeBranch(*ExitingBlock, TB, FB, Cond, false); 623 if (NotAnalyzed) 624 return nullptr; 625 626 MachineBasicBlock *Header = L->getHeader(); 627 // TB must be non-null. If FB is also non-null, one of them must be 628 // the header. Otherwise, branch to TB could be exiting the loop, and 629 // the fall through can go to the header. 630 assert (TB && "Exit block without a branch?"); 631 if (ExitingBlock != Latch && (TB == Latch || FB == Latch)) { 632 MachineBasicBlock *LTB = nullptr, *LFB = nullptr; 633 SmallVector<MachineOperand,2> LCond; 634 bool NotAnalyzed = TII->analyzeBranch(*Latch, LTB, LFB, LCond, false); 635 if (NotAnalyzed) 636 return nullptr; 637 if (TB == Latch) 638 TB = (LTB == Header) ? LTB : LFB; 639 else 640 FB = (LTB == Header) ? LTB: LFB; 641 } 642 assert ((!FB || TB == Header || FB == Header) && "Branches not to header?"); 643 if (!TB || (FB && TB != Header && FB != Header)) 644 return nullptr; 645 646 // Branches of form "if (!P) ..." cause HexagonInstrInfo::analyzeBranch 647 // to put imm(0), followed by P in the vector Cond. 648 // If TB is not the header, it means that the "not-taken" path must lead 649 // to the header. 650 bool Negated = TII->predOpcodeHasNot(Cond) ^ (TB != Header); 651 Register PredReg; 652 unsigned PredPos, PredRegFlags; 653 if (!TII->getPredReg(Cond, PredReg, PredPos, PredRegFlags)) 654 return nullptr; 655 MachineInstr *CondI = MRI->getVRegDef(PredReg); 656 unsigned CondOpc = CondI->getOpcode(); 657 658 Register CmpReg1, CmpReg2; 659 int64_t Mask = 0, ImmValue = 0; 660 bool AnalyzedCmp = 661 TII->analyzeCompare(*CondI, CmpReg1, CmpReg2, Mask, ImmValue); 662 if (!AnalyzedCmp) 663 return nullptr; 664 665 // The comparison operator type determines how we compute the loop 666 // trip count. 667 OldInsts.push_back(CondI); 668 OldInsts.push_back(IVOp); 669 670 // Sadly, the following code gets information based on the position 671 // of the operands in the compare instruction. This has to be done 672 // this way, because the comparisons check for a specific relationship 673 // between the operands (e.g. is-less-than), rather than to find out 674 // what relationship the operands are in (as on PPC). 675 Comparison::Kind Cmp; 676 bool isSwapped = false; 677 const MachineOperand &Op1 = CondI->getOperand(1); 678 const MachineOperand &Op2 = CondI->getOperand(2); 679 const MachineOperand *EndValue = nullptr; 680 681 if (Op1.isReg()) { 682 if (Op2.isImm() || Op1.getReg() == IVReg) 683 EndValue = &Op2; 684 else { 685 EndValue = &Op1; 686 isSwapped = true; 687 } 688 } 689 690 if (!EndValue) 691 return nullptr; 692 693 Cmp = getComparisonKind(CondOpc, InitialValue, EndValue, IVBump); 694 if (!Cmp) 695 return nullptr; 696 if (Negated) 697 Cmp = Comparison::getNegatedComparison(Cmp); 698 if (isSwapped) 699 Cmp = Comparison::getSwappedComparison(Cmp); 700 701 if (InitialValue->isReg()) { 702 Register R = InitialValue->getReg(); 703 MachineBasicBlock *DefBB = MRI->getVRegDef(R)->getParent(); 704 if (!MDT->properlyDominates(DefBB, Header)) { 705 int64_t V; 706 if (!checkForImmediate(*InitialValue, V)) 707 return nullptr; 708 } 709 OldInsts.push_back(MRI->getVRegDef(R)); 710 } 711 if (EndValue->isReg()) { 712 Register R = EndValue->getReg(); 713 MachineBasicBlock *DefBB = MRI->getVRegDef(R)->getParent(); 714 if (!MDT->properlyDominates(DefBB, Header)) { 715 int64_t V; 716 if (!checkForImmediate(*EndValue, V)) 717 return nullptr; 718 } 719 OldInsts.push_back(MRI->getVRegDef(R)); 720 } 721 722 return computeCount(L, InitialValue, EndValue, IVReg, IVBump, Cmp); 723 } 724 725 /// Helper function that returns the expression that represents the 726 /// number of times a loop iterates. The function takes the operands that 727 /// represent the loop start value, loop end value, and induction value. 728 /// Based upon these operands, the function attempts to compute the trip count. 729 CountValue *HexagonHardwareLoops::computeCount(MachineLoop *Loop, 730 const MachineOperand *Start, 731 const MachineOperand *End, 732 Register IVReg, 733 int64_t IVBump, 734 Comparison::Kind Cmp) const { 735 // Cannot handle comparison EQ, i.e. while (A == B). 736 if (Cmp == Comparison::EQ) 737 return nullptr; 738 739 // Check if either the start or end values are an assignment of an immediate. 740 // If so, use the immediate value rather than the register. 741 if (Start->isReg()) { 742 const MachineInstr *StartValInstr = MRI->getVRegDef(Start->getReg()); 743 if (StartValInstr && (StartValInstr->getOpcode() == Hexagon::A2_tfrsi || 744 StartValInstr->getOpcode() == Hexagon::A2_tfrpi)) 745 Start = &StartValInstr->getOperand(1); 746 } 747 if (End->isReg()) { 748 const MachineInstr *EndValInstr = MRI->getVRegDef(End->getReg()); 749 if (EndValInstr && (EndValInstr->getOpcode() == Hexagon::A2_tfrsi || 750 EndValInstr->getOpcode() == Hexagon::A2_tfrpi)) 751 End = &EndValInstr->getOperand(1); 752 } 753 754 if (!Start->isReg() && !Start->isImm()) 755 return nullptr; 756 if (!End->isReg() && !End->isImm()) 757 return nullptr; 758 759 bool CmpLess = Cmp & Comparison::L; 760 bool CmpGreater = Cmp & Comparison::G; 761 bool CmpHasEqual = Cmp & Comparison::EQ; 762 763 // Avoid certain wrap-arounds. This doesn't detect all wrap-arounds. 764 if (CmpLess && IVBump < 0) 765 // Loop going while iv is "less" with the iv value going down. Must wrap. 766 return nullptr; 767 768 if (CmpGreater && IVBump > 0) 769 // Loop going while iv is "greater" with the iv value going up. Must wrap. 770 return nullptr; 771 772 // Phis that may feed into the loop. 773 LoopFeederMap LoopFeederPhi; 774 775 // Check if the initial value may be zero and can be decremented in the first 776 // iteration. If the value is zero, the endloop instruction will not decrement 777 // the loop counter, so we shouldn't generate a hardware loop in this case. 778 if (loopCountMayWrapOrUnderFlow(Start, End, Loop->getLoopPreheader(), Loop, 779 LoopFeederPhi)) 780 return nullptr; 781 782 if (Start->isImm() && End->isImm()) { 783 // Both, start and end are immediates. 