1 //===-- SelectionDAGISel.cpp - Implement the SelectionDAGISel class -------===// 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 implements the SelectionDAGISel class. 11 // 12 //===----------------------------------------------------------------------===// 13 14 #define DEBUG_TYPE "isel" 15 #include "ScheduleDAGSDNodes.h" 16 #include "SelectionDAGBuilder.h" 17 #include "llvm/CodeGen/FunctionLoweringInfo.h" 18 #include "llvm/CodeGen/SelectionDAGISel.h" 19 #include "llvm/Analysis/AliasAnalysis.h" 20 #include "llvm/Analysis/BranchProbabilityInfo.h" 21 #include "llvm/Analysis/DebugInfo.h" 22 #include "llvm/Constants.h" 23 #include "llvm/Function.h" 24 #include "llvm/InlineAsm.h" 25 #include "llvm/Instructions.h" 26 #include "llvm/Intrinsics.h" 27 #include "llvm/IntrinsicInst.h" 28 #include "llvm/LLVMContext.h" 29 #include "llvm/Module.h" 30 #include "llvm/CodeGen/FastISel.h" 31 #include "llvm/CodeGen/GCStrategy.h" 32 #include "llvm/CodeGen/GCMetadata.h" 33 #include "llvm/CodeGen/MachineFrameInfo.h" 34 #include "llvm/CodeGen/MachineFunction.h" 35 #include "llvm/CodeGen/MachineInstrBuilder.h" 36 #include "llvm/CodeGen/MachineModuleInfo.h" 37 #include "llvm/CodeGen/MachineRegisterInfo.h" 38 #include "llvm/CodeGen/ScheduleHazardRecognizer.h" 39 #include "llvm/CodeGen/SchedulerRegistry.h" 40 #include "llvm/CodeGen/SelectionDAG.h" 41 #include "llvm/Target/TargetRegisterInfo.h" 42 #include "llvm/Target/TargetIntrinsicInfo.h" 43 #include "llvm/Target/TargetInstrInfo.h" 44 #include "llvm/Target/TargetLowering.h" 45 #include "llvm/Target/TargetMachine.h" 46 #include "llvm/Target/TargetOptions.h" 47 #include "llvm/Transforms/Utils/BasicBlockUtils.h" 48 #include "llvm/Support/Compiler.h" 49 #include "llvm/Support/Debug.h" 50 #include "llvm/Support/ErrorHandling.h" 51 #include "llvm/Support/Timer.h" 52 #include "llvm/Support/raw_ostream.h" 53 #include "llvm/ADT/PostOrderIterator.h" 54 #include "llvm/ADT/Statistic.h" 55 #include <algorithm> 56 using namespace llvm; 57 58 STATISTIC(NumFastIselFailures, "Number of instructions fast isel failed on"); 59 STATISTIC(NumFastIselSuccess, "Number of instructions fast isel selected"); 60 STATISTIC(NumFastIselBlocks, "Number of blocks selected entirely by fast isel"); 61 STATISTIC(NumDAGBlocks, "Number of blocks selected using DAG"); 62 STATISTIC(NumDAGIselRetries,"Number of times dag isel has to try another path"); 63 64 static cl::opt<bool> 65 EnableFastISelVerbose("fast-isel-verbose", cl::Hidden, 66 cl::desc("Enable verbose messages in the \"fast\" " 67 "instruction selector")); 68 static cl::opt<bool> 69 EnableFastISelAbort("fast-isel-abort", cl::Hidden, 70 cl::desc("Enable abort calls when \"fast\" instruction fails")); 71 72 static cl::opt<bool> 73 UseMBPI("use-mbpi", 74 cl::desc("use Machine Branch Probability Info"), 75 cl::init(true), cl::Hidden); 76 77 #ifndef NDEBUG 78 static cl::opt<bool> 79 ViewDAGCombine1("view-dag-combine1-dags", cl::Hidden, 80 cl::desc("Pop up a window to show dags before the first " 81 "dag combine pass")); 82 static cl::opt<bool> 83 ViewLegalizeTypesDAGs("view-legalize-types-dags", cl::Hidden, 84 cl::desc("Pop up a window to show dags before legalize types")); 85 static cl::opt<bool> 86 ViewLegalizeDAGs("view-legalize-dags", cl::Hidden, 87 cl::desc("Pop up a window to show dags before legalize")); 88 static cl::opt<bool> 89 ViewDAGCombine2("view-dag-combine2-dags", cl::Hidden, 90 cl::desc("Pop up a window to show dags before the second " 91 "dag combine pass")); 92 static cl::opt<bool> 93 ViewDAGCombineLT("view-dag-combine-lt-dags", cl::Hidden, 94 cl::desc("Pop up a window to show dags before the post legalize types" 95 " dag combine pass")); 96 static cl::opt<bool> 97 ViewISelDAGs("view-isel-dags", cl::Hidden, 98 cl::desc("Pop up a window to show isel dags as they are selected")); 99 static cl::opt<bool> 100 ViewSchedDAGs("view-sched-dags", cl::Hidden, 101 cl::desc("Pop up a window to show sched dags as they are processed")); 102 static cl::opt<bool> 103 ViewSUnitDAGs("view-sunit-dags", cl::Hidden, 104 cl::desc("Pop up a window to show SUnit dags after they are processed")); 105 #else 106 static const bool ViewDAGCombine1 = false, 107 ViewLegalizeTypesDAGs = false, ViewLegalizeDAGs = false, 108 ViewDAGCombine2 = false, 109 ViewDAGCombineLT = false, 110 ViewISelDAGs = false, ViewSchedDAGs = false, 111 ViewSUnitDAGs = false; 112 #endif 113 114 //===---------------------------------------------------------------------===// 115 /// 116 /// RegisterScheduler class - Track the registration of instruction schedulers. 117 /// 118 //===---------------------------------------------------------------------===// 119 MachinePassRegistry RegisterScheduler::Registry; 120 121 //===---------------------------------------------------------------------===// 122 /// 123 /// ISHeuristic command line option for instruction schedulers. 124 /// 125 //===---------------------------------------------------------------------===// 126 static cl::opt<RegisterScheduler::FunctionPassCtor, false, 127 RegisterPassParser<RegisterScheduler> > 128 ISHeuristic("pre-RA-sched", 129 cl::init(&createDefaultScheduler), 130 cl::desc("Instruction schedulers available (before register" 131 " allocation):")); 132 133 static RegisterScheduler 134 defaultListDAGScheduler("default", "Best scheduler for the target", 135 createDefaultScheduler); 136 137 namespace llvm { 138 //===--------------------------------------------------------------------===// 139 /// createDefaultScheduler - This creates an instruction scheduler appropriate 140 /// for the target. 141 ScheduleDAGSDNodes* createDefaultScheduler(SelectionDAGISel *IS, 142 CodeGenOpt::Level OptLevel) { 143 const TargetLowering &TLI = IS->getTargetLowering(); 144 145 if (OptLevel == CodeGenOpt::None) 146 return createSourceListDAGScheduler(IS, OptLevel); 147 if (TLI.getSchedulingPreference() == Sched::Latency) 148 return createTDListDAGScheduler(IS, OptLevel); 149 if (TLI.getSchedulingPreference() == Sched::RegPressure) 150 return createBURRListDAGScheduler(IS, OptLevel); 151 if (TLI.getSchedulingPreference() == Sched::Hybrid) 152 return createHybridListDAGScheduler(IS, OptLevel); 153 assert(TLI.getSchedulingPreference() == Sched::ILP && 154 "Unknown sched type!"); 155 return createILPListDAGScheduler(IS, OptLevel); 156 } 157 } 158 159 // EmitInstrWithCustomInserter - This method should be implemented by targets 160 // that mark instructions with the 'usesCustomInserter' flag. These 161 // instructions are special in various ways, which require special support to 162 // insert. The specified MachineInstr is created but not inserted into any 163 // basic blocks, and this method is called to expand it into a sequence of 164 // instructions, potentially also creating new basic blocks and control flow. 165 // When new basic blocks are inserted and the edges from MBB to its successors 166 // are modified, the method should insert pairs of <OldSucc, NewSucc> into the 167 // DenseMap. 168 MachineBasicBlock * 169 TargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI, 170 MachineBasicBlock *MBB) const { 171 #ifndef NDEBUG 172 dbgs() << "If a target marks an instruction with " 173 "'usesCustomInserter', it must implement " 174 "TargetLowering::EmitInstrWithCustomInserter!"; 175 #endif 176 llvm_unreachable(0); 177 return 0; 178 } 179 180 //===----------------------------------------------------------------------===// 181 // SelectionDAGISel code 182 //===----------------------------------------------------------------------===// 183 184 SelectionDAGISel::SelectionDAGISel(const TargetMachine &tm, 185 CodeGenOpt::Level OL) : 186 MachineFunctionPass(ID), TM(tm), TLI(*tm.getTargetLowering()), 187 FuncInfo(new FunctionLoweringInfo(TLI)), 188 CurDAG(new SelectionDAG(tm)), 189 SDB(new SelectionDAGBuilder(*CurDAG, *FuncInfo, OL)), 190 GFI(), 191 OptLevel(OL), 192 DAGSize(0) { 193 initializeGCModuleInfoPass(*PassRegistry::getPassRegistry()); 194 initializeAliasAnalysisAnalysisGroup(*PassRegistry::getPassRegistry()); 195 initializeBranchProbabilityInfoPass(*PassRegistry::getPassRegistry()); 196 } 197 198 SelectionDAGISel::~SelectionDAGISel() { 199 delete SDB; 200 delete CurDAG; 201 delete FuncInfo; 202 } 203 204 void SelectionDAGISel::getAnalysisUsage(AnalysisUsage &AU) const { 205 AU.addRequired<AliasAnalysis>(); 206 AU.addPreserved<AliasAnalysis>(); 207 AU.addRequired<GCModuleInfo>(); 208 AU.addPreserved<GCModuleInfo>(); 209 if (UseMBPI && OptLevel != CodeGenOpt::None) 210 AU.addRequired<BranchProbabilityInfo>(); 211 MachineFunctionPass::getAnalysisUsage(AU); 212 } 213 214 /// SplitCriticalSideEffectEdges - Look for critical edges with a PHI value that 215 /// may trap on it. In this case we have to split the edge so that the path 216 /// through the predecessor block that doesn't go to the phi block doesn't 217 /// execute the possibly trapping instruction. 218 /// 219 /// This is required for correctness, so it must be done at -O0. 220 /// 221 static void SplitCriticalSideEffectEdges(Function &Fn, Pass *SDISel) { 222 // Loop for blocks with phi nodes. 223 for (Function::iterator BB = Fn.begin(), E = Fn.end(); BB != E; ++BB) { 224 PHINode *PN = dyn_cast<PHINode>(BB->begin()); 225 if (PN == 0) continue; 226 227 ReprocessBlock: 228 // For each block with a PHI node, check to see if any of the input values 229 // are potentially trapping constant expressions. Constant expressions are 230 // the only potentially trapping value that can occur as the argument to a 231 // PHI. 232 for (BasicBlock::iterator I = BB->begin(); (PN = dyn_cast<PHINode>(I)); ++I) 233 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) { 234 ConstantExpr *CE = dyn_cast<ConstantExpr>(PN->getIncomingValue(i)); 235 if (CE == 0 || !CE->canTrap()) continue; 236 237 // The only case we have to worry about is when the edge is critical. 238 // Since this block has a PHI Node, we assume it has multiple input 239 // edges: check to see if the pred has multiple successors. 240 BasicBlock *Pred = PN->getIncomingBlock(i); 241 if (Pred->getTerminator()->getNumSuccessors() == 1) 242 continue; 243 244 // Okay, we have to split this edge. 245 SplitCriticalEdge(Pred->getTerminator(), 246 GetSuccessorNumber(Pred, BB), SDISel, true); 247 goto ReprocessBlock; 248 } 249 } 250 } 251 252 bool SelectionDAGISel::runOnMachineFunction(MachineFunction &mf) { 253 // Do some sanity-checking on the command-line options. 254 assert((!EnableFastISelVerbose || EnableFastISel) && 255 "-fast-isel-verbose requires -fast-isel"); 256 assert((!EnableFastISelAbort || EnableFastISel) && 257 "-fast-isel-abort requires -fast-isel"); 258 259 const Function &Fn = *mf.getFunction(); 260 const TargetInstrInfo &TII = *TM.getInstrInfo(); 261 const TargetRegisterInfo &TRI = *TM.getRegisterInfo(); 262 263 MF = &mf; 264 RegInfo = &MF->getRegInfo(); 265 AA = &getAnalysis<AliasAnalysis>(); 266 GFI = Fn.hasGC() ? &getAnalysis<GCModuleInfo>().getFunctionInfo(Fn) : 0; 267 268 DEBUG(dbgs() << "\n\n\n=== " << Fn.getName() << "\n"); 269 270 SplitCriticalSideEffectEdges(const_cast<Function&>(Fn), this); 271 272 CurDAG->init(*MF); 273 FuncInfo->set(Fn, *MF); 274 275 if (UseMBPI && OptLevel != CodeGenOpt::None) 276 FuncInfo->BPI = &getAnalysis<BranchProbabilityInfo>(); 277 else 278 FuncInfo->BPI = 0; 279 280 SDB->init(GFI, *AA); 281 282 SelectAllBasicBlocks(Fn); 283 284 // If the first basic block in the function has live ins that need to be 285 // copied into vregs, emit the copies into the top of the block before 286 // emitting the code for the block. 287 MachineBasicBlock *EntryMBB = MF->begin(); 288 RegInfo->EmitLiveInCopies(EntryMBB, TRI, TII); 289 290 DenseMap<unsigned, unsigned> LiveInMap; 291 if (!FuncInfo->ArgDbgValues.empty()) 292 for (MachineRegisterInfo::livein_iterator LI = RegInfo->livein_begin(), 293 E = RegInfo->livein_end(); LI != E; ++LI) 294 if (LI->second) 295 LiveInMap.insert(std::make_pair(LI->first, LI->second)); 296 297 // Insert DBG_VALUE instructions for function arguments to the entry block. 298 for (unsigned i = 0, e = FuncInfo->ArgDbgValues.size(); i != e; ++i) { 299 MachineInstr *MI = FuncInfo->ArgDbgValues[e-i-1]; 300 unsigned Reg = MI->getOperand(0).getReg(); 301 if (TargetRegisterInfo::isPhysicalRegister(Reg)) 302 EntryMBB->insert(EntryMBB->begin(), MI); 303 else { 304 MachineInstr *Def = RegInfo->getVRegDef(Reg); 305 MachineBasicBlock::iterator InsertPos = Def; 306 // FIXME: VR def may not be in entry block. 307 Def->getParent()->insert(llvm::next(InsertPos), MI); 308 } 309 310 // If Reg is live-in then update debug info to track its copy in a vreg. 311 DenseMap<unsigned, unsigned>::iterator LDI = LiveInMap.find(Reg); 312 if (LDI != LiveInMap.end()) { 313 MachineInstr *Def = RegInfo->getVRegDef(LDI->second); 314 MachineBasicBlock::iterator InsertPos = Def; 315 const MDNode *Variable = 316 MI->getOperand(MI->getNumOperands()-1).getMetadata(); 317 unsigned Offset = MI->getOperand(1).getImm(); 318 // Def is never a terminator here, so it is ok to increment InsertPos. 319 BuildMI(*EntryMBB, ++InsertPos, MI->getDebugLoc(), 320 TII.get(TargetOpcode::DBG_VALUE)) 321 .addReg(LDI->second, RegState::Debug) 322 .addImm(Offset).addMetadata(Variable); 323 324 // If this vreg is directly copied into an exported register then 325 // that COPY instructions also need DBG_VALUE, if it is the only 326 // user of LDI->second. 