1 //===-- AMDILCFGStructurizer.cpp - CFG Structurizer -----------------------===// 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 /// \file 9 //==-----------------------------------------------------------------------===// 10 11 #define DEBUG_TYPE "structcfg" 12 13 #include "AMDGPU.h" 14 #include "AMDGPUInstrInfo.h" 15 #include "R600InstrInfo.h" 16 #include "llvm/Support/Debug.h" 17 #include "llvm/Support/raw_ostream.h" 18 #include "llvm/ADT/SCCIterator.h" 19 #include "llvm/ADT/SmallVector.h" 20 #include "llvm/ADT/Statistic.h" 21 #include "llvm/ADT/DepthFirstIterator.h" 22 #include "llvm/Analysis/DominatorInternals.h" 23 #include "llvm/Analysis/Dominators.h" 24 #include "llvm/CodeGen/MachineDominators.h" 25 #include "llvm/CodeGen/MachineFunction.h" 26 #include "llvm/CodeGen/MachineFunctionAnalysis.h" 27 #include "llvm/CodeGen/MachineFunctionPass.h" 28 #include "llvm/CodeGen/MachineInstrBuilder.h" 29 #include "llvm/CodeGen/MachineJumpTableInfo.h" 30 #include "llvm/CodeGen/MachineLoopInfo.h" 31 #include "llvm/CodeGen/MachinePostDominators.h" 32 #include "llvm/CodeGen/MachineRegisterInfo.h" 33 #include "llvm/Target/TargetInstrInfo.h" 34 #include "llvm/Target/TargetMachine.h" 35 36 using namespace llvm; 37 38 #define DEFAULT_VEC_SLOTS 8 39 40 // TODO: move-begin. 41 42 //===----------------------------------------------------------------------===// 43 // 44 // Statistics for CFGStructurizer. 45 // 46 //===----------------------------------------------------------------------===// 47 48 STATISTIC(numSerialPatternMatch, "CFGStructurizer number of serial pattern " 49 "matched"); 50 STATISTIC(numIfPatternMatch, "CFGStructurizer number of if pattern " 51 "matched"); 52 STATISTIC(numLoopcontPatternMatch, "CFGStructurizer number of loop-continue " 53 "pattern matched"); 54 STATISTIC(numClonedBlock, "CFGStructurizer cloned blocks"); 55 STATISTIC(numClonedInstr, "CFGStructurizer cloned instructions"); 56 57 //===----------------------------------------------------------------------===// 58 // 59 // Miscellaneous utility for CFGStructurizer. 60 // 61 //===----------------------------------------------------------------------===// 62 namespace { 63 #define SHOWNEWINSTR(i) \ 64 DEBUG(dbgs() << "New instr: " << *i << "\n"); 65 66 #define SHOWNEWBLK(b, msg) \ 67 DEBUG( \ 68 dbgs() << msg << "BB" << b->getNumber() << "size " << b->size(); \ 69 dbgs() << "\n"; \ 70 ); 71 72 #define SHOWBLK_DETAIL(b, msg) \ 73 DEBUG( \ 74 if (b) { \ 75 dbgs() << msg << "BB" << b->getNumber() << "size " << b->size(); \ 76 b->print(dbgs()); \ 77 dbgs() << "\n"; \ 78 } \ 79 ); 80 81 #define INVALIDSCCNUM -1 82 83 template<class NodeT> 84 void ReverseVector(SmallVectorImpl<NodeT *> &Src) { 85 size_t sz = Src.size(); 86 for (size_t i = 0; i < sz/2; ++i) { 87 NodeT *t = Src[i]; 88 Src[i] = Src[sz - i - 1]; 89 Src[sz - i - 1] = t; 90 } 91 } 92 93 } // end anonymous namespace 94 95 //===----------------------------------------------------------------------===// 96 // 97 // supporting data structure for CFGStructurizer 98 // 99 //===----------------------------------------------------------------------===// 100 101 102 namespace { 103 104 class BlockInformation { 105 public: 106 bool IsRetired; 107 int SccNum; 108 BlockInformation() : IsRetired(false), SccNum(INVALIDSCCNUM) {} 109 }; 110 111 } // end anonymous namespace 112 113 //===----------------------------------------------------------------------===// 114 // 115 // CFGStructurizer 116 // 117 //===----------------------------------------------------------------------===// 118 119 namespace { 120 class AMDGPUCFGStructurizer : public MachineFunctionPass { 121 public: 122 typedef SmallVector<MachineBasicBlock *, 32> MBBVector; 123 typedef std::map<MachineBasicBlock *, BlockInformation *> MBBInfoMap; 124 typedef std::map<MachineLoop *, MachineBasicBlock *> LoopLandInfoMap; 125 126 enum PathToKind { 127 Not_SinglePath = 0, 128 SinglePath_InPath = 1, 129 SinglePath_NotInPath = 2 130 }; 131 132 static char ID; 133 134 AMDGPUCFGStructurizer(TargetMachine &tm) : 135 MachineFunctionPass(ID), TM(tm), 136 TII(static_cast<const R600InstrInfo *>(tm.getInstrInfo())), 137 TRI(&TII->getRegisterInfo()) { } 138 139 const char *getPassName() const { 140 return "AMD IL Control Flow Graph structurizer Pass"; 141 } 142 143 void getAnalysisUsage(AnalysisUsage &AU) const { 144 AU.addPreserved<MachineFunctionAnalysis>(); 145 AU.addRequired<MachineFunctionAnalysis>(); 146 AU.addRequired<MachineDominatorTree>(); 147 AU.addRequired<MachinePostDominatorTree>(); 148 AU.addRequired<MachineLoopInfo>(); 149 } 150 151 /// Perform the CFG structurization 152 bool run(); 153 154 /// Perform the CFG preparation 155 /// This step will remove every unconditionnal/dead jump instructions and make 156 /// sure all loops have an exit block 157 bool prepare(); 158 159 bool runOnMachineFunction(MachineFunction &MF) { 160 DEBUG(MF.dump();); 161 OrderedBlks.clear(); 162 FuncRep = &MF; 163 MLI = &getAnalysis<MachineLoopInfo>(); 164 DEBUG(dbgs() << "LoopInfo:\n"; PrintLoopinfo(*MLI);); 165 MDT = &getAnalysis<MachineDominatorTree>(); 166 DEBUG(MDT->print(dbgs(), (const llvm::Module*)0);); 167 PDT = &getAnalysis<MachinePostDominatorTree>(); 168 DEBUG(PDT->print(dbgs());); 169 prepare(); 170 run(); 171 DEBUG(MF.dump();); 172 return true; 173 } 174 175 protected: 176 TargetMachine &TM; 177 MachineDominatorTree *MDT; 178 MachinePostDominatorTree *PDT; 179 MachineLoopInfo *MLI; 180 const R600InstrInfo *TII; 181 const AMDGPURegisterInfo *TRI; 182 183 // PRINT FUNCTIONS 184 /// Print the ordered Blocks. 185 void printOrderedBlocks() const { 186 size_t i = 0; 187 for (MBBVector::const_iterator iterBlk = OrderedBlks.begin(), 188 iterBlkEnd = OrderedBlks.end(); iterBlk != iterBlkEnd; ++iterBlk, ++i) { 189 dbgs() << "BB" << (*iterBlk)->getNumber(); 190 dbgs() << "(" << getSCCNum(*iterBlk) << "," << (*iterBlk)->size() << ")"; 191 if (i != 0 && i % 10 == 0) { 192 dbgs() << "\n"; 193 } else { 194 dbgs() << " "; 195 } 196 } 197 } 198 static void PrintLoopinfo(const MachineLoopInfo &LoopInfo) { 199 for (MachineLoop::iterator iter = LoopInfo.begin(), 200 iterEnd = LoopInfo.end(); iter != iterEnd; ++iter) { 201 (*iter)->print(dbgs(), 0); 202 } 203 } 204 205 // UTILITY FUNCTIONS 206 int getSCCNum(MachineBasicBlock *MBB) const; 207 MachineBasicBlock *getLoopLandInfo(MachineLoop *LoopRep) const; 208 bool hasBackEdge(MachineBasicBlock *MBB) const; 209 static unsigned getLoopDepth(MachineLoop *LoopRep); 210 bool isRetiredBlock(MachineBasicBlock *MBB) const; 211 bool isActiveLoophead(MachineBasicBlock *MBB) const; 212 PathToKind singlePathTo(MachineBasicBlock *SrcMBB, MachineBasicBlock *DstMBB, 213 bool AllowSideEntry = true) const; 214 int countActiveBlock(MBBVector::const_iterator It, 215 MBBVector::const_iterator E) const; 216 bool needMigrateBlock(MachineBasicBlock *MBB) const; 217 218 // Utility Functions 219 void reversePredicateSetter(MachineBasicBlock::iterator I); 220 /// Compute the reversed DFS post order of Blocks 221 void orderBlocks(MachineFunction *MF); 222 223 // Function originaly from CFGStructTraits 224 void insertInstrEnd(MachineBasicBlock *MBB, int NewOpcode, 225 DebugLoc DL = DebugLoc()); 226 MachineInstr *insertInstrBefore(MachineBasicBlock *MBB, int NewOpcode, 227 DebugLoc DL = DebugLoc()); 228 MachineInstr *insertInstrBefore(MachineBasicBlock::iterator I, int NewOpcode); 229 void insertCondBranchBefore(MachineBasicBlock::iterator I, int NewOpcode, 230 DebugLoc DL); 231 void insertCondBranchBefore(MachineBasicBlock *MBB, 232 MachineBasicBlock::iterator I, int NewOpcode, int RegNum, 233 DebugLoc DL); 234 void insertCondBranchEnd(MachineBasicBlock *MBB, int NewOpcode, int RegNum); 235 static int getBranchNzeroOpcode(int OldOpcode); 236 static int getBranchZeroOpcode(int OldOpcode); 237 static int getContinueNzeroOpcode(int OldOpcode); 238 static int getContinueZeroOpcode(int OldOpcode); 239 static MachineBasicBlock *getTrueBranch(MachineInstr *MI); 240 static void setTrueBranch(MachineInstr *MI, MachineBasicBlock *MBB); 241 static MachineBasicBlock *getFalseBranch(MachineBasicBlock *MBB, 242 MachineInstr *MI); 243 static bool isCondBranch(MachineInstr *MI); 244 static bool isUncondBranch(MachineInstr *MI); 245 static DebugLoc getLastDebugLocInBB(MachineBasicBlock *MBB); 246 static MachineInstr *getNormalBlockBranchInstr(MachineBasicBlock *MBB); 247 /// The correct naming for this is getPossibleLoopendBlockBranchInstr. 