1 //===- InstCombinePHI.cpp -------------------------------------------------===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 // This file implements the visitPHINode function. 11 // 12 //===----------------------------------------------------------------------===// 13 14 #include "InstCombine.h" 15 #include "llvm/ADT/STLExtras.h" 16 #include "llvm/ADT/SmallPtrSet.h" 17 #include "llvm/Analysis/InstructionSimplify.h" 18 #include "llvm/IR/DataLayout.h" 19 using namespace llvm; 20 21 #define DEBUG_TYPE "instcombine" 22 23 /// FoldPHIArgBinOpIntoPHI - If we have something like phi [add (a,b), add(a,c)] 24 /// and if a/b/c and the add's all have a single use, turn this into a phi 25 /// and a single binop. 26 Instruction *InstCombiner::FoldPHIArgBinOpIntoPHI(PHINode &PN) { 27 Instruction *FirstInst = cast<Instruction>(PN.getIncomingValue(0)); 28 assert(isa<BinaryOperator>(FirstInst) || isa<CmpInst>(FirstInst)); 29 unsigned Opc = FirstInst->getOpcode(); 30 Value *LHSVal = FirstInst->getOperand(0); 31 Value *RHSVal = FirstInst->getOperand(1); 32 33 Type *LHSType = LHSVal->getType(); 34 Type *RHSType = RHSVal->getType(); 35 36 bool isNUW = false, isNSW = false, isExact = false; 37 if (OverflowingBinaryOperator *BO = 38 dyn_cast<OverflowingBinaryOperator>(FirstInst)) { 39 isNUW = BO->hasNoUnsignedWrap(); 40 isNSW = BO->hasNoSignedWrap(); 41 } else if (PossiblyExactOperator *PEO = 42 dyn_cast<PossiblyExactOperator>(FirstInst)) 43 isExact = PEO->isExact(); 44 45 // Scan to see if all operands are the same opcode, and all have one use. 46 for (unsigned i = 1; i != PN.getNumIncomingValues(); ++i) { 47 Instruction *I = dyn_cast<Instruction>(PN.getIncomingValue(i)); 48 if (!I || I->getOpcode() != Opc || !I->hasOneUse() || 49 // Verify type of the LHS matches so we don't fold cmp's of different 50 // types. 51 I->getOperand(0)->getType() != LHSType || 52 I->getOperand(1)->getType() != RHSType) 53 return nullptr; 54 55 // If they are CmpInst instructions, check their predicates 56 if (CmpInst *CI = dyn_cast<CmpInst>(I)) 57 if (CI->getPredicate() != cast<CmpInst>(FirstInst)->getPredicate()) 58 return nullptr; 59 60 if (isNUW) 61 isNUW = cast<OverflowingBinaryOperator>(I)->hasNoUnsignedWrap(); 62 if (isNSW) 63 isNSW = cast<OverflowingBinaryOperator>(I)->hasNoSignedWrap(); 64 if (isExact) 65 isExact = cast<PossiblyExactOperator>(I)->isExact(); 66 67 // Keep track of which operand needs a phi node. 68 if (I->getOperand(0) != LHSVal) LHSVal = nullptr; 69 if (I->getOperand(1) != RHSVal) RHSVal = nullptr; 70 } 71 72 // If both LHS and RHS would need a PHI, don't do this transformation, 73 // because it would increase the number of PHIs entering the block, 74 // which leads to higher register pressure. This is especially 75 // bad when the PHIs are in the header of a loop. 76 if (!LHSVal && !RHSVal) 77 return nullptr; 78 79 // Otherwise, this is safe to transform! 