327 MachineInstr *CopyUseMI = NULL; 328 for (MachineRegisterInfo::use_iterator 329 UI = RegInfo->use_begin(LDI->second); 330 MachineInstr *UseMI = UI.skipInstruction();) { 331 if (UseMI->isDebugValue()) continue; 332 if (UseMI->isCopy() && !CopyUseMI && UseMI->getParent() == EntryMBB) { 333 CopyUseMI = UseMI; continue; 334 } 335 // Otherwise this is another use or second copy use. 336 CopyUseMI = NULL; break; 337 } 338 if (CopyUseMI) { 339 MachineInstr *NewMI = 340 BuildMI(*MF, CopyUseMI->getDebugLoc(), 341 TII.get(TargetOpcode::DBG_VALUE)) 342 .addReg(CopyUseMI->getOperand(0).getReg(), RegState::Debug) 343 .addImm(Offset).addMetadata(Variable); 344 EntryMBB->insertAfter(CopyUseMI, NewMI); 345 } 346 } 347 } 348 349 // Determine if there are any calls in this machine function. 350 MachineFrameInfo *MFI = MF->getFrameInfo(); 351 if (!MFI->hasCalls()) { 352 for (MachineFunction::const_iterator 353 I = MF->begin(), E = MF->end(); I != E; ++I) { 354 const MachineBasicBlock *MBB = I; 355 for (MachineBasicBlock::const_iterator 356 II = MBB->begin(), IE = MBB->end(); II != IE; ++II) { 357 const MCInstrDesc &MCID = TM.getInstrInfo()->get(II->getOpcode()); 358 359 if ((MCID.isCall() && !MCID.isReturn()) || 360 II->isStackAligningInlineAsm()) { 361 MFI->setHasCalls(true); 362 goto done; 363 } 364 } 365 } 366 done:; 367 } 368 369 // Determine if there is a call to setjmp in the machine function. 370 MF->setCallsSetJmp(Fn.callsFunctionThatReturnsTwice()); 371 372 // Replace forward-declared registers with the registers containing 373 // the desired value. 374 MachineRegisterInfo &MRI = MF->getRegInfo(); 375 for (DenseMap<unsigned, unsigned>::iterator 376 I = FuncInfo->RegFixups.begin(), E = FuncInfo->RegFixups.end(); 377 I != E; ++I) { 378 unsigned From = I->first; 379 unsigned To = I->second; 380 // If To is also scheduled to be replaced, find what its ultimate 381 // replacement is. 382 for (;;) { 383 DenseMap<unsigned, unsigned>::iterator J = 384 FuncInfo->RegFixups.find(To); 385 if (J == E) break; 386 To = J->second; 387 } 388 // Replace it. 389 MRI.replaceRegWith(From, To); 390 } 391 392 // Release function-specific state. SDB and CurDAG are already cleared 393 // at this point. 394 FuncInfo->clear(); 395 396 return true; 397 } 398 399 void SelectionDAGISel::SelectBasicBlock(BasicBlock::const_iterator Begin, 400 BasicBlock::const_iterator End, 401 bool &HadTailCall) { 402 // Lower all of the non-terminator instructions. If a call is emitted 403 // as a tail call, cease emitting nodes for this block. Terminators 404 // are handled below. 405 for (BasicBlock::const_iterator I = Begin; I != End && !SDB->HasTailCall; ++I) 406 SDB->visit(*I); 407 408 // Make sure the root of the DAG is up-to-date. 409 CurDAG->setRoot(SDB->getControlRoot()); 410 HadTailCall = SDB->HasTailCall; 411 SDB->clear(); 412 413 // Final step, emit the lowered DAG as machine code. 414 CodeGenAndEmitDAG(); 415 } 416 417 void SelectionDAGISel::ComputeLiveOutVRegInfo() { 418 SmallPtrSet<SDNode*, 128> VisitedNodes; 419 SmallVector<SDNode*, 128> Worklist; 420 421 Worklist.push_back(CurDAG->getRoot().getNode()); 422 423 APInt Mask; 424 APInt KnownZero; 425 APInt KnownOne; 426 427 do { 428 SDNode *N = Worklist.pop_back_val(); 429 430 // If we've already seen this node, ignore it. 431 if (!VisitedNodes.insert(N)) 432 continue; 433 434 // Otherwise, add all chain operands to the worklist. 435 for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) 436 if (N->getOperand(i).getValueType() == MVT::Other) 437 Worklist.push_back(N->getOperand(i).getNode()); 438 439 // If this is a CopyToReg with a vreg dest, process it. 440 if (N->getOpcode() != ISD::CopyToReg) 441 continue; 442 443 unsigned DestReg = cast<RegisterSDNode>(N->getOperand(1))->getReg(); 444 if (!TargetRegisterInfo::isVirtualRegister(DestReg)) 445 continue; 446 447 // Ignore non-scalar or non-integer values. 448 SDValue Src = N->getOperand(2); 449 EVT SrcVT = Src.getValueType(); 450 if (!SrcVT.isInteger() || SrcVT.isVector()) 451 continue; 452 453 unsigned NumSignBits = CurDAG->ComputeNumSignBits(Src); 454 Mask = APInt::getAllOnesValue(SrcVT.getSizeInBits()); 455 CurDAG->ComputeMaskedBits(Src, Mask, KnownZero, KnownOne); 456 FuncInfo->AddLiveOutRegInfo(DestReg, NumSignBits, KnownZero, KnownOne); 457 } while (!Worklist.empty()); 458 } 459 460 void SelectionDAGISel::CodeGenAndEmitDAG() { 461 std::string GroupName; 462 if (TimePassesIsEnabled) 463 GroupName = "Instruction Selection and Scheduling"; 464 std::string BlockName; 465 int BlockNumber = -1; 466 #ifdef NDEBUG 467 if (ViewDAGCombine1 || ViewLegalizeTypesDAGs || ViewLegalizeDAGs || 468 ViewDAGCombine2 || ViewDAGCombineLT || ViewISelDAGs || ViewSchedDAGs || 469 ViewSUnitDAGs) 470 #endif 471 { 472 BlockNumber = FuncInfo->MBB->getNumber(); 473 BlockName = MF->getFunction()->getNameStr() + ":" + 474 FuncInfo->MBB->getBasicBlock()->getNameStr(); 475 } 476 DEBUG(dbgs() << "Initial selection DAG: BB#" << BlockNumber 477 << " '" << BlockName << "'\n"; CurDAG->dump()); 478 479 if (ViewDAGCombine1) CurDAG->viewGraph("dag-combine1 input for " + BlockName); 480 481 // Run the DAG combiner in pre-legalize mode. 482 { 483 NamedRegionTimer T("DAG Combining 1", GroupName, TimePassesIsEnabled); 484 CurDAG->Combine(Unrestricted, *AA, OptLevel); 485 } 486 487 DEBUG(dbgs() << "Optimized lowered selection DAG: BB#" << BlockNumber 488 << " '" << BlockName << "'\n"; CurDAG->dump()); 489 490 // Second step, hack on the DAG until it only uses operations and types that 491 // the target supports. 492 if (ViewLegalizeTypesDAGs) CurDAG->viewGraph("legalize-types input for " + 493 BlockName); 494 495 bool Changed; 496 { 497 NamedRegionTimer T("Type Legalization", GroupName, TimePassesIsEnabled); 498 Changed = CurDAG->LegalizeTypes(); 499 } 500 501 DEBUG(dbgs() << "Type-legalized selection DAG: BB#" << BlockNumber 502 << " '" << BlockName << "'\n"; CurDAG->dump()); 503 504 if (Changed) { 505 if (ViewDAGCombineLT) 506 CurDAG->viewGraph("dag-combine-lt input for " + BlockName); 507 508 // Run the DAG combiner in post-type-legalize mode. 509 { 510 NamedRegionTimer T("DAG Combining after legalize types", GroupName, 511 TimePassesIsEnabled); 512 CurDAG->Combine(NoIllegalTypes, *AA, OptLevel); 513 } 514 515 DEBUG(dbgs() << "Optimized type-legalized selection DAG: BB#" << BlockNumber 516 << " '" << BlockName << "'\n"; CurDAG->dump()); 517 } 518 519 { 520 NamedRegionTimer T("Vector Legalization", GroupName, TimePassesIsEnabled); 521 Changed = CurDAG->LegalizeVectors(); 522 } 523 524 if (Changed) { 525 { 526 NamedRegionTimer T("Type Legalization 2", GroupName, TimePassesIsEnabled); 527 CurDAG->LegalizeTypes(); 528 } 529 530 if (ViewDAGCombineLT) 531 CurDAG->viewGraph("dag-combine-lv input for " + BlockName); 532 533 // Run the DAG combiner in post-type-legalize mode. 534 { 535 NamedRegionTimer T("DAG Combining after legalize vectors", GroupName, 536 TimePassesIsEnabled); 537 CurDAG->Combine(NoIllegalOperations, *AA, OptLevel); 538 } 539 540 DEBUG(dbgs() << "Optimized vector-legalized selection DAG: BB#" 541 << BlockNumber << " '" << BlockName << "'\n"; CurDAG->dump()); 542 } 543 544 if (ViewLegalizeDAGs) CurDAG->viewGraph("legalize input for " + BlockName); 545 546 { 547 NamedRegionTimer T("DAG Legalization", GroupName, TimePassesIsEnabled); 548 CurDAG->Legalize(); 549 } 550 551 DEBUG(dbgs() << "Legalized selection DAG: BB#" << BlockNumber 552 << " '" << BlockName << "'\n"; CurDAG->dump()); 553 554 if (ViewDAGCombine2) CurDAG->viewGraph("dag-combine2 input for " + BlockName); 555 556 // Run the DAG combiner in post-legalize mode. 557 { 558 NamedRegionTimer T("DAG Combining 2", GroupName, TimePassesIsEnabled); 559 CurDAG->Combine(NoIllegalOperations, *AA, OptLevel); 560 } 561 562 DEBUG(dbgs() << "Optimized legalized selection DAG: BB#" << BlockNumber 563 << " '" << BlockName << "'\n"; CurDAG->dump()); 564 565 if (OptLevel != CodeGenOpt::None) 566 ComputeLiveOutVRegInfo(); 567 568 if (ViewISelDAGs) CurDAG->viewGraph("isel input for " + BlockName); 569 570 // Third, instruction select all of the operations to machine code, adding the 571 // code to the MachineBasicBlock. 572 { 573 NamedRegionTimer T("Instruction Selection", GroupName, TimePassesIsEnabled); 574 DoInstructionSelection(); 575 } 576 577 DEBUG(dbgs() << "Selected selection DAG: BB#" << BlockNumber 578 << " '" << BlockName << "'\n"; CurDAG->dump()); 579 580 if (ViewSchedDAGs) CurDAG->viewGraph("scheduler input for " + BlockName); 581 582 // Schedule machine code. 583 ScheduleDAGSDNodes *Scheduler = CreateScheduler(); 584 { 585 NamedRegionTimer T("Instruction Scheduling", GroupName, 586 TimePassesIsEnabled); 587 Scheduler->Run(CurDAG, FuncInfo->MBB, FuncInfo->InsertPt); 588 } 589 590 if (ViewSUnitDAGs) Scheduler->viewGraph(); 591 592 // Emit machine code to BB. This can change 'BB' to the last block being 593 // inserted into. 594 MachineBasicBlock *FirstMBB = FuncInfo->MBB, *LastMBB; 595 { 596 NamedRegionTimer T("Instruction Creation", GroupName, TimePassesIsEnabled); 597 598 LastMBB = FuncInfo->MBB = Scheduler->EmitSchedule(); 599 FuncInfo->InsertPt = Scheduler->InsertPos; 600 } 601 602 // If the block was split, make sure we update any references that are used to 603 // update PHI nodes later on. 604 if (FirstMBB != LastMBB) 605 SDB->UpdateSplitBlock(FirstMBB, LastMBB); 606 607 // Free the scheduler state. 608 { 609 NamedRegionTimer T("Instruction Scheduling Cleanup", GroupName, 610 TimePassesIsEnabled); 611 delete Scheduler; 612 } 613 614 // Free the SelectionDAG state, now that we're finished with it. 615 CurDAG->clear(); 616 } 617 618 void SelectionDAGISel::DoInstructionSelection() { 619 DEBUG(errs() << "===== Instruction selection begins: BB#" 620 << FuncInfo->MBB->getNumber() 621 << " '" << FuncInfo->MBB->getName() << "'\n"); 622 623 PreprocessISelDAG(); 624 625 // Select target instructions for the DAG. 626 { 627 // Number all nodes with a topological order and set DAGSize. 628 DAGSize = CurDAG->AssignTopologicalOrder(); 629 630 // Create a dummy node (which is not added to allnodes), that adds 631 // a reference to the root node, preventing it from being deleted, 632 // and tracking any changes of the root. 633 HandleSDNode Dummy(CurDAG->getRoot()); 634 ISelPosition = SelectionDAG::allnodes_iterator(CurDAG->getRoot().getNode()); 635 ++ISelPosition; 636 637 // The AllNodes list is now topological-sorted. Visit the 638 // nodes by starting at the end of the list (the root of the 639 // graph) and preceding back toward the beginning (the entry 640 // node). 641 while (ISelPosition != CurDAG->allnodes_begin()) { 642 SDNode *Node = --ISelPosition; 643 // Skip dead nodes. DAGCombiner is expected to eliminate all dead nodes, 644 // but there are currently some corner cases that it misses. Also, this 645 // makes it theoretically possible to disable the DAGCombiner. 646 if (Node->use_empty()) 647 continue; 648 649 SDNode *ResNode = Select(Node); 650 651 // FIXME: This is pretty gross. 'Select' should be changed to not return 652 // anything at all and this code should be nuked with a tactical strike. 653 654 // If node should not be replaced, continue with the next one. 655 if (ResNode == Node || Node->getOpcode() == ISD::DELETED_NODE) 656 continue; 657 // Replace node. 658 if (ResNode) 659 ReplaceUses(Node, ResNode); 660 661 // If after the replacement this node is not used any more, 662 // remove this dead node. 663 if (Node->use_empty()) { // Don't delete EntryToken, etc. 664 ISelUpdater ISU(ISelPosition); 665 CurDAG->RemoveDeadNode(Node, &ISU); 666 } 667 } 668 669 CurDAG->setRoot(Dummy.getValue()); 670 } 671 672 DEBUG(errs() << "===== Instruction selection ends:\n"); 673 674 PostprocessISelDAG(); 675 } 676 677 /// PrepareEHLandingPad - Emit an EH_LABEL, set up live-in registers, and 678 /// do other setup for EH landing-pad blocks. 679 void SelectionDAGISel::PrepareEHLandingPad() { 680 // Add a label to mark the beginning of the landing pad. Deletion of the 681 // landing pad can thus be detected via the MachineModuleInfo. 682 MCSymbol *Label = MF->getMMI().addLandingPad(FuncInfo->MBB); 683 684 const MCInstrDesc &II = TM.getInstrInfo()->get(TargetOpcode::EH_LABEL); 685 BuildMI(*FuncInfo->MBB, FuncInfo->InsertPt, SDB->getCurDebugLoc(), II) 686 .addSym(Label); 687 688 // Mark exception register as live in. 689 unsigned Reg = TLI.getExceptionAddressRegister(); 690 if (Reg) FuncInfo->MBB->addLiveIn(Reg); 691 692 // Mark exception selector register as live in. 693 Reg = TLI.getExceptionSelectorRegister(); 694 if (Reg) FuncInfo->MBB->addLiveIn(Reg); 695 696 // FIXME: Hack around an exception handling flaw (PR1508): the personality 697 // function and list of typeids logically belong to the invoke (or, if you 698 // like, the basic block containing the invoke), and need to be associated 699 // with it in the dwarf exception handling tables. Currently however the 700 // information is provided by an intrinsic (eh.selector) that can be moved 701 // to unexpected places by the optimizers: if the unwind edge is critical, 702 // then breaking it can result in the intrinsics being in the successor of 703 // the landing pad, not the landing pad itself. This results 704 // in exceptions not being caught because no typeids are associated with 705 // the invoke. This may not be the only way things can go wrong, but it 706 // is the only way we try to work around for the moment. 