248 /// 249 /// BB with backward-edge could have move instructions after the branch 250 /// instruction. Such move instruction "belong to" the loop backward-edge. 251 MachineInstr *getLoopendBlockBranchInstr(MachineBasicBlock *MBB); 252 static MachineInstr *getReturnInstr(MachineBasicBlock *MBB); 253 static MachineInstr *getContinueInstr(MachineBasicBlock *MBB); 254 static bool isReturnBlock(MachineBasicBlock *MBB); 255 static void cloneSuccessorList(MachineBasicBlock *DstMBB, 256 MachineBasicBlock *SrcMBB) ; 257 static MachineBasicBlock *clone(MachineBasicBlock *MBB); 258 /// MachineBasicBlock::ReplaceUsesOfBlockWith doesn't serve the purpose 259 /// because the AMDGPU instruction is not recognized as terminator fix this 260 /// and retire this routine 261 void replaceInstrUseOfBlockWith(MachineBasicBlock *SrcMBB, 262 MachineBasicBlock *OldMBB, MachineBasicBlock *NewBlk); 263 static void wrapup(MachineBasicBlock *MBB); 264 265 266 int patternMatch(MachineBasicBlock *MBB); 267 int patternMatchGroup(MachineBasicBlock *MBB); 268 int serialPatternMatch(MachineBasicBlock *MBB); 269 int ifPatternMatch(MachineBasicBlock *MBB); 270 int loopendPatternMatch(); 271 int mergeLoop(MachineLoop *LoopRep); 272 int loopcontPatternMatch(MachineLoop *LoopRep, MachineBasicBlock *LoopHeader); 273 274 void handleLoopcontBlock(MachineBasicBlock *ContingMBB, 275 MachineLoop *ContingLoop, MachineBasicBlock *ContMBB, 276 MachineLoop *ContLoop); 277 /// return true iff src1Blk->succ_size() == 0 && src1Blk and src2Blk are in 278 /// the same loop with LoopLandInfo without explicitly keeping track of 279 /// loopContBlks and loopBreakBlks, this is a method to get the information. 280 bool isSameloopDetachedContbreak(MachineBasicBlock *Src1MBB, 281 MachineBasicBlock *Src2MBB); 282 int handleJumpintoIf(MachineBasicBlock *HeadMBB, 283 MachineBasicBlock *TrueMBB, MachineBasicBlock *FalseMBB); 284 int handleJumpintoIfImp(MachineBasicBlock *HeadMBB, 285 MachineBasicBlock *TrueMBB, MachineBasicBlock *FalseMBB); 286 int improveSimpleJumpintoIf(MachineBasicBlock *HeadMBB, 287 MachineBasicBlock *TrueMBB, MachineBasicBlock *FalseMBB, 288 MachineBasicBlock **LandMBBPtr); 289 void showImproveSimpleJumpintoIf(MachineBasicBlock *HeadMBB, 290 MachineBasicBlock *TrueMBB, MachineBasicBlock *FalseMBB, 291 MachineBasicBlock *LandMBB, bool Detail = false); 292 int cloneOnSideEntryTo(MachineBasicBlock *PreMBB, 293 MachineBasicBlock *SrcMBB, MachineBasicBlock *DstMBB); 294 void mergeSerialBlock(MachineBasicBlock *DstMBB, 295 MachineBasicBlock *SrcMBB); 296 297 void mergeIfthenelseBlock(MachineInstr *BranchMI, 298 MachineBasicBlock *MBB, MachineBasicBlock *TrueMBB, 299 MachineBasicBlock *FalseMBB, MachineBasicBlock *LandMBB); 300 void mergeLooplandBlock(MachineBasicBlock *DstMBB, 301 MachineBasicBlock *LandMBB); 302 void mergeLoopbreakBlock(MachineBasicBlock *ExitingMBB, 303 MachineBasicBlock *LandMBB); 304 void settleLoopcontBlock(MachineBasicBlock *ContingMBB, 305 MachineBasicBlock *ContMBB); 306 /// normalizeInfiniteLoopExit change 307 /// B1: 308 /// uncond_br LoopHeader 309 /// 310 /// to 311 /// B1: 312 /// cond_br 1 LoopHeader dummyExit 313 /// and return the newly added dummy exit block 314 MachineBasicBlock *normalizeInfiniteLoopExit(MachineLoop *LoopRep); 315 void removeUnconditionalBranch(MachineBasicBlock *MBB); 316 /// Remove duplicate branches instructions in a block. 317 /// For instance 318 /// B0: 319 /// cond_br X B1 B2 320 /// cond_br X B1 B2 321 /// is transformed to 322 /// B0: 323 /// cond_br X B1 B2 324 void removeRedundantConditionalBranch(MachineBasicBlock *MBB); 325 void addDummyExitBlock(SmallVectorImpl<MachineBasicBlock *> &RetMBB); 326 void removeSuccessor(MachineBasicBlock *MBB); 327 MachineBasicBlock *cloneBlockForPredecessor(MachineBasicBlock *MBB, 328 MachineBasicBlock *PredMBB); 329 void migrateInstruction(MachineBasicBlock *SrcMBB, 330 MachineBasicBlock *DstMBB, MachineBasicBlock::iterator I); 331 void recordSccnum(MachineBasicBlock *MBB, int SCCNum); 332 void retireBlock(MachineBasicBlock *MBB); 333 void setLoopLandBlock(MachineLoop *LoopRep, MachineBasicBlock *MBB = NULL); 334 335 MachineBasicBlock *findNearestCommonPostDom(std::set<MachineBasicBlock *>&); 336 /// This is work around solution for findNearestCommonDominator not avaiable 337 /// to post dom a proper fix should go to Dominators.h. 338 MachineBasicBlock *findNearestCommonPostDom(MachineBasicBlock *MBB1, 339 MachineBasicBlock *MBB2); 340 341 private: 342 MBBInfoMap BlockInfoMap; 343 LoopLandInfoMap LLInfoMap; 344 std::map<MachineLoop *, bool> Visited; 345 MachineFunction *FuncRep; 346 SmallVector<MachineBasicBlock *, DEFAULT_VEC_SLOTS> OrderedBlks; 347 }; 348 349 int AMDGPUCFGStructurizer::getSCCNum(MachineBasicBlock *MBB) const { 350 MBBInfoMap::const_iterator It = BlockInfoMap.find(MBB); 351 if (It == BlockInfoMap.end()) 352 return INVALIDSCCNUM; 353 return (*It).second->SccNum; 354 } 355 356 MachineBasicBlock *AMDGPUCFGStructurizer::getLoopLandInfo(MachineLoop *LoopRep) 357 const { 358 LoopLandInfoMap::const_iterator It = LLInfoMap.find(LoopRep); 359 if (It == LLInfoMap.end()) 360 return NULL; 361 return (*It).second; 362 } 363 364 bool AMDGPUCFGStructurizer::hasBackEdge(MachineBasicBlock *MBB) const { 365 MachineLoop *LoopRep = MLI->getLoopFor(MBB); 366 if (!LoopRep) 367 return false; 368 MachineBasicBlock *LoopHeader = LoopRep->getHeader(); 369 return MBB->isSuccessor(LoopHeader); 370 } 371 372 unsigned AMDGPUCFGStructurizer::getLoopDepth(MachineLoop *LoopRep) { 373 return LoopRep ? LoopRep->getLoopDepth() : 0; 374 } 375 376 bool AMDGPUCFGStructurizer::isRetiredBlock(MachineBasicBlock *MBB) const { 377 MBBInfoMap::const_iterator It = BlockInfoMap.find(MBB); 378 if (It == BlockInfoMap.end()) 379 return false; 380 return (*It).second->IsRetired; 381 } 382 383 bool AMDGPUCFGStructurizer::isActiveLoophead(MachineBasicBlock *MBB) const { 384 MachineLoop *LoopRep = MLI->getLoopFor(MBB); 385 while (LoopRep && LoopRep->getHeader() == MBB) { 386 MachineBasicBlock *LoopLand = getLoopLandInfo(LoopRep); 387 if(!LoopLand) 388 return true; 389 if (!isRetiredBlock(LoopLand)) 390 return true; 391 LoopRep = LoopRep->getParentLoop(); 392 } 393 return false; 394 } 395 AMDGPUCFGStructurizer::PathToKind AMDGPUCFGStructurizer::singlePathTo( 396 MachineBasicBlock *SrcMBB, MachineBasicBlock *DstMBB, 397 bool AllowSideEntry) const { 398 assert(DstMBB); 399 if (SrcMBB == DstMBB) 400 return SinglePath_InPath; 401 while (SrcMBB && SrcMBB->succ_size() == 1) { 402 SrcMBB = *SrcMBB->succ_begin(); 403 if (SrcMBB == DstMBB) 404 return SinglePath_InPath; 405 if (!AllowSideEntry && SrcMBB->pred_size() > 1) 406 return Not_SinglePath; 407 } 408 if (SrcMBB && SrcMBB->succ_size()==0) 409 return SinglePath_NotInPath; 410 return Not_SinglePath; 411 } 412 413 int AMDGPUCFGStructurizer::countActiveBlock(MBBVector::const_iterator It, 414 MBBVector::const_iterator E) const { 415 int Count = 0; 416 while (It != E) { 417 if (!isRetiredBlock(*It)) 418 ++Count; 419 ++It; 