784 int64_t StartV = Start->getImm(); 785 int64_t EndV = End->getImm(); 786 int64_t Dist = EndV - StartV; 787 if (Dist == 0) 788 return nullptr; 789 790 bool Exact = (Dist % IVBump) == 0; 791 792 if (Cmp == Comparison::NE) { 793 if (!Exact) 794 return nullptr; 795 if ((Dist < 0) ^ (IVBump < 0)) 796 return nullptr; 797 } 798 799 // For comparisons that include the final value (i.e. include equality 800 // with the final value), we need to increase the distance by 1. 801 if (CmpHasEqual) 802 Dist = Dist > 0 ? Dist+1 : Dist-1; 803 804 // For the loop to iterate, CmpLess should imply Dist > 0. Similarly, 805 // CmpGreater should imply Dist < 0. These conditions could actually 806 // fail, for example, in unreachable code (which may still appear to be 807 // reachable in the CFG). 808 if ((CmpLess && Dist < 0) || (CmpGreater && Dist > 0)) 809 return nullptr; 810 811 // "Normalized" distance, i.e. with the bump set to +-1. 812 int64_t Dist1 = (IVBump > 0) ? (Dist + (IVBump - 1)) / IVBump 813 : (-Dist + (-IVBump - 1)) / (-IVBump); 814 assert (Dist1 > 0 && "Fishy thing. Both operands have the same sign."); 815 816 uint64_t Count = Dist1; 817 818 if (Count > 0xFFFFFFFFULL) 819 return nullptr; 820 821 return new CountValue(CountValue::CV_Immediate, Count); 822 } 823 824 // A general case: Start and End are some values, but the actual 825 // iteration count may not be available. If it is not, insert 826 // a computation of it into the preheader. 827 828 // If the induction variable bump is not a power of 2, quit. 829 // Othwerise we'd need a general integer division. 830 if (!isPowerOf2_64(std::abs(IVBump))) 831 return nullptr; 832 833 MachineBasicBlock *PH = MLI->findLoopPreheader(Loop, SpecPreheader); 834 assert (PH && "Should have a preheader by now"); 835 MachineBasicBlock::iterator InsertPos = PH->getFirstTerminator(); 836 DebugLoc DL; 837 if (InsertPos != PH->end()) 838 DL = InsertPos->getDebugLoc(); 839 840 // If Start is an immediate and End is a register, the trip count 841 // will be "reg - imm". Hexagon's "subtract immediate" instruction 842 // is actually "reg + -imm". 843 844 // If the loop IV is going downwards, i.e. if the bump is negative, 845 // then the iteration count (computed as End-Start) will need to be 846 // negated. To avoid the negation, just swap Start and End. 847 if (IVBump < 0) { 848 std::swap(Start, End); 849 IVBump = -IVBump; 850 } 851 // Cmp may now have a wrong direction, e.g. LEs may now be GEs. 852 // Signedness, and "including equality" are preserved. 853 854 bool RegToImm = Start->isReg() && End->isImm(); // for (reg..imm) 855 bool RegToReg = Start->isReg() && End->isReg(); // for (reg..reg) 856 857 int64_t StartV = 0, EndV = 0; 858 if (Start->isImm()) 859 StartV = Start->getImm(); 860 if (End->isImm()) 861 EndV = End->getImm(); 862 863 int64_t AdjV = 0; 864 // To compute the iteration count, we would need this computation: 865 // Count = (End - Start + (IVBump-1)) / IVBump 866 // or, when CmpHasEqual: 867 // Count = (End - Start + (IVBump-1)+1) / IVBump 868 // The "IVBump-1" part is the adjustment (AdjV). We can avoid 869 // generating an instruction specifically to add it if we can adjust 870 // the immediate values for Start or End. 871 872 if (CmpHasEqual) { 873 // Need to add 1 to the total iteration count. 874 if (Start->isImm()) 875 StartV--; 876 else if (End->isImm()) 877 EndV++; 878 else 879 AdjV += 1; 880 } 881 882 if (Cmp != Comparison::NE) { 883 if (Start->isImm()) 884 StartV -= (IVBump-1); 885 else if (End->isImm()) 886 EndV += (IVBump-1); 887 else 888 AdjV += (IVBump-1); 889 } 890 891 Register R = 0; 892 unsigned SR = 0; 893 if (Start->isReg()) { 894 R = Start->getReg(); 895 SR = Start->getSubReg(); 896 } else { 897 R = End->getReg(); 898 SR = End->getSubReg(); 899 } 900 const TargetRegisterClass *RC = MRI->getRegClass(R); 901 // Hardware loops cannot handle 64-bit registers. If it's a double 902 // register, it has to have a subregister. 903 if (!SR && RC == &Hexagon::DoubleRegsRegClass) 904 return nullptr; 905 const TargetRegisterClass *IntRC = &Hexagon::IntRegsRegClass; 906 907 // Compute DistR (register with the distance between Start and End). 908 Register DistR; 909 unsigned DistSR; 910 911 // Avoid special case, where the start value is an imm(0). 912 if (Start->isImm() && StartV == 0) { 913 DistR = End->getReg(); 914 DistSR = End->getSubReg(); 915 } else { 916 const MCInstrDesc &SubD = RegToReg ? TII->get(Hexagon::A2_sub) : 917 (RegToImm ? TII->get(Hexagon::A2_subri) : 918 TII->get(Hexagon::A2_addi)); 919 if (RegToReg || RegToImm) { 920 Register SubR = MRI->createVirtualRegister(IntRC); 921 MachineInstrBuilder SubIB = 922 BuildMI(*PH, InsertPos, DL, SubD, SubR); 923 924 if (RegToReg) 925 SubIB.addReg(End->getReg(), 0, End->getSubReg()) 926 .addReg(Start->getReg(), 0, Start->getSubReg()); 927 else 928 SubIB.addImm(EndV) 929 .addReg(Start->getReg(), 0, Start->getSubReg()); 930 DistR = SubR; 931 } else { 932 // If the loop has been unrolled, we should use the original loop count 933 // instead of recalculating the value. This will avoid additional 934 // 'Add' instruction. 935 const MachineInstr *EndValInstr = MRI->getVRegDef(End->getReg()); 936 if (EndValInstr->getOpcode() == Hexagon::A2_addi && 937 EndValInstr->getOperand(1).getSubReg() == 0 && 938 EndValInstr->getOperand(2).getImm() == StartV) { 939 DistR = EndValInstr->getOperand(1).getReg(); 940 } else { 941 Register SubR = MRI->createVirtualRegister(IntRC); 942 MachineInstrBuilder SubIB = 943 BuildMI(*PH, InsertPos, DL, SubD, SubR); 944 SubIB.addReg(End->getReg(), 0, End->getSubReg()) 945 .addImm(-StartV); 946 DistR = SubR; 947 } 948 } 949 DistSR = 0; 950 } 951 952 // From DistR, compute AdjR (register with the adjusted distance). 953 Register AdjR; 954 unsigned AdjSR; 955 956 if (AdjV == 0) { 957 AdjR = DistR; 958 AdjSR = DistSR; 959 } else { 960 // Generate CountR = ADD DistR, AdjVal 961 Register AddR = MRI->createVirtualRegister(IntRC); 962 MCInstrDesc const &AddD = TII->get(Hexagon::A2_addi); 963 BuildMI(*PH, InsertPos, DL, AddD, AddR) 964 .addReg(DistR, 0, DistSR) 965 .addImm(AdjV); 966 967 AdjR = AddR; 968 AdjSR = 0; 969 } 970 971 // From AdjR, compute CountR (register with the final count). 