707 const BasicBlock *LLVMBB = FuncInfo->MBB->getBasicBlock(); 708 const BranchInst *Br = dyn_cast<BranchInst>(LLVMBB->getTerminator()); 709 710 if (Br && Br->isUnconditional()) { // Critical edge? 711 BasicBlock::const_iterator I, E; 712 for (I = LLVMBB->begin(), E = --LLVMBB->end(); I != E; ++I) 713 if (isa<EHSelectorInst>(I)) 714 break; 715 716 if (I == E) 717 // No catch info found - try to extract some from the successor. 718 CopyCatchInfo(Br->getSuccessor(0), LLVMBB, &MF->getMMI(), *FuncInfo); 719 } 720 } 721 722 723 724 /// TryToFoldFastISelLoad - We're checking to see if we can fold the specified 725 /// load into the specified FoldInst. Note that we could have a sequence where 726 /// multiple LLVM IR instructions are folded into the same machineinstr. For 727 /// example we could have: 728 /// A: x = load i32 *P 729 /// B: y = icmp A, 42 730 /// C: br y, ... 731 /// 732 /// In this scenario, LI is "A", and FoldInst is "C". We know about "B" (and 733 /// any other folded instructions) because it is between A and C. 734 /// 735 /// If we succeed in folding the load into the operation, return true. 736 /// 737 bool SelectionDAGISel::TryToFoldFastISelLoad(const LoadInst *LI, 738 const Instruction *FoldInst, 739 FastISel *FastIS) { 740 // We know that the load has a single use, but don't know what it is. If it 741 // isn't one of the folded instructions, then we can't succeed here. Handle 742 // this by scanning the single-use users of the load until we get to FoldInst. 743 unsigned MaxUsers = 6; // Don't scan down huge single-use chains of instrs. 744 745 const Instruction *TheUser = LI->use_back(); 746 while (TheUser != FoldInst && // Scan up until we find FoldInst. 747 // Stay in the right block. 748 TheUser->getParent() == FoldInst->getParent() && 749 --MaxUsers) { // Don't scan too far. 750 // If there are multiple or no uses of this instruction, then bail out. 751 if (!TheUser->hasOneUse()) 752 return false; 753 754 TheUser = TheUser->use_back(); 755 } 756 757 // Don't try to fold volatile loads. Target has to deal with alignment 758 // constraints. 759 if (LI->isVolatile()) return false; 760 761 // Figure out which vreg this is going into. If there is no assigned vreg yet 762 // then there actually was no reference to it. Perhaps the load is referenced 763 // by a dead instruction. 764 unsigned LoadReg = FastIS->getRegForValue(LI); 765 if (LoadReg == 0) 766 return false; 767 768 // Check to see what the uses of this vreg are. If it has no uses, or more 769 // than one use (at the machine instr level) then we can't fold it. 770 MachineRegisterInfo::reg_iterator RI = RegInfo->reg_begin(LoadReg); 771 if (RI == RegInfo->reg_end()) 772 return false; 773 774 // See if there is exactly one use of the vreg. If there are multiple uses, 775 // then the instruction got lowered to multiple machine instructions or the 776 // use of the loaded value ended up being multiple operands of the result, in 777 // either case, we can't fold this. 778 MachineRegisterInfo::reg_iterator PostRI = RI; ++PostRI; 779 if (PostRI != RegInfo->reg_end()) 780 return false; 781 782 assert(RI.getOperand().isUse() && 783 "The only use of the vreg must be a use, we haven't emitted the def!"); 784 785 MachineInstr *User = &*RI; 786 787 // Set the insertion point properly. Folding the load can cause generation of 788 // other random instructions (like sign extends) for addressing modes, make 789 // sure they get inserted in a logical place before the new instruction. 790 FuncInfo->InsertPt = User; 791 FuncInfo->MBB = User->getParent(); 792 793 // Ask the target to try folding the load. 794 return FastIS->TryToFoldLoad(User, RI.getOperandNo(), LI); 795 } 796 797 /// isFoldedOrDeadInstruction - Return true if the specified instruction is 798 /// side-effect free and is either dead or folded into a generated instruction. 799 /// Return false if it needs to be emitted. 800 static bool isFoldedOrDeadInstruction(const Instruction *I, 801 FunctionLoweringInfo *FuncInfo) { 802 return !I->mayWriteToMemory() && // Side-effecting instructions aren't folded. 803 !isa<TerminatorInst>(I) && // Terminators aren't folded. 804 !isa<DbgInfoIntrinsic>(I) && // Debug instructions aren't folded. 805 !FuncInfo->isExportedInst(I); // Exported instrs must be computed. 806 } 807 808 void SelectionDAGISel::SelectAllBasicBlocks(const Function &Fn) { 809 // Initialize the Fast-ISel state, if needed. 810 FastISel *FastIS = 0; 811 if (EnableFastISel) 812 FastIS = TLI.createFastISel(*FuncInfo); 813 814 // Iterate over all basic blocks in the function. 815 ReversePostOrderTraversal<const Function*> RPOT(&Fn); 816 for (ReversePostOrderTraversal<const Function*>::rpo_iterator 817 I = RPOT.begin(), E = RPOT.end(); I != E; ++I) { 818 const BasicBlock *LLVMBB = *I; 819 820 if (OptLevel != CodeGenOpt::None) { 821 bool AllPredsVisited = true; 822 for (const_pred_iterator PI = pred_begin(LLVMBB), PE = pred_end(LLVMBB); 823 PI != PE; ++PI) { 824 if (!FuncInfo->VisitedBBs.count(*PI)) { 825 AllPredsVisited = false; 826 break; 827 } 828 } 829 830 if (AllPredsVisited) { 831 for (BasicBlock::const_iterator I = LLVMBB->begin(); 832 isa<PHINode>(I); ++I) 833 FuncInfo->ComputePHILiveOutRegInfo(cast<PHINode>(I)); 834 } else { 835 for (BasicBlock::const_iterator I = LLVMBB->begin(); 836 isa<PHINode>(I); ++I) 837 FuncInfo->InvalidatePHILiveOutRegInfo(cast<PHINode>(I)); 838 } 839 840 FuncInfo->VisitedBBs.insert(LLVMBB); 841 } 842 843 FuncInfo->MBB = FuncInfo->MBBMap[LLVMBB]; 844 FuncInfo->InsertPt = FuncInfo->MBB->getFirstNonPHI(); 845 846 BasicBlock::const_iterator const Begin = LLVMBB->getFirstNonPHI(); 847 BasicBlock::const_iterator const End = LLVMBB->end(); 848 BasicBlock::const_iterator BI = End; 849 850 FuncInfo->InsertPt = FuncInfo->MBB->getFirstNonPHI(); 851 852 // Setup an EH landing-pad block. 853 if (FuncInfo->MBB->isLandingPad()) 854 PrepareEHLandingPad(); 855 856 // Lower any arguments needed in this block if this is the entry block. 857 if (LLVMBB == &Fn.getEntryBlock()) 858 LowerArguments(LLVMBB); 859 860 // Before doing SelectionDAG ISel, see if FastISel has been requested. 861 if (FastIS) { 862 FastIS->startNewBlock(); 863 864 // Emit code for any incoming arguments. This must happen before 865 // beginning FastISel on the entry block. 866 if (LLVMBB == &Fn.getEntryBlock()) { 867 CurDAG->setRoot(SDB->getControlRoot()); 868 SDB->clear(); 869 CodeGenAndEmitDAG(); 870 871 // If we inserted any instructions at the beginning, make a note of 872 // where they are, so we can be sure to emit subsequent instructions 873 // after them. 874 if (FuncInfo->InsertPt != FuncInfo->MBB->begin()) 875 FastIS->setLastLocalValue(llvm::prior(FuncInfo->InsertPt)); 876 else 877 FastIS->setLastLocalValue(0); 878 } 879 880 // Do FastISel on as many instructions as possible. 881 for (; BI != Begin; --BI) { 882 const Instruction *Inst = llvm::prior(BI); 883 884 // If we no longer require this instruction, skip it. 885 if (isFoldedOrDeadInstruction(Inst, FuncInfo)) 886 continue; 887 888 // Bottom-up: reset the insert pos at the top, after any local-value 889 // instructions. 890 FastIS->recomputeInsertPt(); 891 892 // Try to select the instruction with FastISel. 893 if (FastIS->SelectInstruction(Inst)) { 894 ++NumFastIselSuccess; 895 // If fast isel succeeded, skip over all the folded instructions, and 896 // then see if there is a load right before the selected instructions. 897 // Try to fold the load if so. 898 const Instruction *BeforeInst = Inst; 899 while (BeforeInst != Begin) { 900 BeforeInst = llvm::prior(BasicBlock::const_iterator(BeforeInst)); 901 if (!isFoldedOrDeadInstruction(BeforeInst, FuncInfo)) 902 break; 903 } 904 if (BeforeInst != Inst && isa<LoadInst>(BeforeInst) && 905 BeforeInst->hasOneUse() && 906 TryToFoldFastISelLoad(cast<LoadInst>(BeforeInst), Inst, FastIS)) 907 // If we succeeded, don't re-select the load. 908 BI = llvm::next(BasicBlock::const_iterator(BeforeInst)); 909 continue; 910 } 911 912 // Then handle certain instructions as single-LLVM-Instruction blocks. 913 if (isa<CallInst>(Inst)) { 914 ++NumFastIselFailures; 915 if (EnableFastISelVerbose || EnableFastISelAbort) { 916 dbgs() << "FastISel missed call: "; 917 Inst->dump(); 918 } 919 920 if (!Inst->getType()->isVoidTy() && !Inst->use_empty()) { 921 unsigned &R = FuncInfo->ValueMap[Inst]; 922 if (!R) 923 R = FuncInfo->CreateRegs(Inst->getType()); 924 } 925 926 bool HadTailCall = false; 927 SelectBasicBlock(Inst, BI, HadTailCall); 928 929 // If the call was emitted as a tail call, we're done with the block. 930 if (HadTailCall) { 931 --BI; 932 break; 933 } 934 935 continue; 936 } 937 938 if (isa<TerminatorInst>(Inst) && !isa<BranchInst>(Inst)) { 939 // Don't abort, and use a different message for terminator misses. 940 ++NumFastIselFailures; 941 if (EnableFastISelVerbose || EnableFastISelAbort) { 942 dbgs() << "FastISel missed terminator: "; 943 Inst->dump(); 944 } 945 } else { 946 ++NumFastIselFailures; 947 if (EnableFastISelVerbose || EnableFastISelAbort) { 948 dbgs() << "FastISel miss: "; 949 Inst->dump(); 950 } 951 if (EnableFastISelAbort) 952 // The "fast" selector couldn't handle something and bailed. 953 // For the purpose of debugging, just abort. 954 llvm_unreachable("FastISel didn't select the entire block"); 955 } 956 break; 957 } 958 959 FastIS->recomputeInsertPt(); 960 } 961 962 if (Begin != BI) 963 ++NumDAGBlocks; 964 else 965 ++NumFastIselBlocks; 966 967 if (Begin != BI) { 968 // Run SelectionDAG instruction selection on the remainder of the block 969 // not handled by FastISel. If FastISel is not run, this is the entire 970 // block. 971 bool HadTailCall; 972 SelectBasicBlock(Begin, BI, HadTailCall); 973 } 974 975 FinishBasicBlock(); 976 FuncInfo->PHINodesToUpdate.clear(); 977 } 978 979 delete FastIS; 980 SDB->clearDanglingDebugInfo(); 981 } 982 983 void 984 SelectionDAGISel::FinishBasicBlock() { 985 986 DEBUG(dbgs() << "Total amount of phi nodes to update: " 987 << FuncInfo->PHINodesToUpdate.size() << "\n"; 988 for (unsigned i = 0, e = FuncInfo->PHINodesToUpdate.size(); i != e; ++i) 989 dbgs() << "Node " << i << " : (" 990 << FuncInfo->PHINodesToUpdate[i].first 991 << ", " << FuncInfo->PHINodesToUpdate[i].second << ")\n"); 992 993 // Next, now that we know what the last MBB the LLVM BB expanded is, update 994 // PHI nodes in successors. 995 if (SDB->SwitchCases.empty() && 996 SDB->JTCases.empty() && 997 SDB->BitTestCases.empty()) { 998 for (unsigned i = 0, e = FuncInfo->PHINodesToUpdate.size(); i != e; ++i) { 999 MachineInstr *PHI = FuncInfo->PHINodesToUpdate[i].first; 1000 assert(PHI->isPHI() && 1001 "This is not a machine PHI node that we are updating!"); 1002 if (!FuncInfo->MBB->isSuccessor(PHI->getParent())) 1003 continue; 1004 PHI->addOperand( 1005 MachineOperand::CreateReg(FuncInfo->PHINodesToUpdate[i].second, false)); 1006 PHI->addOperand(MachineOperand::CreateMBB(FuncInfo->MBB)); 1007 } 1008 return; 1009 } 1010 1011 for (unsigned i = 0, e = SDB->BitTestCases.size(); i != e; ++i) { 1012 // Lower header first, if it wasn't already lowered 1013 if (!SDB->BitTestCases[i].Emitted) { 1014 // Set the current basic block to the mbb we wish to insert the code into 1015 FuncInfo->MBB = SDB->BitTestCases[i].Parent; 1016 FuncInfo->InsertPt = FuncInfo->MBB->end(); 1017 // Emit the code 1018 SDB->visitBitTestHeader(SDB->BitTestCases[i], FuncInfo->MBB); 1019 CurDAG->setRoot(SDB->getRoot()); 1020 SDB->clear(); 1021 CodeGenAndEmitDAG(); 1022 } 1023 1024 for (unsigned j = 0, ej = SDB->BitTestCases[i].Cases.size(); j != ej; ++j) { 1025 // Set the current basic block to the mbb we wish to insert the code into 1026 FuncInfo->MBB = SDB->BitTestCases[i].Cases[j].ThisBB; 1027 FuncInfo->InsertPt = FuncInfo->MBB->end(); 1028 // Emit the code 1029 if (j+1 != ej) 1030 SDB->visitBitTestCase(SDB->BitTestCases[i], 1031 SDB->BitTestCases[i].Cases[j+1].ThisBB, 1032 SDB->BitTestCases[i].Reg, 1033 SDB->BitTestCases[i].Cases[j], 1034 FuncInfo->MBB); 1035 else 1036 SDB->visitBitTestCase(SDB->BitTestCases[i], 1037 SDB->BitTestCases[i].Default, 1038 SDB->BitTestCases[i].Reg, 1039 SDB->BitTestCases[i].Cases[j], 1040 FuncInfo->MBB); 1041 1042 1043 CurDAG->setRoot(SDB->getRoot()); 1044 SDB->clear(); 1045 CodeGenAndEmitDAG(); 1046 } 1047 1048 // Update PHI Nodes 1049 for (unsigned pi = 0, pe = FuncInfo->PHINodesToUpdate.size(); 1050 pi != pe; ++pi) { 1051 MachineInstr *PHI = FuncInfo->PHINodesToUpdate[pi].first; 1052 MachineBasicBlock *PHIBB = PHI->getParent(); 1053 assert(PHI->isPHI() && 1054 "This is not a machine PHI node that we are updating!"); 1055 // This is "default" BB. We have two jumps to it. From "header" BB and 1056 // from last "case" BB. 1057 if (PHIBB == SDB->BitTestCases[i].Default) { 1058 PHI->addOperand(MachineOperand:: 1059 CreateReg(FuncInfo->PHINodesToUpdate[pi].second, 1060 false)); 1061 PHI->addOperand(MachineOperand::CreateMBB(SDB->BitTestCases[i].Parent)); 1062 PHI->addOperand(MachineOperand:: 1063 CreateReg(FuncInfo->PHINodesToUpdate[pi].second, 1064 false)); 1065 PHI->addOperand(MachineOperand::CreateMBB(SDB->BitTestCases[i].Cases. 