420 } 421 return Count; 422 } 423 424 bool AMDGPUCFGStructurizer::needMigrateBlock(MachineBasicBlock *MBB) const { 425 unsigned BlockSizeThreshold = 30; 426 unsigned CloneInstrThreshold = 100; 427 bool MultiplePreds = MBB && (MBB->pred_size() > 1); 428 429 if(!MultiplePreds) 430 return false; 431 unsigned BlkSize = MBB->size(); 432 return ((BlkSize > BlockSizeThreshold) && 433 (BlkSize * (MBB->pred_size() - 1) > CloneInstrThreshold)); 434 } 435 436 void AMDGPUCFGStructurizer::reversePredicateSetter( 437 MachineBasicBlock::iterator I) { 438 while (I--) { 439 if (I->getOpcode() == AMDGPU::PRED_X) { 440 switch (static_cast<MachineInstr *>(I)->getOperand(2).getImm()) { 441 case OPCODE_IS_ZERO_INT: 442 static_cast<MachineInstr *>(I)->getOperand(2) 443 .setImm(OPCODE_IS_NOT_ZERO_INT); 444 return; 445 case OPCODE_IS_NOT_ZERO_INT: 446 static_cast<MachineInstr *>(I)->getOperand(2) 447 .setImm(OPCODE_IS_ZERO_INT); 448 return; 449 case OPCODE_IS_ZERO: 450 static_cast<MachineInstr *>(I)->getOperand(2) 451 .setImm(OPCODE_IS_NOT_ZERO); 452 return; 453 case OPCODE_IS_NOT_ZERO: 454 static_cast<MachineInstr *>(I)->getOperand(2) 455 .setImm(OPCODE_IS_ZERO); 456 return; 457 default: 458 llvm_unreachable("PRED_X Opcode invalid!"); 459 } 460 } 461 } 462 } 463 464 void AMDGPUCFGStructurizer::insertInstrEnd(MachineBasicBlock *MBB, 465 int NewOpcode, DebugLoc DL) { 466 MachineInstr *MI = MBB->getParent() 467 ->CreateMachineInstr(TII->get(NewOpcode), DL); 468 MBB->push_back(MI); 469 //assume the instruction doesn't take any reg operand ... 470 SHOWNEWINSTR(MI); 471 } 472 473 MachineInstr *AMDGPUCFGStructurizer::insertInstrBefore(MachineBasicBlock *MBB, 474 int NewOpcode, DebugLoc DL) { 475 MachineInstr *MI = 476 MBB->getParent()->CreateMachineInstr(TII->get(NewOpcode), DL); 477 if (MBB->begin() != MBB->end()) 478 MBB->insert(MBB->begin(), MI); 479 else 480 MBB->push_back(MI); 481 SHOWNEWINSTR(MI); 482 return MI; 483 } 484 485 MachineInstr *AMDGPUCFGStructurizer::insertInstrBefore( 486 MachineBasicBlock::iterator I, int NewOpcode) { 487 MachineInstr *OldMI = &(*I); 488 MachineBasicBlock *MBB = OldMI->getParent(); 489 MachineInstr *NewMBB = 490 MBB->getParent()->CreateMachineInstr(TII->get(NewOpcode), DebugLoc()); 491 MBB->insert(I, NewMBB); 492 //assume the instruction doesn't take any reg operand ... 493 SHOWNEWINSTR(NewMBB); 494 return NewMBB; 495 } 496 497 void AMDGPUCFGStructurizer::insertCondBranchBefore( 498 MachineBasicBlock::iterator I, int NewOpcode, DebugLoc DL) { 499 MachineInstr *OldMI = &(*I); 500 MachineBasicBlock *MBB = OldMI->getParent(); 501 MachineFunction *MF = MBB->getParent(); 502 MachineInstr *NewMI = MF->CreateMachineInstr(TII->get(NewOpcode), DL); 503 MBB->insert(I, NewMI); 504 MachineInstrBuilder MIB(*MF, NewMI); 505 MIB.addReg(OldMI->getOperand(1).getReg(), false); 506 SHOWNEWINSTR(NewMI); 507 //erase later oldInstr->eraseFromParent(); 508 } 509 510 void AMDGPUCFGStructurizer::insertCondBranchBefore(MachineBasicBlock *blk, 511 MachineBasicBlock::iterator I, int NewOpcode, int RegNum, 512 DebugLoc DL) { 513 MachineFunction *MF = blk->getParent(); 514 MachineInstr *NewInstr = MF->CreateMachineInstr(TII->get(NewOpcode), DL); 515 //insert before 516 blk->insert(I, NewInstr); 517 MachineInstrBuilder(*MF, NewInstr).addReg(RegNum, false); 518 SHOWNEWINSTR(NewInstr); 519 } 520 521 void AMDGPUCFGStructurizer::insertCondBranchEnd(MachineBasicBlock *MBB, 522 int NewOpcode, int RegNum) { 523 MachineFunction *MF = MBB->getParent(); 524 MachineInstr *NewInstr = 525 MF->CreateMachineInstr(TII->get(NewOpcode), DebugLoc()); 526 MBB->push_back(NewInstr); 527 MachineInstrBuilder(*MF, NewInstr).addReg(RegNum, false); 528 SHOWNEWINSTR(NewInstr); 529 } 530 531 int AMDGPUCFGStructurizer::getBranchNzeroOpcode(int OldOpcode) { 532 switch(OldOpcode) { 533 case AMDGPU::JUMP_COND: 534 case AMDGPU::JUMP: return AMDGPU::IF_PREDICATE_SET; 535 case AMDGPU::BRANCH_COND_i32: 536 case AMDGPU::BRANCH_COND_f32: return AMDGPU::IF_LOGICALNZ_f32; 537 default: llvm_unreachable("internal error"); 538 } 539 return -1; 540 } 541 542 int AMDGPUCFGStructurizer::getBranchZeroOpcode(int OldOpcode) { 543 switch(OldOpcode) { 544 case AMDGPU::JUMP_COND: 545 case AMDGPU::JUMP: return AMDGPU::IF_PREDICATE_SET; 546 case AMDGPU::BRANCH_COND_i32: 547 case AMDGPU::BRANCH_COND_f32: return AMDGPU::IF_LOGICALZ_f32; 548 default: llvm_unreachable("internal error"); 549 } 550 return -1; 551 } 552 553 int AMDGPUCFGStructurizer::getContinueNzeroOpcode(int OldOpcode) { 554 switch(OldOpcode) { 555 case AMDGPU::JUMP_COND: 556 case AMDGPU::JUMP: return AMDGPU::CONTINUE_LOGICALNZ_i32; 557 default: llvm_unreachable("internal error"); 558 }; 559 return -1; 560 } 561 562 int AMDGPUCFGStructurizer::getContinueZeroOpcode(int OldOpcode) { 563 switch(OldOpcode) { 564 case AMDGPU::JUMP_COND: 565 case AMDGPU::JUMP: return AMDGPU::CONTINUE_LOGICALZ_i32; 566 default: llvm_unreachable("internal error"); 567 } 568 return -1; 569 } 570 571 MachineBasicBlock *AMDGPUCFGStructurizer::getTrueBranch(MachineInstr *MI) { 572 return MI->getOperand(0).getMBB(); 573 } 574 575 void AMDGPUCFGStructurizer::setTrueBranch(MachineInstr *MI, 576 MachineBasicBlock *MBB) { 577 MI->getOperand(0).setMBB(MBB); 578 } 579 580 MachineBasicBlock * 581 AMDGPUCFGStructurizer::getFalseBranch(MachineBasicBlock *MBB, 582 MachineInstr *MI) { 583 assert(MBB->succ_size() == 2); 584 MachineBasicBlock *TrueBranch = getTrueBranch(MI); 585 MachineBasicBlock::succ_iterator It = MBB->succ_begin(); 586 MachineBasicBlock::succ_iterator Next = It; 587 ++Next; 588 return (*It == TrueBranch) ? *Next : *It; 589 } 590 591 bool AMDGPUCFGStructurizer::isCondBranch(MachineInstr *MI) { 592 switch (MI->getOpcode()) { 593 case AMDGPU::JUMP_COND: 594 case AMDGPU::BRANCH_COND_i32: 595 case AMDGPU::BRANCH_COND_f32: return true; 596 default: 597 return false; 598 } 599 return false; 600 } 601 602 bool AMDGPUCFGStructurizer::isUncondBranch(MachineInstr *MI) { 603 switch (MI->getOpcode()) { 604 case AMDGPU::JUMP: 605 case AMDGPU::BRANCH: 606 return true; 607 default: 608 return false; 609 } 610 return false; 611 } 612 613 DebugLoc AMDGPUCFGStructurizer::getLastDebugLocInBB(MachineBasicBlock *MBB) { 614 //get DebugLoc from the first MachineBasicBlock instruction with debug info 615 DebugLoc DL; 616 for (MachineBasicBlock::iterator It = MBB->begin(); It != MBB->end(); 617 ++It) { 618 MachineInstr *instr = &(*It); 619 if (instr->getDebugLoc().isUnknown() == false) 620 DL = instr->getDebugLoc(); 621 } 622 return DL; 623 } 624 625 MachineInstr *AMDGPUCFGStructurizer::getNormalBlockBranchInstr( 626 MachineBasicBlock *MBB) { 627 MachineBasicBlock::reverse_iterator It = MBB->rbegin(); 628 MachineInstr *MI = &*It; 629 if (MI && (isCondBranch(MI) || isUncondBranch(MI))) 630 return MI; 631 return NULL; 632 } 633 634 MachineInstr *AMDGPUCFGStructurizer::getLoopendBlockBranchInstr( 635 MachineBasicBlock *MBB) { 636 for (MachineBasicBlock::reverse_iterator It = MBB->rbegin(), E = MBB->rend(); 637 It != E; ++It) { 638 // FIXME: Simplify 639 MachineInstr *MI = &*It; 640 if (MI) { 641 if (isCondBranch(MI) || isUncondBranch(MI)) 642 return MI; 643 else if (!TII->isMov(MI->getOpcode())) 644 break; 645 } 646 } 647 return NULL; 648 } 649 650 MachineInstr *AMDGPUCFGStructurizer::getReturnInstr(MachineBasicBlock *MBB) { 651 MachineBasicBlock::reverse_iterator It = MBB->rbegin(); 652 if (It != MBB->rend()) { 653 MachineInstr *instr = &(*It); 654 if (instr->getOpcode() == AMDGPU::RETURN) 655 return instr; 656 } 657 return NULL; 658 } 659 660 MachineInstr *AMDGPUCFGStructurizer::getContinueInstr(MachineBasicBlock *MBB) { 661 MachineBasicBlock::reverse_iterator It = MBB->rbegin(); 662 if (It != MBB->rend()) { 663 MachineInstr *MI = &(*It); 664 if (MI->getOpcode() == AMDGPU::CONTINUE) 665 return MI; 666 } 667 return NULL; 668 } 669 670 bool AMDGPUCFGStructurizer::isReturnBlock(MachineBasicBlock *MBB) { 671 MachineInstr *MI = getReturnInstr(MBB); 672 bool IsReturn = (MBB->succ_size() == 0); 673 if (MI) 674 assert(IsReturn); 675 else if (IsReturn) 676 DEBUG( 677 dbgs() << "BB" << MBB->getNumber() 678 <<" is return block without RETURN instr\n";); 679 return IsReturn; 680 } 681 682 void AMDGPUCFGStructurizer::cloneSuccessorList(MachineBasicBlock *DstMBB, 683 MachineBasicBlock *SrcMBB) { 684 for (MachineBasicBlock::succ_iterator It = SrcMBB->succ_begin(), 685 iterEnd = SrcMBB->succ_end(); It != iterEnd; ++It) 686 DstMBB->addSuccessor(*It); // *iter's predecessor is also taken care of 687 } 688 689 MachineBasicBlock *AMDGPUCFGStructurizer::clone(MachineBasicBlock *MBB) { 690 MachineFunction *Func = MBB->getParent(); 691 MachineBasicBlock *NewMBB = Func->CreateMachineBasicBlock(); 692 Func->push_back(NewMBB); //insert to function 693 for (MachineBasicBlock::iterator It = MBB->begin(), E = MBB->end(); 694 It != E; ++It) { 695 MachineInstr *MI = Func->CloneMachineInstr(It); 696 NewMBB->push_back(MI); 697 } 698 return NewMBB; 699 } 700 701 void AMDGPUCFGStructurizer::replaceInstrUseOfBlockWith( 702 MachineBasicBlock *SrcMBB, MachineBasicBlock *OldMBB, 703 MachineBasicBlock *NewBlk) { 704 MachineInstr *BranchMI = getLoopendBlockBranchInstr(SrcMBB); 705 if (BranchMI && isCondBranch(BranchMI) && 706 getTrueBranch(BranchMI) == OldMBB) 707 setTrueBranch(BranchMI, NewBlk); 708 } 709 710 void AMDGPUCFGStructurizer::wrapup(MachineBasicBlock *MBB) { 711 assert((!MBB->getParent()->getJumpTableInfo() 712 || MBB->getParent()->getJumpTableInfo()->isEmpty()) 713 && "found a jump table"); 714 715 //collect continue right before endloop 716 SmallVector<MachineInstr *, DEFAULT_VEC_SLOTS> ContInstr; 717 MachineBasicBlock::iterator Pre = MBB->begin(); 718 MachineBasicBlock::iterator E = MBB->end(); 719 MachineBasicBlock::iterator It = Pre; 720 while (It != E) { 721 if (Pre->getOpcode() == AMDGPU::CONTINUE 722 && It->getOpcode() == AMDGPU::ENDLOOP) 723 ContInstr.push_back(Pre); 724 Pre = It; 725 ++It; 726 } 727 728 //delete continue right before endloop 729 for (unsigned i = 0; i < ContInstr.size(); ++i) 730 ContInstr[i]->eraseFromParent(); 731 732 // TODO to fix up jump table so later phase won't be confused. if 733 // (jumpTableInfo->isEmpty() == false) { need to clean the jump table, but 734 // there isn't such an interface yet. alternatively, replace all the other 735 // blocks in the jump table with the entryBlk //} 736 737 } 738 739 740 bool AMDGPUCFGStructurizer::prepare() { 741 bool Changed = false; 742 743 //FIXME: if not reducible flow graph, make it so ??? 744 745 DEBUG(dbgs() << "AMDGPUCFGStructurizer::prepare\n";); 746 747 orderBlocks(FuncRep); 748 749 SmallVector<MachineBasicBlock *, DEFAULT_VEC_SLOTS> RetBlks; 750 751 // Add an ExitBlk to loop that don't have one 752 for (MachineLoopInfo::iterator It = MLI->begin(), 753 E = MLI->end(); It != E; ++It) { 754 MachineLoop *LoopRep = (*It); 755 MBBVector ExitingMBBs; 756 LoopRep->getExitingBlocks(ExitingMBBs); 757 758 if (ExitingMBBs.size() == 0) { 759 MachineBasicBlock* DummyExitBlk = normalizeInfiniteLoopExit(LoopRep); 760 if (DummyExitBlk) 761 RetBlks.push_back(DummyExitBlk); 762 } 763 } 764 765 // Remove unconditional branch instr. 766 // Add dummy exit block iff there are multiple returns. 767 for (SmallVectorImpl<MachineBasicBlock *>::const_iterator 768 It = OrderedBlks.begin(), E = OrderedBlks.end(); It != E; ++It) { 769 MachineBasicBlock *MBB = *It; 770 removeUnconditionalBranch(MBB); 771 removeRedundantConditionalBranch(MBB); 772 if (isReturnBlock(MBB)) { 773 RetBlks.push_back(MBB); 774 } 775 assert(MBB->succ_size() <= 2); 776 } 777 778 if (RetBlks.size() >= 2) { 779 addDummyExitBlock(RetBlks); 780 Changed = true; 781 } 782 783 return Changed; 784 } 785 786 bool AMDGPUCFGStructurizer::run() { 787 788 //Assume reducible CFG... 789 DEBUG(dbgs() << "AMDGPUCFGStructurizer::run\n";FuncRep->viewCFG();); 790 791 #ifdef STRESSTEST 792 //Use the worse block ordering to test the algorithm. 793 ReverseVector(orderedBlks); 794 #endif 795 796 DEBUG(dbgs() << "Ordered blocks:\n"; printOrderedBlocks();); 797 int NumIter = 0; 798 bool Finish = false; 799 MachineBasicBlock *MBB; 800 bool MakeProgress = false; 801 int NumRemainedBlk = countActiveBlock(OrderedBlks.begin(), 802 OrderedBlks.end()); 803 804 do { 805 ++NumIter; 806 DEBUG( 807 dbgs() << "numIter = " << NumIter 808 << ", numRemaintedBlk = " << NumRemainedBlk << "\n"; 809 ); 810 811 SmallVectorImpl<MachineBasicBlock *>::const_iterator It = 812 OrderedBlks.begin(); 813 SmallVectorImpl<MachineBasicBlock *>::const_iterator E = 814 OrderedBlks.end(); 815 816 SmallVectorImpl<MachineBasicBlock *>::const_iterator SccBeginIter = 817 It; 818 MachineBasicBlock *SccBeginMBB = NULL; 819 int SccNumBlk = 0; // The number of active blocks, init to a 820 // maximum possible number. 821 int SccNumIter; // Number of iteration in this SCC. 822 823 while (It != E) { 824 MBB = *It; 825 826 if (!SccBeginMBB) { 827 SccBeginIter = It; 828 SccBeginMBB = MBB; 829 SccNumIter = 0; 830 SccNumBlk = NumRemainedBlk; // Init to maximum possible number. 831 DEBUG( 832 dbgs() << "start processing SCC" << getSCCNum(SccBeginMBB); 833 dbgs() << "\n"; 834 ); 835 } 836 837 if (!isRetiredBlock(MBB)) 838 patternMatch(MBB); 839 840 ++It; 841 842 bool ContNextScc = true; 843 if (It == E 844 || getSCCNum(SccBeginMBB) != getSCCNum(*It)) { 845 // Just finish one scc. 846 ++SccNumIter; 847 int sccRemainedNumBlk = countActiveBlock(SccBeginIter, It); 848 if (sccRemainedNumBlk != 1 && sccRemainedNumBlk >= SccNumBlk) { 849 DEBUG( 850 dbgs() << "Can't reduce SCC " << getSCCNum(MBB) 851 << ", sccNumIter = " << SccNumIter; 852 dbgs() << "doesn't make any progress\n"; 853 ); 854 ContNextScc = true; 855 } else if (sccRemainedNumBlk != 1 && sccRemainedNumBlk < SccNumBlk) { 856 SccNumBlk = sccRemainedNumBlk; 857 It = SccBeginIter; 858 ContNextScc = false; 859 DEBUG( 860 dbgs() << "repeat processing SCC" << getSCCNum(MBB) 861 << "sccNumIter = " << SccNumIter << "\n"; 862 FuncRep->viewCFG(); 863 ); 864 } else { 865 // Finish the current scc. 866 ContNextScc = true; 867 } 868 } else { 869 // Continue on next component in the current scc. 870 ContNextScc = false; 871 } 872 873 if (ContNextScc) 874 SccBeginMBB = NULL; 875 } //while, "one iteration" over the function. 876 877 MachineBasicBlock *EntryMBB = 878 GraphTraits<MachineFunction *>::nodes_begin(FuncRep); 879 if (EntryMBB->succ_size() == 0) { 880 Finish = true; 881 DEBUG( 882 dbgs() << "Reduce to one block\n"; 883 ); 884 } else { 885 int NewnumRemainedBlk 886 = countActiveBlock(OrderedBlks.begin(), OrderedBlks.end()); 887 // consider cloned blocks ?? 888 if (NewnumRemainedBlk == 1 || NewnumRemainedBlk < NumRemainedBlk) { 889 MakeProgress = true; 890 NumRemainedBlk = NewnumRemainedBlk; 891 } else { 892 MakeProgress = false; 893 DEBUG( 894 dbgs() << "No progress\n"; 895 ); 896 } 897 } 898 } while (!Finish && MakeProgress); 899 900 // Misc wrap up to maintain the consistency of the Function representation. 901 wrapup(GraphTraits<MachineFunction *>::nodes_begin(FuncRep)); 902 903 // Detach retired Block, release memory. 904 for (MBBInfoMap::iterator It = BlockInfoMap.begin(), E = BlockInfoMap.end(); 905 It != E; ++It) { 906 if ((*It).second && (*It).second->IsRetired) { 907 assert(((*It).first)->getNumber() != -1); 908 DEBUG( 909 dbgs() << "Erase BB" << ((*It).first)->getNumber() << "\n"; 910 ); 911 (*It).first->eraseFromParent(); //Remove from the parent Function. 