1//===-- HexagonHardwareLoops.cpp - Identify and generate hardware loops ---===// 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 pass identifies loops where we can generate the Hexagon hardware 11// loop instruction. The hardware loop can perform loop branches with a 12// zero-cycle overhead. 13// 14// The pattern that defines the induction variable can changed depending on 15// prior optimizations. For example, the IndVarSimplify phase run by 'opt' 16// normalizes induction variables, and the Loop Strength Reduction pass 17// run by 'llc' may also make changes to the induction variable. 18// The pattern detected by this phase is due to running Strength Reduction. 19// 20// Criteria for hardware loops: 21// - Countable loops (w/ ind. var for a trip count) 22// - Assumes loops are normalized by IndVarSimplify 23// - Try inner-most loops first 24// - No function calls in loops. 25// 26//===----------------------------------------------------------------------===// 27 28#include "llvm/ADT/SmallSet.h" 29#include "Hexagon.h" 30#include "HexagonSubtarget.h" 31#include "llvm/ADT/Statistic.h" 32#include "llvm/CodeGen/MachineDominators.h" 33#include "llvm/CodeGen/MachineFunction.h" 34#include "llvm/CodeGen/MachineFunctionPass.h" 35#include "llvm/CodeGen/MachineInstrBuilder.h" 36#include "llvm/CodeGen/MachineLoopInfo.h" 37#include "llvm/CodeGen/MachineRegisterInfo.h" 38#include "llvm/PassSupport.h" 39#include "llvm/Support/CommandLine.h" 40#include "llvm/Support/Debug.h" 41#include "llvm/Support/raw_ostream.h" 42#include "llvm/Target/TargetInstrInfo.h" 43#include <algorithm> 44#include <vector> 45 46using namespace llvm; 47 48#define DEBUG_TYPE "hwloops" 49 50#ifndef NDEBUG 51static cl::opt<int> HWLoopLimit("hexagon-max-hwloop", cl::Hidden, cl::init(-1)); 52 53// Option to create preheader only for a specific function. 54static cl::opt<std::string> PHFn("hexagon-hwloop-phfn", cl::Hidden, 55 cl::init("")); 56#endif 57 58// Option to create a preheader if one doesn't exist. 59static cl::opt<bool> HWCreatePreheader("hexagon-hwloop-preheader", 60 cl::Hidden, cl::init(true), 61 cl::desc("Add a preheader to a hardware loop if one doesn't exist")); 62 63STATISTIC(NumHWLoops, "Number of loops converted to hardware loops"); 64 65namespace llvm { 66 FunctionPass *createHexagonHardwareLoops(); 67 void initializeHexagonHardwareLoopsPass(PassRegistry&); 68} 69 70namespace { 71 class CountValue; 72 struct HexagonHardwareLoops : public MachineFunctionPass { 73 MachineLoopInfo *MLI; 74 MachineRegisterInfo *MRI; 75 MachineDominatorTree *MDT; 76 const HexagonInstrInfo *TII; 77#ifndef NDEBUG 78 static int Counter; 79#endif 80 81 public: 82 static char ID; 83 84 HexagonHardwareLoops() : MachineFunctionPass(ID) { 85 initializeHexagonHardwareLoopsPass(*PassRegistry::getPassRegistry()); 86 } 87 88 bool runOnMachineFunction(MachineFunction &MF) override; 89 90 const char *getPassName() const override { return "Hexagon Hardware Loops"; } 91 92 void getAnalysisUsage(AnalysisUsage &AU) const override { 93 AU.addRequired<MachineDominatorTree>(); 94 AU.addRequired<MachineLoopInfo>(); 95 MachineFunctionPass::getAnalysisUsage(AU); 96 } 97 98 private: 99 typedef std::map<unsigned, MachineInstr *> LoopFeederMap; 100 101 /// Kinds of comparisons in the compare instructions. 102 struct Comparison { 103 enum Kind { 104 EQ = 0x01, 105 NE = 0x02, 106 L = 0x04, 107 G = 0x08, 108 U = 0x40, 109 LTs = L, 110 LEs = L | EQ, 111 GTs = G, 112 GEs = G | EQ, 113 LTu = L | U, 114 LEu = L | EQ | U, 115 GTu = G | U, 116 GEu = G | EQ | U 117 }; 118 119 static Kind getSwappedComparison(Kind Cmp) { 120 assert ((!((Cmp & L) && (Cmp & G))) && "Malformed comparison operator"); 121 if ((Cmp & L) || (Cmp & G)) 122 return (Kind)(Cmp ^ (L|G)); 123 return Cmp; 124 } 125 126 static Kind getNegatedComparison(Kind Cmp) { 127 if ((Cmp & L) || (Cmp & G)) 128 return (Kind)((Cmp ^ (L | G)) ^ EQ); 129 if ((Cmp & NE) || (Cmp & EQ)) 130 return (Kind)(Cmp ^ (EQ | NE)); 131 return (Kind)0; 132 } 133 134 static bool isSigned(Kind Cmp) { 135 return (Cmp & (L | G) && !(Cmp & U)); 136 } 137 138 static bool isUnsigned(Kind Cmp) { 139 return (Cmp & U); 140 } 141 142 }; 143 144 /// \brief Find the register that contains the loop controlling 145 /// induction variable. 146 /// If successful, it will return true and set the \p Reg, \p IVBump 147 /// and \p IVOp arguments. Otherwise it will return false. 148 /// The returned induction register is the register R that follows the 149 /// following induction pattern: 150 /// loop: 151 /// R = phi ..., [ R.next, LatchBlock ] 152 /// R.next = R + #bump 153 /// if (R.next < #N) goto loop 154 /// IVBump is the immediate value added to R, and IVOp is the instruction 155 /// "R.next = R + #bump". 156 bool findInductionRegister(MachineLoop *L, unsigned &Reg, 157 int64_t &IVBump, MachineInstr *&IVOp) const; 158 159 /// \brief Return the comparison kind for the specified opcode. 160 Comparison::Kind getComparisonKind(unsigned CondOpc, 161 MachineOperand *InitialValue, 162 const MachineOperand *Endvalue, 163 int64_t IVBump) const; 164 165 /// \brief Analyze the statements in a loop to determine if the loop 166 /// has a computable trip count and, if so, return a value that represents 167 /// the trip count expression. 168 CountValue *getLoopTripCount(MachineLoop *L, 169 SmallVectorImpl<MachineInstr *> &OldInsts); 170 171 /// \brief Return the expression that represents the number of times 172 /// a loop iterates. The function takes the operands that represent the 173 /// loop start value, loop end value, and induction value. Based upon 174 /// these operands, the function attempts to compute the trip count. 175 /// If the trip count is not directly available (as an immediate value, 176 /// or a register), the function will attempt to insert computation of it 177 /// to the loop's preheader. 178 CountValue *computeCount(MachineLoop *Loop, const MachineOperand *Start, 179 const MachineOperand *End, unsigned IVReg, 180 int64_t IVBump, Comparison::Kind Cmp) const; 181 182 /// \brief Return true if the instruction is not valid within a hardware 183 /// loop. 184 bool isInvalidLoopOperation(const MachineInstr *MI, 185 bool IsInnerHWLoop) const; 186 187 /// \brief Return true if the loop contains an instruction that inhibits 188 /// using the hardware loop. 189 bool containsInvalidInstruction(MachineLoop *L, bool IsInnerHWLoop) const; 190 191 /// \brief Given a loop, check if we can convert it to a hardware loop. 192 /// If so, then perform the conversion and return true. 193 bool convertToHardwareLoop(MachineLoop *L, bool &L0used, bool &L1used); 194 195 /// \brief Return true if the instruction is now dead. 196 bool isDead(const MachineInstr *MI, 197 SmallVectorImpl<MachineInstr *> &DeadPhis) const; 198 199 /// \brief Remove the instruction if it is now dead. 200 void removeIfDead(MachineInstr *MI); 201 202 /// \brief Make sure that the "bump" instruction executes before the 203 /// compare. We need that for the IV fixup, so that the compare 204 /// instruction would not use a bumped value that has not yet been 205 /// defined. If the instructions are out of order, try to reorder them. 206 bool orderBumpCompare(MachineInstr *BumpI, MachineInstr *CmpI); 207 208 /// \brief Return true if MO and MI pair is visited only once. If visited 209 /// more than once, this indicates there is recursion. In such a case, 210 /// return false. 211 bool isLoopFeeder(MachineLoop *L, MachineBasicBlock *A, MachineInstr *MI, 212 const MachineOperand *MO, 213 LoopFeederMap &LoopFeederPhi) const; 214 215 /// \brief Return true if the Phi may generate a value that may underflow, 216 /// or may wrap. 217 bool phiMayWrapOrUnderflow(MachineInstr *Phi, const MachineOperand *EndVal, 218 MachineBasicBlock *MBB, MachineLoop *L, 219 LoopFeederMap &LoopFeederPhi) const; 220 221 /// \brief Return true if the induction variable may underflow an unsigned 222 /// value in the first iteration. 223 bool loopCountMayWrapOrUnderFlow(const MachineOperand *InitVal, 224 const MachineOperand *EndVal, 225 MachineBasicBlock *MBB, MachineLoop *L, 226 LoopFeederMap &LoopFeederPhi) const; 227 228 /// \brief Check if the given operand has a compile-time known constant 229 /// value. Return true if yes, and false otherwise. When returning true, set 230 /// Val to the corresponding constant value. 231 bool checkForImmediate(const MachineOperand &MO, int64_t &Val) const; 232 233 /// \brief Check if the operand has a compile-time known constant value. 234 bool isImmediate(const MachineOperand &MO) const { 235 int64_t V; 236 return checkForImmediate(MO, V); 237 } 238 239 /// \brief Return the immediate for the specified operand. 