1066 back().ThisBB)); 1067 } 1068 // One of "cases" BB. 1069 for (unsigned j = 0, ej = SDB->BitTestCases[i].Cases.size(); 1070 j != ej; ++j) { 1071 MachineBasicBlock* cBB = SDB->BitTestCases[i].Cases[j].ThisBB; 1072 if (cBB->isSuccessor(PHIBB)) { 1073 PHI->addOperand(MachineOperand:: 1074 CreateReg(FuncInfo->PHINodesToUpdate[pi].second, 1075 false)); 1076 PHI->addOperand(MachineOperand::CreateMBB(cBB)); 1077 } 1078 } 1079 } 1080 } 1081 SDB->BitTestCases.clear(); 1082 1083 // If the JumpTable record is filled in, then we need to emit a jump table. 1084 // Updating the PHI nodes is tricky in this case, since we need to determine 1085 // whether the PHI is a successor of the range check MBB or the jump table MBB 1086 for (unsigned i = 0, e = SDB->JTCases.size(); i != e; ++i) { 1087 // Lower header first, if it wasn't already lowered 1088 if (!SDB->JTCases[i].first.Emitted) { 1089 // Set the current basic block to the mbb we wish to insert the code into 1090 FuncInfo->MBB = SDB->JTCases[i].first.HeaderBB; 1091 FuncInfo->InsertPt = FuncInfo->MBB->end(); 1092 // Emit the code 1093 SDB->visitJumpTableHeader(SDB->JTCases[i].second, SDB->JTCases[i].first, 1094 FuncInfo->MBB); 1095 CurDAG->setRoot(SDB->getRoot()); 1096 SDB->clear(); 1097 CodeGenAndEmitDAG(); 1098 } 1099 1100 // Set the current basic block to the mbb we wish to insert the code into 1101 FuncInfo->MBB = SDB->JTCases[i].second.MBB; 1102 FuncInfo->InsertPt = FuncInfo->MBB->end(); 1103 // Emit the code 1104 SDB->visitJumpTable(SDB->JTCases[i].second); 1105 CurDAG->setRoot(SDB->getRoot()); 1106 SDB->clear(); 1107 CodeGenAndEmitDAG(); 1108 1109 // Update PHI Nodes 1110 for (unsigned pi = 0, pe = FuncInfo->PHINodesToUpdate.size(); 1111 pi != pe; ++pi) { 1112 MachineInstr *PHI = FuncInfo->PHINodesToUpdate[pi].first; 1113 MachineBasicBlock *PHIBB = PHI->getParent(); 1114 assert(PHI->isPHI() && 1115 "This is not a machine PHI node that we are updating!"); 1116 // "default" BB. We can go there only from header BB. 1117 if (PHIBB == SDB->JTCases[i].second.Default) { 1118 PHI->addOperand 1119 (MachineOperand::CreateReg(FuncInfo->PHINodesToUpdate[pi].second, 1120 false)); 1121 PHI->addOperand 1122 (MachineOperand::CreateMBB(SDB->JTCases[i].first.HeaderBB)); 1123 } 1124 // JT BB. Just iterate over successors here 1125 if (FuncInfo->MBB->isSuccessor(PHIBB)) { 1126 PHI->addOperand 1127 (MachineOperand::CreateReg(FuncInfo->PHINodesToUpdate[pi].second, 1128 false)); 1129 PHI->addOperand(MachineOperand::CreateMBB(FuncInfo->MBB)); 1130 } 1131 } 1132 } 1133 SDB->JTCases.clear(); 1134 1135 // If the switch block involved a branch to one of the actual successors, we 1136 // need to update PHI nodes in that block. 1137 for (unsigned i = 0, e = FuncInfo->PHINodesToUpdate.size(); i != e; ++i) { 1138 MachineInstr *PHI = FuncInfo->PHINodesToUpdate[i].first; 1139 assert(PHI->isPHI() && 1140 "This is not a machine PHI node that we are updating!"); 1141 if (FuncInfo->MBB->isSuccessor(PHI->getParent())) { 1142 PHI->addOperand( 1143 MachineOperand::CreateReg(FuncInfo->PHINodesToUpdate[i].second, false)); 1144 PHI->addOperand(MachineOperand::CreateMBB(FuncInfo->MBB)); 1145 } 1146 } 1147 1148 // If we generated any switch lowering information, build and codegen any 1149 // additional DAGs necessary. 1150 for (unsigned i = 0, e = SDB->SwitchCases.size(); i != e; ++i) { 1151 // Set the current basic block to the mbb we wish to insert the code into 1152 FuncInfo->MBB = SDB->SwitchCases[i].ThisBB; 1153 FuncInfo->InsertPt = FuncInfo->MBB->end(); 1154 1155 // Determine the unique successors. 1156 SmallVector<MachineBasicBlock *, 2> Succs; 1157 Succs.push_back(SDB->SwitchCases[i].TrueBB); 1158 if (SDB->SwitchCases[i].TrueBB != SDB->SwitchCases[i].FalseBB) 1159 Succs.push_back(SDB->SwitchCases[i].FalseBB); 1160 1161 // Emit the code. Note that this could result in FuncInfo->MBB being split. 1162 SDB->visitSwitchCase(SDB->SwitchCases[i], FuncInfo->MBB); 1163 CurDAG->setRoot(SDB->getRoot()); 1164 SDB->clear(); 1165 CodeGenAndEmitDAG(); 1166 1167 // Remember the last block, now that any splitting is done, for use in 1168 // populating PHI nodes in successors. 1169 MachineBasicBlock *ThisBB = FuncInfo->MBB; 1170 1171 // Handle any PHI nodes in successors of this chunk, as if we were coming 1172 // from the original BB before switch expansion. Note that PHI nodes can 1173 // occur multiple times in PHINodesToUpdate. We have to be very careful to 1174 // handle them the right number of times. 1175 for (unsigned i = 0, e = Succs.size(); i != e; ++i) { 1176 FuncInfo->MBB = Succs[i]; 1177 FuncInfo->InsertPt = FuncInfo->MBB->end(); 1178 // FuncInfo->MBB may have been removed from the CFG if a branch was 1179 // constant folded. 1180 if (ThisBB->isSuccessor(FuncInfo->MBB)) { 1181 for (MachineBasicBlock::iterator Phi = FuncInfo->MBB->begin(); 1182 Phi != FuncInfo->MBB->end() && Phi->isPHI(); 1183 ++Phi) { 1184 // This value for this PHI node is recorded in PHINodesToUpdate. 1185 for (unsigned pn = 0; ; ++pn) { 1186 assert(pn != FuncInfo->PHINodesToUpdate.size() && 1187 "Didn't find PHI entry!"); 1188 if (FuncInfo->PHINodesToUpdate[pn].first == Phi) { 1189 Phi->addOperand(MachineOperand:: 1190 CreateReg(FuncInfo->PHINodesToUpdate[pn].second, 1191 false)); 1192 Phi->addOperand(MachineOperand::CreateMBB(ThisBB)); 1193 break; 1194 } 1195 } 1196 } 1197 } 1198 } 1199 } 1200 SDB->SwitchCases.clear(); 1201 } 1202 1203 1204 /// Create the scheduler. If a specific scheduler was specified 1205 /// via the SchedulerRegistry, use it, otherwise select the 1206 /// one preferred by the target. 1207 /// 1208 ScheduleDAGSDNodes *SelectionDAGISel::CreateScheduler() { 1209 RegisterScheduler::FunctionPassCtor Ctor = RegisterScheduler::getDefault(); 1210 1211 if (!Ctor) { 1212 Ctor = ISHeuristic; 1213 RegisterScheduler::setDefault(Ctor); 1214 } 1215 1216 return Ctor(this, OptLevel); 1217 } 1218 1219 //===----------------------------------------------------------------------===// 1220 // Helper functions used by the generated instruction selector. 1221 //===----------------------------------------------------------------------===// 1222 // Calls to these methods are generated by tblgen. 1223 1224 /// CheckAndMask - The isel is trying to match something like (and X, 255). If 1225 /// the dag combiner simplified the 255, we still want to match. RHS is the 1226 /// actual value in the DAG on the RHS of an AND, and DesiredMaskS is the value 1227 /// specified in the .td file (e.g. 255). 1228 bool SelectionDAGISel::CheckAndMask(SDValue LHS, ConstantSDNode *RHS, 1229 int64_t DesiredMaskS) const { 1230 const APInt &ActualMask = RHS->getAPIntValue(); 1231 const APInt &DesiredMask = APInt(LHS.getValueSizeInBits(), DesiredMaskS); 1232 1233 // If the actual mask exactly matches, success! 1234 if (ActualMask == DesiredMask) 1235 return true; 1236 1237 // If the actual AND mask is allowing unallowed bits, this doesn't match. 1238 if (ActualMask.intersects(~DesiredMask)) 1239 return false; 1240 1241 // Otherwise, the DAG Combiner may have proven that the value coming in is 1242 // either already zero or is not demanded. Check for known zero input bits. 1243 APInt NeededMask = DesiredMask & ~ActualMask; 1244 if (CurDAG->MaskedValueIsZero(LHS, NeededMask)) 1245 return true; 1246 1247 // TODO: check to see if missing bits are just not demanded. 1248 1249 // Otherwise, this pattern doesn't match. 1250 return false; 1251 } 1252 1253 /// CheckOrMask - The isel is trying to match something like (or X, 255). If 1254 /// the dag combiner simplified the 255, we still want to match. RHS is the 1255 /// actual value in the DAG on the RHS of an OR, and DesiredMaskS is the value 1256 /// specified in the .td file (e.g. 255). 1257 bool SelectionDAGISel::CheckOrMask(SDValue LHS, ConstantSDNode *RHS, 1258 int64_t DesiredMaskS) const { 1259 const APInt &ActualMask = RHS->getAPIntValue(); 1260 const APInt &DesiredMask = APInt(LHS.getValueSizeInBits(), DesiredMaskS); 1261 1262 // If the actual mask exactly matches, success! 1263 if (ActualMask == DesiredMask) 1264 return true; 1265 1266 // If the actual AND mask is allowing unallowed bits, this doesn't match. 1267 if (ActualMask.intersects(~DesiredMask)) 1268 return false; 1269 1270 // Otherwise, the DAG Combiner may have proven that the value coming in is 1271 // either already zero or is not demanded. Check for known zero input bits. 1272 APInt NeededMask = DesiredMask & ~ActualMask; 1273 1274 APInt KnownZero, KnownOne; 1275 CurDAG->ComputeMaskedBits(LHS, NeededMask, KnownZero, KnownOne); 1276 1277 // If all the missing bits in the or are already known to be set, match! 1278 if ((NeededMask & KnownOne) == NeededMask) 1279 return true; 1280 1281 // TODO: check to see if missing bits are just not demanded. 1282 1283 // Otherwise, this pattern doesn't match. 1284 return false; 1285 } 1286 1287 1288 /// SelectInlineAsmMemoryOperands - Calls to this are automatically generated 1289 /// by tblgen. Others should not call it. 1290 void SelectionDAGISel:: 1291 SelectInlineAsmMemoryOperands(std::vector<SDValue> &Ops) { 1292 std::vector<SDValue> InOps; 1293 std::swap(InOps, Ops); 1294 1295 Ops.push_back(InOps[InlineAsm::Op_InputChain]); // 0 1296 Ops.push_back(InOps[InlineAsm::Op_AsmString]); // 1 1297 Ops.push_back(InOps[InlineAsm::Op_MDNode]); // 2, !srcloc 1298 Ops.push_back(InOps[InlineAsm::Op_ExtraInfo]); // 3 (SideEffect, AlignStack) 1299 1300 unsigned i = InlineAsm::Op_FirstOperand, e = InOps.size(); 1301 if (InOps[e-1].getValueType() == MVT::Glue) 1302 --e; // Don't process a glue operand if it is here. 1303 1304 while (i != e) { 1305 unsigned Flags = cast<ConstantSDNode>(InOps[i])->getZExtValue(); 1306 if (!InlineAsm::isMemKind(Flags)) { 1307 // Just skip over this operand, copying the operands verbatim. 1308 Ops.insert(Ops.end(), InOps.begin()+i, 1309 InOps.begin()+i+InlineAsm::getNumOperandRegisters(Flags) + 1); 1310 i += InlineAsm::getNumOperandRegisters(Flags) + 1; 1311 } else { 1312 assert(InlineAsm::getNumOperandRegisters(Flags) == 1 && 1313 "Memory operand with multiple values?"); 1314 // Otherwise, this is a memory operand. Ask the target to select it. 1315 std::vector<SDValue> SelOps; 1316 if (SelectInlineAsmMemoryOperand(InOps[i+1], 'm', SelOps)) 1317 report_fatal_error("Could not match memory address. Inline asm" 1318 " failure!"); 1319 1320 // Add this to the output node. 1321 unsigned NewFlags = 1322 InlineAsm::getFlagWord(InlineAsm::Kind_Mem, SelOps.size()); 1323 Ops.push_back(CurDAG->getTargetConstant(NewFlags, MVT::i32)); 1324 Ops.insert(Ops.end(), SelOps.begin(), SelOps.end()); 1325 i += 2; 1326 } 1327 } 1328 1329 // Add the glue input back if present. 1330 if (e != InOps.size()) 1331 Ops.push_back(InOps.back()); 1332 } 1333 1334 /// findGlueUse - Return use of MVT::Glue value produced by the specified 1335 /// SDNode. 1336 /// 1337 static SDNode *findGlueUse(SDNode *N) { 1338 unsigned FlagResNo = N->getNumValues()-1; 1339 for (SDNode::use_iterator I = N->use_begin(), E = N->use_end(); I != E; ++I) { 1340 SDUse &Use = I.getUse(); 1341 if (Use.getResNo() == FlagResNo) 1342 return Use.getUser(); 1343 } 1344 return NULL; 1345 } 1346 1347 /// findNonImmUse - Return true if "Use" is a non-immediate use of "Def". 1348 /// This function recursively traverses up the operand chain, ignoring 1349 /// certain nodes. 1350 static bool findNonImmUse(SDNode *Use, SDNode* Def, SDNode *ImmedUse, 1351 SDNode *Root, SmallPtrSet<SDNode*, 16> &Visited, 1352 bool IgnoreChains) { 1353 // The NodeID's are given uniques ID's where a node ID is guaranteed to be 1354 // greater than all of its (recursive) operands. If we scan to a point where 1355 // 'use' is smaller than the node we're scanning for, then we know we will 1356 // never find it. 1357 // 1358 // The Use may be -1 (unassigned) if it is a newly allocated node. This can 1359 // happen because we scan down to newly selected nodes in the case of glue 1360 // uses. 1361 if ((Use->getNodeId() < Def->getNodeId() && Use->getNodeId() != -1)) 1362 return false; 1363 1364 // Don't revisit nodes if we already scanned it and didn't fail, we know we 1365 // won't fail if we scan it again. 1366 if (!Visited.insert(Use)) 1367 return false; 1368 1369 for (unsigned i = 0, e = Use->getNumOperands(); i != e; ++i) { 1370 // Ignore chain uses, they are validated by HandleMergeInputChains. 1371 if (Use->getOperand(i).getValueType() == MVT::Other && IgnoreChains) 1372 continue; 1373 1374 SDNode *N = Use->getOperand(i).getNode(); 1375 if (N == Def) { 1376 if (Use == ImmedUse || Use == Root) 1377 continue; // We are not looking for immediate use. 1378 assert(N != Root); 1379 return true; 1380 } 1381 1382 // Traverse up the operand chain. 1383 if (findNonImmUse(N, Def, ImmedUse, Root, Visited, IgnoreChains)) 1384 return true; 1385 } 1386 return false; 1387 } 1388 1389 /// IsProfitableToFold - Returns true if it's profitable to fold the specific 1390 /// operand node N of U during instruction selection that starts at Root. 1391 bool SelectionDAGISel::IsProfitableToFold(SDValue N, SDNode *U, 1392 SDNode *Root) const { 1393 if (OptLevel == CodeGenOpt::None) return false; 1394 return N.hasOneUse(); 1395 } 1396 1397 /// IsLegalToFold - Returns true if the specific operand node N of 1398 /// U can be folded during instruction selection that starts at Root. 1399 bool SelectionDAGISel::IsLegalToFold(SDValue N, SDNode *U, SDNode *Root, 1400 CodeGenOpt::Level OptLevel, 1401 bool IgnoreChains) { 1402 if (OptLevel == CodeGenOpt::None) return false; 1403 1404 // If Root use can somehow reach N through a path that that doesn't contain 1405 // U then folding N would create a cycle. e.g. In the following 1406 // diagram, Root can reach N through X. If N is folded into into Root, then 1407 // X is both a predecessor and a successor of U. 1408 // 1409 // [N*] // 1410 // ^ ^ // 1411 // / \ // 1412 // [U*] [X]? // 1413 // ^ ^ // 1414 // \ / // 1415 // \ / // 1416 // [Root*] // 1417 // 1418 // * indicates nodes to be folded together. 