972 Register CountR; 973 unsigned CountSR; 974 975 if (IVBump == 1) { 976 CountR = AdjR; 977 CountSR = AdjSR; 978 } else { 979 // The IV bump is a power of two. Log_2(IV bump) is the shift amount. 980 unsigned Shift = Log2_32(IVBump); 981 982 // Generate NormR = LSR DistR, Shift. 983 Register LsrR = MRI->createVirtualRegister(IntRC); 984 const MCInstrDesc &LsrD = TII->get(Hexagon::S2_lsr_i_r); 985 BuildMI(*PH, InsertPos, DL, LsrD, LsrR) 986 .addReg(AdjR, 0, AdjSR) 987 .addImm(Shift); 988 989 CountR = LsrR; 990 CountSR = 0; 991 } 992 993 return new CountValue(CountValue::CV_Register, CountR, CountSR); 994 } 995 996 /// Return true if the operation is invalid within hardware loop. 997 bool HexagonHardwareLoops::isInvalidLoopOperation(const MachineInstr *MI, 998 bool IsInnerHWLoop) const { 999 // Call is not allowed because the callee may use a hardware loop except for 1000 // the case when the call never returns. 1001 if (MI->getDesc().isCall()) 1002 return !TII->doesNotReturn(*MI); 1003 1004 // Check if the instruction defines a hardware loop register. 1005 using namespace Hexagon; 1006 1007 static const Register Regs01[] = { LC0, SA0, LC1, SA1 }; 1008 static const Register Regs1[] = { LC1, SA1 }; 1009 auto CheckRegs = IsInnerHWLoop ? ArrayRef(Regs01, std::size(Regs01)) 1010 : ArrayRef(Regs1, std::size(Regs1)); 1011 for (Register R : CheckRegs) 1012 if (MI->modifiesRegister(R, TRI)) 1013 return true; 1014 1015 return false; 1016 } 1017 1018 /// Return true if the loop contains an instruction that inhibits 1019 /// the use of the hardware loop instruction. 1020 bool HexagonHardwareLoops::containsInvalidInstruction(MachineLoop *L, 1021 bool IsInnerHWLoop) const { 1022 LLVM_DEBUG(dbgs() << "\nhw_loop head, " 1023 << printMBBReference(**L->block_begin())); 1024 for (MachineBasicBlock *MBB : L->getBlocks()) { 1025 for (const MachineInstr &MI : *MBB) { 1026 if (isInvalidLoopOperation(&MI, IsInnerHWLoop)) { 1027 LLVM_DEBUG(dbgs() << "\nCannot convert to hw_loop due to:"; 1028 MI.dump();); 1029 return true; 1030 } 1031 } 1032 } 1033 return false; 1034 } 1035 1036 /// Returns true if the instruction is dead. This was essentially 1037 /// copied from DeadMachineInstructionElim::isDead, but with special cases 1038 /// for inline asm, physical registers and instructions with side effects 1039 /// removed. 1040 bool HexagonHardwareLoops::isDead(const MachineInstr *MI, 1041 SmallVectorImpl<MachineInstr *> &DeadPhis) const { 1042 // Examine each operand. 1043 for (const MachineOperand &MO : MI->operands()) { 1044 if (!MO.isReg() || !MO.isDef()) 1045 continue; 1046 1047 Register Reg = MO.getReg(); 1048 if (MRI->use_nodbg_empty(Reg)) 1049 continue; 1050 1051 using use_nodbg_iterator = MachineRegisterInfo::use_nodbg_iterator; 1052 1053 // This instruction has users, but if the only user is the phi node for the 1054 // parent block, and the only use of that phi node is this instruction, then 1055 // this instruction is dead: both it (and the phi node) can be removed. 1056 use_nodbg_iterator I = MRI->use_nodbg_begin(Reg); 1057 use_nodbg_iterator End = MRI->use_nodbg_end(); 1058 if (std::next(I) != End || !I->getParent()->isPHI()) 1059 return false; 1060 1061 MachineInstr *OnePhi = I->getParent(); 1062 for (const MachineOperand &OPO : OnePhi->operands()) { 1063 if (!OPO.isReg() || !OPO.isDef()) 1064 continue; 1065 1066 Register OPReg = OPO.getReg(); 1067 use_nodbg_iterator nextJ; 1068 for (use_nodbg_iterator J = MRI->use_nodbg_begin(OPReg); 1069 J != End; J = nextJ) { 1070 nextJ = std::next(J); 1071 MachineOperand &Use = *J; 1072 MachineInstr *UseMI = Use.getParent(); 1073 1074 // If the phi node has a user that is not MI, bail. 1075 if (MI != UseMI) 1076 return false; 1077 } 1078 } 1079 DeadPhis.push_back(OnePhi); 1080 } 1081 1082 // If there are no defs with uses, the instruction is dead. 1083 return true; 1084 } 1085 1086 void HexagonHardwareLoops::removeIfDead(MachineInstr *MI) { 1087 // This procedure was essentially copied from DeadMachineInstructionElim. 1088 1089 SmallVector<MachineInstr*, 1> DeadPhis; 1090 if (isDead(MI, DeadPhis)) { 1091 LLVM_DEBUG(dbgs() << "HW looping will remove: " << *MI); 1092 1093 // It is possible that some DBG_VALUE instructions refer to this 1094 // instruction. Examine each def operand for such references; 1095 // if found, mark the DBG_VALUE as undef (but don't delete it). 1096 for (const MachineOperand &MO : MI->operands()) { 1097 if (!MO.isReg() || !MO.isDef()) 1098 continue; 1099 Register Reg = MO.getReg(); 1100 // We use make_early_inc_range here because setReg below invalidates the 1101 // iterator. 1102 for (MachineOperand &MO : 1103 llvm::make_early_inc_range(MRI->use_operands(Reg))) { 1104 MachineInstr *UseMI = MO.getParent(); 1105 if (UseMI == MI) 1106 continue; 1107 if (MO.isDebug()) 1108 MO.setReg(0U); 1109 } 1110 } 1111 1112 MI->eraseFromParent(); 1113 for (unsigned i = 0; i < DeadPhis.size(); ++i) 1114 DeadPhis[i]->eraseFromParent(); 1115 } 1116 } 1117 1118 /// Check if the loop is a candidate for converting to a hardware 1119 /// loop. If so, then perform the transformation. 1120 /// 1121 /// This function works on innermost loops first. A loop can be converted 1122 /// if it is a counting loop; either a register value or an immediate. 1123 /// 1124 /// The code makes several assumptions about the representation of the loop 1125 /// in llvm. 1126 bool HexagonHardwareLoops::convertToHardwareLoop(MachineLoop *L, 1127 bool &RecL0used, 1128 bool &RecL1used) { 1129 // This is just to confirm basic correctness. 1130 assert(L->getHeader() && "Loop without a header?"); 1131 1132 bool Changed = false; 1133 bool L0Used = false; 1134 bool L1Used = false; 1135 1136 // Process nested loops first. 1137 for (MachineLoop *I : *L) { 1138 Changed |= convertToHardwareLoop(I, RecL0used, RecL1used); 1139 L0Used |= RecL0used; 1140 L1Used |= RecL1used; 1141 } 1142 1143 // If a nested loop has been converted, then we can't convert this loop. 1144 if (Changed && L0Used && L1Used) 1145 return Changed; 1146 1147 unsigned LOOP_i; 1148 unsigned LOOP_r; 1149 unsigned ENDLOOP; 1150 1151 // Flag used to track loopN instruction: 1152 // 1 - Hardware loop is being generated for the inner most loop. 1153 // 0 - Hardware loop is being generated for the outer loop. 1154 unsigned IsInnerHWLoop = 1; 1155 1156 if (L0Used) { 1157 LOOP_i = Hexagon::J2_loop1i; 1158 LOOP_r = Hexagon::J2_loop1r; 1159 ENDLOOP = Hexagon::ENDLOOP1; 1160 IsInnerHWLoop = 0; 1161 } else { 1162 LOOP_i = Hexagon::J2_loop0i; 1163 LOOP_r = Hexagon::J2_loop0r; 1164 ENDLOOP = Hexagon::ENDLOOP0; 1165 } 1166 1167 #ifndef NDEBUG 1168 // Stop trying after reaching the limit (if any). 