240 int64_t getImmediate(const MachineOperand &MO) const { 241 int64_t V; 242 if (!checkForImmediate(MO, V)) 243 llvm_unreachable("Invalid operand"); 244 return V; 245 } 246 247 /// \brief Reset the given machine operand to now refer to a new immediate 248 /// value. Assumes that the operand was already referencing an immediate 249 /// value, either directly, or via a register. 250 void setImmediate(MachineOperand &MO, int64_t Val); 251 252 /// \brief Fix the data flow of the induction varible. 253 /// The desired flow is: phi ---> bump -+-> comparison-in-latch. 254 /// | 255 /// +-> back to phi 256 /// where "bump" is the increment of the induction variable: 257 /// iv = iv + #const. 258 /// Due to some prior code transformations, the actual flow may look 259 /// like this: 260 /// phi -+-> bump ---> back to phi 261 /// | 262 /// +-> comparison-in-latch (against upper_bound-bump), 263 /// i.e. the comparison that controls the loop execution may be using 264 /// the value of the induction variable from before the increment. 265 /// 266 /// Return true if the loop's flow is the desired one (i.e. it's 267 /// either been fixed, or no fixing was necessary). 268 /// Otherwise, return false. This can happen if the induction variable 269 /// couldn't be identified, or if the value in the latch's comparison 270 /// cannot be adjusted to reflect the post-bump value. 271 bool fixupInductionVariable(MachineLoop *L); 272 273 /// \brief Given a loop, if it does not have a preheader, create one. 274 /// Return the block that is the preheader. 275 MachineBasicBlock *createPreheaderForLoop(MachineLoop *L); 276 }; 277 278 char HexagonHardwareLoops::ID = 0; 279#ifndef NDEBUG 280 int HexagonHardwareLoops::Counter = 0; 281#endif 282 283 /// \brief Abstraction for a trip count of a loop. A smaller version 284 /// of the MachineOperand class without the concerns of changing the 285 /// operand representation. 286 class CountValue { 287 public: 288 enum CountValueType { 289 CV_Register, 290 CV_Immediate 291 }; 292 private: 293 CountValueType Kind; 294 union Values { 295 struct { 296 unsigned Reg; 297 unsigned Sub; 298 } R; 299 unsigned ImmVal; 300 } Contents; 301 302 public: 303 explicit CountValue(CountValueType t, unsigned v, unsigned u = 0) { 304 Kind = t; 305 if (Kind == CV_Register) { 306 Contents.R.Reg = v; 307 Contents.R.Sub = u; 308 } else { 309 Contents.ImmVal = v; 310 } 311 } 312 bool isReg() const { return Kind == CV_Register; } 313 bool isImm() const { return Kind == CV_Immediate; } 314 315 unsigned getReg() const { 316 assert(isReg() && "Wrong CountValue accessor"); 317 return Contents.R.Reg; 318 } 319 unsigned getSubReg() const { 320 assert(isReg() && "Wrong CountValue accessor"); 321 return Contents.R.Sub; 322 } 323 unsigned getImm() const { 324 assert(isImm() && "Wrong CountValue accessor"); 325 return Contents.ImmVal; 326 } 327 328 void print(raw_ostream &OS, const TargetRegisterInfo *TRI = nullptr) const { 329 if (isReg()) { OS << PrintReg(Contents.R.Reg, TRI, Contents.R.Sub); } 330 if (isImm()) { OS << Contents.ImmVal; } 331 } 332 }; 333} // end anonymous namespace 334 335 336INITIALIZE_PASS_BEGIN(HexagonHardwareLoops, "hwloops", 337 "Hexagon Hardware Loops", false, false) 338INITIALIZE_PASS_DEPENDENCY(MachineDominatorTree) 339INITIALIZE_PASS_DEPENDENCY(MachineLoopInfo) 340INITIALIZE_PASS_END(HexagonHardwareLoops, "hwloops", 341 "Hexagon Hardware Loops", false, false) 342 343FunctionPass *llvm::createHexagonHardwareLoops() { 344 return new HexagonHardwareLoops(); 345} 346 347bool HexagonHardwareLoops::runOnMachineFunction(MachineFunction &MF) { 348 DEBUG(dbgs() << "********* Hexagon Hardware Loops *********\n"); 349 350 bool Changed = false; 351 352 MLI = &getAnalysis<MachineLoopInfo>(); 353 MRI = &MF.getRegInfo(); 354 MDT = &getAnalysis<MachineDominatorTree>(); 355 TII = MF.getSubtarget<HexagonSubtarget>().getInstrInfo(); 356 357 for (auto &L : *MLI) 358 if (!L->getParentLoop()) { 359 bool L0Used = false; 360 bool L1Used = false; 361 Changed |= convertToHardwareLoop(L, L0Used, L1Used); 362 } 363 364 return Changed; 365} 366 367/// \brief Return the latch block if it's one of the exiting blocks. Otherwise, 368/// return the exiting block. Return 'null' when multiple exiting blocks are 369/// present. 370static MachineBasicBlock* getExitingBlock(MachineLoop *L) { 371 if (MachineBasicBlock *Latch = L->getLoopLatch()) { 372 if (L->isLoopExiting(Latch)) 373 return Latch; 374 else 375 return L->getExitingBlock(); 376 } 377 return nullptr; 378} 379 380bool HexagonHardwareLoops::findInductionRegister(MachineLoop *L, 381 unsigned &Reg, 382 int64_t &IVBump, 383 MachineInstr *&IVOp 384 ) const { 385 MachineBasicBlock *Header = L->getHeader(); 386 MachineBasicBlock *Preheader = L->getLoopPreheader(); 387 MachineBasicBlock *Latch = L->getLoopLatch(); 388 MachineBasicBlock *ExitingBlock = getExitingBlock(L); 389 if (!Header || !Preheader || !Latch || !ExitingBlock) 390 return false; 391 392 // This pair represents an induction register together with an immediate 393 // value that will be added to it in each loop iteration. 394 typedef std::pair<unsigned,int64_t> RegisterBump; 395 396 // Mapping: R.next -> (R, bump), where R, R.next and bump are derived 397 // from an induction operation 398 // R.next = R + bump 399 // where bump is an immediate value. 400 typedef std::map<unsigned,RegisterBump> InductionMap; 401 402 InductionMap IndMap; 403 404 typedef MachineBasicBlock::instr_iterator instr_iterator; 405 for (instr_iterator I = Header->instr_begin(), E = Header->instr_end(); 406 I != E && I->isPHI(); ++I) { 407 MachineInstr *Phi = &*I; 408 409 // Have a PHI instruction. Get the operand that corresponds to the 410 // latch block, and see if is a result of an addition of form "reg+imm", 411 // where the "reg" is defined by the PHI node we are looking at. 412 for (unsigned i = 1, n = Phi->getNumOperands(); i < n; i += 2) { 413 if (Phi->getOperand(i+1).getMBB() != Latch) 414 continue; 415 416 unsigned PhiOpReg = Phi->getOperand(i).getReg(); 417 MachineInstr *DI = MRI->getVRegDef(PhiOpReg); 418 unsigned UpdOpc = DI->getOpcode(); 419 bool isAdd = (UpdOpc == Hexagon::A2_addi || UpdOpc == Hexagon::A2_addp); 420 421 if (isAdd) { 422 // If the register operand to the add is the PHI we're looking at, this 423 // meets the induction pattern. 424 unsigned IndReg = DI->getOperand(1).getReg(); 425 MachineOperand &Opnd2 = DI->getOperand(2); 426 int64_t V; 427 if (MRI->getVRegDef(IndReg) == Phi && checkForImmediate(Opnd2, V)) { 428 unsigned UpdReg = DI->getOperand(0).getReg(); 429 IndMap.insert(std::make_pair(UpdReg, std::make_pair(IndReg, V))); 430 } 431 } 432 } // for (i) 433 } // for (instr) 434 435 SmallVector<MachineOperand,2> Cond; 436 MachineBasicBlock *TB = nullptr, *FB = nullptr; 437 bool NotAnalyzed = TII->AnalyzeBranch(*ExitingBlock, TB, FB, Cond, false); 438 if (NotAnalyzed) 439 return false; 440 441 unsigned PredR, PredPos, PredRegFlags; 442 if (!TII->getPredReg(Cond, PredR, PredPos, PredRegFlags)) 443 return false; 444 445 MachineInstr *PredI = MRI->getVRegDef(PredR); 446 if (!PredI->isCompare()) 447 return false; 448 449 unsigned CmpReg1 = 0, CmpReg2 = 0; 450 int CmpImm = 0, CmpMask = 0; 451 bool CmpAnalyzed = TII->analyzeCompare(PredI, CmpReg1, CmpReg2, 452 CmpMask, CmpImm); 453 // Fail if the compare was not analyzed, or it's not comparing a register 454 // with an immediate value. Not checking the mask here, since we handle 455 // the individual compare opcodes (including A4_cmpb*) later on. 456 if (!CmpAnalyzed) 457 return false; 458 459 // Exactly one of the input registers to the comparison should be among 460 // the induction registers. 461 InductionMap::iterator IndMapEnd = IndMap.end(); 462 InductionMap::iterator F = IndMapEnd; 463 if (CmpReg1 != 0) { 464 InductionMap::iterator F1 = IndMap.find(CmpReg1); 465 if (F1 != IndMapEnd) 466 F = F1; 467 } 468 if (CmpReg2 != 0) { 469 InductionMap::iterator F2 = IndMap.find(CmpReg2); 470 if (F2 != IndMapEnd) { 471 if (F != IndMapEnd) 472 return false; 473 F = F2; 474 } 475 } 476 if (F == IndMapEnd) 477 return false; 478 479 Reg = F->second.first; 480 IVBump = F->second.second; 481 IVOp = MRI->getVRegDef(F->first); 482 return true; 483} 484 485// Return the comparison kind for the specified opcode. 486HexagonHardwareLoops::Comparison::Kind 487HexagonHardwareLoops::getComparisonKind(unsigned CondOpc, 488 MachineOperand *InitialValue, 489 const MachineOperand *EndValue, 490 int64_t IVBump) const { 491 Comparison::Kind Cmp = (Comparison::Kind)0; 492 switch (CondOpc) { 493 case Hexagon::C2_cmpeqi: 494 case Hexagon::C2_cmpeq: 495 case Hexagon::C2_cmpeqp: 496 Cmp = Comparison::EQ; 497 break; 498 case Hexagon::C4_cmpneq: 499 case Hexagon::C4_cmpneqi: 500 Cmp = Comparison::NE; 501 break; 502 case Hexagon::C4_cmplte: 503 Cmp = Comparison::LEs; 504 break; 505 case Hexagon::C4_cmplteu: 506 Cmp = Comparison::LEu; 507 break; 508 case Hexagon::C2_cmpgtui: 509 case Hexagon::C2_cmpgtu: 510 case Hexagon::C2_cmpgtup: 511 Cmp = Comparison::GTu; 512 break; 513 case Hexagon::C2_cmpgti: 514 case Hexagon::C2_cmpgt: 515 case Hexagon::C2_cmpgtp: 516 Cmp = Comparison::GTs; 517 break; 518 default: 519 return (Comparison::Kind)0; 520 } 521 return Cmp; 522} 523 524/// \brief Analyze the statements in a loop to determine if the loop has 525/// a computable trip count and, if so, return a value that represents 526/// the trip count expression. 