1419 // 1420 // If Root produces glue, then it gets (even more) interesting. Since it 1421 // will be "glued" together with its glue use in the scheduler, we need to 1422 // check if it might reach N. 1423 // 1424 // [N*] // 1425 // ^ ^ // 1426 // / \ // 1427 // [U*] [X]? // 1428 // ^ ^ // 1429 // \ \ // 1430 // \ | // 1431 // [Root*] | // 1432 // ^ | // 1433 // f | // 1434 // | / // 1435 // [Y] / // 1436 // ^ / // 1437 // f / // 1438 // | / // 1439 // [GU] // 1440 // 1441 // If GU (glue use) indirectly reaches N (the load), and Root folds N 1442 // (call it Fold), then X is a predecessor of GU and a successor of 1443 // Fold. But since Fold and GU are glued together, this will create 1444 // a cycle in the scheduling graph. 1445 1446 // If the node has glue, walk down the graph to the "lowest" node in the 1447 // glueged set. 1448 EVT VT = Root->getValueType(Root->getNumValues()-1); 1449 while (VT == MVT::Glue) { 1450 SDNode *GU = findGlueUse(Root); 1451 if (GU == NULL) 1452 break; 1453 Root = GU; 1454 VT = Root->getValueType(Root->getNumValues()-1); 1455 1456 // If our query node has a glue result with a use, we've walked up it. If 1457 // the user (which has already been selected) has a chain or indirectly uses 1458 // the chain, our WalkChainUsers predicate will not consider it. Because of 1459 // this, we cannot ignore chains in this predicate. 1460 IgnoreChains = false; 1461 } 1462 1463 1464 SmallPtrSet<SDNode*, 16> Visited; 1465 return !findNonImmUse(Root, N.getNode(), U, Root, Visited, IgnoreChains); 1466 } 1467 1468 SDNode *SelectionDAGISel::Select_INLINEASM(SDNode *N) { 1469 std::vector<SDValue> Ops(N->op_begin(), N->op_end()); 1470 SelectInlineAsmMemoryOperands(Ops); 1471 1472 std::vector<EVT> VTs; 1473 VTs.push_back(MVT::Other); 1474 VTs.push_back(MVT::Glue); 1475 SDValue New = CurDAG->getNode(ISD::INLINEASM, N->getDebugLoc(), 1476 VTs, &Ops[0], Ops.size()); 1477 New->setNodeId(-1); 1478 return New.getNode(); 1479 } 1480 1481 SDNode *SelectionDAGISel::Select_UNDEF(SDNode *N) { 1482 return CurDAG->SelectNodeTo(N, TargetOpcode::IMPLICIT_DEF,N->getValueType(0)); 1483 } 1484 1485 /// GetVBR - decode a vbr encoding whose top bit is set. 1486 LLVM_ATTRIBUTE_ALWAYS_INLINE static uint64_t 1487 GetVBR(uint64_t Val, const unsigned char *MatcherTable, unsigned &Idx) { 1488 assert(Val >= 128 && "Not a VBR"); 1489 Val &= 127; // Remove first vbr bit. 1490 1491 unsigned Shift = 7; 1492 uint64_t NextBits; 1493 do { 1494 NextBits = MatcherTable[Idx++]; 1495 Val |= (NextBits&127) << Shift; 1496 Shift += 7; 1497 } while (NextBits & 128); 1498 1499 return Val; 1500 } 1501 1502 1503 /// UpdateChainsAndGlue - When a match is complete, this method updates uses of 1504 /// interior glue and chain results to use the new glue and chain results. 1505 void SelectionDAGISel:: 1506 UpdateChainsAndGlue(SDNode *NodeToMatch, SDValue InputChain, 1507 const SmallVectorImpl<SDNode*> &ChainNodesMatched, 1508 SDValue InputGlue, 1509 const SmallVectorImpl<SDNode*> &GlueResultNodesMatched, 1510 bool isMorphNodeTo) { 1511 SmallVector<SDNode*, 4> NowDeadNodes; 1512 1513 ISelUpdater ISU(ISelPosition); 1514 1515 // Now that all the normal results are replaced, we replace the chain and 1516 // glue results if present. 1517 if (!ChainNodesMatched.empty()) { 1518 assert(InputChain.getNode() != 0 && 1519 "Matched input chains but didn't produce a chain"); 1520 // Loop over all of the nodes we matched that produced a chain result. 1521 // Replace all the chain results with the final chain we ended up with. 1522 for (unsigned i = 0, e = ChainNodesMatched.size(); i != e; ++i) { 1523 SDNode *ChainNode = ChainNodesMatched[i]; 1524 1525 // If this node was already deleted, don't look at it. 1526 if (ChainNode->getOpcode() == ISD::DELETED_NODE) 1527 continue; 1528 1529 // Don't replace the results of the root node if we're doing a 1530 // MorphNodeTo. 1531 if (ChainNode == NodeToMatch && isMorphNodeTo) 1532 continue; 1533 1534 SDValue ChainVal = SDValue(ChainNode, ChainNode->getNumValues()-1); 1535 if (ChainVal.getValueType() == MVT::Glue) 1536 ChainVal = ChainVal.getValue(ChainVal->getNumValues()-2); 1537 assert(ChainVal.getValueType() == MVT::Other && "Not a chain?"); 1538 CurDAG->ReplaceAllUsesOfValueWith(ChainVal, InputChain, &ISU); 1539 1540 // If the node became dead and we haven't already seen it, delete it. 1541 if (ChainNode->use_empty() && 1542 !std::count(NowDeadNodes.begin(), NowDeadNodes.end(), ChainNode)) 1543 NowDeadNodes.push_back(ChainNode); 1544 } 1545 } 1546 1547 // If the result produces glue, update any glue results in the matched 1548 // pattern with the glue result. 1549 if (InputGlue.getNode() != 0) { 1550 // Handle any interior nodes explicitly marked. 1551 for (unsigned i = 0, e = GlueResultNodesMatched.size(); i != e; ++i) { 1552 SDNode *FRN = GlueResultNodesMatched[i]; 1553 1554 // If this node was already deleted, don't look at it. 1555 if (FRN->getOpcode() == ISD::DELETED_NODE) 1556 continue; 1557 1558 assert(FRN->getValueType(FRN->getNumValues()-1) == MVT::Glue && 1559 "Doesn't have a glue result"); 1560 CurDAG->ReplaceAllUsesOfValueWith(SDValue(FRN, FRN->getNumValues()-1), 1561 InputGlue, &ISU); 1562 1563 // If the node became dead and we haven't already seen it, delete it. 1564 if (FRN->use_empty() && 1565 !std::count(NowDeadNodes.begin(), NowDeadNodes.end(), FRN)) 1566 NowDeadNodes.push_back(FRN); 1567 } 1568 } 1569 1570 if (!NowDeadNodes.empty()) 1571 CurDAG->RemoveDeadNodes(NowDeadNodes, &ISU); 1572 1573 DEBUG(errs() << "ISEL: Match complete!\n"); 1574 } 1575 1576 enum ChainResult { 1577 CR_Simple, 1578 CR_InducesCycle, 1579 CR_LeadsToInteriorNode 1580 }; 1581 1582 /// WalkChainUsers - Walk down the users of the specified chained node that is 1583 /// part of the pattern we're matching, looking at all of the users we find. 1584 /// This determines whether something is an interior node, whether we have a 1585 /// non-pattern node in between two pattern nodes (which prevent folding because 1586 /// it would induce a cycle) and whether we have a TokenFactor node sandwiched 1587 /// between pattern nodes (in which case the TF becomes part of the pattern). 1588 /// 1589 /// The walk we do here is guaranteed to be small because we quickly get down to 1590 /// already selected nodes "below" us. 1591 static ChainResult 1592 WalkChainUsers(SDNode *ChainedNode, 1593 SmallVectorImpl<SDNode*> &ChainedNodesInPattern, 1594 SmallVectorImpl<SDNode*> &InteriorChainedNodes) { 1595 ChainResult Result = CR_Simple; 1596 1597 for (SDNode::use_iterator UI = ChainedNode->use_begin(), 1598 E = ChainedNode->use_end(); UI != E; ++UI) { 1599 // Make sure the use is of the chain, not some other value we produce. 1600 if (UI.getUse().getValueType() != MVT::Other) continue; 1601 1602 SDNode *User = *UI; 1603 1604 // If we see an already-selected machine node, then we've gone beyond the 1605 // pattern that we're selecting down into the already selected chunk of the 1606 // DAG. 1607 if (User->isMachineOpcode() || 1608 User->getOpcode() == ISD::HANDLENODE) // Root of the graph. 1609 continue; 1610 1611 if (User->getOpcode() == ISD::CopyToReg || 1612 User->getOpcode() == ISD::CopyFromReg || 1613 User->getOpcode() == ISD::INLINEASM || 1614 User->getOpcode() == ISD::EH_LABEL) { 1615 // If their node ID got reset to -1 then they've already been selected. 1616 // Treat them like a MachineOpcode. 1617 if (User->getNodeId() == -1) 1618 continue; 1619 } 1620 1621 // If we have a TokenFactor, we handle it specially. 1622 if (User->getOpcode() != ISD::TokenFactor) { 1623 // If the node isn't a token factor and isn't part of our pattern, then it 1624 // must be a random chained node in between two nodes we're selecting. 1625 // This happens when we have something like: 1626 // x = load ptr 1627 // call 1628 // y = x+4 1629 // store y -> ptr 1630 // Because we structurally match the load/store as a read/modify/write, 1631 // but the call is chained between them. We cannot fold in this case 1632 // because it would induce a cycle in the graph. 1633 if (!std::count(ChainedNodesInPattern.begin(), 1634 ChainedNodesInPattern.end(), User)) 1635 return CR_InducesCycle; 1636 1637 // Otherwise we found a node that is part of our pattern. For example in: 1638 // x = load ptr 1639 // y = x+4 1640 // store y -> ptr 1641 // This would happen when we're scanning down from the load and see the 1642 // store as a user. Record that there is a use of ChainedNode that is 1643 // part of the pattern and keep scanning uses. 1644 Result = CR_LeadsToInteriorNode; 1645 InteriorChainedNodes.push_back(User); 1646 continue; 1647 } 1648 1649 // If we found a TokenFactor, there are two cases to consider: first if the 1650 // TokenFactor is just hanging "below" the pattern we're matching (i.e. no 1651 // uses of the TF are in our pattern) we just want to ignore it. Second, 1652 // the TokenFactor can be sandwiched in between two chained nodes, like so: 1653 // [Load chain] 1654 // ^ 1655 // | 1656 // [Load] 1657 // ^ ^ 1658 // | \ DAG's like cheese 1659 // / \ do you? 1660 // / | 1661 // [TokenFactor] [Op] 1662 // ^ ^ 1663 // | | 1664 // \ / 1665 // \ / 1666 // [Store] 1667 // 1668 // In this case, the TokenFactor becomes part of our match and we rewrite it 1669 // as a new TokenFactor. 1670 // 1671 // To distinguish these two cases, do a recursive walk down the uses. 1672 switch (WalkChainUsers(User, ChainedNodesInPattern, InteriorChainedNodes)) { 1673 case CR_Simple: 1674 // If the uses of the TokenFactor are just already-selected nodes, ignore 1675 // it, it is "below" our pattern. 1676 continue; 1677 case CR_InducesCycle: 1678 // If the uses of the TokenFactor lead to nodes that are not part of our 1679 // pattern that are not selected, folding would turn this into a cycle, 1680 // bail out now. 1681 return CR_InducesCycle; 1682 case CR_LeadsToInteriorNode: 1683 break; // Otherwise, keep processing. 1684 } 1685 1686 // Okay, we know we're in the interesting interior case. The TokenFactor 1687 // is now going to be considered part of the pattern so that we rewrite its 1688 // uses (it may have uses that are not part of the pattern) with the 1689 // ultimate chain result of the generated code. We will also add its chain 1690 // inputs as inputs to the ultimate TokenFactor we create. 1691 Result = CR_LeadsToInteriorNode; 1692 ChainedNodesInPattern.push_back(User); 1693 InteriorChainedNodes.push_back(User); 1694 continue; 1695 } 1696 1697 return Result; 1698 } 1699 1700 /// HandleMergeInputChains - This implements the OPC_EmitMergeInputChains 1701 /// operation for when the pattern matched at least one node with a chains. The 1702 /// input vector contains a list of all of the chained nodes that we match. We 1703 /// must determine if this is a valid thing to cover (i.e. matching it won't 1704 /// induce cycles in the DAG) and if so, creating a TokenFactor node. that will 1705 /// be used as the input node chain for the generated nodes. 1706 static SDValue 1707 HandleMergeInputChains(SmallVectorImpl<SDNode*> &ChainNodesMatched, 1708 SelectionDAG *CurDAG) { 1709 // Walk all of the chained nodes we've matched, recursively scanning down the 1710 // users of the chain result. This adds any TokenFactor nodes that are caught 1711 // in between chained nodes to the chained and interior nodes list. 1712 SmallVector<SDNode*, 3> InteriorChainedNodes; 1713 for (unsigned i = 0, e = ChainNodesMatched.size(); i != e; ++i) { 1714 if (WalkChainUsers(ChainNodesMatched[i], ChainNodesMatched, 1715 InteriorChainedNodes) == CR_InducesCycle) 1716 return SDValue(); // Would induce a cycle. 1717 } 1718 1719 // Okay, we have walked all the matched nodes and collected TokenFactor nodes 1720 // that we are interested in. Form our input TokenFactor node. 1721 SmallVector<SDValue, 3> InputChains; 1722 for (unsigned i = 0, e = ChainNodesMatched.size(); i != e; ++i) { 1723 // Add the input chain of this node to the InputChains list (which will be 1724 // the operands of the generated TokenFactor) if it's not an interior node. 1725 SDNode *N = ChainNodesMatched[i]; 1726 if (N->getOpcode() != ISD::TokenFactor) { 1727 if (std::count(InteriorChainedNodes.begin(),InteriorChainedNodes.end(),N)) 1728 continue; 1729 1730 // Otherwise, add the input chain. 1731 SDValue InChain = ChainNodesMatched[i]->getOperand(0); 1732 assert(InChain.getValueType() == MVT::Other && "Not a chain"); 1733 InputChains.push_back(InChain); 1734 continue; 1735 } 1736 1737 // If we have a token factor, we want to add all inputs of the token factor 1738 // that are not part of the pattern we're matching. 1739 for (unsigned op = 0, e = N->getNumOperands(); op != e; ++op) { 1740 if (!std::count(ChainNodesMatched.begin(), ChainNodesMatched.end(), 1741 N->getOperand(op).getNode())) 1742 InputChains.push_back(N->getOperand(op)); 1743 } 1744 } 1745 1746 SDValue Res; 1747 if (InputChains.size() == 1) 1748 return InputChains[0]; 1749 return CurDAG->getNode(ISD::TokenFactor, ChainNodesMatched[0]->getDebugLoc(), 1750 MVT::Other, &InputChains[0], InputChains.size()); 1751 } 1752 1753 /// MorphNode - Handle morphing a node in place for the selector. 