1169 int Limit = HWLoopLimit; 1170 if (Limit >= 0) { 1171 if (Counter >= HWLoopLimit) 1172 return false; 1173 Counter++; 1174 } 1175 #endif 1176 1177 // Does the loop contain any invalid instructions? 1178 if (containsInvalidInstruction(L, IsInnerHWLoop)) 1179 return false; 1180 1181 MachineBasicBlock *LastMBB = L->findLoopControlBlock(); 1182 // Don't generate hw loop if the loop has more than one exit. 1183 if (!LastMBB) 1184 return false; 1185 1186 MachineBasicBlock::iterator LastI = LastMBB->getFirstTerminator(); 1187 if (LastI == LastMBB->end()) 1188 return false; 1189 1190 // Is the induction variable bump feeding the latch condition? 1191 if (!fixupInductionVariable(L)) 1192 return false; 1193 1194 // Ensure the loop has a preheader: the loop instruction will be 1195 // placed there. 1196 MachineBasicBlock *Preheader = MLI->findLoopPreheader(L, SpecPreheader); 1197 if (!Preheader) { 1198 Preheader = createPreheaderForLoop(L); 1199 if (!Preheader) 1200 return false; 1201 } 1202 1203 MachineBasicBlock::iterator InsertPos = Preheader->getFirstTerminator(); 1204 1205 SmallVector<MachineInstr*, 2> OldInsts; 1206 // Are we able to determine the trip count for the loop? 1207 CountValue *TripCount = getLoopTripCount(L, OldInsts); 1208 if (!TripCount) 1209 return false; 1210 1211 // Is the trip count available in the preheader? 1212 if (TripCount->isReg()) { 1213 // There will be a use of the register inserted into the preheader, 1214 // so make sure that the register is actually defined at that point. 1215 MachineInstr *TCDef = MRI->getVRegDef(TripCount->getReg()); 1216 MachineBasicBlock *BBDef = TCDef->getParent(); 1217 if (!MDT->dominates(BBDef, Preheader)) 1218 return false; 1219 } 1220 1221 // Determine the loop start. 1222 MachineBasicBlock *TopBlock = L->getTopBlock(); 1223 MachineBasicBlock *ExitingBlock = L->findLoopControlBlock(); 1224 MachineBasicBlock *LoopStart = nullptr; 1225 if (ExitingBlock != L->getLoopLatch()) { 1226 MachineBasicBlock *TB = nullptr, *FB = nullptr; 1227 SmallVector<MachineOperand, 2> Cond; 1228 1229 if (TII->analyzeBranch(*ExitingBlock, TB, FB, Cond, false)) 1230 return false; 1231 1232 if (L->contains(TB)) 1233 LoopStart = TB; 1234 else if (L->contains(FB)) 1235 LoopStart = FB; 1236 else 1237 return false; 1238 } 1239 else 1240 LoopStart = TopBlock; 1241 1242 // Convert the loop to a hardware loop. 1243 LLVM_DEBUG(dbgs() << "Change to hardware loop at "; L->dump()); 1244 DebugLoc DL; 1245 if (InsertPos != Preheader->end()) 1246 DL = InsertPos->getDebugLoc(); 1247 1248 if (TripCount->isReg()) { 1249 // Create a copy of the loop count register. 1250 Register CountReg = MRI->createVirtualRegister(&Hexagon::IntRegsRegClass); 1251 BuildMI(*Preheader, InsertPos, DL, TII->get(TargetOpcode::COPY), CountReg) 1252 .addReg(TripCount->getReg(), 0, TripCount->getSubReg()); 1253 // Add the Loop instruction to the beginning of the loop. 1254 BuildMI(*Preheader, InsertPos, DL, TII->get(LOOP_r)).addMBB(LoopStart) 1255 .addReg(CountReg); 1256 } else { 1257 assert(TripCount->isImm() && "Expecting immediate value for trip count"); 1258 // Add the Loop immediate instruction to the beginning of the loop, 1259 // if the immediate fits in the instructions. Otherwise, we need to 1260 // create a new virtual register. 1261 int64_t CountImm = TripCount->getImm(); 1262 if (!TII->isValidOffset(LOOP_i, CountImm, TRI)) { 1263 Register CountReg = MRI->createVirtualRegister(&Hexagon::IntRegsRegClass); 1264 BuildMI(*Preheader, InsertPos, DL, TII->get(Hexagon::A2_tfrsi), CountReg) 1265 .addImm(CountImm); 1266 BuildMI(*Preheader, InsertPos, DL, TII->get(LOOP_r)) 1267 .addMBB(LoopStart).addReg(CountReg); 1268 } else 1269 BuildMI(*Preheader, InsertPos, DL, TII->get(LOOP_i)) 1270 .addMBB(LoopStart).addImm(CountImm); 1271 } 1272 1273 // Make sure the loop start always has a reference in the CFG. 1274 LoopStart->setMachineBlockAddressTaken(); 1275 1276 // Replace the loop branch with an endloop instruction. 1277 DebugLoc LastIDL = LastI->getDebugLoc(); 1278 BuildMI(*LastMBB, LastI, LastIDL, TII->get(ENDLOOP)).addMBB(LoopStart); 1279 1280 // The loop ends with either: 1281 // - a conditional branch followed by an unconditional branch, or 1282 // - a conditional branch to the loop start. 1283 if (LastI->getOpcode() == Hexagon::J2_jumpt || 1284 LastI->getOpcode() == Hexagon::J2_jumpf) { 1285 // Delete one and change/add an uncond. branch to out of the loop. 1286 MachineBasicBlock *BranchTarget = LastI->getOperand(1).getMBB(); 1287 LastI = LastMBB->erase(LastI); 1288 if (!L->contains(BranchTarget)) { 1289 if (LastI != LastMBB->end()) 1290 LastI = LastMBB->erase(LastI); 1291 SmallVector<MachineOperand, 0> Cond; 1292 TII->insertBranch(*LastMBB, BranchTarget, nullptr, Cond, LastIDL); 1293 } 1294 } else { 1295 // Conditional branch to loop start; just delete it. 1296 LastMBB->erase(LastI); 1297 } 1298 delete TripCount; 1299 1300 // The induction operation and the comparison may now be 1301 // unneeded. If these are unneeded, then remove them. 1302 for (unsigned i = 0; i < OldInsts.size(); ++i) 1303 removeIfDead(OldInsts[i]); 1304 1305 ++NumHWLoops; 1306 1307 // Set RecL1used and RecL0used only after hardware loop has been 1308 // successfully generated. Doing it earlier can cause wrong loop instruction 1309 // to be used. 1310 if (L0Used) // Loop0 was already used. So, the correct loop must be loop1. 1311 RecL1used = true; 1312 else 1313 RecL0used = true; 1314 1315 return true; 1316 } 1317 1318 bool HexagonHardwareLoops::orderBumpCompare(MachineInstr *BumpI, 1319 MachineInstr *CmpI) { 1320 assert (BumpI != CmpI && "Bump and compare in the same instruction?"); 1321 1322 MachineBasicBlock *BB = BumpI->getParent(); 1323 if (CmpI->getParent() != BB) 1324 return false; 1325 1326 using instr_iterator = MachineBasicBlock::instr_iterator; 1327 1328 // Check if things are in order to begin with. 1329 for (instr_iterator I(BumpI), E = BB->instr_end(); I != E; ++I) 1330 if (&*I == CmpI) 1331 return true; 1332 1333 // Out of order. 