527/// 528/// This function iterates over the phi nodes in the loop to check for 529/// induction variable patterns that are used in the calculation for 530/// the number of time the loop is executed. 531CountValue *HexagonHardwareLoops::getLoopTripCount(MachineLoop *L, 532 SmallVectorImpl<MachineInstr *> &OldInsts) { 533 MachineBasicBlock *TopMBB = L->getTopBlock(); 534 MachineBasicBlock::pred_iterator PI = TopMBB->pred_begin(); 535 assert(PI != TopMBB->pred_end() && 536 "Loop must have more than one incoming edge!"); 537 MachineBasicBlock *Backedge = *PI++; 538 if (PI == TopMBB->pred_end()) // dead loop? 539 return nullptr; 540 MachineBasicBlock *Incoming = *PI++; 541 if (PI != TopMBB->pred_end()) // multiple backedges? 542 return nullptr; 543 544 // Make sure there is one incoming and one backedge and determine which 545 // is which. 546 if (L->contains(Incoming)) { 547 if (L->contains(Backedge)) 548 return nullptr; 549 std::swap(Incoming, Backedge); 550 } else if (!L->contains(Backedge)) 551 return nullptr; 552 553 // Look for the cmp instruction to determine if we can get a useful trip 554 // count. The trip count can be either a register or an immediate. The 555 // location of the value depends upon the type (reg or imm). 556 MachineBasicBlock *ExitingBlock = getExitingBlock(L); 557 if (!ExitingBlock) 558 return nullptr; 559 560 unsigned IVReg = 0; 561 int64_t IVBump = 0; 562 MachineInstr *IVOp; 563 bool FoundIV = findInductionRegister(L, IVReg, IVBump, IVOp); 564 if (!FoundIV) 565 return nullptr; 566 567 MachineBasicBlock *Preheader = L->getLoopPreheader(); 568 569 MachineOperand *InitialValue = nullptr; 570 MachineInstr *IV_Phi = MRI->getVRegDef(IVReg); 571 MachineBasicBlock *Latch = L->getLoopLatch(); 572 for (unsigned i = 1, n = IV_Phi->getNumOperands(); i < n; i += 2) { 573 MachineBasicBlock *MBB = IV_Phi->getOperand(i+1).getMBB(); 574 if (MBB == Preheader) 575 InitialValue = &IV_Phi->getOperand(i); 576 else if (MBB == Latch) 577 IVReg = IV_Phi->getOperand(i).getReg(); // Want IV reg after bump. 578 } 579 if (!InitialValue) 580 return nullptr; 581 582 SmallVector<MachineOperand,2> Cond; 583 MachineBasicBlock *TB = nullptr, *FB = nullptr; 584 bool NotAnalyzed = TII->AnalyzeBranch(*ExitingBlock, TB, FB, Cond, false); 585 if (NotAnalyzed) 586 return nullptr; 587 588 MachineBasicBlock *Header = L->getHeader(); 589 // TB must be non-null. If FB is also non-null, one of them must be 590 // the header. Otherwise, branch to TB could be exiting the loop, and 591 // the fall through can go to the header. 592 assert (TB && "Exit block without a branch?"); 593 if (ExitingBlock != Latch && (TB == Latch || FB == Latch)) { 594 MachineBasicBlock *LTB = 0, *LFB = 0; 595 SmallVector<MachineOperand,2> LCond; 596 bool NotAnalyzed = TII->AnalyzeBranch(*Latch, LTB, LFB, LCond, false); 597 if (NotAnalyzed) 598 return nullptr; 599 if (TB == Latch) 600 TB = (LTB == Header) ? LTB : LFB; 601 else 602 FB = (LTB == Header) ? LTB: LFB; 603 } 604 assert ((!FB || TB == Header || FB == Header) && "Branches not to header?"); 605 if (!TB || (FB && TB != Header && FB != Header)) 606 return nullptr; 607 608 // Branches of form "if (!P) ..." cause HexagonInstrInfo::AnalyzeBranch 609 // to put imm(0), followed by P in the vector Cond. 610 // If TB is not the header, it means that the "not-taken" path must lead 611 // to the header. 612 bool Negated = TII->predOpcodeHasNot(Cond) ^ (TB != Header); 613 unsigned PredReg, PredPos, PredRegFlags; 614 if (!TII->getPredReg(Cond, PredReg, PredPos, PredRegFlags)) 615 return nullptr; 616 MachineInstr *CondI = MRI->getVRegDef(PredReg); 617 unsigned CondOpc = CondI->getOpcode(); 618 619 unsigned CmpReg1 = 0, CmpReg2 = 0; 620 int Mask = 0, ImmValue = 0; 621 bool AnalyzedCmp = TII->analyzeCompare(CondI, CmpReg1, CmpReg2, 622 Mask, ImmValue); 623 if (!AnalyzedCmp) 624 return nullptr; 625 626 // The comparison operator type determines how we compute the loop 627 // trip count. 628 OldInsts.push_back(CondI); 629 OldInsts.push_back(IVOp); 630 631 // Sadly, the following code gets information based on the position 632 // of the operands in the compare instruction. This has to be done 633 // this way, because the comparisons check for a specific relationship 634 // between the operands (e.g. is-less-than), rather than to find out 635 // what relationship the operands are in (as on PPC). 636 Comparison::Kind Cmp; 637 bool isSwapped = false; 638 const MachineOperand &Op1 = CondI->getOperand(1); 639 const MachineOperand &Op2 = CondI->getOperand(2); 640 const MachineOperand *EndValue = nullptr; 641 642 if (Op1.isReg()) { 643 if (Op2.isImm() || Op1.getReg() == IVReg) 644 EndValue = &Op2; 645 else { 646 EndValue = &Op1; 647 isSwapped = true; 648 } 649 } 650 651 if (!EndValue) 652 return nullptr; 653 654 Cmp = getComparisonKind(CondOpc, InitialValue, EndValue, IVBump); 655 if (!Cmp) 656 return nullptr; 657 if (Negated) 658 Cmp = Comparison::getNegatedComparison(Cmp); 659 if (isSwapped) 660 Cmp = Comparison::getSwappedComparison(Cmp); 661 662 if (InitialValue->isReg()) { 663 unsigned R = InitialValue->getReg(); 664 MachineBasicBlock *DefBB = MRI->getVRegDef(R)->getParent(); 665 if (!MDT->properlyDominates(DefBB, Header)) 666 return nullptr; 667 OldInsts.push_back(MRI->getVRegDef(R)); 668 } 669 if (EndValue->isReg()) { 670 unsigned R = EndValue->getReg(); 671 MachineBasicBlock *DefBB = MRI->getVRegDef(R)->getParent(); 672 if (!MDT->properlyDominates(DefBB, Header)) 673 return nullptr; 674 OldInsts.push_back(MRI->getVRegDef(R)); 675 } 676 677 return computeCount(L, InitialValue, EndValue, IVReg, IVBump, Cmp); 678} 679 680/// \brief Helper function that returns the expression that represents the 681/// number of times a loop iterates. The function takes the operands that 682/// represent the loop start value, loop end value, and induction value. 683/// Based upon these operands, the function attempts to compute the trip count. 684CountValue *HexagonHardwareLoops::computeCount(MachineLoop *Loop, 685 const MachineOperand *Start, 686 const MachineOperand *End, 687 unsigned IVReg, 688 int64_t IVBump, 689 Comparison::Kind Cmp) const { 690 // Cannot handle comparison EQ, i.e. while (A == B). 691 if (Cmp == Comparison::EQ) 692 return nullptr; 693 694 // Check if either the start or end values are an assignment of an immediate. 695 // If so, use the immediate value rather than the register. 696 if (Start->isReg()) { 697 const MachineInstr *StartValInstr = MRI->getVRegDef(Start->getReg()); 698 if (StartValInstr && (StartValInstr->getOpcode() == Hexagon::A2_tfrsi || 699 StartValInstr->getOpcode() == Hexagon::A2_tfrpi)) 700 Start = &StartValInstr->getOperand(1); 701 } 702 if (End->isReg()) { 703 const MachineInstr *EndValInstr = MRI->getVRegDef(End->getReg()); 704 if (EndValInstr && (EndValInstr->getOpcode() == Hexagon::A2_tfrsi || 705 EndValInstr->getOpcode() == Hexagon::A2_tfrpi)) 706 End = &EndValInstr->getOperand(1); 707 } 708 709 if (!Start->isReg() && !Start->isImm()) 710 return nullptr; 711 if (!End->isReg() && !End->isImm()) 712 return nullptr; 713 714 bool CmpLess = Cmp & Comparison::L; 715 bool CmpGreater = Cmp & Comparison::G; 716 bool CmpHasEqual = Cmp & Comparison::EQ; 717 718 // Avoid certain wrap-arounds. This doesn't detect all wrap-arounds. 719 if (CmpLess && IVBump < 0) 720 // Loop going while iv is "less" with the iv value going down. Must wrap. 721 return nullptr; 722 723 if (CmpGreater && IVBump > 0) 724 // Loop going while iv is "greater" with the iv value going up. Must wrap. 725 return nullptr; 726 727 // Phis that may feed into the loop. 728 LoopFeederMap LoopFeederPhi; 729 730 // Check if the initial value may be zero and can be decremented in the first 731 // iteration. If the value is zero, the endloop instruction will not decrement 732 // the loop counter, so we shouldn't generate a hardware loop in this case. 733 if (loopCountMayWrapOrUnderFlow(Start, End, Loop->getLoopPreheader(), Loop, 734 LoopFeederPhi)) 735 return nullptr; 736 737 if (Start->isImm() && End->isImm()) { 738 // Both, start and end are immediates. 739 int64_t StartV = Start->getImm(); 740 int64_t EndV = End->getImm(); 741 int64_t Dist = EndV - StartV; 742 if (Dist == 0) 743 return nullptr; 744 745 bool Exact = (Dist % IVBump) == 0; 746 747 if (Cmp == Comparison::NE) { 748 if (!Exact) 749 return nullptr; 750 if ((Dist < 0) ^ (IVBump < 0)) 751 return nullptr; 752 } 753 754 // For comparisons that include the final value (i.e. include equality 755 // with the final value), we need to increase the distance by 1. 