912 } 913 delete (*It).second; 914 } 915 BlockInfoMap.clear(); 916 LLInfoMap.clear(); 917 918 DEBUG( 919 FuncRep->viewCFG(); 920 ); 921 922 if (!Finish) 923 llvm_unreachable("IRREDUCIBL_CF"); 924 925 return true; 926 } 927 928 929 930 void AMDGPUCFGStructurizer::orderBlocks(MachineFunction *MF) { 931 int SccNum = 0; 932 MachineBasicBlock *MBB; 933 for (scc_iterator<MachineFunction *> It = scc_begin(MF), E = scc_end(MF); 934 It != E; ++It, ++SccNum) { 935 std::vector<MachineBasicBlock *> &SccNext = *It; 936 for (std::vector<MachineBasicBlock *>::const_iterator 937 blockIter = SccNext.begin(), blockEnd = SccNext.end(); 938 blockIter != blockEnd; ++blockIter) { 939 MBB = *blockIter; 940 OrderedBlks.push_back(MBB); 941 recordSccnum(MBB, SccNum); 942 } 943 } 944 945 //walk through all the block in func to check for unreachable 946 typedef GraphTraits<MachineFunction *> GTM; 947 MachineFunction::iterator It = GTM::nodes_begin(MF), E = GTM::nodes_end(MF); 948 for (; It != E; ++It) { 949 MachineBasicBlock *MBB = &(*It); 950 SccNum = getSCCNum(MBB); 951 if (SccNum == INVALIDSCCNUM) 952 dbgs() << "unreachable block BB" << MBB->getNumber() << "\n"; 953 } 954 } 955 956 int AMDGPUCFGStructurizer::patternMatch(MachineBasicBlock *MBB) { 957 int NumMatch = 0; 958 int CurMatch; 959 960 DEBUG( 961 dbgs() << "Begin patternMatch BB" << MBB->getNumber() << "\n"; 962 ); 963 964 while ((CurMatch = patternMatchGroup(MBB)) > 0) 965 NumMatch += CurMatch; 966 967 DEBUG( 968 dbgs() << "End patternMatch BB" << MBB->getNumber() 969 << ", numMatch = " << NumMatch << "\n"; 970 ); 971 972 return NumMatch; 973 } 974 975 int AMDGPUCFGStructurizer::patternMatchGroup(MachineBasicBlock *MBB) { 976 int NumMatch = 0; 977 NumMatch += loopendPatternMatch(); 978 NumMatch += serialPatternMatch(MBB); 979 NumMatch += ifPatternMatch(MBB); 980 return NumMatch; 981 } 982 983 984 int AMDGPUCFGStructurizer::serialPatternMatch(MachineBasicBlock *MBB) { 985 if (MBB->succ_size() != 1) 986 return 0; 987 988 MachineBasicBlock *childBlk = *MBB->succ_begin(); 989 if (childBlk->pred_size() != 1 || isActiveLoophead(childBlk)) 990 return 0; 991 992 mergeSerialBlock(MBB, childBlk); 993 ++numSerialPatternMatch; 994 return 1; 995 } 996 997 int AMDGPUCFGStructurizer::ifPatternMatch(MachineBasicBlock *MBB) { 998 //two edges 999 if (MBB->succ_size() != 2) 1000 return 0; 1001 if (hasBackEdge(MBB)) 1002 return 0; 1003 MachineInstr *BranchMI = getNormalBlockBranchInstr(MBB); 1004 if (!BranchMI) 1005 return 0; 1006 1007 assert(isCondBranch(BranchMI)); 1008 1009 MachineBasicBlock *TrueMBB = getTrueBranch(BranchMI); 1010 serialPatternMatch(TrueMBB); 1011 ifPatternMatch(TrueMBB); 1012 MachineBasicBlock *FalseMBB = getFalseBranch(MBB, BranchMI); 1013 serialPatternMatch(FalseMBB); 1014 ifPatternMatch(FalseMBB); 1015 MachineBasicBlock *LandBlk; 1016 int Cloned = 0; 1017 1018 assert (!TrueMBB->succ_empty() || !FalseMBB->succ_empty()); 1019 // TODO: Simplify 1020 if (TrueMBB->succ_size() == 1 && FalseMBB->succ_size() == 1 1021 && *TrueMBB->succ_begin() == *FalseMBB->succ_begin()) { 1022 // Diamond pattern 1023 LandBlk = *TrueMBB->succ_begin(); 1024 } else if (TrueMBB->succ_size() == 1 && *TrueMBB->succ_begin() == FalseMBB) { 1025 // Triangle pattern, false is empty 1026 LandBlk = FalseMBB; 1027 FalseMBB = NULL; 1028 } else if (FalseMBB->succ_size() == 1 1029 && *FalseMBB->succ_begin() == TrueMBB) { 1030 // Triangle pattern, true is empty 1031 // We reverse the predicate to make a triangle, empty false pattern; 1032 std::swap(TrueMBB, FalseMBB); 1033 reversePredicateSetter(MBB->end()); 1034 LandBlk = FalseMBB; 1035 FalseMBB = NULL; 1036 } else if (FalseMBB->succ_size() == 1 1037 && isSameloopDetachedContbreak(TrueMBB, FalseMBB)) { 1038 LandBlk = *FalseMBB->succ_begin(); 1039 } else if (TrueMBB->succ_size() == 1 1040 && isSameloopDetachedContbreak(FalseMBB, TrueMBB)) { 1041 LandBlk = *TrueMBB->succ_begin(); 1042 } else { 1043 return handleJumpintoIf(MBB, TrueMBB, FalseMBB); 1044 } 1045 1046 // improveSimpleJumpinfoIf can handle the case where landBlk == NULL but the 1047 // new BB created for landBlk==NULL may introduce new challenge to the 1048 // reduction process. 1049 if (LandBlk && 1050 ((TrueMBB && TrueMBB->pred_size() > 1) 1051 || (FalseMBB && FalseMBB->pred_size() > 1))) { 1052 Cloned += improveSimpleJumpintoIf(MBB, TrueMBB, FalseMBB, &LandBlk); 1053 } 1054 1055 if (TrueMBB && TrueMBB->pred_size() > 1) { 1056 TrueMBB = cloneBlockForPredecessor(TrueMBB, MBB); 1057 ++Cloned; 1058 } 1059 1060 if (FalseMBB && FalseMBB->pred_size() > 1) { 1061 FalseMBB = cloneBlockForPredecessor(FalseMBB, MBB); 1062 ++Cloned; 1063 } 1064 1065 mergeIfthenelseBlock(BranchMI, MBB, TrueMBB, FalseMBB, LandBlk); 1066 1067 ++numIfPatternMatch; 1068 1069 numClonedBlock += Cloned; 1070 1071 return 1 + Cloned; 1072 } 1073 1074 int AMDGPUCFGStructurizer::loopendPatternMatch() { 1075 std::vector<MachineLoop *> NestedLoops; 1076 for (MachineLoopInfo::iterator It = MLI->begin(), E = MLI->end(); 1077 It != E; ++It) { 1078 df_iterator<MachineLoop *> LpIt = df_begin(*It), 1079 LpE = df_end(*It); 1080 for (; LpIt != LpE; ++LpIt) 1081 NestedLoops.push_back(*LpIt); 1082 } 1083 if (NestedLoops.size() == 0) 1084 return 0; 1085 1086 // Process nested loop outside->inside, so "continue" to a outside loop won't 1087 // be mistaken as "break" of the current loop. 1088 int Num = 0; 1089 for (std::vector<MachineLoop *>::reverse_iterator It = NestedLoops.rbegin(), 1090 E = NestedLoops.rend(); It != E; ++It) { 1091 MachineLoop *ExaminedLoop = *It; 1092 if (ExaminedLoop->getNumBlocks() == 0 || Visited[ExaminedLoop]) 1093 continue; 1094 DEBUG(dbgs() << "Processing:\n"; ExaminedLoop->dump();); 1095 int NumBreak = mergeLoop(ExaminedLoop); 1096 if (NumBreak == -1) 1097 break; 1098 Num += NumBreak; 1099 } 1100 return Num; 1101 } 1102 1103 int AMDGPUCFGStructurizer::mergeLoop(MachineLoop *LoopRep) { 1104 MachineBasicBlock *LoopHeader = LoopRep->getHeader(); 1105 MBBVector ExitingMBBs; 1106 LoopRep->getExitingBlocks(ExitingMBBs); 1107 assert(!ExitingMBBs.empty() && "Infinite Loop not supported"); 1108 DEBUG(dbgs() << "Loop has " << ExitingMBBs.size() << " exiting blocks\n";); 1109 // We assume a single ExitBlk 1110 MBBVector ExitBlks; 1111 LoopRep->getExitBlocks(ExitBlks); 1112 SmallPtrSet<MachineBasicBlock *, 2> ExitBlkSet; 1113 for (unsigned i = 0, e = ExitBlks.size(); i < e; ++i) 1114 ExitBlkSet.insert(ExitBlks[i]); 1115 assert(ExitBlkSet.size() == 1); 1116 MachineBasicBlock *ExitBlk = *ExitBlks.begin(); 1117 assert(ExitBlk && "Loop has several exit block"); 1118 MBBVector LatchBlks; 1119 typedef GraphTraits<Inverse<MachineBasicBlock*> > InvMBBTraits; 1120 InvMBBTraits::ChildIteratorType PI = InvMBBTraits::child_begin(LoopHeader), 1121 PE = InvMBBTraits::child_end(LoopHeader); 1122 for (; PI != PE; PI++) { 1123 if (LoopRep->contains(*PI)) 1124 LatchBlks.push_back(*PI); 1125 } 1126 1127 for (unsigned i = 0, e = ExitingMBBs.size(); i < e; ++i) 1128 mergeLoopbreakBlock(ExitingMBBs[i], ExitBlk); 1129 for (unsigned i = 0, e = LatchBlks.size(); i < e; ++i) 1130 settleLoopcontBlock(LatchBlks[i], LoopHeader); 1131 int Match = 0; 1132 do { 1133 Match = 0; 1134 Match += serialPatternMatch(LoopHeader); 1135 Match += ifPatternMatch(LoopHeader); 1136 } while (Match > 0); 1137 mergeLooplandBlock(LoopHeader, ExitBlk); 1138 MachineLoop *ParentLoop = LoopRep->getParentLoop(); 1139 if (ParentLoop) 1140 MLI->changeLoopFor(LoopHeader, ParentLoop); 1141 else 1142 MLI->removeBlock(LoopHeader); 1143 Visited[LoopRep] = true; 1144 return 1; 1145 } 1146 1147 int AMDGPUCFGStructurizer::loopcontPatternMatch(MachineLoop *LoopRep, 1148 MachineBasicBlock *LoopHeader) { 1149 int NumCont = 0; 1150 SmallVector<MachineBasicBlock *, DEFAULT_VEC_SLOTS> ContMBB; 1151 typedef GraphTraits<Inverse<MachineBasicBlock *> > GTIM; 1152 