1334 Register PredR = CmpI->getOperand(0).getReg(); 1335 bool FoundBump = false; 1336 instr_iterator CmpIt = CmpI->getIterator(), NextIt = std::next(CmpIt); 1337 for (instr_iterator I = NextIt, E = BB->instr_end(); I != E; ++I) { 1338 MachineInstr *In = &*I; 1339 for (unsigned i = 0, n = In->getNumOperands(); i < n; ++i) { 1340 MachineOperand &MO = In->getOperand(i); 1341 if (MO.isReg() && MO.isUse()) { 1342 if (MO.getReg() == PredR) // Found an intervening use of PredR. 1343 return false; 1344 } 1345 } 1346 1347 if (In == BumpI) { 1348 BB->splice(++BumpI->getIterator(), BB, CmpI->getIterator()); 1349 FoundBump = true; 1350 break; 1351 } 1352 } 1353 assert (FoundBump && "Cannot determine instruction order"); 1354 return FoundBump; 1355 } 1356 1357 /// This function is required to break recursion. Visiting phis in a loop may 1358 /// result in recursion during compilation. We break the recursion by making 1359 /// sure that we visit a MachineOperand and its definition in a 1360 /// MachineInstruction only once. If we attempt to visit more than once, then 1361 /// there is recursion, and will return false. 1362 bool HexagonHardwareLoops::isLoopFeeder(MachineLoop *L, MachineBasicBlock *A, 1363 MachineInstr *MI, 1364 const MachineOperand *MO, 1365 LoopFeederMap &LoopFeederPhi) const { 1366 if (LoopFeederPhi.find(MO->getReg()) == LoopFeederPhi.end()) { 1367 LLVM_DEBUG(dbgs() << "\nhw_loop head, " 1368 << printMBBReference(**L->block_begin())); 1369 // Ignore all BBs that form Loop. 1370 if (llvm::is_contained(L->getBlocks(), A)) 1371 return false; 1372 MachineInstr *Def = MRI->getVRegDef(MO->getReg()); 1373 LoopFeederPhi.insert(std::make_pair(MO->getReg(), Def)); 1374 return true; 1375 } else 1376 // Already visited node. 1377 return false; 1378 } 1379 1380 /// Return true if a Phi may generate a value that can underflow. 1381 /// This function calls loopCountMayWrapOrUnderFlow for each Phi operand. 1382 bool HexagonHardwareLoops::phiMayWrapOrUnderflow( 1383 MachineInstr *Phi, const MachineOperand *EndVal, MachineBasicBlock *MBB, 1384 MachineLoop *L, LoopFeederMap &LoopFeederPhi) const { 1385 assert(Phi->isPHI() && "Expecting a Phi."); 1386 // Walk through each Phi, and its used operands. Make sure that 1387 // if there is recursion in Phi, we won't generate hardware loops. 1388 for (int i = 1, n = Phi->getNumOperands(); i < n; i += 2) 1389 if (isLoopFeeder(L, MBB, Phi, &(Phi->getOperand(i)), LoopFeederPhi)) 1390 if (loopCountMayWrapOrUnderFlow(&(Phi->getOperand(i)), EndVal, 1391 Phi->getParent(), L, LoopFeederPhi)) 1392 return true; 1393 return false; 1394 } 1395 1396 /// Return true if the induction variable can underflow in the first iteration. 1397 /// An example, is an initial unsigned value that is 0 and is decrement in the 1398 /// first itertion of a do-while loop. In this case, we cannot generate a 1399 /// hardware loop because the endloop instruction does not decrement the loop 1400 /// counter if it is <= 1. We only need to perform this analysis if the 1401 /// initial value is a register. 1402 /// 1403 /// This function assumes the initial value may underfow unless proven 1404 /// otherwise. If the type is signed, then we don't care because signed 1405 /// underflow is undefined. We attempt to prove the initial value is not 1406 /// zero by perfoming a crude analysis of the loop counter. This function 1407 /// checks if the initial value is used in any comparison prior to the loop 1408 /// and, if so, assumes the comparison is a range check. This is inexact, 1409 /// but will catch the simple cases. 1410 bool HexagonHardwareLoops::loopCountMayWrapOrUnderFlow( 1411 const MachineOperand *InitVal, const MachineOperand *EndVal, 1412 MachineBasicBlock *MBB, MachineLoop *L, 1413 LoopFeederMap &LoopFeederPhi) const { 1414 // Only check register values since they are unknown. 1415 if (!InitVal->isReg()) 1416 return false; 1417 1418 if (!EndVal->isImm()) 1419 return false; 1420 1421 // A register value that is assigned an immediate is a known value, and it 1422 // won't underflow in the first iteration. 1423 int64_t Imm; 1424 if (checkForImmediate(*InitVal, Imm)) 1425 return (EndVal->getImm() == Imm); 1426 1427 Register Reg = InitVal->getReg(); 1428 1429 // We don't know the value of a physical register. 1430 if (!Reg.isVirtual()) 1431 return true; 1432 1433 MachineInstr *Def = MRI->getVRegDef(Reg); 1434 if (!Def) 1435 return true; 1436 1437 // If the initial value is a Phi or copy and the operands may not underflow, 1438 // then the definition cannot be underflow either. 1439 if (Def->isPHI() && !phiMayWrapOrUnderflow(Def, EndVal, Def->getParent(), 1440 L, LoopFeederPhi)) 1441 return false; 1442 if (Def->isCopy() && !loopCountMayWrapOrUnderFlow(&(Def->getOperand(1)), 1443 EndVal, Def->getParent(), 1444 L, LoopFeederPhi)) 1445 return false; 1446 1447 // Iterate over the uses of the initial value. If the initial value is used 1448 // in a compare, then we assume this is a range check that ensures the loop 1449 // doesn't underflow. This is not an exact test and should be improved. 1450 for (MachineRegisterInfo::use_instr_nodbg_iterator I = MRI->use_instr_nodbg_begin(Reg), 1451 E = MRI->use_instr_nodbg_end(); I != E; ++I) { 1452 MachineInstr *MI = &*I; 1453 Register CmpReg1, CmpReg2; 1454 int64_t CmpMask = 0, CmpValue = 0; 1455 1456 if (!TII->analyzeCompare(*MI, CmpReg1, CmpReg2, CmpMask, CmpValue)) 1457 continue; 1458 1459 MachineBasicBlock *TBB = nullptr, *FBB = nullptr; 1460 SmallVector<MachineOperand, 2> Cond; 1461 if (TII->analyzeBranch(*MI->getParent(), TBB, FBB, Cond, false)) 1462 continue; 1463 1464 Comparison::Kind Cmp = 1465 getComparisonKind(MI->getOpcode(), nullptr, nullptr, 0); 1466 if (Cmp == 0) 1467 continue; 1468 if (TII->predOpcodeHasNot(Cond) ^ (TBB != MBB)) 1469 Cmp = Comparison::getNegatedComparison(Cmp); 1470 if (CmpReg2 != 0 && CmpReg2 == Reg) 1471 Cmp = Comparison::getSwappedComparison(Cmp); 1472 1473 // Signed underflow is undefined. 