756 if (CmpHasEqual) 757 Dist = Dist > 0 ? Dist+1 : Dist-1; 758 759 // For the loop to iterate, CmpLess should imply Dist > 0. Similarly, 760 // CmpGreater should imply Dist < 0. These conditions could actually 761 // fail, for example, in unreachable code (which may still appear to be 762 // reachable in the CFG). 763 if ((CmpLess && Dist < 0) || (CmpGreater && Dist > 0)) 764 return nullptr; 765 766 // "Normalized" distance, i.e. with the bump set to +-1. 767 int64_t Dist1 = (IVBump > 0) ? (Dist + (IVBump - 1)) / IVBump 768 : (-Dist + (-IVBump - 1)) / (-IVBump); 769 assert (Dist1 > 0 && "Fishy thing. Both operands have the same sign."); 770 771 uint64_t Count = Dist1; 772 773 if (Count > 0xFFFFFFFFULL) 774 return nullptr; 775 776 return new CountValue(CountValue::CV_Immediate, Count); 777 } 778 779 // A general case: Start and End are some values, but the actual 780 // iteration count may not be available. If it is not, insert 781 // a computation of it into the preheader. 782 783 // If the induction variable bump is not a power of 2, quit. 784 // Othwerise we'd need a general integer division. 785 if (!isPowerOf2_64(std::abs(IVBump))) 786 return nullptr; 787 788 MachineBasicBlock *PH = Loop->getLoopPreheader(); 789 assert (PH && "Should have a preheader by now"); 790 MachineBasicBlock::iterator InsertPos = PH->getFirstTerminator(); 791 DebugLoc DL; 792 if (InsertPos != PH->end()) 793 DL = InsertPos->getDebugLoc(); 794 795 // If Start is an immediate and End is a register, the trip count 796 // will be "reg - imm". Hexagon's "subtract immediate" instruction 797 // is actually "reg + -imm". 798 799 // If the loop IV is going downwards, i.e. if the bump is negative, 800 // then the iteration count (computed as End-Start) will need to be 801 // negated. To avoid the negation, just swap Start and End. 802 if (IVBump < 0) { 803 std::swap(Start, End); 804 IVBump = -IVBump; 805 } 806 // Cmp may now have a wrong direction, e.g. LEs may now be GEs. 807 // Signedness, and "including equality" are preserved. 808 809 bool RegToImm = Start->isReg() && End->isImm(); // for (reg..imm) 810 bool RegToReg = Start->isReg() && End->isReg(); // for (reg..reg) 811 812 int64_t StartV = 0, EndV = 0; 813 if (Start->isImm()) 814 StartV = Start->getImm(); 815 if (End->isImm()) 816 EndV = End->getImm(); 817 818 int64_t AdjV = 0; 819 // To compute the iteration count, we would need this computation: 820 // Count = (End - Start + (IVBump-1)) / IVBump 821 // or, when CmpHasEqual: 822 // Count = (End - Start + (IVBump-1)+1) / IVBump 823 // The "IVBump-1" part is the adjustment (AdjV). We can avoid 824 // generating an instruction specifically to add it if we can adjust 825 // the immediate values for Start or End. 826 827 if (CmpHasEqual) { 828 // Need to add 1 to the total iteration count. 829 if (Start->isImm()) 830 StartV--; 831 else if (End->isImm()) 832 EndV++; 833 else 834 AdjV += 1; 835 } 836 837 if (Cmp != Comparison::NE) { 838 if (Start->isImm()) 839 StartV -= (IVBump-1); 840 else if (End->isImm()) 841 EndV += (IVBump-1); 842 else 843 AdjV += (IVBump-1); 844 } 845 846 unsigned R = 0, SR = 0; 847 if (Start->isReg()) { 848 R = Start->getReg(); 849 SR = Start->getSubReg(); 850 } else { 851 R = End->getReg(); 852 SR = End->getSubReg(); 853 } 854 const TargetRegisterClass *RC = MRI->getRegClass(R); 855 // Hardware loops cannot handle 64-bit registers. If it's a double 856 // register, it has to have a subregister. 857 if (!SR && RC == &Hexagon::DoubleRegsRegClass) 858 return nullptr; 859 const TargetRegisterClass *IntRC = &Hexagon::IntRegsRegClass; 860 861 // Compute DistR (register with the distance between Start and End). 862 unsigned DistR, DistSR; 863 864 // Avoid special case, where the start value is an imm(0). 865 if (Start->isImm() && StartV == 0) { 866 DistR = End->getReg(); 867 DistSR = End->getSubReg(); 868 } else { 869 const MCInstrDesc &SubD = RegToReg ? TII->get(Hexagon::A2_sub) : 870 (RegToImm ? TII->get(Hexagon::A2_subri) : 871 TII->get(Hexagon::A2_addi)); 872 if (RegToReg || RegToImm) { 873 unsigned SubR = MRI->createVirtualRegister(IntRC); 874 MachineInstrBuilder SubIB = 875 BuildMI(*PH, InsertPos, DL, SubD, SubR); 876 877 if (RegToReg) 878 SubIB.addReg(End->getReg(), 0, End->getSubReg()) 879 .addReg(Start->getReg(), 0, Start->getSubReg()); 880 else 881 SubIB.addImm(EndV) 882 .addReg(Start->getReg(), 0, Start->getSubReg()); 883 DistR = SubR; 884 } else { 885 // If the loop has been unrolled, we should use the original loop count 886 // instead of recalculating the value. This will avoid additional 887 // 'Add' instruction. 888 const MachineInstr *EndValInstr = MRI->getVRegDef(End->getReg()); 889 if (EndValInstr->getOpcode() == Hexagon::A2_addi && 890 EndValInstr->getOperand(2).getImm() == StartV) { 891 DistR = EndValInstr->getOperand(1).getReg(); 892 } else { 893 unsigned SubR = MRI->createVirtualRegister(IntRC); 894 MachineInstrBuilder SubIB = 895 BuildMI(*PH, InsertPos, DL, SubD, SubR); 896 SubIB.addReg(End->getReg(), 0, End->getSubReg()) 897 .addImm(-StartV); 898 DistR = SubR; 899 } 900 } 901 DistSR = 0; 902 } 903 904 // From DistR, compute AdjR (register with the adjusted distance). 905 unsigned AdjR, AdjSR; 906 907 if (AdjV == 0) { 908 AdjR = DistR; 909 AdjSR = DistSR; 910 } else { 911 // Generate CountR = ADD DistR, AdjVal 912 unsigned AddR = MRI->createVirtualRegister(IntRC); 913 MCInstrDesc const &AddD = TII->get(Hexagon::A2_addi); 914 BuildMI(*PH, InsertPos, DL, AddD, AddR) 915 .addReg(DistR, 0, DistSR) 916 .addImm(AdjV); 917 918 AdjR = AddR; 919 AdjSR = 0; 920 } 921 922 // From AdjR, compute CountR (register with the final count). 923 unsigned CountR, CountSR; 924 925 if (IVBump == 1) { 926 CountR = AdjR; 927 CountSR = AdjSR; 928 } else { 929 // The IV bump is a power of two. Log_2(IV bump) is the shift amount. 930 unsigned Shift = Log2_32(IVBump); 931 932 // Generate NormR = LSR DistR, Shift. 933 unsigned LsrR = MRI->createVirtualRegister(IntRC); 934 const MCInstrDesc &LsrD = TII->get(Hexagon::S2_lsr_i_r); 935 BuildMI(*PH, InsertPos, DL, LsrD, LsrR) 936 .addReg(AdjR, 0, AdjSR) 937 .addImm(Shift); 938 939 CountR = LsrR; 940 CountSR = 0; 941 } 942 943 return new CountValue(CountValue::CV_Register, CountR, CountSR); 944} 945 946/// \brief Return true if the operation is invalid within hardware loop. 947bool HexagonHardwareLoops::isInvalidLoopOperation(const MachineInstr *MI, 948 bool IsInnerHWLoop) const { 949 950 // Call is not allowed because the callee may use a hardware loop except for 951 // the case when the call never returns. 952 if (MI->getDesc().isCall() && MI->getOpcode() != Hexagon::CALLv3nr) 953 return true; 954 955 // Check if the instruction defines a hardware loop register. 956 for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) { 957 const MachineOperand &MO = MI->getOperand(i); 958 if (!MO.isReg() || !MO.isDef()) 959 continue; 960 unsigned R = MO.getReg(); 961 if (IsInnerHWLoop && (R == Hexagon::LC0 || R == Hexagon::SA0 || 962 R == Hexagon::LC1 || R == Hexagon::SA1)) 963 return true; 964 if (!IsInnerHWLoop && (R == Hexagon::LC1 || R == Hexagon::SA1)) 965 return true; 966 } 967 return false; 968} 969 970/// \brief Return true if the loop contains an instruction that inhibits 971/// the use of the hardware loop instruction. 972bool HexagonHardwareLoops::containsInvalidInstruction(MachineLoop *L, 973 bool IsInnerHWLoop) const { 974 const std::vector<MachineBasicBlock *> &Blocks = L->getBlocks(); 975 DEBUG(dbgs() << "\nhw_loop head, BB#" << Blocks[0]->getNumber();); 976 for (unsigned i = 0, e = Blocks.size(); i != e; ++i) { 977 MachineBasicBlock *MBB = Blocks[i]; 978 for (MachineBasicBlock::iterator 979 MII = MBB->begin(), E = MBB->end(); MII != E; ++MII) { 980 const MachineInstr *MI = &*MII; 981 if (isInvalidLoopOperation(MI, IsInnerHWLoop)) { 982 DEBUG(dbgs()<< "\nCannot convert to hw_loop due to:"; MI->dump();); 983 return true; 984 } 985 } 986 } 987 return false; 988} 989 990/// \brief Returns true if the instruction is dead. This was essentially 991/// copied from DeadMachineInstructionElim::isDead, but with special cases 992/// for inline asm, physical registers and instructions with side effects 993/// removed. 994bool HexagonHardwareLoops::isDead(const MachineInstr *MI, 995 SmallVectorImpl<MachineInstr *> &DeadPhis) const { 996 // Examine each operand. 997 for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) { 998 const MachineOperand &MO = MI->getOperand(i); 999 if (!MO.isReg() || !MO.isDef()) 1000 continue; 1001 1002 unsigned Reg = MO.getReg(); 1003 if (MRI->use_nodbg_empty(Reg)) 1004 continue; 1005 1006 typedef MachineRegisterInfo::use_nodbg_iterator use_nodbg_iterator; 1007 1008 // This instruction has users, but if the only user is the phi node for the 1009 // parent block, and the only use of that phi node is this instruction, then 1010 // this instruction is dead: both it (and the phi node) can be removed. 