80 81 Value *InLHS = FirstInst->getOperand(0); 82 Value *InRHS = FirstInst->getOperand(1); 83 PHINode *NewLHS = nullptr, *NewRHS = nullptr; 84 if (!LHSVal) { 85 NewLHS = PHINode::Create(LHSType, PN.getNumIncomingValues(), 86 FirstInst->getOperand(0)->getName() + ".pn"); 87 NewLHS->addIncoming(InLHS, PN.getIncomingBlock(0)); 88 InsertNewInstBefore(NewLHS, PN); 89 LHSVal = NewLHS; 90 } 91 92 if (!RHSVal) { 93 NewRHS = PHINode::Create(RHSType, PN.getNumIncomingValues(), 94 FirstInst->getOperand(1)->getName() + ".pn"); 95 NewRHS->addIncoming(InRHS, PN.getIncomingBlock(0)); 96 InsertNewInstBefore(NewRHS, PN); 97 RHSVal = NewRHS; 98 } 99 100 // Add all operands to the new PHIs. 101 if (NewLHS || NewRHS) { 102 for (unsigned i = 1, e = PN.getNumIncomingValues(); i != e; ++i) { 103 Instruction *InInst = cast<Instruction>(PN.getIncomingValue(i)); 104 if (NewLHS) { 105 Value *NewInLHS = InInst->getOperand(0); 106 NewLHS->addIncoming(NewInLHS, PN.getIncomingBlock(i)); 107 } 108 if (NewRHS) { 109 Value *NewInRHS = InInst->getOperand(1); 110 NewRHS->addIncoming(NewInRHS, PN.getIncomingBlock(i)); 111 } 112 } 113 } 114 115 if (CmpInst *CIOp = dyn_cast<CmpInst>(FirstInst)) { 116 CmpInst *NewCI = CmpInst::Create(CIOp->getOpcode(), CIOp->getPredicate(), 117 LHSVal, RHSVal); 118 NewCI->setDebugLoc(FirstInst->getDebugLoc()); 119 return NewCI; 120 } 121 122 BinaryOperator *BinOp = cast<BinaryOperator>(FirstInst); 123 BinaryOperator *NewBinOp = 124 BinaryOperator::Create(BinOp->getOpcode(), LHSVal, RHSVal); 125 if (isNUW) NewBinOp->setHasNoUnsignedWrap(); 126 if (isNSW) NewBinOp->setHasNoSignedWrap(); 127 if (isExact) NewBinOp->setIsExact(); 128 NewBinOp->setDebugLoc(FirstInst->getDebugLoc()); 129 return NewBinOp; 130 } 131 132 Instruction *InstCombiner::FoldPHIArgGEPIntoPHI(PHINode &PN) { 133 GetElementPtrInst *FirstInst =cast<GetElementPtrInst>(PN.getIncomingValue(0)); 134 135 SmallVector<Value*, 16> FixedOperands(FirstInst->op_begin(), 136 FirstInst->op_end()); 137 // This is true if all GEP bases are allocas and if all indices into them are 138 // constants. 139 bool AllBasePointersAreAllocas = true; 140 141 // We don't want to replace this phi if the replacement would require 142 // more than one phi, which leads to higher register pressure. This is 143 // especially bad when the PHIs are in the header of a loop. 144 bool NeededPhi = false; 145 146 bool AllInBounds = true; 147 148 // Scan to see if all operands are the same opcode, and all have one use. 149 for (unsigned i = 1; i != PN.getNumIncomingValues(); ++i) { 150 GetElementPtrInst *GEP= dyn_cast<GetElementPtrInst>(PN.getIncomingValue(i)); 151 if (!GEP || !GEP->hasOneUse() || GEP->getType() != FirstInst->getType() || 152 GEP->getNumOperands() != FirstInst->getNumOperands()) 153 return nullptr; 154 155 AllInBounds &= GEP->isInBounds(); 156 157 // Keep track of whether or not all GEPs are of alloca pointers. 158 if (AllBasePointersAreAllocas && 159 (!isa<AllocaInst>(GEP->getOperand(0)) || 160 !GEP->hasAllConstantIndices())) 161 AllBasePointersAreAllocas = false; 162 163 // Compare the operand lists. 164 for (unsigned op = 0, e = FirstInst->getNumOperands(); op != e; ++op) { 165 if (FirstInst->getOperand(op) == GEP->getOperand(op)) 166 continue; 167 168 // Don't merge two GEPs when two operands differ (introducing phi nodes) 169 // if one of the PHIs has a constant for the index. The index may be 170 // substantially cheaper to compute for the constants, so making it a 171 // variable index could pessimize the path. This also handles the case 172 // for struct indices, which must always be constant. 173 if (isa<ConstantInt>(FirstInst->getOperand(op)) || 174 isa<ConstantInt>(GEP->getOperand(op))) 175 return nullptr; 176 177 if (FirstInst->getOperand(op)->getType() !