1474 if (Comparison::isSigned(Cmp)) 1475 return false; 1476 1477 // Check if there is a comparison of the initial value. If the initial value 1478 // is greater than or not equal to another value, then assume this is a 1479 // range check. 1480 if ((Cmp & Comparison::G) || Cmp == Comparison::NE) 1481 return false; 1482 } 1483 1484 // OK - this is a hack that needs to be improved. We really need to analyze 1485 // the instructions performed on the initial value. This works on the simplest 1486 // cases only. 1487 if (!Def->isCopy() && !Def->isPHI()) 1488 return false; 1489 1490 return true; 1491 } 1492 1493 bool HexagonHardwareLoops::checkForImmediate(const MachineOperand &MO, 1494 int64_t &Val) const { 1495 if (MO.isImm()) { 1496 Val = MO.getImm(); 1497 return true; 1498 } 1499 if (!MO.isReg()) 1500 return false; 1501 1502 // MO is a register. Check whether it is defined as an immediate value, 1503 // and if so, get the value of it in TV. That value will then need to be 1504 // processed to handle potential subregisters in MO. 1505 int64_t TV; 1506 1507 Register R = MO.getReg(); 1508 if (!R.isVirtual()) 1509 return false; 1510 MachineInstr *DI = MRI->getVRegDef(R); 1511 unsigned DOpc = DI->getOpcode(); 1512 switch (DOpc) { 1513 case TargetOpcode::COPY: 1514 case Hexagon::A2_tfrsi: 1515 case Hexagon::A2_tfrpi: 1516 case Hexagon::CONST32: 1517 case Hexagon::CONST64: 1518 // Call recursively to avoid an extra check whether operand(1) is 1519 // indeed an immediate (it could be a global address, for example), 1520 // plus we can handle COPY at the same time. 1521 if (!checkForImmediate(DI->getOperand(1), TV)) 1522 return false; 1523 break; 1524 case Hexagon::A2_combineii: 1525 case Hexagon::A4_combineir: 1526 case Hexagon::A4_combineii: 1527 case Hexagon::A4_combineri: 1528 case Hexagon::A2_combinew: { 1529 const MachineOperand &S1 = DI->getOperand(1); 1530 const MachineOperand &S2 = DI->getOperand(2); 1531 int64_t V1, V2; 1532 if (!checkForImmediate(S1, V1) || !checkForImmediate(S2, V2)) 1533 return false; 1534 TV = V2 | (static_cast<uint64_t>(V1) << 32); 1535 break; 1536 } 1537 case TargetOpcode::REG_SEQUENCE: { 1538 const MachineOperand &S1 = DI->getOperand(1); 1539 const MachineOperand &S3 = DI->getOperand(3); 1540 int64_t V1, V3; 1541 if (!checkForImmediate(S1, V1) || !checkForImmediate(S3, V3)) 1542 return false; 1543 unsigned Sub2 = DI->getOperand(2).getImm(); 1544 unsigned Sub4 = DI->getOperand(4).getImm(); 1545 if (Sub2 == Hexagon::isub_lo && Sub4 == Hexagon::isub_hi) 1546 TV = V1 | (V3 << 32); 1547 else if (Sub2 == Hexagon::isub_hi && Sub4 == Hexagon::isub_lo) 1548 TV = V3 | (V1 << 32); 1549 else 1550 llvm_unreachable("Unexpected form of REG_SEQUENCE"); 1551 break; 1552 } 1553 1554 default: 1555 return false; 1556 } 1557 1558 // By now, we should have successfully obtained the immediate value defining 1559 // the register referenced in MO. Handle a potential use of a subregister. 1560 switch (MO.getSubReg()) { 1561 case Hexagon::isub_lo: 1562 Val = TV & 0xFFFFFFFFULL; 1563 break; 1564 case Hexagon::isub_hi: 1565 Val = (TV >> 32) & 0xFFFFFFFFULL; 1566 break; 1567 default: 1568 Val = TV; 1569 break; 1570 } 1571 return true; 1572 } 1573 1574 void HexagonHardwareLoops::setImmediate(MachineOperand &MO, int64_t Val) { 1575 if (MO.isImm()) { 1576 MO.setImm(Val); 1577 return; 1578 } 1579 1580 assert(MO.isReg()); 1581 Register R = MO.getReg(); 1582 MachineInstr *DI = MRI->getVRegDef(R); 1583 1584 const TargetRegisterClass *RC = MRI->getRegClass(R); 1585 Register NewR = MRI->createVirtualRegister(RC); 1586 MachineBasicBlock &B = *DI->getParent(); 1587 DebugLoc DL = DI->getDebugLoc(); 1588 BuildMI(B, DI, DL, TII->get(DI->getOpcode()), NewR).addImm(Val); 1589 MO.setReg(NewR); 1590 } 1591 1592 bool HexagonHardwareLoops::fixupInductionVariable(MachineLoop *L) { 1593 MachineBasicBlock *Header = L->getHeader(); 1594 MachineBasicBlock *Latch = L->getLoopLatch(); 1595 MachineBasicBlock *ExitingBlock = L->findLoopControlBlock(); 1596 1597 if (!(Header && Latch && ExitingBlock)) 1598 return false; 1599 1600 // These data structures follow the same concept as the corresponding 1601 // ones in findInductionRegister (where some comments are). 1602 using RegisterBump = std::pair<Register, int64_t>; 1603 using RegisterInduction = std::pair<Register, RegisterBump>; 1604 using RegisterInductionSet = std::set<RegisterInduction>; 1605 1606 // Register candidates for induction variables, with their associated bumps. 1607 RegisterInductionSet IndRegs; 1608 1609 // Look for induction patterns: 1610 // %1 = PHI ..., [ latch, %2 ] 1611 // %2 = ADD %1, imm 1612 using instr_iterator = MachineBasicBlock::instr_iterator; 1613 1614 for (instr_iterator I = Header->instr_begin(), E = Header->instr_end(); 1615 I != E && I->isPHI(); ++I) { 1616 MachineInstr *Phi = &*I; 1617 1618 // Have a PHI instruction. 1619 for (unsigned i = 1, n = Phi->getNumOperands(); i < n; i += 2) { 1620 if (Phi->getOperand(i+1).getMBB() != Latch) 1621 continue; 1622 1623 Register PhiReg = Phi->getOperand(i).getReg(); 1624 MachineInstr *DI = MRI->getVRegDef(PhiReg); 1625 1626 if (DI->getDesc().isAdd()) { 1627 // If the register operand to the add/sub is the PHI we are looking 1628 // at, this meets the induction pattern. 1629 Register IndReg = DI->getOperand(1).getReg(); 1630 MachineOperand &Opnd2 = DI->getOperand(2); 1631 int64_t V; 1632 if (MRI->getVRegDef(IndReg) == Phi && checkForImmediate(Opnd2, V)) { 1633 Register UpdReg = DI->getOperand(0).getReg(); 1634 IndRegs.insert(std::make_pair(UpdReg, std::make_pair(IndReg, V))); 1635 } 1636 } 1637 } // for (i) 1638 } // for (instr) 1639 1640 if (IndRegs.empty()) 1641 return false; 1642 1643 MachineBasicBlock *TB = nullptr, *FB = nullptr; 1644 SmallVector<MachineOperand,2> Cond; 1645 // analyzeBranch returns true if it fails to analyze branch. 