1011 use_nodbg_iterator I = MRI->use_nodbg_begin(Reg); 1012 use_nodbg_iterator End = MRI->use_nodbg_end(); 1013 if (std::next(I) != End || !I->getParent()->isPHI()) 1014 return false; 1015 1016 MachineInstr *OnePhi = I->getParent(); 1017 for (unsigned j = 0, f = OnePhi->getNumOperands(); j != f; ++j) { 1018 const MachineOperand &OPO = OnePhi->getOperand(j); 1019 if (!OPO.isReg() || !OPO.isDef()) 1020 continue; 1021 1022 unsigned OPReg = OPO.getReg(); 1023 use_nodbg_iterator nextJ; 1024 for (use_nodbg_iterator J = MRI->use_nodbg_begin(OPReg); 1025 J != End; J = nextJ) { 1026 nextJ = std::next(J); 1027 MachineOperand &Use = *J; 1028 MachineInstr *UseMI = Use.getParent(); 1029 1030 // If the phi node has a user that is not MI, bail. 1031 if (MI != UseMI) 1032 return false; 1033 } 1034 } 1035 DeadPhis.push_back(OnePhi); 1036 } 1037 1038 // If there are no defs with uses, the instruction is dead. 1039 return true; 1040} 1041 1042void HexagonHardwareLoops::removeIfDead(MachineInstr *MI) { 1043 // This procedure was essentially copied from DeadMachineInstructionElim. 1044 1045 SmallVector<MachineInstr*, 1> DeadPhis; 1046 if (isDead(MI, DeadPhis)) { 1047 DEBUG(dbgs() << "HW looping will remove: " << *MI); 1048 1049 // It is possible that some DBG_VALUE instructions refer to this 1050 // instruction. Examine each def operand for such references; 1051 // if found, mark the DBG_VALUE as undef (but don't delete it). 1052 for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) { 1053 const MachineOperand &MO = MI->getOperand(i); 1054 if (!MO.isReg() || !MO.isDef()) 1055 continue; 1056 unsigned Reg = MO.getReg(); 1057 MachineRegisterInfo::use_iterator nextI; 1058 for (MachineRegisterInfo::use_iterator I = MRI->use_begin(Reg), 1059 E = MRI->use_end(); I != E; I = nextI) { 1060 nextI = std::next(I); // I is invalidated by the setReg 1061 MachineOperand &Use = *I; 1062 MachineInstr *UseMI = I->getParent(); 1063 if (UseMI == MI) 1064 continue; 1065 if (Use.isDebug()) 1066 UseMI->getOperand(0).setReg(0U); 1067 } 1068 } 1069 1070 MI->eraseFromParent(); 1071 for (unsigned i = 0; i < DeadPhis.size(); ++i) 1072 DeadPhis[i]->eraseFromParent(); 1073 } 1074} 1075 1076/// \brief Check if the loop is a candidate for converting to a hardware 1077/// loop. If so, then perform the transformation. 1078/// 1079/// This function works on innermost loops first. A loop can be converted 1080/// if it is a counting loop; either a register value or an immediate. 1081/// 1082/// The code makes several assumptions about the representation of the loop 1083/// in llvm. 1084bool HexagonHardwareLoops::convertToHardwareLoop(MachineLoop *L, 1085 bool &RecL0used, 1086 bool &RecL1used) { 1087 // This is just for sanity. 1088 assert(L->getHeader() && "Loop without a header?"); 1089 1090 bool Changed = false; 1091 bool L0Used = false; 1092 bool L1Used = false; 1093 1094 // Process nested loops first. 1095 for (MachineLoop::iterator I = L->begin(), E = L->end(); I != E; ++I) { 1096 Changed |= convertToHardwareLoop(*I, RecL0used, RecL1used); 1097 L0Used |= RecL0used; 1098 L1Used |= RecL1used; 1099 } 1100 1101 // If a nested loop has been converted, then we can't convert this loop. 1102 if (Changed && L0Used && L1Used) 1103 return Changed; 1104 1105 unsigned LOOP_i; 1106 unsigned LOOP_r; 1107 unsigned ENDLOOP; 1108 1109 // Flag used to track loopN instruction: 1110 // 1 - Hardware loop is being generated for the inner most loop. 1111 // 0 - Hardware loop is being generated for the outer loop. 1112 unsigned IsInnerHWLoop = 1; 1113 1114 if (L0Used) { 1115 LOOP_i = Hexagon::J2_loop1i; 1116 LOOP_r = Hexagon::J2_loop1r; 1117 ENDLOOP = Hexagon::ENDLOOP1; 1118 IsInnerHWLoop = 0; 1119 } else { 1120 LOOP_i = Hexagon::J2_loop0i; 1121 LOOP_r = Hexagon::J2_loop0r; 1122 ENDLOOP = Hexagon::ENDLOOP0; 1123 } 1124 1125#ifndef NDEBUG 1126 // Stop trying after reaching the limit (if any). 1127 int Limit = HWLoopLimit; 1128 if (Limit >= 0) { 1129 if (Counter >= HWLoopLimit) 1130 return false; 1131 Counter++; 1132 } 1133#endif 1134 1135 // Does the loop contain any invalid instructions? 1136 if (containsInvalidInstruction(L, IsInnerHWLoop)) 1137 return false; 1138 1139 MachineBasicBlock *LastMBB = getExitingBlock(L); 1140 // Don't generate hw loop if the loop has more than one exit. 1141 if (!LastMBB) 1142 return false; 1143 1144 MachineBasicBlock::iterator LastI = LastMBB->getFirstTerminator(); 1145 if (LastI == LastMBB->end()) 1146 return false; 1147 1148 // Is the induction variable bump feeding the latch condition? 1149 if (!fixupInductionVariable(L)) 1150 return false; 1151 1152 // Ensure the loop has a preheader: the loop instruction will be 1153 // placed there. 1154 MachineBasicBlock *Preheader = L->getLoopPreheader(); 1155 if (!Preheader) { 1156 Preheader = createPreheaderForLoop(L); 1157 if (!Preheader) 1158 return false; 1159 } 1160 1161 MachineBasicBlock::iterator InsertPos = Preheader->getFirstTerminator(); 1162 1163 SmallVector<MachineInstr*, 2> OldInsts; 1164 // Are we able to determine the trip count for the loop? 1165 CountValue *TripCount = getLoopTripCount(L, OldInsts); 1166 if (!TripCount) 1167 return false; 1168 1169 // Is the trip count available in the preheader? 1170 if (TripCount->isReg()) { 1171 // There will be a use of the register inserted into the preheader, 1172 // so make sure that the register is actually defined at that point. 1173 MachineInstr *TCDef = MRI->getVRegDef(TripCount->getReg()); 1174 MachineBasicBlock *BBDef = TCDef->getParent(); 1175 if (!MDT->dominates(BBDef, Preheader)) 1176 return false; 1177 } 1178 1179 // Determine the loop start. 1180 MachineBasicBlock *TopBlock = L->getTopBlock(); 1181 MachineBasicBlock *ExitingBlock = getExitingBlock(L); 1182 MachineBasicBlock *LoopStart = 0; 1183 if (ExitingBlock != L->getLoopLatch()) { 1184 MachineBasicBlock *TB = 0, *FB = 0; 1185 SmallVector<MachineOperand, 2> Cond; 1186 1187 if (TII->AnalyzeBranch(*ExitingBlock, TB, FB, Cond, false)) 1188 return false; 1189 1190 if (L->contains(TB)) 1191 LoopStart = TB; 1192 else if (L->contains(FB)) 1193 LoopStart = FB; 1194 else 1195 return false; 1196 } 1197 else 1198 LoopStart = TopBlock; 1199 1200 // Convert the loop to a hardware loop. 1201 DEBUG(dbgs() << "Change to hardware loop at "; L->dump()); 1202 DebugLoc DL; 1203 if (InsertPos != Preheader->end()) 1204 DL = InsertPos->getDebugLoc(); 1205 1206 if (TripCount->isReg()) { 1207 // Create a copy of the loop count register. 1208 unsigned CountReg = MRI->createVirtualRegister(&Hexagon::IntRegsRegClass); 1209 BuildMI(*Preheader, InsertPos, DL, TII->get(TargetOpcode::COPY), CountReg) 1210 .addReg(TripCount->getReg(), 0, TripCount->getSubReg()); 1211 // Add the Loop instruction to the beginning of the loop. 1212 BuildMI(*Preheader, InsertPos, DL, TII->get(LOOP_r)).addMBB(LoopStart) 1213 .addReg(CountReg); 1214 } else { 1215 assert(TripCount->isImm() && "Expecting immediate value for trip count"); 1216 // Add the Loop immediate instruction to the beginning of the loop, 1217 // if the immediate fits in the instructions. Otherwise, we need to 1218 // create a new virtual register. 1219 int64_t CountImm = TripCount->getImm(); 1220 if (!TII->isValidOffset(LOOP_i, CountImm)) { 1221 unsigned CountReg = MRI->createVirtualRegister(&Hexagon::IntRegsRegClass); 1222 BuildMI(*Preheader, InsertPos, DL, TII->get(Hexagon::A2_tfrsi), CountReg) 1223 .addImm(CountImm); 1224 BuildMI(*Preheader, InsertPos, DL, TII->get(LOOP_r)) 1225 .addMBB(LoopStart).addReg(CountReg); 1226 } else 1227 BuildMI(*Preheader, InsertPos, DL, TII->get(LOOP_i)) 1228 .addMBB(LoopStart).addImm(CountImm); 1229 } 1230 1231 // Make sure the loop start always has a reference in the CFG. We need 1232 // to create a BlockAddress operand to get this mechanism to work both the 1233 // MachineBasicBlock and BasicBlock objects need the flag set. 1234 LoopStart->setHasAddressTaken(); 1235 // This line is needed to set the hasAddressTaken flag on the BasicBlock 1236 // object. 1237 BlockAddress::get(const_cast<BasicBlock *>(LoopStart->getBasicBlock())); 1238 1239 // Replace the loop branch with an endloop instruction. 1240 DebugLoc LastIDL = LastI->getDebugLoc(); 1241 BuildMI(*LastMBB, LastI, LastIDL, TII->get(ENDLOOP)).addMBB(LoopStart); 1242 1243 // The loop ends with either: 1244 // - a conditional branch followed by an unconditional branch, or 1245 // - a conditional branch to the loop start. 