1754 SDNode *SelectionDAGISel:: 1755 MorphNode(SDNode *Node, unsigned TargetOpc, SDVTList VTList, 1756 const SDValue *Ops, unsigned NumOps, unsigned EmitNodeInfo) { 1757 // It is possible we're using MorphNodeTo to replace a node with no 1758 // normal results with one that has a normal result (or we could be 1759 // adding a chain) and the input could have glue and chains as well. 1760 // In this case we need to shift the operands down. 1761 // FIXME: This is a horrible hack and broken in obscure cases, no worse 1762 // than the old isel though. 1763 int OldGlueResultNo = -1, OldChainResultNo = -1; 1764 1765 unsigned NTMNumResults = Node->getNumValues(); 1766 if (Node->getValueType(NTMNumResults-1) == MVT::Glue) { 1767 OldGlueResultNo = NTMNumResults-1; 1768 if (NTMNumResults != 1 && 1769 Node->getValueType(NTMNumResults-2) == MVT::Other) 1770 OldChainResultNo = NTMNumResults-2; 1771 } else if (Node->getValueType(NTMNumResults-1) == MVT::Other) 1772 OldChainResultNo = NTMNumResults-1; 1773 1774 // Call the underlying SelectionDAG routine to do the transmogrification. Note 1775 // that this deletes operands of the old node that become dead. 1776 SDNode *Res = CurDAG->MorphNodeTo(Node, ~TargetOpc, VTList, Ops, NumOps); 1777 1778 // MorphNodeTo can operate in two ways: if an existing node with the 1779 // specified operands exists, it can just return it. Otherwise, it 1780 // updates the node in place to have the requested operands. 1781 if (Res == Node) { 1782 // If we updated the node in place, reset the node ID. To the isel, 1783 // this should be just like a newly allocated machine node. 1784 Res->setNodeId(-1); 1785 } 1786 1787 unsigned ResNumResults = Res->getNumValues(); 1788 // Move the glue if needed. 1789 if ((EmitNodeInfo & OPFL_GlueOutput) && OldGlueResultNo != -1 && 1790 (unsigned)OldGlueResultNo != ResNumResults-1) 1791 CurDAG->ReplaceAllUsesOfValueWith(SDValue(Node, OldGlueResultNo), 1792 SDValue(Res, ResNumResults-1)); 1793 1794 if ((EmitNodeInfo & OPFL_GlueOutput) != 0) 1795 --ResNumResults; 1796 1797 // Move the chain reference if needed. 1798 if ((EmitNodeInfo & OPFL_Chain) && OldChainResultNo != -1 && 1799 (unsigned)OldChainResultNo != ResNumResults-1) 1800 CurDAG->ReplaceAllUsesOfValueWith(SDValue(Node, OldChainResultNo), 1801 SDValue(Res, ResNumResults-1)); 1802 1803 // Otherwise, no replacement happened because the node already exists. Replace 1804 // Uses of the old node with the new one. 1805 if (Res != Node) 1806 CurDAG->ReplaceAllUsesWith(Node, Res); 1807 1808 return Res; 1809 } 1810 1811 /// CheckPatternPredicate - Implements OP_CheckPatternPredicate. 1812 LLVM_ATTRIBUTE_ALWAYS_INLINE static bool 1813 CheckSame(const unsigned char *MatcherTable, unsigned &MatcherIndex, 1814 SDValue N, 1815 const SmallVectorImpl<std::pair<SDValue, SDNode*> > &RecordedNodes) { 1816 // Accept if it is exactly the same as a previously recorded node. 1817 unsigned RecNo = MatcherTable[MatcherIndex++]; 1818 assert(RecNo < RecordedNodes.size() && "Invalid CheckSame"); 1819 return N == RecordedNodes[RecNo].first; 1820 } 1821 1822 /// CheckPatternPredicate - Implements OP_CheckPatternPredicate. 1823 LLVM_ATTRIBUTE_ALWAYS_INLINE static bool 1824 CheckPatternPredicate(const unsigned char *MatcherTable, unsigned &MatcherIndex, 1825 SelectionDAGISel &SDISel) { 1826 return SDISel.CheckPatternPredicate(MatcherTable[MatcherIndex++]); 1827 } 1828 1829 /// CheckNodePredicate - Implements OP_CheckNodePredicate. 1830 LLVM_ATTRIBUTE_ALWAYS_INLINE static bool 1831 CheckNodePredicate(const unsigned char *MatcherTable, unsigned &MatcherIndex, 1832 SelectionDAGISel &SDISel, SDNode *N) { 1833 return SDISel.CheckNodePredicate(N, MatcherTable[MatcherIndex++]); 1834 } 1835 1836 LLVM_ATTRIBUTE_ALWAYS_INLINE static bool 1837 CheckOpcode(const unsigned char *MatcherTable, unsigned &MatcherIndex, 1838 SDNode *N) { 1839 uint16_t Opc = MatcherTable[MatcherIndex++]; 1840 Opc |= (unsigned short)MatcherTable[MatcherIndex++] << 8; 1841 return N->getOpcode() == Opc; 1842 } 1843 1844 LLVM_ATTRIBUTE_ALWAYS_INLINE static bool 1845 CheckType(const unsigned char *MatcherTable, unsigned &MatcherIndex, 1846 SDValue N, const TargetLowering &TLI) { 1847 MVT::SimpleValueType VT = (MVT::SimpleValueType)MatcherTable[MatcherIndex++]; 1848 if (N.getValueType() == VT) return true; 1849 1850 // Handle the case when VT is iPTR. 1851 return VT == MVT::iPTR && N.getValueType() == TLI.getPointerTy(); 1852 } 1853 1854 LLVM_ATTRIBUTE_ALWAYS_INLINE static bool 1855 CheckChildType(const unsigned char *MatcherTable, unsigned &MatcherIndex, 1856 SDValue N, const TargetLowering &TLI, 1857 unsigned ChildNo) { 1858 if (ChildNo >= N.getNumOperands()) 1859 return false; // Match fails if out of range child #. 1860 return ::CheckType(MatcherTable, MatcherIndex, N.getOperand(ChildNo), TLI); 1861 } 1862 1863 1864 LLVM_ATTRIBUTE_ALWAYS_INLINE static bool 1865 CheckCondCode(const unsigned char *MatcherTable, unsigned &MatcherIndex, 1866 SDValue N) { 1867 return cast<CondCodeSDNode>(N)->get() == 1868 (ISD::CondCode)MatcherTable[MatcherIndex++]; 1869 } 1870 1871 LLVM_ATTRIBUTE_ALWAYS_INLINE static bool 1872 CheckValueType(const unsigned char *MatcherTable, unsigned &MatcherIndex, 1873 SDValue N, const TargetLowering &TLI) { 1874 MVT::SimpleValueType VT = (MVT::SimpleValueType)MatcherTable[MatcherIndex++]; 1875 if (cast<VTSDNode>(N)->getVT() == VT) 1876 return true; 1877 1878 // Handle the case when VT is iPTR. 1879 return VT == MVT::iPTR && cast<VTSDNode>(N)->getVT() == TLI.getPointerTy(); 1880 } 1881 1882 LLVM_ATTRIBUTE_ALWAYS_INLINE static bool 1883 CheckInteger(const unsigned char *MatcherTable, unsigned &MatcherIndex, 1884 SDValue N) { 1885 int64_t Val = MatcherTable[MatcherIndex++]; 1886 if (Val & 128) 1887 Val = GetVBR(Val, MatcherTable, MatcherIndex); 1888 1889 ConstantSDNode *C = dyn_cast<ConstantSDNode>(N); 1890 return C != 0 && C->getSExtValue() == Val; 1891 } 1892 1893 LLVM_ATTRIBUTE_ALWAYS_INLINE static bool 1894 CheckAndImm(const unsigned char *MatcherTable, unsigned &MatcherIndex, 1895 SDValue N, SelectionDAGISel &SDISel) { 1896 int64_t Val = MatcherTable[MatcherIndex++]; 1897 if (Val & 128) 1898 Val = GetVBR(Val, MatcherTable, MatcherIndex); 1899 1900 if (N->getOpcode() != ISD::AND) return false; 1901 1902 ConstantSDNode *C = dyn_cast<ConstantSDNode>(N->getOperand(1)); 1903 return C != 0 && SDISel.CheckAndMask(N.getOperand(0), C, Val); 1904 } 1905 1906 LLVM_ATTRIBUTE_ALWAYS_INLINE static bool 1907 CheckOrImm(const unsigned char *MatcherTable, unsigned &MatcherIndex, 1908 SDValue N, SelectionDAGISel &SDISel) { 1909 int64_t Val = MatcherTable[MatcherIndex++]; 1910 if (Val & 128) 1911 Val = GetVBR(Val, MatcherTable, MatcherIndex); 1912 1913 if (N->getOpcode() != ISD::OR) return false; 1914 1915 ConstantSDNode *C = dyn_cast<ConstantSDNode>(N->getOperand(1)); 1916 return C != 0 && SDISel.CheckOrMask(N.getOperand(0), C, Val); 1917 } 1918 1919 /// IsPredicateKnownToFail - If we know how and can do so without pushing a 1920 /// scope, evaluate the current node. If the current predicate is known to 1921 /// fail, set Result=true and return anything. If the current predicate is 1922 /// known to pass, set Result=false and return the MatcherIndex to continue 1923 /// with. If the current predicate is unknown, set Result=false and return the 1924 /// MatcherIndex to continue with. 1925 static unsigned IsPredicateKnownToFail(const unsigned char *Table, 1926 unsigned Index, SDValue N, 1927 bool &Result, SelectionDAGISel &SDISel, 1928 SmallVectorImpl<std::pair<SDValue, SDNode*> > &RecordedNodes) { 1929 switch (Table[Index++]) { 1930 default: 1931 Result = false; 1932 return Index-1; // Could not evaluate this predicate. 1933 case SelectionDAGISel::OPC_CheckSame: 1934 Result = !::CheckSame(Table, Index, N, RecordedNodes); 1935 return Index; 1936 case SelectionDAGISel::OPC_CheckPatternPredicate: 1937 Result = !::CheckPatternPredicate(Table, Index, SDISel); 1938 return Index; 1939 case SelectionDAGISel::OPC_CheckPredicate: 1940 Result = !::CheckNodePredicate(Table, Index, SDISel, N.getNode()); 1941 return Index; 1942 case SelectionDAGISel::OPC_CheckOpcode: 1943 Result = !::CheckOpcode(Table, Index, N.getNode()); 1944 return Index; 1945 case SelectionDAGISel::OPC_CheckType: 1946 Result = !::CheckType(Table, Index, N, SDISel.TLI); 1947 return Index; 1948 case SelectionDAGISel::OPC_CheckChild0Type: 1949 case SelectionDAGISel::OPC_CheckChild1Type: 1950 case SelectionDAGISel::OPC_CheckChild2Type: 1951 case SelectionDAGISel::OPC_CheckChild3Type: 1952 case SelectionDAGISel::OPC_CheckChild4Type: 1953 case SelectionDAGISel::OPC_CheckChild5Type: 1954 case SelectionDAGISel::OPC_CheckChild6Type: 1955 case SelectionDAGISel::OPC_CheckChild7Type: 1956 Result = !::CheckChildType(Table, Index, N, SDISel.TLI, 1957 Table[Index-1] - SelectionDAGISel::OPC_CheckChild0Type); 1958 return Index; 1959 case SelectionDAGISel::OPC_CheckCondCode: 1960 Result = !::CheckCondCode(Table, Index, N); 1961 return Index; 1962 case SelectionDAGISel::OPC_CheckValueType: 1963 Result = !::CheckValueType(Table, Index, N, SDISel.TLI); 1964 return Index; 1965 case SelectionDAGISel::OPC_CheckInteger: 1966 Result = !::CheckInteger(Table, Index, N); 1967 return Index; 1968 case SelectionDAGISel::OPC_CheckAndImm: 1969 Result = !::CheckAndImm(Table, Index, N, SDISel); 1970 return Index; 1971 case SelectionDAGISel::OPC_CheckOrImm: 1972 Result = !::CheckOrImm(Table, Index, N, SDISel); 1973 return Index; 1974 } 1975 } 1976 1977 namespace { 1978 1979 struct MatchScope { 1980 /// FailIndex - If this match fails, this is the index to continue with. 1981 unsigned FailIndex; 1982 1983 /// NodeStack - The node stack when the scope was formed. 1984 SmallVector<SDValue, 4> NodeStack; 1985 1986 /// NumRecordedNodes - The number of recorded nodes when the scope was formed. 1987 unsigned NumRecordedNodes; 1988 1989 /// NumMatchedMemRefs - The number of matched memref entries. 1990 unsigned NumMatchedMemRefs; 1991 1992 /// InputChain/InputGlue - The current chain/glue 1993 SDValue InputChain, InputGlue; 1994 1995 /// HasChainNodesMatched - True if the ChainNodesMatched list is non-empty. 1996 bool HasChainNodesMatched, HasGlueResultNodesMatched; 1997 }; 1998 1999 } 2000 2001 SDNode *SelectionDAGISel:: 2002 SelectCodeCommon(SDNode *NodeToMatch, const unsigned char *MatcherTable, 2003 unsigned TableSize) { 2004 // FIXME: Should these even be selected? Handle these cases in the caller? 2005 switch (NodeToMatch->getOpcode()) { 2006 default: 2007 break; 2008 case ISD::EntryToken: // These nodes remain the same. 2009 case ISD::BasicBlock: 2010 case ISD::Register: 2011 //case ISD::VALUETYPE: 2012 //case ISD::CONDCODE: 2013 case ISD::HANDLENODE: 2014 case ISD::MDNODE_SDNODE: 2015 case ISD::TargetConstant: 2016 case ISD::TargetConstantFP: 2017 case ISD::TargetConstantPool: 2018 case ISD::TargetFrameIndex: 2019 case ISD::TargetExternalSymbol: 2020 case ISD::TargetBlockAddress: 2021 case ISD::TargetJumpTable: 2022 case ISD::TargetGlobalTLSAddress: 2023 case ISD::TargetGlobalAddress: 2024 case ISD::TokenFactor: 2025 case ISD::CopyFromReg: 2026 case ISD::CopyToReg: 2027 case ISD::EH_LABEL: 2028 NodeToMatch->setNodeId(-1); // Mark selected. 2029 return 0; 2030 case ISD::AssertSext: 2031 case ISD::AssertZext: 2032 CurDAG->ReplaceAllUsesOfValueWith(SDValue(NodeToMatch, 0), 2033 NodeToMatch->getOperand(0)); 2034 return 0; 2035 case ISD::INLINEASM: return Select_INLINEASM(NodeToMatch); 2036 case ISD::UNDEF: return Select_UNDEF(NodeToMatch); 2037 } 2038 2039 assert(!NodeToMatch->isMachineOpcode() && "Node already selected!"); 2040 2041 // Set up the node stack with NodeToMatch as the only node on the stack. 2042 SmallVector<SDValue, 8> NodeStack; 2043 SDValue N = SDValue(NodeToMatch, 0); 2044 NodeStack.push_back(N); 2045 2046 // MatchScopes - Scopes used when matching, if a match failure happens, this 2047 // indicates where to continue checking. 2048 SmallVector<MatchScope, 8> MatchScopes; 2049 2050 // RecordedNodes - This is the set of nodes that have been recorded by the 2051 // state machine. The second value is the parent of the node, or null if the 2052 // root is recorded. 2053 SmallVector<std::pair<SDValue, SDNode*>, 8> RecordedNodes; 2054 2055 // MatchedMemRefs - This is the set of MemRef's we've seen in the input 2056 // pattern. 2057 SmallVector<MachineMemOperand*, 2> MatchedMemRefs; 2058 2059 // These are the current input chain and glue for use when generating nodes. 2060 // Various Emit operations change these. For example, emitting a copytoreg 2061 // uses and updates these. 2062 SDValue InputChain, InputGlue; 2063 2064 // ChainNodesMatched - If a pattern matches nodes that have input/output 2065 // chains, the OPC_EmitMergeInputChains operation is emitted which indicates 2066 // which ones they are. The result is captured into this list so that we can 2067 // update the chain results when the pattern is complete. 