1646 bool NotAnalyzed = TII->analyzeBranch(*ExitingBlock, TB, FB, Cond, false); 1647 if (NotAnalyzed || Cond.empty()) 1648 return false; 1649 1650 if (ExitingBlock != Latch && (TB == Latch || FB == Latch)) { 1651 MachineBasicBlock *LTB = nullptr, *LFB = nullptr; 1652 SmallVector<MachineOperand,2> LCond; 1653 bool NotAnalyzed = TII->analyzeBranch(*Latch, LTB, LFB, LCond, false); 1654 if (NotAnalyzed) 1655 return false; 1656 1657 // Since latch is not the exiting block, the latch branch should be an 1658 // unconditional branch to the loop header. 1659 if (TB == Latch) 1660 TB = (LTB == Header) ? LTB : LFB; 1661 else 1662 FB = (LTB == Header) ? LTB : LFB; 1663 } 1664 if (TB != Header) { 1665 if (FB != Header) { 1666 // The latch/exit block does not go back to the header. 1667 return false; 1668 } 1669 // FB is the header (i.e., uncond. jump to branch header) 1670 // In this case, the LoopBody -> TB should not be a back edge otherwise 1671 // it could result in an infinite loop after conversion to hw_loop. 1672 // This case can happen when the Latch has two jumps like this: 1673 // Jmp_c OuterLoopHeader <-- TB 1674 // Jmp InnerLoopHeader <-- FB 1675 if (MDT->dominates(TB, FB)) 1676 return false; 1677 } 1678 1679 // Expecting a predicate register as a condition. It won't be a hardware 1680 // predicate register at this point yet, just a vreg. 1681 // HexagonInstrInfo::analyzeBranch for negated branches inserts imm(0) 1682 // into Cond, followed by the predicate register. For non-negated branches 1683 // it's just the register. 1684 unsigned CSz = Cond.size(); 1685 if (CSz != 1 && CSz != 2) 1686 return false; 1687 1688 if (!Cond[CSz-1].isReg()) 1689 return false; 1690 1691 Register P = Cond[CSz - 1].getReg(); 1692 MachineInstr *PredDef = MRI->getVRegDef(P); 1693 1694 if (!PredDef->isCompare()) 1695 return false; 1696 1697 SmallSet<Register,2> CmpRegs; 1698 MachineOperand *CmpImmOp = nullptr; 1699 1700 // Go over all operands to the compare and look for immediate and register 1701 // operands. Assume that if the compare has a single register use and a 1702 // single immediate operand, then the register is being compared with the 1703 // immediate value. 1704 for (MachineOperand &MO : PredDef->operands()) { 1705 if (MO.isReg()) { 1706 // Skip all implicit references. In one case there was: 1707 // %140 = FCMPUGT32_rr %138, %139, implicit %usr 1708 if (MO.isImplicit()) 1709 continue; 1710 if (MO.isUse()) { 1711 if (!isImmediate(MO)) { 1712 CmpRegs.insert(MO.getReg()); 1713 continue; 1714 } 1715 // Consider the register to be the "immediate" operand. 1716 if (CmpImmOp) 1717 return false; 1718 CmpImmOp = &MO; 1719 } 1720 } else if (MO.isImm()) { 1721 if (CmpImmOp) // A second immediate argument? Confusing. Bail out. 1722 return false; 1723 CmpImmOp = &MO; 1724 } 1725 } 1726 1727 if (CmpRegs.empty()) 1728 return false; 1729 1730 // Check if the compared register follows the order we want. Fix if needed. 1731 for (RegisterInductionSet::iterator I = IndRegs.begin(), E = IndRegs.end(); 1732 I != E; ++I) { 1733 // This is a success. If the register used in the comparison is one that 1734 // we have identified as a bumped (updated) induction register, there is 1735 // nothing to do. 1736 if (CmpRegs.count(I->first)) 1737 return true; 1738 1739 // Otherwise, if the register being compared comes out of a PHI node, 1740 // and has been recognized as following the induction pattern, and is 1741 // compared against an immediate, we can fix it. 1742 const RegisterBump &RB = I->second; 1743 if (CmpRegs.count(RB.first)) { 1744 if (!CmpImmOp) { 1745 // If both operands to the compare instruction are registers, see if 1746 // it can be changed to use induction register as one of the operands. 1747 MachineInstr *IndI = nullptr; 1748 MachineInstr *nonIndI = nullptr; 1749 MachineOperand *IndMO = nullptr; 1750 MachineOperand *nonIndMO = nullptr; 1751 1752 for (unsigned i = 1, n = PredDef->getNumOperands(); i < n; ++i) { 1753 MachineOperand &MO = PredDef->getOperand(i); 1754 if (MO.isReg() && MO.getReg() == RB.first) { 1755 LLVM_DEBUG(dbgs() << "\n DefMI(" << i 1756 << ") = " << *(MRI->getVRegDef(I->first))); 1757 if (IndI) 1758 return false; 1759 1760 IndI = MRI->getVRegDef(I->first); 1761 IndMO = &MO; 1762 } else if (MO.isReg()) { 1763 LLVM_DEBUG(dbgs() << "\n DefMI(" << i 1764 << ") = " << *(MRI->getVRegDef(MO.getReg()))); 1765 if (nonIndI) 1766 return false; 1767 1768 nonIndI = MRI->getVRegDef(MO.getReg()); 1769 nonIndMO = &MO; 1770 } 1771 } 1772 if (IndI && nonIndI && 1773 nonIndI->getOpcode() == Hexagon::A2_addi && 1774 nonIndI->getOperand(2).isImm() && 1775 nonIndI->getOperand(2).getImm() == - RB.second) { 1776 bool Order = orderBumpCompare(IndI, PredDef); 1777 if (Order) { 1778 IndMO->setReg(I->first); 1779 nonIndMO->setReg(nonIndI->getOperand(1).getReg()); 1780 return true; 1781 } 1782 } 1783 return false; 1784 } 1785 1786 // It is not valid to do this transformation on an unsigned comparison 1787 // because it may underflow. 1788 Comparison::Kind Cmp = 1789 getComparisonKind(PredDef->getOpcode(), nullptr, nullptr, 0); 1790 if (!Cmp || Comparison::isUnsigned(Cmp)) 1791 return false; 1792 1793 // If the register is being compared against an immediate, try changing 1794 // the compare instruction to use induction register and adjust the 1795 // immediate operand. 1796 int64_t CmpImm = getImmediate(*CmpImmOp); 1797 int64_t V = RB.second; 1798 // Handle Overflow (64-bit). 1799 if (((V > 0) && (CmpImm > INT64_MAX - V)) || 1800 ((V < 0) && (CmpImm < INT64_MIN - V))) 1801 return false; 1802 CmpImm += V; 1803 // Most comparisons of register against an immediate value allow 1804 // the immediate to be constant-extended. There are some exceptions 1805 // though. Make sure the new combination will work. 1806 if (CmpImmOp->isImm() && !TII->isExtendable(*PredDef) && 1807 !TII->isValidOffset(PredDef->getOpcode(), CmpImm, TRI, false)) 1808 return false; 1809 1810 // Make sure that the compare happens after the bump. Otherwise, 1811 // after the fixup, the compare would use a yet-undefined register. 1812 MachineInstr *BumpI = MRI->getVRegDef(I->first); 1813 bool Order = orderBumpCompare(BumpI, PredDef); 1814 if (!Order) 1815 return false; 1816 1817 // Finally, fix the compare instruction. 