1246 if (LastI->getOpcode() == Hexagon::J2_jumpt || 1247 LastI->getOpcode() == Hexagon::J2_jumpf) { 1248 // Delete one and change/add an uncond. branch to out of the loop. 1249 MachineBasicBlock *BranchTarget = LastI->getOperand(1).getMBB(); 1250 LastI = LastMBB->erase(LastI); 1251 if (!L->contains(BranchTarget)) { 1252 if (LastI != LastMBB->end()) 1253 LastI = LastMBB->erase(LastI); 1254 SmallVector<MachineOperand, 0> Cond; 1255 TII->InsertBranch(*LastMBB, BranchTarget, nullptr, Cond, LastIDL); 1256 } 1257 } else { 1258 // Conditional branch to loop start; just delete it. 1259 LastMBB->erase(LastI); 1260 } 1261 delete TripCount; 1262 1263 // The induction operation and the comparison may now be 1264 // unneeded. If these are unneeded, then remove them. 1265 for (unsigned i = 0; i < OldInsts.size(); ++i) 1266 removeIfDead(OldInsts[i]); 1267 1268 ++NumHWLoops; 1269 1270 // Set RecL1used and RecL0used only after hardware loop has been 1271 // successfully generated. Doing it earlier can cause wrong loop instruction 1272 // to be used. 1273 if (L0Used) // Loop0 was already used. So, the correct loop must be loop1. 1274 RecL1used = true; 1275 else 1276 RecL0used = true; 1277 1278 return true; 1279} 1280 1281bool HexagonHardwareLoops::orderBumpCompare(MachineInstr *BumpI, 1282 MachineInstr *CmpI) { 1283 assert (BumpI != CmpI && "Bump and compare in the same instruction?"); 1284 1285 MachineBasicBlock *BB = BumpI->getParent(); 1286 if (CmpI->getParent() != BB) 1287 return false; 1288 1289 typedef MachineBasicBlock::instr_iterator instr_iterator; 1290 // Check if things are in order to begin with. 1291 for (instr_iterator I(BumpI), E = BB->instr_end(); I != E; ++I) 1292 if (&*I == CmpI) 1293 return true; 1294 1295 // Out of order. 1296 unsigned PredR = CmpI->getOperand(0).getReg(); 1297 bool FoundBump = false; 1298 instr_iterator CmpIt = CmpI->getIterator(), NextIt = std::next(CmpIt); 1299 for (instr_iterator I = NextIt, E = BB->instr_end(); I != E; ++I) { 1300 MachineInstr *In = &*I; 1301 for (unsigned i = 0, n = In->getNumOperands(); i < n; ++i) { 1302 MachineOperand &MO = In->getOperand(i); 1303 if (MO.isReg() && MO.isUse()) { 1304 if (MO.getReg() == PredR) // Found an intervening use of PredR. 1305 return false; 1306 } 1307 } 1308 1309 if (In == BumpI) { 1310 BB->splice(++BumpI->getIterator(), BB, CmpI->getIterator()); 1311 FoundBump = true; 1312 break; 1313 } 1314 } 1315 assert (FoundBump && "Cannot determine instruction order"); 1316 return FoundBump; 1317} 1318 1319/// This function is required to break recursion. Visiting phis in a loop may 1320/// result in recursion during compilation. We break the recursion by making 1321/// sure that we visit a MachineOperand and its definition in a 1322/// MachineInstruction only once. If we attempt to visit more than once, then 1323/// there is recursion, and will return false. 1324bool HexagonHardwareLoops::isLoopFeeder(MachineLoop *L, MachineBasicBlock *A, 1325 MachineInstr *MI, 1326 const MachineOperand *MO, 1327 LoopFeederMap &LoopFeederPhi) const { 1328 if (LoopFeederPhi.find(MO->getReg()) == LoopFeederPhi.end()) { 1329 const std::vector<MachineBasicBlock *> &Blocks = L->getBlocks(); 1330 DEBUG(dbgs() << "\nhw_loop head, BB#" << Blocks[0]->getNumber();); 1331 // Ignore all BBs that form Loop. 1332 for (unsigned i = 0, e = Blocks.size(); i != e; ++i) { 1333 MachineBasicBlock *MBB = Blocks[i]; 1334 if (A == MBB) 1335 return false; 1336 } 1337 MachineInstr *Def = MRI->getVRegDef(MO->getReg()); 1338 LoopFeederPhi.insert(std::make_pair(MO->getReg(), Def)); 1339 return true; 1340 } else 1341 // Already visited node. 1342 return false; 1343} 1344 1345/// Return true if a Phi may generate a value that can underflow. 1346/// This function calls loopCountMayWrapOrUnderFlow for each Phi operand. 1347bool HexagonHardwareLoops::phiMayWrapOrUnderflow( 1348 MachineInstr *Phi, const MachineOperand *EndVal, MachineBasicBlock *MBB, 1349 MachineLoop *L, LoopFeederMap &LoopFeederPhi) const { 1350 assert(Phi->isPHI() && "Expecting a Phi."); 1351 // Walk through each Phi, and its used operands. Make sure that 1352 // if there is recursion in Phi, we won't generate hardware loops. 1353 for (int i = 1, n = Phi->getNumOperands(); i < n; i += 2) 1354 if (isLoopFeeder(L, MBB, Phi, &(Phi->getOperand(i)), LoopFeederPhi)) 1355 if (loopCountMayWrapOrUnderFlow(&(Phi->getOperand(i)), EndVal, 1356 Phi->getParent(), L, LoopFeederPhi)) 1357 return true; 1358 return false; 1359} 1360 1361/// Return true if the induction variable can underflow in the first iteration. 1362/// An example, is an initial unsigned value that is 0 and is decrement in the 1363/// first itertion of a do-while loop. In this case, we cannot generate a 1364/// hardware loop because the endloop instruction does not decrement the loop 1365/// counter if it is <= 1. We only need to perform this analysis if the 1366/// initial value is a register. 1367/// 1368/// This function assumes the initial value may underfow unless proven 1369/// otherwise. If the type is signed, then we don't care because signed 1370/// underflow is undefined. We attempt to prove the initial value is not 1371/// zero by perfoming a crude analysis of the loop counter. This function 1372/// checks if the initial value is used in any comparison prior to the loop 1373/// and, if so, assumes the comparison is a range check. This is inexact, 1374/// but will catch the simple cases. 1375bool HexagonHardwareLoops::loopCountMayWrapOrUnderFlow( 1376 const MachineOperand *InitVal, const MachineOperand *EndVal, 1377 MachineBasicBlock *MBB, MachineLoop *L, 1378 LoopFeederMap &LoopFeederPhi) const { 1379 // Only check register values since they are unknown. 1380 if (!InitVal->isReg()) 1381 return false; 1382 1383 if (!EndVal->isImm()) 1384 return false; 1385 1386 // A register value that is assigned an immediate is a known value, and it 1387 // won't underflow in the first iteration. 1388 int64_t Imm; 1389 if (checkForImmediate(*InitVal, Imm)) 1390 return (EndVal->getImm() == Imm); 1391 1392 unsigned Reg = InitVal->getReg(); 1393 1394 // We don't know the value of a physical register. 1395 if (!TargetRegisterInfo::isVirtualRegister(Reg)) 1396 return true; 1397 1398 MachineInstr *Def = MRI->getVRegDef(Reg); 1399 if (!Def) 1400 return true; 1401 1402 // If the initial value is a Phi or copy and the operands may not underflow, 1403 // then the definition cannot be underflow either. 1404 if (Def->isPHI() && !phiMayWrapOrUnderflow(Def, EndVal, Def->getParent(), 1405 L, LoopFeederPhi)) 1406 return false; 1407 if (Def->isCopy() && !loopCountMayWrapOrUnderFlow(&(Def->getOperand(1)), 1408 EndVal, Def->getParent(), 1409 L, LoopFeederPhi)) 1410 return false; 1411 1412 // Iterate over the uses of the initial value. If the initial value is used 1413 // in a compare, then we assume this is a range check that ensures the loop 1414 // doesn't underflow. This is not an exact test and should be improved. 1415 for (MachineRegisterInfo::use_instr_nodbg_iterator I = MRI->use_instr_nodbg_begin(Reg), 1416 E = MRI->use_instr_nodbg_end(); I != E; ++I) { 1417 MachineInstr *MI = &*I; 1418 unsigned CmpReg1 = 0, CmpReg2 = 0; 1419 int CmpMask = 0, CmpValue = 0; 1420 1421 if (!TII->analyzeCompare(MI, CmpReg1, CmpReg2, CmpMask, CmpValue)) 1422 continue; 1423 1424 MachineBasicBlock *TBB = 0, *FBB = 0; 1425 SmallVector<MachineOperand, 2> Cond; 1426 if (TII->AnalyzeBranch(*MI->getParent(), TBB, FBB, Cond, false)) 1427 continue; 1428 1429 Comparison::Kind Cmp = getComparisonKind(MI->getOpcode(), 0, 0, 0); 1430 if (Cmp == 0) 1431 continue; 1432 if (TII->predOpcodeHasNot(Cond) ^ (TBB != MBB)) 1433 Cmp = Comparison::getNegatedComparison(Cmp); 1434 if (CmpReg2 != 0 && CmpReg2 == Reg) 1435 Cmp = Comparison::getSwappedComparison(Cmp); 1436 1437 // Signed underflow is undefined. 1438 if (Comparison::isSigned(Cmp)) 1439 return false; 1440 1441 // Check if there is a comparison of the initial value. If the initial value 1442 // is greater than or not equal to another value, then assume this is a 1443 // range check. 