GTIM::ChildIteratorType It = GTIM::child_begin(LoopHeader), 1153 E = GTIM::child_end(LoopHeader); 1154 for (; It != E; ++It) { 1155 MachineBasicBlock *MBB = *It; 1156 if (LoopRep->contains(MBB)) { 1157 handleLoopcontBlock(MBB, MLI->getLoopFor(MBB), 1158 LoopHeader, LoopRep); 1159 ContMBB.push_back(MBB); 1160 ++NumCont; 1161 } 1162 } 1163 1164 for (SmallVectorImpl<MachineBasicBlock *>::iterator It = ContMBB.begin(), 1165 E = ContMBB.end(); It != E; ++It) { 1166 (*It)->removeSuccessor(LoopHeader); 1167 } 1168 1169 numLoopcontPatternMatch += NumCont; 1170 1171 return NumCont; 1172 } 1173 1174 1175 bool AMDGPUCFGStructurizer::isSameloopDetachedContbreak( 1176 MachineBasicBlock *Src1MBB, MachineBasicBlock *Src2MBB) { 1177 if (Src1MBB->succ_size() == 0) { 1178 MachineLoop *LoopRep = MLI->getLoopFor(Src1MBB); 1179 if (LoopRep&& LoopRep == MLI->getLoopFor(Src2MBB)) { 1180 MachineBasicBlock *&TheEntry = LLInfoMap[LoopRep]; 1181 if (TheEntry) { 1182 DEBUG( 1183 dbgs() << "isLoopContBreakBlock yes src1 = BB" 1184 << Src1MBB->getNumber() 1185 << " src2 = BB" << Src2MBB->getNumber() << "\n"; 1186 ); 1187 return true; 1188 } 1189 } 1190 } 1191 return false; 1192 } 1193 1194 int AMDGPUCFGStructurizer::handleJumpintoIf(MachineBasicBlock *HeadMBB, 1195 MachineBasicBlock *TrueMBB, MachineBasicBlock *FalseMBB) { 1196 int Num = handleJumpintoIfImp(HeadMBB, TrueMBB, FalseMBB); 1197 if (Num == 0) { 1198 DEBUG( 1199 dbgs() << "handleJumpintoIf swap trueBlk and FalseBlk" << "\n"; 1200 ); 1201 Num = handleJumpintoIfImp(HeadMBB, FalseMBB, TrueMBB); 1202 } 1203 return Num; 1204 } 1205 1206 int AMDGPUCFGStructurizer::handleJumpintoIfImp(MachineBasicBlock *HeadMBB, 1207 MachineBasicBlock *TrueMBB, MachineBasicBlock *FalseMBB) { 1208 int Num = 0; 1209 MachineBasicBlock *DownBlk; 1210 1211 //trueBlk could be the common post dominator 1212 DownBlk = TrueMBB; 1213 1214 DEBUG( 1215 dbgs() << "handleJumpintoIfImp head = BB" << HeadMBB->getNumber() 1216 << " true = BB" << TrueMBB->getNumber() 1217 << ", numSucc=" << TrueMBB->succ_size() 1218 << " false = BB" << FalseMBB->getNumber() << "\n"; 1219 ); 1220 1221 while (DownBlk) { 1222 DEBUG( 1223 dbgs() << "check down = BB" << DownBlk->getNumber(); 1224 ); 1225 1226 if (singlePathTo(FalseMBB, DownBlk) == SinglePath_InPath) { 1227 DEBUG( 1228 dbgs() << " working\n"; 1229 ); 1230 1231 Num += cloneOnSideEntryTo(HeadMBB, TrueMBB, DownBlk); 1232 Num += cloneOnSideEntryTo(HeadMBB, FalseMBB, DownBlk); 1233 1234 numClonedBlock += Num; 1235 Num += serialPatternMatch(*HeadMBB->succ_begin()); 1236 Num += serialPatternMatch(*(++HeadMBB->succ_begin())); 1237 Num += ifPatternMatch(HeadMBB); 1238 assert(Num > 0); 1239 1240 break; 1241 } 1242 DEBUG( 1243 dbgs() << " not working\n"; 1244 ); 1245 DownBlk = (DownBlk->succ_size() == 1) ? (*DownBlk->succ_begin()) : NULL; 1246 } // walk down the postDomTree 1247 1248 return Num; 1249 } 1250 1251 void AMDGPUCFGStructurizer::showImproveSimpleJumpintoIf( 1252 MachineBasicBlock *HeadMBB, MachineBasicBlock *TrueMBB, 1253 MachineBasicBlock *FalseMBB, MachineBasicBlock *LandMBB, bool Detail) { 1254 dbgs() << "head = BB" << HeadMBB->getNumber() 1255 << " size = " << HeadMBB->size(); 1256 if (Detail) { 1257 dbgs() << "\n"; 1258 HeadMBB->print(dbgs()); 1259 dbgs() << "\n"; 1260 } 1261 1262 if (TrueMBB) { 1263 dbgs() << ", true = BB" << TrueMBB->getNumber() << " size = " 1264 << TrueMBB->size() << " numPred = " << TrueMBB->pred_size(); 1265 if (Detail) { 1266 dbgs() << "\n"; 1267 TrueMBB->print(dbgs()); 1268 dbgs() << "\n"; 1269 } 1270 } 1271 if (FalseMBB) { 1272 dbgs() << ", false = BB" << FalseMBB->getNumber() << " size = " 1273 << FalseMBB->size() << " numPred = " << FalseMBB->pred_size(); 1274 if (Detail) { 1275 dbgs() << "\n"; 1276 FalseMBB->print(dbgs()); 1277 dbgs() << "\n"; 1278 } 1279 } 1280 if (LandMBB) { 1281 dbgs() << ", land = BB" << LandMBB->getNumber() << " size = " 1282 << LandMBB->size() << " numPred = " << LandMBB->pred_size(); 1283 if (Detail) { 1284 dbgs() << "\n"; 1285 LandMBB->print(dbgs()); 1286 dbgs() << "\n"; 1287 } 1288 } 1289 1290 dbgs() << "\n"; 1291 } 1292 1293 int AMDGPUCFGStructurizer::improveSimpleJumpintoIf(MachineBasicBlock *HeadMBB, 1294 MachineBasicBlock *TrueMBB, MachineBasicBlock *FalseMBB, 1295 MachineBasicBlock **LandMBBPtr) { 1296 bool MigrateTrue = false; 1297 bool MigrateFalse = false; 1298 1299 MachineBasicBlock *LandBlk = *LandMBBPtr; 1300 1301 assert((!TrueMBB || TrueMBB->succ_size() <= 1) 1302 && (!FalseMBB || FalseMBB->succ_size() <= 1)); 1303 1304 if (TrueMBB == FalseMBB) 1305 return 0; 1306 1307 MigrateTrue = needMigrateBlock(TrueMBB); 1308 MigrateFalse = needMigrateBlock(FalseMBB); 1309 1310 if (!MigrateTrue && !MigrateFalse) 1311 return 0; 1312 1313 // If we need to migrate either trueBlk and falseBlk, migrate the rest that 1314 // have more than one predecessors. without doing this, its predecessor 1315 // rather than headBlk will have undefined value in initReg. 1316 if (!MigrateTrue && TrueMBB && TrueMBB->pred_size() > 1) 1317 MigrateTrue = true; 1318 if (!MigrateFalse && FalseMBB && FalseMBB->pred_size() > 1) 1319 MigrateFalse = true; 1320 1321 DEBUG( 1322 dbgs() << "before improveSimpleJumpintoIf: "; 1323 showImproveSimpleJumpintoIf(HeadMBB, TrueMBB, FalseMBB, LandBlk, 0); 1324 ); 1325 1326 // org: headBlk => if () {trueBlk} else {falseBlk} => landBlk 1327 // 1328 // new: headBlk => if () {initReg = 1; org trueBlk branch} else 1329 // {initReg = 0; org falseBlk branch } 1330 // => landBlk => if (initReg) {org trueBlk} else {org falseBlk} 1331 // => org landBlk 1332 // if landBlk->pred_size() > 2, put the about if-else inside 1333 // if (initReg !=2) {...} 1334 // 1335 // add initReg = initVal to headBlk 1336 1337 const TargetRegisterClass * I32RC = TRI->getCFGStructurizerRegClass(MVT::i32); 1338 unsigned InitReg = 1339 HeadMBB->getParent()->getRegInfo().createVirtualRegister(I32RC); 1340 if (!MigrateTrue || !MigrateFalse) 1341 llvm_unreachable("Extra register needed to handle CFG"); 1342 1343 int NumNewBlk = 0; 1344 1345 if (!LandBlk) { 1346 LandBlk = HeadMBB->getParent()->CreateMachineBasicBlock(); 1347 HeadMBB->getParent()->push_back(LandBlk); //insert to function 1348 1349 if (TrueMBB) { 1350 TrueMBB->addSuccessor(LandBlk); 1351 } else { 1352 HeadMBB->addSuccessor(LandBlk); 1353 } 1354 1355 if (FalseMBB) { 1356 FalseMBB->addSuccessor(LandBlk); 1357 } else { 1358 HeadMBB->addSuccessor(LandBlk); 1359 } 1360 1361 NumNewBlk ++; 1362 } 1363 1364 bool LandBlkHasOtherPred = (LandBlk->pred_size() > 2); 1365 1366 //insert AMDGPU::ENDIF to avoid special case "input landBlk == NULL" 1367 MachineBasicBlock::iterator I = insertInstrBefore(LandBlk, AMDGPU::ENDIF); 1368 1369 if (LandBlkHasOtherPred) { 1370 llvm_unreachable("Extra register needed to handle CFG"); 1371 unsigned CmpResReg = 1372 HeadMBB->getParent()->getRegInfo().createVirtualRegister(I32RC); 1373 llvm_unreachable("Extra compare instruction needed to handle CFG"); 1374 insertCondBranchBefore(LandBlk, I, AMDGPU::IF_PREDICATE_SET, 1375 CmpResReg, DebugLoc()); 1376 } 1377 1378 insertCondBranchBefore(LandBlk, I, AMDGPU::IF_PREDICATE_SET, InitReg, 1379 DebugLoc()); 1380 1381 if (MigrateTrue) { 1382 migrateInstruction(TrueMBB, LandBlk, I); 1383 // need to uncondionally insert the assignment to ensure a path from its 1384 // predecessor rather than headBlk has valid value in initReg if 1385 // (initVal != 1). 