1818 setImmediate(*CmpImmOp, CmpImm); 1819 for (MachineOperand &MO : PredDef->operands()) { 1820 if (MO.isReg() && MO.getReg() == RB.first) { 1821 MO.setReg(I->first); 1822 return true; 1823 } 1824 } 1825 } 1826 } 1827 1828 return false; 1829 } 1830 1831 /// createPreheaderForLoop - Create a preheader for a given loop. 1832 MachineBasicBlock *HexagonHardwareLoops::createPreheaderForLoop( 1833 MachineLoop *L) { 1834 if (MachineBasicBlock *TmpPH = MLI->findLoopPreheader(L, SpecPreheader)) 1835 return TmpPH; 1836 if (!HWCreatePreheader) 1837 return nullptr; 1838 1839 MachineBasicBlock *Header = L->getHeader(); 1840 MachineBasicBlock *Latch = L->getLoopLatch(); 1841 MachineBasicBlock *ExitingBlock = L->findLoopControlBlock(); 1842 MachineFunction *MF = Header->getParent(); 1843 DebugLoc DL; 1844 1845 #ifndef NDEBUG 1846 if ((!PHFn.empty()) && (PHFn != MF->getName())) 1847 return nullptr; 1848 #endif 1849 1850 if (!Latch || !ExitingBlock || Header->hasAddressTaken()) 1851 return nullptr; 1852 1853 using instr_iterator = MachineBasicBlock::instr_iterator; 1854 1855 // Verify that all existing predecessors have analyzable branches 1856 // (or no branches at all). 1857 using MBBVector = std::vector<MachineBasicBlock *>; 1858 1859 MBBVector Preds(Header->pred_begin(), Header->pred_end()); 1860 SmallVector<MachineOperand,2> Tmp1; 1861 MachineBasicBlock *TB = nullptr, *FB = nullptr; 1862 1863 if (TII->analyzeBranch(*ExitingBlock, TB, FB, Tmp1, false)) 1864 return nullptr; 1865 1866 for (MachineBasicBlock *PB : Preds) { 1867 bool NotAnalyzed = TII->analyzeBranch(*PB, TB, FB, Tmp1, false); 1868 if (NotAnalyzed) 1869 return nullptr; 1870 } 1871 1872 MachineBasicBlock *NewPH = MF->CreateMachineBasicBlock(); 1873 MF->insert(Header->getIterator(), NewPH); 1874 1875 if (Header->pred_size() > 2) { 1876 // Ensure that the header has only two predecessors: the preheader and 1877 // the loop latch. Any additional predecessors of the header should 1878 // join at the newly created preheader. Inspect all PHI nodes from the 1879 // header and create appropriate corresponding PHI nodes in the preheader. 1880 1881 for (instr_iterator I = Header->instr_begin(), E = Header->instr_end(); 1882 I != E && I->isPHI(); ++I) { 1883 MachineInstr *PN = &*I; 1884 1885 const MCInstrDesc &PD = TII->get(TargetOpcode::PHI); 1886 MachineInstr *NewPN = MF->CreateMachineInstr(PD, DL); 1887 NewPH->insert(NewPH->end(), NewPN); 1888 1889 Register PR = PN->getOperand(0).getReg(); 1890 const TargetRegisterClass *RC = MRI->getRegClass(PR); 1891 Register NewPR = MRI->createVirtualRegister(RC); 1892 NewPN->addOperand(MachineOperand::CreateReg(NewPR, true)); 1893 1894 // Copy all non-latch operands of a header's PHI node to the newly 1895 // created PHI node in the preheader. 1896 for (unsigned i = 1, n = PN->getNumOperands(); i < n; i += 2) { 1897 Register PredR = PN->getOperand(i).getReg(); 1898 unsigned PredRSub = PN->getOperand(i).getSubReg(); 1899 MachineBasicBlock *PredB = PN->getOperand(i+1).getMBB(); 1900 if (PredB == Latch) 1901 continue; 1902 1903 MachineOperand MO = MachineOperand::CreateReg(PredR, false); 1904 MO.setSubReg(PredRSub); 1905 NewPN->addOperand(MO); 1906 NewPN->addOperand(MachineOperand::CreateMBB(PredB)); 1907 } 1908 1909 // Remove copied operands from the old PHI node and add the value 1910 // coming from the preheader's PHI. 1911 for (int i = PN->getNumOperands()-2; i > 0; i -= 2) { 1912 MachineBasicBlock *PredB = PN->getOperand(i+1).getMBB(); 1913 if (PredB != Latch) { 1914 PN->removeOperand(i+1); 1915 PN->removeOperand(i); 1916 } 1917 } 1918 PN->addOperand(MachineOperand::CreateReg(NewPR, false)); 1919 PN->addOperand(MachineOperand::CreateMBB(NewPH)); 1920 } 1921 } else { 1922 assert(Header->pred_size() == 2); 1923 1924 // The header has only two predecessors, but the non-latch predecessor 1925 // is not a preheader (e.g. it has other successors, etc.) 1926 // In such a case we don't need any extra PHI nodes in the new preheader, 1927 // all we need is to adjust existing PHIs in the header to now refer to 1928 // the new preheader. 1929 for (instr_iterator I = Header->instr_begin(), E = Header->instr_end(); 1930 I != E && I->isPHI(); ++I) { 1931 MachineInstr *PN = &*I; 1932 for (unsigned i = 1, n = PN->getNumOperands(); i < n; i += 2) { 1933 MachineOperand &MO = PN->getOperand(i+1); 1934 if (MO.getMBB() != Latch) 1935 MO.setMBB(NewPH); 1936 } 1937 } 1938 } 1939 1940 // "Reroute" the CFG edges to link in the new preheader. 1941 // If any of the predecessors falls through to the header, insert a branch 1942 // to the new preheader in that place. 1943 SmallVector<MachineOperand,1> Tmp2; 1944 SmallVector<MachineOperand,1> EmptyCond; 1945 1946 TB = FB = nullptr; 1947 1948 for (MachineBasicBlock *PB : Preds) { 1949 if (PB != Latch) { 1950 Tmp2.clear(); 1951 bool NotAnalyzed = TII->analyzeBranch(*PB, TB, FB, Tmp2, false); 1952 (void)NotAnalyzed; // suppress compiler warning 1953 assert (!NotAnalyzed && "Should be analyzable!"); 1954 if (TB != Header && (Tmp2.empty() || FB != Header)) 1955 TII->insertBranch(*PB, NewPH, nullptr, EmptyCond, DL); 1956 PB->ReplaceUsesOfBlockWith(Header, NewPH); 1957 } 1958 } 1959 1960 // It can happen that the latch block will fall through into the header. 1961 // Insert an unconditional branch to the header. 1962 TB = FB = nullptr; 1963 bool LatchNotAnalyzed = TII->analyzeBranch(*Latch, TB, FB, Tmp2, false); 1964 (void)LatchNotAnalyzed; // suppress compiler warning 1965 assert (!LatchNotAnalyzed && "Should be analyzable!"); 1966 if (!TB && !FB) 1967 TII->insertBranch(*Latch, Header, nullptr, EmptyCond, DL); 1968 1969 // Finally, the branch from the preheader to the header. 1970 TII->insertBranch(*NewPH, Header, nullptr, EmptyCond, DL); 1971 NewPH->addSuccessor(Header); 1972 1973 MachineLoop *ParentLoop = L->getParentLoop(); 1974 if (ParentLoop) 1975 ParentLoop->addBasicBlockToLoop(NewPH, MLI->getBase()); 1976 1977 // Update the dominator information with the new preheader. 1978 if (MDT) { 1979 if (MachineDomTreeNode *HN = MDT->getNode(Header)) { 1980 if (MachineDomTreeNode *DHN = HN->getIDom()) { 1981 MDT->addNewBlock(NewPH, DHN->getBlock()); 1982 MDT->changeImmediateDominator(Header, NewPH); 1983 } 1984 } 1985 } 1986 1987 return NewPH; 1988 } 1989