1444 if ((Cmp & Comparison::G) || Cmp == Comparison::NE) 1445 return false; 1446 } 1447 1448 // OK - this is a hack that needs to be improved. We really need to analyze 1449 // the instructions performed on the initial value. This works on the simplest 1450 // cases only. 1451 if (!Def->isCopy() && !Def->isPHI()) 1452 return false; 1453 1454 return true; 1455} 1456 1457bool HexagonHardwareLoops::checkForImmediate(const MachineOperand &MO, 1458 int64_t &Val) const { 1459 if (MO.isImm()) { 1460 Val = MO.getImm(); 1461 return true; 1462 } 1463 if (!MO.isReg()) 1464 return false; 1465 1466 // MO is a register. Check whether it is defined as an immediate value, 1467 // and if so, get the value of it in TV. That value will then need to be 1468 // processed to handle potential subregisters in MO. 1469 int64_t TV; 1470 1471 unsigned R = MO.getReg(); 1472 if (!TargetRegisterInfo::isVirtualRegister(R)) 1473 return false; 1474 MachineInstr *DI = MRI->getVRegDef(R); 1475 unsigned DOpc = DI->getOpcode(); 1476 switch (DOpc) { 1477 case TargetOpcode::COPY: 1478 case Hexagon::A2_tfrsi: 1479 case Hexagon::A2_tfrpi: 1480 case Hexagon::CONST32_Int_Real: 1481 case Hexagon::CONST64_Int_Real: { 1482 // Call recursively to avoid an extra check whether operand(1) is 1483 // indeed an immediate (it could be a global address, for example), 1484 // plus we can handle COPY at the same time. 1485 if (!checkForImmediate(DI->getOperand(1), TV)) 1486 return false; 1487 break; 1488 } 1489 case Hexagon::A2_combineii: 1490 case Hexagon::A4_combineir: 1491 case Hexagon::A4_combineii: 1492 case Hexagon::A4_combineri: 1493 case Hexagon::A2_combinew: { 1494 const MachineOperand &S1 = DI->getOperand(1); 1495 const MachineOperand &S2 = DI->getOperand(2); 1496 int64_t V1, V2; 1497 if (!checkForImmediate(S1, V1) || !checkForImmediate(S2, V2)) 1498 return false; 1499 TV = V2 | (V1 << 32); 1500 break; 1501 } 1502 case TargetOpcode::REG_SEQUENCE: { 1503 const MachineOperand &S1 = DI->getOperand(1); 1504 const MachineOperand &S3 = DI->getOperand(3); 1505 int64_t V1, V3; 1506 if (!checkForImmediate(S1, V1) || !checkForImmediate(S3, V3)) 1507 return false; 1508 unsigned Sub2 = DI->getOperand(2).getImm(); 1509 unsigned Sub4 = DI->getOperand(4).getImm(); 1510 if (Sub2 == Hexagon::subreg_loreg && Sub4 == Hexagon::subreg_hireg) 1511 TV = V1 | (V3 << 32); 1512 else if (Sub2 == Hexagon::subreg_hireg && Sub4 == Hexagon::subreg_loreg) 1513 TV = V3 | (V1 << 32); 1514 else 1515 llvm_unreachable("Unexpected form of REG_SEQUENCE"); 1516 break; 1517 } 1518 1519 default: 1520 return false; 1521 } 1522 1523 // By now, we should have successfuly obtained the immediate value defining 1524 // the register referenced in MO. Handle a potential use of a subregister. 1525 switch (MO.getSubReg()) { 1526 case Hexagon::subreg_loreg: 1527 Val = TV & 0xFFFFFFFFULL; 1528 break; 1529 case Hexagon::subreg_hireg: 1530 Val = (TV >> 32) & 0xFFFFFFFFULL; 1531 break; 1532 default: 1533 Val = TV; 1534 break; 1535 } 1536 return true; 1537} 1538 1539void HexagonHardwareLoops::setImmediate(MachineOperand &MO, int64_t Val) { 1540 if (MO.isImm()) { 1541 MO.setImm(Val); 1542 return; 1543 } 1544 1545 assert(MO.isReg()); 1546 unsigned R = MO.getReg(); 1547 MachineInstr *DI = MRI->getVRegDef(R); 1548 1549 const TargetRegisterClass *RC = MRI->getRegClass(R); 1550 unsigned NewR = MRI->createVirtualRegister(RC); 1551 MachineBasicBlock &B = *DI->getParent(); 1552 DebugLoc DL = DI->getDebugLoc(); 1553 BuildMI(B, DI, DL, TII->get(DI->getOpcode()), NewR).addImm(Val); 1554 MO.setReg(NewR); 1555} 1556 1557static bool isImmValidForOpcode(unsigned CmpOpc, int64_t Imm) { 1558 // These two instructions are not extendable. 1559 if (CmpOpc == Hexagon::A4_cmpbeqi) 1560 return isUInt<8>(Imm); 1561 if (CmpOpc == Hexagon::A4_cmpbgti) 1562 return isInt<8>(Imm); 1563 // The rest of the comparison-with-immediate instructions are extendable. 1564 return true; 1565} 1566 1567bool HexagonHardwareLoops::fixupInductionVariable(MachineLoop *L) { 1568 MachineBasicBlock *Header = L->getHeader(); 1569 MachineBasicBlock *Latch = L->getLoopLatch(); 1570 MachineBasicBlock *ExitingBlock = getExitingBlock(L); 1571 1572 if (!(Header && Latch && ExitingBlock)) 1573 return false; 1574 1575 // These data structures follow the same concept as the corresponding 1576 // ones in findInductionRegister (where some comments are). 1577 typedef std::pair<unsigned,int64_t> RegisterBump; 1578 typedef std::pair<unsigned,RegisterBump> RegisterInduction; 1579 typedef std::set<RegisterInduction> RegisterInductionSet; 1580 1581 // Register candidates for induction variables, with their associated bumps. 1582 RegisterInductionSet IndRegs; 1583 1584 // Look for induction patterns: 1585 // vreg1 = PHI ..., [ latch, vreg2 ] 1586 // vreg2 = ADD vreg1, imm 1587 typedef MachineBasicBlock::instr_iterator instr_iterator; 1588 for (instr_iterator I = Header->instr_begin(), E = Header->instr_end(); 1589 I != E && I->isPHI(); ++I) { 1590 MachineInstr *Phi = &*I; 1591 1592 // Have a PHI instruction. 1593 for (unsigned i = 1, n = Phi->getNumOperands(); i < n; i += 2) { 1594 if (Phi->getOperand(i+1).getMBB() != Latch) 1595 continue; 1596 1597 unsigned PhiReg = Phi->getOperand(i).getReg(); 1598 MachineInstr *DI = MRI->getVRegDef(PhiReg); 1599 unsigned UpdOpc = DI->getOpcode(); 1600 bool isAdd = (UpdOpc == Hexagon::A2_addi || UpdOpc == Hexagon::A2_addp); 1601 1602 if (isAdd) { 1603 // If the register operand to the add/sub is the PHI we are looking 1604 // at, this meets the induction pattern. 1605 unsigned IndReg = DI->getOperand(1).getReg(); 1606 MachineOperand &Opnd2 = DI->getOperand(2); 1607 int64_t V; 1608 if (MRI->getVRegDef(IndReg) == Phi && checkForImmediate(Opnd2, V)) { 1609 unsigned UpdReg = DI->getOperand(0).getReg(); 1610 IndRegs.insert(std::make_pair(UpdReg, std::make_pair(IndReg, V))); 1611 } 1612 } 1613 } // for (i) 1614 } // for (instr) 1615 1616 if (IndRegs.empty()) 1617 return false; 1618 1619 MachineBasicBlock *TB = nullptr, *FB = nullptr; 1620 SmallVector<MachineOperand,2> Cond; 1621 // AnalyzeBranch returns true if it fails to analyze branch. 1622 bool NotAnalyzed = TII->AnalyzeBranch(*ExitingBlock, TB, FB, Cond, false); 1623 if (NotAnalyzed || Cond.empty()) 1624 return false; 1625 1626 if (ExitingBlock != Latch && (TB == Latch || FB == Latch)) { 1627 MachineBasicBlock *LTB = 0, *LFB = 0; 1628 SmallVector<MachineOperand,2> LCond; 1629 bool NotAnalyzed = TII->AnalyzeBranch(*Latch, LTB, LFB, LCond, false); 1630 if (NotAnalyzed) 1631 return false; 1632 1633 // Since latch is not the exiting block, the latch branch should be an 1634 // unconditional branch to the loop header. 1635 if (TB == Latch) 1636 TB = (LTB == Header) ? LTB : LFB; 1637 else 1638 FB = (LTB == Header) ? LTB : LFB; 1639 } 1640 if (TB != Header) { 1641 if (FB != Header) { 1642 // The latch/exit block does not go back to the header. 1643 return false; 1644 } 1645 // FB is the header (i.e., uncond. jump to branch header) 1646 // In this case, the LoopBody -> TB should not be a back edge otherwise 1647 // it could result in an infinite loop after conversion to hw_loop. 1648 // This case can happen when the Latch has two jumps like this: 1649 // Jmp_c OuterLoopHeader <-- TB 1650 // Jmp InnerLoopHeader <-- FB 1651 if (MDT->dominates(TB, FB)) 1652 return false; 1653 } 1654 1655 // Expecting a predicate register as a condition. It won't be a hardware 1656 // predicate register at this point yet, just a vreg. 1657 // HexagonInstrInfo::AnalyzeBranch for negated branches inserts imm(0) 1658 // into Cond, followed by the predicate register. For non-negated branches 1659 // it's just the register. 1660 unsigned CSz = Cond.size(); 1661 if (CSz != 1 && CSz != 2) 1662 return false; 1663 1664 if (!Cond[CSz-1].isReg()) 1665 return false; 1666 1667 unsigned P = Cond[CSz-1].getReg(); 1668 MachineInstr *PredDef = MRI->getVRegDef(P); 1669 1670 if (!PredDef->isCompare()) 1671 return false; 1672 1673 SmallSet<unsigned,2> CmpRegs; 1674 MachineOperand *CmpImmOp = nullptr; 1675 1676 // Go over all operands to the compare and look for immediate and register 1677 // operands. Assume that if the compare has a single register use and a 1678 // single immediate operand, then the register is being compared with the 1679 // immediate value. 1680 for (unsigned i = 0, n = PredDef->getNumOperands(); i < n; ++i) { 1681 MachineOperand &MO = PredDef->getOperand(i); 1682 if (MO.isReg()) { 1683 // Skip all implicit references. In one case there was: 1684 // %vreg140<def> = FCMPUGT32_rr %vreg138, %vreg139, %USR<imp-use> 1685 if (MO.isImplicit()) 1686 continue; 1687 if (MO.isUse()) { 1688 if (!isImmediate(MO)) { 1689 CmpRegs.insert(MO.getReg()); 1690 continue; 1691 } 1692 // Consider the register to be the "immediate" operand. 1693 if (CmpImmOp) 1694 return false; 1695 CmpImmOp = &MO; 1696 } 1697 } else if (MO.isImm()) { 1698 if (CmpImmOp) // A second immediate argument? Confusing. Bail out. 1699 return false; 1700 CmpImmOp = &MO; 1701 } 1702 } 1703 1704 if (CmpRegs.empty()) 1705 return false; 1706 1707 // Check if the compared register follows the order we want. Fix if needed. 1708 for (RegisterInductionSet::iterator I = IndRegs.begin(), E = IndRegs.end(); 1709 I != E; ++I) { 1710 // This is a success. If the register used in the comparison is one that 1711 // we have identified as a bumped (updated) induction register, there is 1712 // nothing to do. 1713 if (CmpRegs.count(I->first)) 1714 return true; 1715 1716 // Otherwise, if the register being compared comes out of a PHI node, 1717 // and has been recognized as following the induction pattern, and is 1718 // compared against an immediate, we can fix it. 1719 const RegisterBump &RB = I->second; 1720 if (CmpRegs.count(RB.first)) { 1721 if (!CmpImmOp) { 1722 // If both operands to the compare instruction are registers, see if 1723 // it can be changed to use induction register as one of the operands. 