=GEP->getOperand(op)->getType()) 178 return nullptr; 179 180 // If we already needed a PHI for an earlier operand, and another operand 181 // also requires a PHI, we'd be introducing more PHIs than we're 182 // eliminating, which increases register pressure on entry to the PHI's 183 // block. 184 if (NeededPhi) 185 return nullptr; 186 187 FixedOperands[op] = nullptr; // Needs a PHI. 188 NeededPhi = true; 189 } 190 } 191 192 // If all of the base pointers of the PHI'd GEPs are from allocas, don't 193 // bother doing this transformation. At best, this will just save a bit of 194 // offset calculation, but all the predecessors will have to materialize the 195 // stack address into a register anyway. We'd actually rather *clone* the 196 // load up into the predecessors so that we have a load of a gep of an alloca, 197 // which can usually all be folded into the load. 198 if (AllBasePointersAreAllocas) 199 return nullptr; 200 201 // Otherwise, this is safe to transform. Insert PHI nodes for each operand 202 // that is variable. 203 SmallVector<PHINode*, 16> OperandPhis(FixedOperands.size()); 204 205 bool HasAnyPHIs = false; 206 for (unsigned i = 0, e = FixedOperands.size(); i != e; ++i) { 207 if (FixedOperands[i]) continue; // operand doesn't need a phi. 208 Value *FirstOp = FirstInst->getOperand(i); 209 PHINode *NewPN = PHINode::Create(FirstOp->getType(), e, 210 FirstOp->getName()+".pn"); 211 InsertNewInstBefore(NewPN, PN); 212 213 NewPN->addIncoming(FirstOp, PN.getIncomingBlock(0)); 214 OperandPhis[i] = NewPN; 215 FixedOperands[i] = NewPN; 216 HasAnyPHIs = true; 217 } 218 219 220 // Add all operands to the new PHIs. 221 if (HasAnyPHIs) { 222 for (unsigned i = 1, e = PN.getNumIncomingValues(); i != e; ++i) { 223 GetElementPtrInst *InGEP =cast<GetElementPtrInst>(PN.getIncomingValue(i)); 224 BasicBlock *InBB = PN.getIncomingBlock(i); 225 226 for (unsigned op = 0, e = OperandPhis.size(); op != e; ++op) 227 if (PHINode *OpPhi = OperandPhis[op]) 228 OpPhi->addIncoming(InGEP->getOperand(op), InBB); 229 } 230 } 231 232 Value *Base = FixedOperands[0]; 233 GetElementPtrInst *NewGEP = 234 GetElementPtrInst::Create(Base, makeArrayRef(FixedOperands).slice(1)); 235 if (AllInBounds) NewGEP->setIsInBounds(); 236 NewGEP->setDebugLoc(FirstInst->getDebugLoc()); 237 return NewGEP; 238 } 239 240 241 /// isSafeAndProfitableToSinkLoad - Return true if we know that it is safe to 242 /// sink the load out of the block that defines it. This means that it must be 243 /// obvious the value of the load is not changed from the point of the load to 244 /// the end of the block it is in. 245 /// 246 /// Finally, it is safe, but not profitable, to sink a load targeting a 247 /// non-address-taken alloca. Doing so will cause us to not promote the alloca 248 /// to a register. 249 static bool isSafeAndProfitableToSinkLoad(LoadInst *L) { 250 BasicBlock::iterator BBI = L, E = L->getParent()->end(); 251 252 for (++BBI; BBI != E; ++BBI) 253 if (BBI->mayWriteToMemory()) 254 return false; 255 256 // Check for non-address taken alloca. If not address-taken already, it isn't 257 // profitable to do this xform. 258 if (AllocaInst *AI = dyn_cast<AllocaInst>(L->getOperand(0))) { 259 bool isAddressTaken = false; 260 for (User *U : AI->users()) { 261 if (isa<LoadInst>(U)) continue; 262 if (StoreInst *SI = dyn_cast<StoreInst>(U)) { 263 // If storing TO the alloca, then the address isn't taken. 