1386 llvm_unreachable("Extra register needed to handle CFG"); 1387 } 1388 insertInstrBefore(I, AMDGPU::ELSE); 1389 1390 if (MigrateFalse) { 1391 migrateInstruction(FalseMBB, LandBlk, I); 1392 // need to uncondionally insert the assignment to ensure a path from its 1393 // predecessor rather than headBlk has valid value in initReg if 1394 // (initVal != 0) 1395 llvm_unreachable("Extra register needed to handle CFG"); 1396 } 1397 1398 if (LandBlkHasOtherPred) { 1399 // add endif 1400 insertInstrBefore(I, AMDGPU::ENDIF); 1401 1402 // put initReg = 2 to other predecessors of landBlk 1403 for (MachineBasicBlock::pred_iterator PI = LandBlk->pred_begin(), 1404 PE = LandBlk->pred_end(); PI != PE; ++PI) { 1405 MachineBasicBlock *MBB = *PI; 1406 if (MBB != TrueMBB && MBB != FalseMBB) 1407 llvm_unreachable("Extra register needed to handle CFG"); 1408 } 1409 } 1410 DEBUG( 1411 dbgs() << "result from improveSimpleJumpintoIf: "; 1412 showImproveSimpleJumpintoIf(HeadMBB, TrueMBB, FalseMBB, LandBlk, 0); 1413 ); 1414 1415 // update landBlk 1416 *LandMBBPtr = LandBlk; 1417 1418 return NumNewBlk; 1419 } 1420 1421 void AMDGPUCFGStructurizer::handleLoopcontBlock(MachineBasicBlock *ContingMBB, 1422 MachineLoop *ContingLoop, MachineBasicBlock *ContMBB, 1423 MachineLoop *ContLoop) { 1424 DEBUG(dbgs() << "loopcontPattern cont = BB" << ContingMBB->getNumber() 1425 << " header = BB" << ContMBB->getNumber() << "\n"; 1426 dbgs() << "Trying to continue loop-depth = " 1427 << getLoopDepth(ContLoop) 1428 << " from loop-depth = " << getLoopDepth(ContingLoop) << "\n";); 1429 settleLoopcontBlock(ContingMBB, ContMBB); 1430 } 1431 1432 void AMDGPUCFGStructurizer::mergeSerialBlock(MachineBasicBlock *DstMBB, 1433 MachineBasicBlock *SrcMBB) { 1434 DEBUG( 1435 dbgs() << "serialPattern BB" << DstMBB->getNumber() 1436 << " <= BB" << SrcMBB->getNumber() << "\n"; 1437 ); 1438 DstMBB->splice(DstMBB->end(), SrcMBB, SrcMBB->begin(), SrcMBB->end()); 1439 1440 DstMBB->removeSuccessor(SrcMBB); 1441 cloneSuccessorList(DstMBB, SrcMBB); 1442 1443 removeSuccessor(SrcMBB); 1444 MLI->removeBlock(SrcMBB); 1445 retireBlock(SrcMBB); 1446 } 1447 1448 void AMDGPUCFGStructurizer::mergeIfthenelseBlock(MachineInstr *BranchMI, 1449 MachineBasicBlock *MBB, MachineBasicBlock *TrueMBB, 1450 MachineBasicBlock *FalseMBB, MachineBasicBlock *LandMBB) { 1451 assert (TrueMBB); 1452 DEBUG( 1453 dbgs() << "ifPattern BB" << MBB->getNumber(); 1454 dbgs() << "{ "; 1455 if (TrueMBB) { 1456 dbgs() << "BB" << TrueMBB->getNumber(); 1457 } 1458 dbgs() << " } else "; 1459 dbgs() << "{ "; 1460 if (FalseMBB) { 1461 dbgs() << "BB" << FalseMBB->getNumber(); 1462 } 1463 dbgs() << " }\n "; 1464 dbgs() << "landBlock: "; 1465 if (!LandMBB) { 1466 dbgs() << "NULL"; 1467 } else { 1468 dbgs() << "BB" << LandMBB->getNumber(); 1469 } 1470 dbgs() << "\n"; 1471 ); 1472 1473 int OldOpcode = BranchMI->getOpcode(); 1474 DebugLoc BranchDL = BranchMI->getDebugLoc(); 1475 1476 // transform to 1477 // if cond 1478 // trueBlk 1479 // else 1480 // falseBlk 1481 // endif 1482 // landBlk 1483 1484 MachineBasicBlock::iterator I = BranchMI; 1485 insertCondBranchBefore(I, getBranchNzeroOpcode(OldOpcode), 1486 BranchDL); 1487 1488 if (TrueMBB) { 1489 MBB->splice(I, TrueMBB, TrueMBB->begin(), TrueMBB->end()); 1490 MBB->removeSuccessor(TrueMBB); 1491 if (LandMBB && TrueMBB->succ_size()!=0) 1492 TrueMBB->removeSuccessor(LandMBB); 1493 retireBlock(TrueMBB); 1494 MLI->removeBlock(TrueMBB); 1495 } 1496 1497 if (FalseMBB) { 1498 insertInstrBefore(I, AMDGPU::ELSE); 1499 MBB->splice(I, FalseMBB, FalseMBB->begin(), 1500 FalseMBB->end()); 1501 MBB->removeSuccessor(FalseMBB); 1502 if (LandMBB && FalseMBB->succ_size() != 0) 1503 FalseMBB->removeSuccessor(LandMBB); 1504 retireBlock(FalseMBB); 1505 MLI->removeBlock(FalseMBB); 1506 } 1507 insertInstrBefore(I, AMDGPU::ENDIF); 1508 1509 BranchMI->eraseFromParent(); 1510 1511 if (LandMBB && TrueMBB && FalseMBB) 1512 MBB->addSuccessor(LandMBB); 1513 1514 } 1515 1516 void AMDGPUCFGStructurizer::mergeLooplandBlock(MachineBasicBlock *DstBlk, 1517 MachineBasicBlock *LandMBB) { 1518 DEBUG(dbgs() << "loopPattern header = BB" << DstBlk->getNumber() 1519 << " land = BB" << LandMBB->getNumber() << "\n";); 1520 1521 insertInstrBefore(DstBlk, AMDGPU::WHILELOOP, DebugLoc()); 1522 insertInstrEnd(DstBlk, AMDGPU::ENDLOOP, DebugLoc()); 1523 DstBlk->addSuccessor(LandMBB); 1524 DstBlk->removeSuccessor(DstBlk); 1525 } 1526 1527 1528 void AMDGPUCFGStructurizer::mergeLoopbreakBlock(MachineBasicBlock *ExitingMBB, 1529 MachineBasicBlock *LandMBB) { 1530 DEBUG(dbgs() << "loopbreakPattern exiting = BB" << ExitingMBB->getNumber() 1531 << " land = BB" << LandMBB->getNumber() << "\n";); 1532 MachineInstr *BranchMI = getLoopendBlockBranchInstr(ExitingMBB); 1533 assert(BranchMI && isCondBranch(BranchMI)); 1534 DebugLoc DL = BranchMI->getDebugLoc(); 1535 MachineBasicBlock *TrueBranch = getTrueBranch(BranchMI); 1536 MachineBasicBlock::iterator I = BranchMI; 1537 if (TrueBranch != LandMBB) 1538 reversePredicateSetter(I); 1539 insertCondBranchBefore(ExitingMBB, I, AMDGPU::IF_PREDICATE_SET, AMDGPU::PREDICATE_BIT, DL); 1540 insertInstrBefore(I, AMDGPU::BREAK); 1541 insertInstrBefore(I, AMDGPU::ENDIF); 1542 //now branchInst can be erase safely 1543 BranchMI->eraseFromParent(); 1544 //now take care of successors, retire blocks 1545 ExitingMBB->removeSuccessor(LandMBB); 1546 } 1547 1548 void AMDGPUCFGStructurizer::settleLoopcontBlock(MachineBasicBlock *ContingMBB, 1549 MachineBasicBlock *ContMBB) { 1550 DEBUG(dbgs() << "settleLoopcontBlock conting = BB" 1551 << ContingMBB->getNumber() 1552 << ", cont = BB" << ContMBB->getNumber() << "\n";); 1553 1554 MachineInstr *MI = getLoopendBlockBranchInstr(ContingMBB); 1555 if (MI) { 1556 assert(isCondBranch(MI)); 1557 MachineBasicBlock::iterator I = MI; 1558 MachineBasicBlock *TrueBranch = getTrueBranch(MI); 1559 int OldOpcode = MI->getOpcode(); 1560 DebugLoc DL = MI->getDebugLoc(); 1561 1562 bool UseContinueLogical = ((&*ContingMBB->rbegin()) == MI); 1563 1564 if (UseContinueLogical == false) { 1565 int BranchOpcode = 1566 TrueBranch == ContMBB ? getBranchNzeroOpcode(OldOpcode) : 1567 getBranchZeroOpcode(OldOpcode); 1568 insertCondBranchBefore(I, BranchOpcode, DL); 1569 // insertEnd to ensure phi-moves, if exist, go before the continue-instr. 1570 insertInstrEnd(ContingMBB, AMDGPU::CONTINUE, DL); 1571 insertInstrEnd(ContingMBB, AMDGPU::ENDIF, DL); 1572 } else { 1573 int BranchOpcode = 1574 TrueBranch == ContMBB ? getContinueNzeroOpcode(OldOpcode) : 1575 getContinueZeroOpcode(OldOpcode); 1576 insertCondBranchBefore(I, BranchOpcode, DL); 1577 } 1578 1579 MI->eraseFromParent(); 1580 } else { 1581 // if we've arrived here then we've already erased the branch instruction 1582 // travel back up the basic block to see the last reference of our debug 1583 // location we've just inserted that reference here so it should be 1584 // representative insertEnd to ensure phi-moves, if exist, go before the 1585 // continue-instr. 