1724 MachineInstr *IndI = nullptr; 1725 MachineInstr *nonIndI = nullptr; 1726 MachineOperand *IndMO = nullptr; 1727 MachineOperand *nonIndMO = nullptr; 1728 1729 for (unsigned i = 1, n = PredDef->getNumOperands(); i < n; ++i) { 1730 MachineOperand &MO = PredDef->getOperand(i); 1731 if (MO.isReg() && MO.getReg() == RB.first) { 1732 DEBUG(dbgs() << "\n DefMI(" << i << ") = " 1733 << *(MRI->getVRegDef(I->first))); 1734 if (IndI) 1735 return false; 1736 1737 IndI = MRI->getVRegDef(I->first); 1738 IndMO = &MO; 1739 } else if (MO.isReg()) { 1740 DEBUG(dbgs() << "\n DefMI(" << i << ") = " 1741 << *(MRI->getVRegDef(MO.getReg()))); 1742 if (nonIndI) 1743 return false; 1744 1745 nonIndI = MRI->getVRegDef(MO.getReg()); 1746 nonIndMO = &MO; 1747 } 1748 } 1749 if (IndI && nonIndI && 1750 nonIndI->getOpcode() == Hexagon::A2_addi && 1751 nonIndI->getOperand(2).isImm() && 1752 nonIndI->getOperand(2).getImm() == - RB.second) { 1753 bool Order = orderBumpCompare(IndI, PredDef); 1754 if (Order) { 1755 IndMO->setReg(I->first); 1756 nonIndMO->setReg(nonIndI->getOperand(1).getReg()); 1757 return true; 1758 } 1759 } 1760 return false; 1761 } 1762 1763 // It is not valid to do this transformation on an unsigned comparison 1764 // because it may underflow. 1765 Comparison::Kind Cmp = getComparisonKind(PredDef->getOpcode(), 0, 0, 0); 1766 if (!Cmp || Comparison::isUnsigned(Cmp)) 1767 return false; 1768 1769 // If the register is being compared against an immediate, try changing 1770 // the compare instruction to use induction register and adjust the 1771 // immediate operand. 1772 int64_t CmpImm = getImmediate(*CmpImmOp); 1773 int64_t V = RB.second; 1774 // Handle Overflow (64-bit). 1775 if (((V > 0) && (CmpImm > INT64_MAX - V)) || 1776 ((V < 0) && (CmpImm < INT64_MIN - V))) 1777 return false; 1778 CmpImm += V; 1779 // Most comparisons of register against an immediate value allow 1780 // the immediate to be constant-extended. There are some exceptions 1781 // though. Make sure the new combination will work. 1782 if (CmpImmOp->isImm()) 1783 if (!isImmValidForOpcode(PredDef->getOpcode(), CmpImm)) 1784 return false; 1785 1786 // Make sure that the compare happens after the bump. Otherwise, 1787 // after the fixup, the compare would use a yet-undefined register. 1788 MachineInstr *BumpI = MRI->getVRegDef(I->first); 1789 bool Order = orderBumpCompare(BumpI, PredDef); 1790 if (!Order) 1791 return false; 1792 1793 // Finally, fix the compare instruction. 1794 setImmediate(*CmpImmOp, CmpImm); 1795 for (unsigned i = 0, n = PredDef->getNumOperands(); i < n; ++i) { 1796 MachineOperand &MO = PredDef->getOperand(i); 1797 if (MO.isReg() && MO.getReg() == RB.first) { 1798 MO.setReg(I->first); 1799 return true; 1800 } 1801 } 1802 } 1803 } 1804 1805 return false; 1806} 1807 1808/// \brief Create a preheader for a given loop. 1809MachineBasicBlock *HexagonHardwareLoops::createPreheaderForLoop( 1810 MachineLoop *L) { 1811 if (MachineBasicBlock *TmpPH = L->getLoopPreheader()) 1812 return TmpPH; 1813 1814 if (!HWCreatePreheader) 1815 return nullptr; 1816 1817 MachineBasicBlock *Header = L->getHeader(); 1818 MachineBasicBlock *Latch = L->getLoopLatch(); 1819 MachineBasicBlock *ExitingBlock = getExitingBlock(L); 1820 MachineFunction *MF = Header->getParent(); 1821 DebugLoc DL; 1822 1823#ifndef NDEBUG 1824 if ((PHFn != "") && (PHFn != MF->getName())) 1825 return nullptr; 1826#endif 1827 1828 if (!Latch || !ExitingBlock || Header->hasAddressTaken()) 1829 return nullptr; 1830 1831 typedef MachineBasicBlock::instr_iterator instr_iterator; 1832 1833 // Verify that all existing predecessors have analyzable branches 1834 // (or no branches at all). 1835 typedef std::vector<MachineBasicBlock*> MBBVector; 1836 MBBVector Preds(Header->pred_begin(), Header->pred_end()); 1837 SmallVector<MachineOperand,2> Tmp1; 1838 MachineBasicBlock *TB = nullptr, *FB = nullptr; 1839 1840 if (TII->AnalyzeBranch(*ExitingBlock, TB, FB, Tmp1, false)) 1841 return nullptr; 1842 1843 for (MBBVector::iterator I = Preds.begin(), E = Preds.end(); I != E; ++I) { 1844 MachineBasicBlock *PB = *I; 1845 bool NotAnalyzed = TII->AnalyzeBranch(*PB, TB, FB, Tmp1, false); 1846 if (NotAnalyzed) 1847 return nullptr; 1848 } 1849 1850 MachineBasicBlock *NewPH = MF->CreateMachineBasicBlock(); 1851 MF->insert(Header->getIterator(), NewPH); 1852 1853 if (Header->pred_size() > 2) { 1854 // Ensure that the header has only two predecessors: the preheader and 1855 // the loop latch. Any additional predecessors of the header should 1856 // join at the newly created preheader. Inspect all PHI nodes from the 1857 // header and create appropriate corresponding PHI nodes in the preheader. 1858 1859 for (instr_iterator I = Header->instr_begin(), E = Header->instr_end(); 1860 I != E && I->isPHI(); ++I) { 1861 MachineInstr *PN = &*I; 1862 1863 const MCInstrDesc &PD = TII->get(TargetOpcode::PHI); 1864 MachineInstr *NewPN = MF->CreateMachineInstr(PD, DL); 1865 NewPH->insert(NewPH->end(), NewPN); 1866 1867 unsigned PR = PN->getOperand(0).getReg(); 1868 const TargetRegisterClass *RC = MRI->getRegClass(PR); 1869 unsigned NewPR = MRI->createVirtualRegister(RC); 1870 NewPN->addOperand(MachineOperand::CreateReg(NewPR, true)); 1871 1872 // Copy all non-latch operands of a header's PHI node to the newly 1873 // created PHI node in the preheader. 1874 for (unsigned i = 1, n = PN->getNumOperands(); i < n; i += 2) { 1875 unsigned PredR = PN->getOperand(i).getReg(); 1876 unsigned PredRSub = PN->getOperand(i).getSubReg(); 1877 MachineBasicBlock *PredB = PN->getOperand(i+1).getMBB(); 1878 if (PredB == Latch) 1879 continue; 1880 1881 MachineOperand MO = MachineOperand::CreateReg(PredR, false); 1882 MO.setSubReg(PredRSub); 1883 NewPN->addOperand(MO); 1884 NewPN->addOperand(MachineOperand::CreateMBB(PredB)); 1885 } 1886 1887 // Remove copied operands from the old PHI node and add the value 1888 // coming from the preheader's PHI. 1889 for (int i = PN->getNumOperands()-2; i > 0; i -= 2) { 1890 MachineBasicBlock *PredB = PN->getOperand(i+1).getMBB(); 1891 if (PredB != Latch) { 1892 PN->RemoveOperand(i+1); 1893 PN->RemoveOperand(i); 1894 } 1895 } 1896 PN->addOperand(MachineOperand::CreateReg(NewPR, false)); 1897 PN->addOperand(MachineOperand::CreateMBB(NewPH)); 1898 } 1899 1900 } else { 1901 assert(Header->pred_size() == 2); 1902 1903 // The header has only two predecessors, but the non-latch predecessor 1904 // is not a preheader (e.g. it has other successors, etc.) 1905 // In such a case we don't need any extra PHI nodes in the new preheader, 1906 // all we need is to adjust existing PHIs in the header to now refer to 1907 // the new preheader. 1908 for (instr_iterator I = Header->instr_begin(), E = Header->instr_end(); 1909 I != E && I->isPHI(); ++I) { 1910 MachineInstr *PN = &*I; 1911 for (unsigned i = 1, n = PN->getNumOperands(); i < n; i += 2) { 1912 MachineOperand &MO = PN->getOperand(i+1); 1913 if (MO.getMBB() != Latch) 1914 MO.setMBB(NewPH); 1915 } 1916 } 1917 } 1918 1919 // "Reroute" the CFG edges to link in the new preheader. 1920 // If any of the predecessors falls through to the header, insert a branch 1921 // to the new preheader in that place. 1922 SmallVector<MachineOperand,1> Tmp2; 1923 SmallVector<MachineOperand,1> EmptyCond; 1924 1925 TB = FB = nullptr; 1926 1927 for (MBBVector::iterator I = Preds.begin(), E = Preds.end(); I != E; ++I) { 1928 MachineBasicBlock *PB = *I; 1929 if (PB != Latch) { 1930 Tmp2.clear(); 1931 bool NotAnalyzed = TII->AnalyzeBranch(*PB, TB, FB, Tmp2, false); 1932 (void)NotAnalyzed; // suppress compiler warning 1933 assert (!NotAnalyzed && "Should be analyzable!"); 1934 if (TB != Header && (Tmp2.empty() || FB != Header)) 1935 TII->InsertBranch(*PB, NewPH, nullptr, EmptyCond, DL); 1936 PB->ReplaceUsesOfBlockWith(Header, NewPH); 1937 } 1938 } 1939 1940 // It can happen that the latch block will fall through into the header. 1941 // Insert an unconditional branch to the header. 1942 TB = FB = nullptr; 1943 bool LatchNotAnalyzed = TII->AnalyzeBranch(*Latch, TB, FB, Tmp2, false); 1944 (void)LatchNotAnalyzed; // suppress compiler warning 1945 assert (!LatchNotAnalyzed && "Should be analyzable!"); 1946 if (!TB && !FB) 1947 TII->InsertBranch(*Latch, Header, nullptr, EmptyCond, DL); 1948 1949 // Finally, the branch from the preheader to the header. 1950 TII->InsertBranch(*NewPH, Header, nullptr, EmptyCond, DL); 1951 NewPH->addSuccessor(Header); 1952 1953 MachineLoop *ParentLoop = L->getParentLoop(); 1954 if (ParentLoop) 1955 ParentLoop->addBasicBlockToLoop(NewPH, MLI->getBase()); 1956 1957 // Update the dominator information with the new preheader. 1958 if (MDT) { 1959 MachineDomTreeNode *HDom = MDT->getNode(Header); 1960 MDT->addNewBlock(NewPH, HDom->getIDom()->getBlock()); 1961 MDT->changeImmediateDominator(Header, NewPH); 1962 } 1963 1964 return NewPH; 1965} 1966