2068 SmallVector<SDNode*, 3> ChainNodesMatched; 2069 SmallVector<SDNode*, 3> GlueResultNodesMatched; 2070 2071 DEBUG(errs() << "ISEL: Starting pattern match on root node: "; 2072 NodeToMatch->dump(CurDAG); 2073 errs() << '\n'); 2074 2075 // Determine where to start the interpreter. Normally we start at opcode #0, 2076 // but if the state machine starts with an OPC_SwitchOpcode, then we 2077 // accelerate the first lookup (which is guaranteed to be hot) with the 2078 // OpcodeOffset table. 2079 unsigned MatcherIndex = 0; 2080 2081 if (!OpcodeOffset.empty()) { 2082 // Already computed the OpcodeOffset table, just index into it. 2083 if (N.getOpcode() < OpcodeOffset.size()) 2084 MatcherIndex = OpcodeOffset[N.getOpcode()]; 2085 DEBUG(errs() << " Initial Opcode index to " << MatcherIndex << "\n"); 2086 2087 } else if (MatcherTable[0] == OPC_SwitchOpcode) { 2088 // Otherwise, the table isn't computed, but the state machine does start 2089 // with an OPC_SwitchOpcode instruction. Populate the table now, since this 2090 // is the first time we're selecting an instruction. 2091 unsigned Idx = 1; 2092 while (1) { 2093 // Get the size of this case. 2094 unsigned CaseSize = MatcherTable[Idx++]; 2095 if (CaseSize & 128) 2096 CaseSize = GetVBR(CaseSize, MatcherTable, Idx); 2097 if (CaseSize == 0) break; 2098 2099 // Get the opcode, add the index to the table. 2100 uint16_t Opc = MatcherTable[Idx++]; 2101 Opc |= (unsigned short)MatcherTable[Idx++] << 8; 2102 if (Opc >= OpcodeOffset.size()) 2103 OpcodeOffset.resize((Opc+1)*2); 2104 OpcodeOffset[Opc] = Idx; 2105 Idx += CaseSize; 2106 } 2107 2108 // Okay, do the lookup for the first opcode. 2109 if (N.getOpcode() < OpcodeOffset.size()) 2110 MatcherIndex = OpcodeOffset[N.getOpcode()]; 2111 } 2112 2113 while (1) { 2114 assert(MatcherIndex < TableSize && "Invalid index"); 2115 #ifndef NDEBUG 2116 unsigned CurrentOpcodeIndex = MatcherIndex; 2117 #endif 2118 BuiltinOpcodes Opcode = (BuiltinOpcodes)MatcherTable[MatcherIndex++]; 2119 switch (Opcode) { 2120 case OPC_Scope: { 2121 // Okay, the semantics of this operation are that we should push a scope 2122 // then evaluate the first child. However, pushing a scope only to have 2123 // the first check fail (which then pops it) is inefficient. If we can 2124 // determine immediately that the first check (or first several) will 2125 // immediately fail, don't even bother pushing a scope for them. 2126 unsigned FailIndex; 2127 2128 while (1) { 2129 unsigned NumToSkip = MatcherTable[MatcherIndex++]; 2130 if (NumToSkip & 128) 2131 NumToSkip = GetVBR(NumToSkip, MatcherTable, MatcherIndex); 2132 // Found the end of the scope with no match. 2133 if (NumToSkip == 0) { 2134 FailIndex = 0; 2135 break; 2136 } 2137 2138 FailIndex = MatcherIndex+NumToSkip; 2139 2140 unsigned MatcherIndexOfPredicate = MatcherIndex; 2141 (void)MatcherIndexOfPredicate; // silence warning. 2142 2143 // If we can't evaluate this predicate without pushing a scope (e.g. if 2144 // it is a 'MoveParent') or if the predicate succeeds on this node, we 2145 // push the scope and evaluate the full predicate chain. 2146 bool Result; 2147 MatcherIndex = IsPredicateKnownToFail(MatcherTable, MatcherIndex, N, 2148 Result, *this, RecordedNodes); 2149 if (!Result) 2150 break; 2151 2152 DEBUG(errs() << " Skipped scope entry (due to false predicate) at " 2153 << "index " << MatcherIndexOfPredicate 2154 << ", continuing at " << FailIndex << "\n"); 2155 ++NumDAGIselRetries; 2156 2157 // Otherwise, we know that this case of the Scope is guaranteed to fail, 2158 // move to the next case. 2159 MatcherIndex = FailIndex; 2160 } 2161 2162 // If the whole scope failed to match, bail. 2163 if (FailIndex == 0) break; 2164 2165 // Push a MatchScope which indicates where to go if the first child fails 2166 // to match. 2167 MatchScope NewEntry; 2168 NewEntry.FailIndex = FailIndex; 2169 NewEntry.NodeStack.append(NodeStack.begin(), NodeStack.end()); 2170 NewEntry.NumRecordedNodes = RecordedNodes.size(); 2171 NewEntry.NumMatchedMemRefs = MatchedMemRefs.size(); 2172 NewEntry.InputChain = InputChain; 2173 NewEntry.InputGlue = InputGlue; 2174 NewEntry.HasChainNodesMatched = !ChainNodesMatched.empty(); 2175 NewEntry.HasGlueResultNodesMatched = !GlueResultNodesMatched.empty(); 2176 MatchScopes.push_back(NewEntry); 2177 continue; 2178 } 2179 case OPC_RecordNode: { 2180 // Remember this node, it may end up being an operand in the pattern. 2181 SDNode *Parent = 0; 2182 if (NodeStack.size() > 1) 2183 Parent = NodeStack[NodeStack.size()-2].getNode(); 2184 RecordedNodes.push_back(std::make_pair(N, Parent)); 2185 continue; 2186 } 2187 2188 case OPC_RecordChild0: case OPC_RecordChild1: 2189 case OPC_RecordChild2: case OPC_RecordChild3: 2190 case OPC_RecordChild4: case OPC_RecordChild5: 2191 case OPC_RecordChild6: case OPC_RecordChild7: { 2192 unsigned ChildNo = Opcode-OPC_RecordChild0; 2193 if (ChildNo >= N.getNumOperands()) 2194 break; // Match fails if out of range child #. 2195 2196 RecordedNodes.push_back(std::make_pair(N->getOperand(ChildNo), 2197 N.getNode())); 2198 continue; 2199 } 2200 case OPC_RecordMemRef: 2201 MatchedMemRefs.push_back(cast<MemSDNode>(N)->getMemOperand()); 2202 continue; 2203 2204 case OPC_CaptureGlueInput: 2205 // If the current node has an input glue, capture it in InputGlue. 2206 if (N->getNumOperands() != 0 && 2207 N->getOperand(N->getNumOperands()-1).getValueType() == MVT::Glue) 2208 InputGlue = N->getOperand(N->getNumOperands()-1); 2209 continue; 2210 2211 case OPC_MoveChild: { 2212 unsigned ChildNo = MatcherTable[MatcherIndex++]; 2213 if (ChildNo >= N.getNumOperands()) 2214 break; // Match fails if out of range child #. 2215 N = N.getOperand(ChildNo); 2216 NodeStack.push_back(N); 2217 continue; 2218 } 2219 2220 case OPC_MoveParent: 2221 // Pop the current node off the NodeStack. 2222 NodeStack.pop_back(); 2223 assert(!NodeStack.empty() && "Node stack imbalance!"); 2224 N = NodeStack.back(); 2225 continue; 2226 2227 case OPC_CheckSame: 2228 if (!::CheckSame(MatcherTable, MatcherIndex, N, RecordedNodes)) break; 2229 continue; 2230 case OPC_CheckPatternPredicate: 2231 if (!::CheckPatternPredicate(MatcherTable, MatcherIndex, *this)) break; 2232 continue; 2233 case OPC_CheckPredicate: 2234 if (!::CheckNodePredicate(MatcherTable, MatcherIndex, *this, 2235 N.getNode())) 2236 break; 2237 continue; 2238 case OPC_CheckComplexPat: { 2239 unsigned CPNum = MatcherTable[MatcherIndex++]; 2240 unsigned RecNo = MatcherTable[MatcherIndex++]; 2241 assert(RecNo < RecordedNodes.size() && "Invalid CheckComplexPat"); 2242 if (!CheckComplexPattern(NodeToMatch, RecordedNodes[RecNo].second, 2243 RecordedNodes[RecNo].first, CPNum, 2244 RecordedNodes)) 2245 break; 2246 continue; 2247 } 2248 case OPC_CheckOpcode: 2249 if (!::CheckOpcode(MatcherTable, MatcherIndex, N.getNode())) break; 2250 continue; 2251 2252 case OPC_CheckType: 2253 if (!::CheckType(MatcherTable, MatcherIndex, N, TLI)) break; 2254 continue; 2255 2256 case OPC_SwitchOpcode: { 2257 unsigned CurNodeOpcode = N.getOpcode(); 2258 unsigned SwitchStart = MatcherIndex-1; (void)SwitchStart; 2259 unsigned CaseSize; 2260 while (1) { 2261 // Get the size of this case. 2262 CaseSize = MatcherTable[MatcherIndex++]; 2263 if (CaseSize & 128) 2264 CaseSize = GetVBR(CaseSize, MatcherTable, MatcherIndex); 2265 if (CaseSize == 0) break; 2266 2267 uint16_t Opc = MatcherTable[MatcherIndex++]; 2268 Opc |= (unsigned short)MatcherTable[MatcherIndex++] << 8; 2269 2270 // If the opcode matches, then we will execute this case. 2271 if (CurNodeOpcode == Opc) 2272 break; 2273 2274 // Otherwise, skip over this case. 2275 MatcherIndex += CaseSize; 2276 } 2277 2278 // If no cases matched, bail out. 2279 if (CaseSize == 0) break; 2280 2281 // Otherwise, execute the case we found. 2282 DEBUG(errs() << " OpcodeSwitch from " << SwitchStart 2283 << " to " << MatcherIndex << "\n"); 2284 continue; 2285 } 2286 2287 case OPC_SwitchType: { 2288 MVT CurNodeVT = N.getValueType().getSimpleVT(); 2289 unsigned SwitchStart = MatcherIndex-1; (void)SwitchStart; 2290 unsigned CaseSize; 2291 while (1) { 2292 // Get the size of this case. 2293 CaseSize = MatcherTable[MatcherIndex++]; 2294 if (CaseSize & 128) 2295 CaseSize = GetVBR(CaseSize, MatcherTable, MatcherIndex); 2296 if (CaseSize == 0) break; 2297 2298 MVT CaseVT = (MVT::SimpleValueType)MatcherTable[MatcherIndex++]; 2299 if (CaseVT == MVT::iPTR) 2300 CaseVT = TLI.getPointerTy(); 2301 2302 // If the VT matches, then we will execute this case. 2303 if (CurNodeVT == CaseVT) 2304 break; 2305 2306 // Otherwise, skip over this case. 2307 MatcherIndex += CaseSize; 2308 } 2309 2310 // If no cases matched, bail out. 2311 if (CaseSize == 0) break; 2312 2313 // Otherwise, execute the case we found. 2314 DEBUG(errs() << " TypeSwitch[" << EVT(CurNodeVT).getEVTString() 2315 << "] from " << SwitchStart << " to " << MatcherIndex<<'\n'); 2316 continue; 2317 } 2318 case OPC_CheckChild0Type: case OPC_CheckChild1Type: 2319 case OPC_CheckChild2Type: case OPC_CheckChild3Type: 2320 case OPC_CheckChild4Type: case OPC_CheckChild5Type: 2321 case OPC_CheckChild6Type: case OPC_CheckChild7Type: 2322 if (!::CheckChildType(MatcherTable, MatcherIndex, N, TLI, 2323 Opcode-OPC_CheckChild0Type)) 2324 break; 2325 continue; 2326 case OPC_CheckCondCode: 2327 if (!::CheckCondCode(MatcherTable, MatcherIndex, N)) break; 2328 continue; 2329 case OPC_CheckValueType: 2330 if (!::CheckValueType(MatcherTable, MatcherIndex, N, TLI)) break; 2331 continue; 2332 case OPC_CheckInteger: 2333 if (!::CheckInteger(MatcherTable, MatcherIndex, N)) break; 2334 continue; 2335 case OPC_CheckAndImm: 2336 if (!::CheckAndImm(MatcherTable, MatcherIndex, N, *this)) break; 2337 continue; 2338 case OPC_CheckOrImm: 2339 if (!::CheckOrImm(MatcherTable, MatcherIndex, N, *this)) break; 2340 continue; 2341 2342 case OPC_CheckFoldableChainNode: { 2343 assert(NodeStack.size() != 1 && "No parent node"); 2344 // Verify that all intermediate nodes between the root and this one have 2345 // a single use. 2346 bool HasMultipleUses = false; 2347 for (unsigned i = 1, e = NodeStack.size()-1; i != e; ++i) 2348 if (!NodeStack[i].hasOneUse()) { 2349 HasMultipleUses = true; 2350 break; 2351 } 2352 if (HasMultipleUses) break; 2353 2354 // Check to see that the target thinks this is profitable to fold and that 2355 // we can fold it without inducing cycles in the graph. 2356 if (!IsProfitableToFold(N, NodeStack[NodeStack.size()-2].getNode(), 2357 NodeToMatch) || 2358 !IsLegalToFold(N, NodeStack[NodeStack.size()-2].getNode(), 2359 NodeToMatch, OptLevel, 2360 true/*We validate our own chains*/)) 2361 break; 2362 2363 continue; 2364 } 2365 case OPC_EmitInteger: { 2366 MVT::SimpleValueType VT = 2367 (MVT::SimpleValueType)MatcherTable[MatcherIndex++]; 2368 int64_t Val = MatcherTable[MatcherIndex++]; 2369 if (Val & 128) 2370 Val = GetVBR(Val, MatcherTable, MatcherIndex); 2371 RecordedNodes.push_back(std::pair<SDValue, SDNode*>( 2372 CurDAG->getTargetConstant(Val, VT), (SDNode*)0)); 2373 continue; 2374 } 2375 case OPC_EmitRegister: { 2376 MVT::SimpleValueType VT = 2377 (MVT::SimpleValueType)MatcherTable[MatcherIndex++]; 2378 unsigned RegNo = MatcherTable[MatcherIndex++]; 2379 RecordedNodes.push_back(std::pair<SDValue, SDNode*>( 2380 CurDAG->getRegister(RegNo, VT), (SDNode*)0)); 2381 continue; 2382 } 2383 case OPC_EmitRegister2: { 2384 // For targets w/ more than 256 register names, the register enum 2385 // values are stored in two bytes in the matcher table (just like 2386 // opcodes). 2387 MVT::SimpleValueType VT = 2388 (MVT::SimpleValueType)MatcherTable[MatcherIndex++]; 2389 unsigned RegNo = MatcherTable[MatcherIndex++]; 2390 RegNo |= MatcherTable[MatcherIndex++] << 8; 2391 RecordedNodes.push_back(std::pair<SDValue, SDNode*>( 2392 CurDAG->getRegister(RegNo, VT), (SDNode*)0)); 2393 continue; 2394 } 2395 2396 case OPC_EmitConvertToTarget: { 2397 // Convert from IMM/FPIMM to target version. 2398 unsigned RecNo = MatcherTable[MatcherIndex++]; 2399 assert(RecNo < RecordedNodes.size() && "Invalid CheckSame"); 2400 SDValue Imm = RecordedNodes[RecNo].first; 2401 2402 if (Imm->getOpcode() == ISD::Constant) { 2403 int64_t Val = cast<ConstantSDNode>(Imm)->getZExtValue(); 2404 Imm = CurDAG->getTargetConstant(Val, Imm.getValueType()); 2405 } else if (Imm->getOpcode() == ISD::ConstantFP) { 2406 const ConstantFP *Val=cast<ConstantFPSDNode>(Imm)->getConstantFPValue(); 2407 Imm = CurDAG->getTargetConstantFP(*Val, Imm.getValueType()); 2408 } 2409 2410 RecordedNodes.push_back(std::make_pair(Imm, RecordedNodes[RecNo].second)); 2411 continue; 2412 } 2413 2414 case OPC_EmitMergeInputChains1_0: // OPC_EmitMergeInputChains, 1, 0 2415 case OPC_EmitMergeInputChains1_1: { // OPC_EmitMergeInputChains, 1, 1 2416 // These are space-optimized forms of OPC_EmitMergeInputChains. 2417 assert(InputChain.getNode() == 0 && 2418 "EmitMergeInputChains should be the first chain producing node"); 2419 assert(ChainNodesMatched.empty() && 2420 "Should only have one EmitMergeInputChains per match"); 2421 2422 // Read all of the chained nodes. 2423 unsigned RecNo = Opcode == OPC_EmitMergeInputChains1_1; 2424 assert(RecNo < RecordedNodes.size() && "Invalid CheckSame"); 2425 ChainNodesMatched.push_back(RecordedNodes[RecNo].first.getNode()); 2426 2427 // FIXME: What if other value results of the node have uses not matched 2428 // by this pattern? 2429 if (ChainNodesMatched.back() != NodeToMatch && 2430 !RecordedNodes[RecNo].first.hasOneUse()) { 2431 ChainNodesMatched.clear(); 2432 break; 2433 } 2434 2435 // Merge the input chains if they are not intra-pattern references. 