1586 insertInstrEnd(ContingMBB, AMDGPU::CONTINUE, 1587 getLastDebugLocInBB(ContingMBB)); 1588 } 1589 } 1590 1591 int AMDGPUCFGStructurizer::cloneOnSideEntryTo(MachineBasicBlock *PreMBB, 1592 MachineBasicBlock *SrcMBB, MachineBasicBlock *DstMBB) { 1593 int Cloned = 0; 1594 assert(PreMBB->isSuccessor(SrcMBB)); 1595 while (SrcMBB && SrcMBB != DstMBB) { 1596 assert(SrcMBB->succ_size() == 1); 1597 if (SrcMBB->pred_size() > 1) { 1598 SrcMBB = cloneBlockForPredecessor(SrcMBB, PreMBB); 1599 ++Cloned; 1600 } 1601 1602 PreMBB = SrcMBB; 1603 SrcMBB = *SrcMBB->succ_begin(); 1604 } 1605 1606 return Cloned; 1607 } 1608 1609 MachineBasicBlock * 1610 AMDGPUCFGStructurizer::cloneBlockForPredecessor(MachineBasicBlock *MBB, 1611 MachineBasicBlock *PredMBB) { 1612 assert(PredMBB->isSuccessor(MBB) && 1613 "succBlk is not a prececessor of curBlk"); 1614 1615 MachineBasicBlock *CloneMBB = clone(MBB); //clone instructions 1616 replaceInstrUseOfBlockWith(PredMBB, MBB, CloneMBB); 1617 //srcBlk, oldBlk, newBlk 1618 1619 PredMBB->removeSuccessor(MBB); 1620 PredMBB->addSuccessor(CloneMBB); 1621 1622 // add all successor to cloneBlk 1623 cloneSuccessorList(CloneMBB, MBB); 1624 1625 numClonedInstr += MBB->size(); 1626 1627 DEBUG( 1628 dbgs() << "Cloned block: " << "BB" 1629 << MBB->getNumber() << "size " << MBB->size() << "\n"; 1630 ); 1631 1632 SHOWNEWBLK(CloneMBB, "result of Cloned block: "); 1633 1634 return CloneMBB; 1635 } 1636 1637 void AMDGPUCFGStructurizer::migrateInstruction(MachineBasicBlock *SrcMBB, 1638 MachineBasicBlock *DstMBB, MachineBasicBlock::iterator I) { 1639 MachineBasicBlock::iterator SpliceEnd; 1640 //look for the input branchinstr, not the AMDGPU branchinstr 1641 MachineInstr *BranchMI = getNormalBlockBranchInstr(SrcMBB); 1642 if (!BranchMI) { 1643 DEBUG( 1644 dbgs() << "migrateInstruction don't see branch instr\n" ; 1645 ); 1646 SpliceEnd = SrcMBB->end(); 1647 } else { 1648 DEBUG( 1649 dbgs() << "migrateInstruction see branch instr\n" ; 1650 BranchMI->dump(); 1651 ); 1652 SpliceEnd = BranchMI; 1653 } 1654 DEBUG( 1655 dbgs() << "migrateInstruction before splice dstSize = " << DstMBB->size() 1656 << "srcSize = " << SrcMBB->size() << "\n"; 1657 ); 1658 1659 //splice insert before insertPos 1660 DstMBB->splice(I, SrcMBB, SrcMBB->begin(), SpliceEnd); 1661 1662 DEBUG( 1663 dbgs() << "migrateInstruction after splice dstSize = " << DstMBB->size() 1664 << "srcSize = " << SrcMBB->size() << "\n"; 1665 ); 1666 } 1667 1668 MachineBasicBlock * 1669 AMDGPUCFGStructurizer::normalizeInfiniteLoopExit(MachineLoop* LoopRep) { 1670 MachineBasicBlock *LoopHeader = LoopRep->getHeader(); 1671 MachineBasicBlock *LoopLatch = LoopRep->getLoopLatch(); 1672 const TargetRegisterClass * I32RC = TRI->getCFGStructurizerRegClass(MVT::i32); 1673 1674 if (!LoopHeader || !LoopLatch) 1675 return NULL; 1676 MachineInstr *BranchMI = getLoopendBlockBranchInstr(LoopLatch); 1677 // Is LoopRep an infinite loop ? 1678 if (!BranchMI || !isUncondBranch(BranchMI)) 1679 return NULL; 1680 1681 MachineBasicBlock *DummyExitBlk = FuncRep->CreateMachineBasicBlock(); 1682 FuncRep->push_back(DummyExitBlk); //insert to function 1683 SHOWNEWBLK(DummyExitBlk, "DummyExitBlock to normalize infiniteLoop: "); 1684 DEBUG(dbgs() << "Old branch instr: " << *BranchMI << "\n";); 1685 MachineBasicBlock::iterator I = BranchMI; 1686 unsigned ImmReg = FuncRep->getRegInfo().createVirtualRegister(I32RC); 1687 llvm_unreachable("Extra register needed to handle CFG"); 1688 MachineInstr *NewMI = insertInstrBefore(I, AMDGPU::BRANCH_COND_i32); 1689 MachineInstrBuilder MIB(*FuncRep, NewMI); 1690 MIB.addMBB(LoopHeader); 1691 MIB.addReg(ImmReg, false); 1692 SHOWNEWINSTR(NewMI); 1693 BranchMI->eraseFromParent(); 1694 LoopLatch->addSuccessor(DummyExitBlk); 1695 1696 return DummyExitBlk; 1697 } 1698 1699 void AMDGPUCFGStructurizer::removeUnconditionalBranch(MachineBasicBlock *MBB) { 1700 MachineInstr *BranchMI; 1701 1702 // I saw two unconditional branch in one basic block in example 1703 // test_fc_do_while_or.c need to fix the upstream on this to remove the loop. 1704 while ((BranchMI = getLoopendBlockBranchInstr(MBB)) 1705 && isUncondBranch(BranchMI)) { 1706 DEBUG(dbgs() << "Removing uncond branch instr"; BranchMI->dump();); 1707 BranchMI->eraseFromParent(); 1708 } 1709 } 1710 1711 void AMDGPUCFGStructurizer::removeRedundantConditionalBranch( 1712 MachineBasicBlock *MBB) { 1713 if (MBB->succ_size() != 2) 1714 return; 1715 MachineBasicBlock *MBB1 = *MBB->succ_begin(); 1716 MachineBasicBlock *MBB2 = *(++MBB->succ_begin()); 1717 if (MBB1 != MBB2) 1718 return; 1719 1720 MachineInstr *BranchMI = getNormalBlockBranchInstr(MBB); 1721 assert(BranchMI && isCondBranch(BranchMI)); 1722 DEBUG(dbgs() << "Removing unneeded cond branch instr"; BranchMI->dump();); 1723 BranchMI->eraseFromParent(); 1724 SHOWNEWBLK(MBB1, "Removing redundant successor"); 1725 MBB->removeSuccessor(MBB1); 1726 } 1727 1728 void AMDGPUCFGStructurizer::addDummyExitBlock( 1729 SmallVectorImpl<MachineBasicBlock*> &RetMBB) { 1730 MachineBasicBlock *DummyExitBlk = FuncRep->CreateMachineBasicBlock(); 1731 FuncRep->push_back(DummyExitBlk); //insert to function 1732 insertInstrEnd(DummyExitBlk, AMDGPU::RETURN); 1733 1734 for (SmallVectorImpl<MachineBasicBlock *>::iterator It = RetMBB.begin(), 1735 E = RetMBB.end(); It != E; ++It) { 1736 MachineBasicBlock *MBB = *It; 1737 MachineInstr *MI = getReturnInstr(MBB); 1738 if (MI) 1739 MI->eraseFromParent(); 1740 MBB->addSuccessor(DummyExitBlk); 1741 DEBUG( 1742 dbgs() << "Add dummyExitBlock to BB" << MBB->getNumber() 1743 << " successors\n"; 1744 ); 1745 } 1746 SHOWNEWBLK(DummyExitBlk, "DummyExitBlock: "); 1747 } 1748 1749 void AMDGPUCFGStructurizer::removeSuccessor(MachineBasicBlock *MBB) { 1750 while (MBB->succ_size()) 1751 MBB->removeSuccessor(*MBB->succ_begin()); 1752 } 1753 1754 void AMDGPUCFGStructurizer::recordSccnum(MachineBasicBlock *MBB, 1755 int SccNum) { 1756 BlockInformation *&srcBlkInfo = BlockInfoMap[MBB]; 1757 if (!srcBlkInfo) 1758 srcBlkInfo = new BlockInformation(); 1759 srcBlkInfo->SccNum = SccNum; 1760 } 1761 1762 void AMDGPUCFGStructurizer::retireBlock(MachineBasicBlock *MBB) { 1763 DEBUG( 1764 dbgs() << "Retiring BB" << MBB->getNumber() << "\n"; 1765 ); 1766 1767 BlockInformation *&SrcBlkInfo = BlockInfoMap[MBB]; 1768 1769 if (!SrcBlkInfo) 1770 SrcBlkInfo = new BlockInformation(); 1771 1772 SrcBlkInfo->IsRetired = true; 1773 assert(MBB->succ_size() == 0 && MBB->pred_size() == 0 1774 && "can't retire block yet"); 1775 } 1776 1777 void AMDGPUCFGStructurizer::setLoopLandBlock(MachineLoop *loopRep, 1778 MachineBasicBlock *MBB) { 1779 MachineBasicBlock *&TheEntry = LLInfoMap[loopRep]; 1780 if (!MBB) { 1781 MBB = FuncRep->CreateMachineBasicBlock(); 1782 FuncRep->push_back(MBB); //insert to function 1783 SHOWNEWBLK(MBB, "DummyLandingBlock for loop without break: "); 1784 } 1785 TheEntry = MBB; 1786 DEBUG( 1787 dbgs() << "setLoopLandBlock loop-header = BB" 1788 << loopRep->getHeader()->getNumber() 1789 << " landing-block = BB" << MBB->getNumber() << "\n"; 1790 ); 1791 } 1792 1793 MachineBasicBlock * 1794 AMDGPUCFGStructurizer::findNearestCommonPostDom(MachineBasicBlock *MBB1, 1795 MachineBasicBlock *MBB2) { 1796 1797 if (PDT->dominates(MBB1, MBB2)) 1798 return MBB1; 1799 if (PDT->dominates(MBB2, MBB1)) 1800 return MBB2; 1801 1802 MachineDomTreeNode *Node1 = PDT->getNode(MBB1); 1803 MachineDomTreeNode *Node2 = PDT->getNode(MBB2); 1804 1805 // Handle newly cloned node. 1806 if (!Node1 && MBB1->succ_size() == 1) 1807 return findNearestCommonPostDom(*MBB1->succ_begin(), MBB2); 1808 if (!Node2 && MBB2->succ_size() == 1) 1809 return findNearestCommonPostDom(MBB1, *MBB2->succ_begin()); 1810 1811 if (!Node1 || !Node2) 1812 return NULL; 1813 1814 Node1 = Node1->getIDom(); 1815 while (Node1) { 1816 if (PDT->dominates(Node1, Node2)) 1817 return Node1->getBlock(); 1818 Node1 = Node1->getIDom(); 1819 } 1820 1821 return NULL; 1822 } 1823 1824 MachineBasicBlock * 1825 AMDGPUCFGStructurizer::findNearestCommonPostDom( 1826 std::set<MachineBasicBlock *> &MBBs) { 1827 MachineBasicBlock *CommonDom; 1828 std::set<MachineBasicBlock *>::const_iterator It = MBBs.begin(); 1829 std::set<MachineBasicBlock *>::const_iterator E = MBBs.end(); 1830 for (CommonDom = *It; It != E && CommonDom; ++It) { 1831 MachineBasicBlock *MBB = *It; 1832 if (MBB != CommonDom) 1833 CommonDom = findNearestCommonPostDom(MBB, CommonDom); 1834 } 1835 1836 DEBUG( 1837 dbgs() << "Common post dominator for exit blocks is "; 1838 if (CommonDom) 1839 dbgs() << "BB" << CommonDom->getNumber() << "\n"; 1840 else 1841 dbgs() << "NULL\n"; 1842 ); 1843 1844 return CommonDom; 1845 } 1846 1847 char AMDGPUCFGStructurizer::ID = 0; 1848 1849 } // end anonymous namespace 1850 1851 1852 FunctionPass *llvm::createAMDGPUCFGStructurizerPass(TargetMachine &tm) { 1853 return new AMDGPUCFGStructurizer(tm); 1854 } 1855