2436 InputChain = HandleMergeInputChains(ChainNodesMatched, CurDAG); 2437 2438 if (InputChain.getNode() == 0) 2439 break; // Failed to merge. 2440 continue; 2441 } 2442 2443 case OPC_EmitMergeInputChains: { 2444 assert(InputChain.getNode() == 0 && 2445 "EmitMergeInputChains should be the first chain producing node"); 2446 // This node gets a list of nodes we matched in the input that have 2447 // chains. We want to token factor all of the input chains to these nodes 2448 // together. However, if any of the input chains is actually one of the 2449 // nodes matched in this pattern, then we have an intra-match reference. 2450 // Ignore these because the newly token factored chain should not refer to 2451 // the old nodes. 2452 unsigned NumChains = MatcherTable[MatcherIndex++]; 2453 assert(NumChains != 0 && "Can't TF zero chains"); 2454 2455 assert(ChainNodesMatched.empty() && 2456 "Should only have one EmitMergeInputChains per match"); 2457 2458 // Read all of the chained nodes. 2459 for (unsigned i = 0; i != NumChains; ++i) { 2460 unsigned RecNo = MatcherTable[MatcherIndex++]; 2461 assert(RecNo < RecordedNodes.size() && "Invalid CheckSame"); 2462 ChainNodesMatched.push_back(RecordedNodes[RecNo].first.getNode()); 2463 2464 // FIXME: What if other value results of the node have uses not matched 2465 // by this pattern? 2466 if (ChainNodesMatched.back() != NodeToMatch && 2467 !RecordedNodes[RecNo].first.hasOneUse()) { 2468 ChainNodesMatched.clear(); 2469 break; 2470 } 2471 } 2472 2473 // If the inner loop broke out, the match fails. 2474 if (ChainNodesMatched.empty()) 2475 break; 2476 2477 // Merge the input chains if they are not intra-pattern references. 2478 InputChain = HandleMergeInputChains(ChainNodesMatched, CurDAG); 2479 2480 if (InputChain.getNode() == 0) 2481 break; // Failed to merge. 2482 2483 continue; 2484 } 2485 2486 case OPC_EmitCopyToReg: { 2487 unsigned RecNo = MatcherTable[MatcherIndex++]; 2488 assert(RecNo < RecordedNodes.size() && "Invalid CheckSame"); 2489 unsigned DestPhysReg = MatcherTable[MatcherIndex++]; 2490 2491 if (InputChain.getNode() == 0) 2492 InputChain = CurDAG->getEntryNode(); 2493 2494 InputChain = CurDAG->getCopyToReg(InputChain, NodeToMatch->getDebugLoc(), 2495 DestPhysReg, RecordedNodes[RecNo].first, 2496 InputGlue); 2497 2498 InputGlue = InputChain.getValue(1); 2499 continue; 2500 } 2501 2502 case OPC_EmitNodeXForm: { 2503 unsigned XFormNo = MatcherTable[MatcherIndex++]; 2504 unsigned RecNo = MatcherTable[MatcherIndex++]; 2505 assert(RecNo < RecordedNodes.size() && "Invalid CheckSame"); 2506 SDValue Res = RunSDNodeXForm(RecordedNodes[RecNo].first, XFormNo); 2507 RecordedNodes.push_back(std::pair<SDValue,SDNode*>(Res, (SDNode*) 0)); 2508 continue; 2509 } 2510 2511 case OPC_EmitNode: 2512 case OPC_MorphNodeTo: { 2513 uint16_t TargetOpc = MatcherTable[MatcherIndex++]; 2514 TargetOpc |= (unsigned short)MatcherTable[MatcherIndex++] << 8; 2515 unsigned EmitNodeInfo = MatcherTable[MatcherIndex++]; 2516 // Get the result VT list. 2517 unsigned NumVTs = MatcherTable[MatcherIndex++]; 2518 SmallVector<EVT, 4> VTs; 2519 for (unsigned i = 0; i != NumVTs; ++i) { 2520 MVT::SimpleValueType VT = 2521 (MVT::SimpleValueType)MatcherTable[MatcherIndex++]; 2522 if (VT == MVT::iPTR) VT = TLI.getPointerTy().SimpleTy; 2523 VTs.push_back(VT); 2524 } 2525 2526 if (EmitNodeInfo & OPFL_Chain) 2527 VTs.push_back(MVT::Other); 2528 if (EmitNodeInfo & OPFL_GlueOutput) 2529 VTs.push_back(MVT::Glue); 2530 2531 // This is hot code, so optimize the two most common cases of 1 and 2 2532 // results. 2533 SDVTList VTList; 2534 if (VTs.size() == 1) 2535 VTList = CurDAG->getVTList(VTs[0]); 2536 else if (VTs.size() == 2) 2537 VTList = CurDAG->getVTList(VTs[0], VTs[1]); 2538 else 2539 VTList = CurDAG->getVTList(VTs.data(), VTs.size()); 2540 2541 // Get the operand list. 2542 unsigned NumOps = MatcherTable[MatcherIndex++]; 2543 SmallVector<SDValue, 8> Ops; 2544 for (unsigned i = 0; i != NumOps; ++i) { 2545 unsigned RecNo = MatcherTable[MatcherIndex++]; 2546 if (RecNo & 128) 2547 RecNo = GetVBR(RecNo, MatcherTable, MatcherIndex); 2548 2549 assert(RecNo < RecordedNodes.size() && "Invalid EmitNode"); 2550 Ops.push_back(RecordedNodes[RecNo].first); 2551 } 2552 2553 // If there are variadic operands to add, handle them now. 2554 if (EmitNodeInfo & OPFL_VariadicInfo) { 2555 // Determine the start index to copy from. 2556 unsigned FirstOpToCopy = getNumFixedFromVariadicInfo(EmitNodeInfo); 2557 FirstOpToCopy += (EmitNodeInfo & OPFL_Chain) ? 1 : 0; 2558 assert(NodeToMatch->getNumOperands() >= FirstOpToCopy && 2559 "Invalid variadic node"); 2560 // Copy all of the variadic operands, not including a potential glue 2561 // input. 2562 for (unsigned i = FirstOpToCopy, e = NodeToMatch->getNumOperands(); 2563 i != e; ++i) { 2564 SDValue V = NodeToMatch->getOperand(i); 2565 if (V.getValueType() == MVT::Glue) break; 2566 Ops.push_back(V); 2567 } 2568 } 2569 2570 // If this has chain/glue inputs, add them. 2571 if (EmitNodeInfo & OPFL_Chain) 2572 Ops.push_back(InputChain); 2573 if ((EmitNodeInfo & OPFL_GlueInput) && InputGlue.getNode() != 0) 2574 Ops.push_back(InputGlue); 2575 2576 // Create the node. 2577 SDNode *Res = 0; 2578 if (Opcode != OPC_MorphNodeTo) { 2579 // If this is a normal EmitNode command, just create the new node and 2580 // add the results to the RecordedNodes list. 2581 Res = CurDAG->getMachineNode(TargetOpc, NodeToMatch->getDebugLoc(), 2582 VTList, Ops.data(), Ops.size()); 2583 2584 // Add all the non-glue/non-chain results to the RecordedNodes list. 2585 for (unsigned i = 0, e = VTs.size(); i != e; ++i) { 2586 if (VTs[i] == MVT::Other || VTs[i] == MVT::Glue) break; 2587 RecordedNodes.push_back(std::pair<SDValue,SDNode*>(SDValue(Res, i), 2588 (SDNode*) 0)); 2589 } 2590 2591 } else { 2592 Res = MorphNode(NodeToMatch, TargetOpc, VTList, Ops.data(), Ops.size(), 2593 EmitNodeInfo); 2594 } 2595 2596 // If the node had chain/glue results, update our notion of the current 2597 // chain and glue. 2598 if (EmitNodeInfo & OPFL_GlueOutput) { 2599 InputGlue = SDValue(Res, VTs.size()-1); 2600 if (EmitNodeInfo & OPFL_Chain) 2601 InputChain = SDValue(Res, VTs.size()-2); 2602 } else if (EmitNodeInfo & OPFL_Chain) 2603 InputChain = SDValue(Res, VTs.size()-1); 2604 2605 // If the OPFL_MemRefs glue is set on this node, slap all of the 2606 // accumulated memrefs onto it. 2607 // 2608 // FIXME: This is vastly incorrect for patterns with multiple outputs 2609 // instructions that access memory and for ComplexPatterns that match 2610 // loads. 2611 if (EmitNodeInfo & OPFL_MemRefs) { 2612 // Only attach load or store memory operands if the generated 2613 // instruction may load or store. 2614 const MCInstrDesc &MCID = TM.getInstrInfo()->get(TargetOpc); 2615 bool mayLoad = MCID.mayLoad(); 2616 bool mayStore = MCID.mayStore(); 2617 2618 unsigned NumMemRefs = 0; 2619 for (SmallVector<MachineMemOperand*, 2>::const_iterator I = 2620 MatchedMemRefs.begin(), E = MatchedMemRefs.end(); I != E; ++I) { 2621 if ((*I)->isLoad()) { 2622 if (mayLoad) 2623 ++NumMemRefs; 2624 } else if ((*I)->isStore()) { 2625 if (mayStore) 2626 ++NumMemRefs; 2627 } else { 2628 ++NumMemRefs; 2629 } 2630 } 2631 2632 MachineSDNode::mmo_iterator MemRefs = 2633 MF->allocateMemRefsArray(NumMemRefs); 2634 2635 MachineSDNode::mmo_iterator MemRefsPos = MemRefs; 2636 for (SmallVector<MachineMemOperand*, 2>::const_iterator I = 2637 MatchedMemRefs.begin(), E = MatchedMemRefs.end(); I != E; ++I) { 2638 if ((*I)->isLoad()) { 2639 if (mayLoad) 2640 *MemRefsPos++ = *I; 2641 } else if ((*I)->isStore()) { 2642 if (mayStore) 2643 *MemRefsPos++ = *I; 2644 } else { 2645 *MemRefsPos++ = *I; 2646 } 2647 } 2648 2649 cast<MachineSDNode>(Res) 2650 ->setMemRefs(MemRefs, MemRefs + NumMemRefs); 2651 } 2652 2653 DEBUG(errs() << " " 2654 << (Opcode == OPC_MorphNodeTo ? "Morphed" : "Created") 2655 << " node: "; Res->dump(CurDAG); errs() << "\n"); 2656 2657 // If this was a MorphNodeTo then we're completely done! 2658 if (Opcode == OPC_MorphNodeTo) { 2659 // Update chain and glue uses. 2660 UpdateChainsAndGlue(NodeToMatch, InputChain, ChainNodesMatched, 2661 InputGlue, GlueResultNodesMatched, true); 2662 return Res; 2663 } 2664 2665 continue; 2666 } 2667 2668 case OPC_MarkGlueResults: { 2669 unsigned NumNodes = MatcherTable[MatcherIndex++]; 2670 2671 // Read and remember all the glue-result nodes. 2672 for (unsigned i = 0; i != NumNodes; ++i) { 2673 unsigned RecNo = MatcherTable[MatcherIndex++]; 2674 if (RecNo & 128) 2675 RecNo = GetVBR(RecNo, MatcherTable, MatcherIndex); 2676 2677 assert(RecNo < RecordedNodes.size() && "Invalid CheckSame"); 2678 GlueResultNodesMatched.push_back(RecordedNodes[RecNo].first.getNode()); 2679 } 2680 continue; 2681 } 2682 2683 case OPC_CompleteMatch: { 2684 // The match has been completed, and any new nodes (if any) have been 2685 // created. Patch up references to the matched dag to use the newly 2686 // created nodes. 2687 unsigned NumResults = MatcherTable[MatcherIndex++]; 2688 2689 for (unsigned i = 0; i != NumResults; ++i) { 2690 unsigned ResSlot = MatcherTable[MatcherIndex++]; 2691 if (ResSlot & 128) 2692 ResSlot = GetVBR(ResSlot, MatcherTable, MatcherIndex); 2693 2694 assert(ResSlot < RecordedNodes.size() && "Invalid CheckSame"); 2695 SDValue Res = RecordedNodes[ResSlot].first; 2696 2697 assert(i < NodeToMatch->getNumValues() && 2698 NodeToMatch->getValueType(i) != MVT::Other && 2699 NodeToMatch->getValueType(i) != MVT::Glue && 2700 "Invalid number of results to complete!"); 2701 assert((NodeToMatch->getValueType(i) == Res.getValueType() || 2702 NodeToMatch->getValueType(i) == MVT::iPTR || 2703 Res.getValueType() == MVT::iPTR || 2704 NodeToMatch->getValueType(i).getSizeInBits() == 2705 Res.getValueType().getSizeInBits()) && 2706 "invalid replacement"); 2707 CurDAG->ReplaceAllUsesOfValueWith(SDValue(NodeToMatch, i), Res); 2708 } 2709 2710 // If the root node defines glue, add it to the glue nodes to update list. 2711 if (NodeToMatch->getValueType(NodeToMatch->getNumValues()-1) == MVT::Glue) 2712 GlueResultNodesMatched.push_back(NodeToMatch); 2713 2714 // Update chain and glue uses. 2715 UpdateChainsAndGlue(NodeToMatch, InputChain, ChainNodesMatched, 2716 InputGlue, GlueResultNodesMatched, false); 2717 2718 assert(NodeToMatch->use_empty() && 2719 "Didn't replace all uses of the node?"); 2720 2721 // FIXME: We just return here, which interacts correctly with SelectRoot 2722 // above. We should fix this to not return an SDNode* anymore. 2723 return 0; 2724 } 2725 } 2726 2727 // If the code reached this point, then the match failed. See if there is 2728 // another child to try in the current 'Scope', otherwise pop it until we 2729 // find a case to check. 2730 DEBUG(errs() << " Match failed at index " << CurrentOpcodeIndex << "\n"); 2731 ++NumDAGIselRetries; 2732 while (1) { 2733 if (MatchScopes.empty()) { 2734 CannotYetSelect(NodeToMatch); 2735 return 0; 2736 } 2737 2738 // Restore the interpreter state back to the point where the scope was 2739 // formed. 2740 MatchScope &LastScope = MatchScopes.back(); 2741 RecordedNodes.resize(LastScope.NumRecordedNodes); 2742 NodeStack.clear(); 2743 NodeStack.append(LastScope.NodeStack.begin(), LastScope.NodeStack.end()); 2744 N = NodeStack.back(); 2745 2746 if (LastScope.NumMatchedMemRefs != MatchedMemRefs.size()) 2747 MatchedMemRefs.resize(LastScope.NumMatchedMemRefs); 2748 MatcherIndex = LastScope.FailIndex; 2749 2750 DEBUG(errs() << " Continuing at " << MatcherIndex << "\n"); 2751 2752 InputChain = LastScope.InputChain; 2753 InputGlue = LastScope.InputGlue; 2754 if (!LastScope.HasChainNodesMatched) 2755 ChainNodesMatched.clear(); 2756 if (!LastScope.HasGlueResultNodesMatched) 2757 GlueResultNodesMatched.clear(); 2758 2759 // Check to see what the offset is at the new MatcherIndex. If it is zero 2760 // we have reached the end of this scope, otherwise we have another child 2761 // in the current scope to try. 2762 unsigned NumToSkip = MatcherTable[MatcherIndex++]; 2763 if (NumToSkip & 128) 2764 NumToSkip = GetVBR(NumToSkip, MatcherTable, MatcherIndex); 2765 2766 // If we have another child in this scope to match, update FailIndex and 2767 // try it. 2768 if (NumToSkip != 0) { 2769 LastScope.FailIndex = MatcherIndex+NumToSkip; 2770 break; 2771 } 2772 2773 // End of this scope, pop it and try the next child in the containing 2774 // scope. 2775 MatchScopes.pop_back(); 2776 } 2777 } 2778 } 2779 2780 2781 2782 void SelectionDAGISel::CannotYetSelect(SDNode *N) { 2783 std::string msg; 2784 raw_string_ostream Msg(msg); 2785 Msg << "Cannot select: "; 2786 2787 if (N->getOpcode() != ISD::INTRINSIC_W_CHAIN && 2788 N->getOpcode() != ISD::INTRINSIC_WO_CHAIN && 2789 N->getOpcode() != ISD::INTRINSIC_VOID) { 2790 N->printrFull(Msg, CurDAG); 2791 } else { 2792 bool HasInputChain = N->getOperand(0).getValueType() == MVT::Other; 2793 unsigned iid = 2794 cast<ConstantSDNode>(N->getOperand(HasInputChain))->getZExtValue(); 2795 if (iid < Intrinsic::num_intrinsics) 2796 Msg << "intrinsic %" << Intrinsic::getName((Intrinsic::ID)iid); 2797 else if (const TargetIntrinsicInfo *TII = TM.getIntrinsicInfo()) 2798 Msg << "target intrinsic %" << TII->getName(iid); 2799 else 2800 Msg << "unknown intrinsic #" << iid; 2801 } 2802 report_fatal_error(Msg.str()); 2803 } 2804 2805 char SelectionDAGISel::ID = 0; 2806