264 if (SI->getOperand(1) == AI) continue; 265 } 266 isAddressTaken = true; 267 break; 268 } 269 270 if (!isAddressTaken && AI->isStaticAlloca()) 271 return false; 272 } 273 274 // If this load is a load from a GEP with a constant offset from an alloca, 275 // then we don't want to sink it. In its present form, it will be 276 // load [constant stack offset]. Sinking it will cause us to have to 277 // materialize the stack addresses in each predecessor in a register only to 278 // do a shared load from register in the successor. 279 if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(L->getOperand(0))) 280 if (AllocaInst *AI = dyn_cast<AllocaInst>(GEP->getOperand(0))) 281 if (AI->isStaticAlloca() && GEP->hasAllConstantIndices()) 282 return false; 283 284 return true; 285 } 286 287 Instruction *InstCombiner::FoldPHIArgLoadIntoPHI(PHINode &PN) { 288 LoadInst *FirstLI = cast<LoadInst>(PN.getIncomingValue(0)); 289 290 // FIXME: This is overconservative; this transform is allowed in some cases 291 // for atomic operations. 292 if (FirstLI->isAtomic()) 293 return nullptr; 294 295 // When processing loads, we need to propagate two bits of information to the 296 // sunk load: whether it is volatile, and what its alignment is. We currently 297 // don't sink loads when some have their alignment specified and some don't. 298 // visitLoadInst will propagate an alignment onto the load when TD is around, 299 // and if TD isn't around, we can't handle the mixed case. 300 bool isVolatile = FirstLI->isVolatile(); 301 unsigned LoadAlignment = FirstLI->getAlignment(); 302 unsigned LoadAddrSpace = FirstLI->getPointerAddressSpace(); 303 304 // We can't sink the load if the loaded value could be modified between the 305 // load and the PHI. 306 if (FirstLI->getParent() != PN.getIncomingBlock(0) || 307 !isSafeAndProfitableToSinkLoad(FirstLI)) 308 return nullptr; 309 310 // If the PHI is of volatile loads and the load block has multiple 311 // successors, sinking it would remove a load of the volatile value from 312 // the path through the other successor. 313 if (isVolatile && 314 FirstLI->getParent()->getTerminator()->getNumSuccessors() != 1) 315 return nullptr; 316 317 // Check to see if all arguments are the same operation. 318 for (unsigned i = 1, e = PN.getNumIncomingValues(); i != e; ++i) { 319 LoadInst *LI = dyn_cast<LoadInst>(PN.getIncomingValue(i)); 320 if (!LI || !LI->hasOneUse()) 321 return nullptr; 322 323 // We can't sink the load if the loaded value could be modified between 324 // the load and the PHI. 325 if (LI->isVolatile() != isVolatile || 326 LI->getParent() != PN.getIncomingBlock(i) || 327 LI->getPointerAddressSpace() != LoadAddrSpace || 328 !isSafeAndProfitableToSinkLoad(LI)) 329 return nullptr; 330 331 // If some of the loads have an alignment specified but not all of them, 332 // we can't do the transformation. 333 if ((LoadAlignment != 0) != (LI->getAlignment() != 0)) 334 return nullptr; 335 336 LoadAlignment = std::min(LoadAlignment, LI->getAlignment()); 337 338 // If the PHI is of volatile loads and the load block has multiple 339 // successors, sinking it would remove a load of the volatile value from 340 // the path through the other successor. 341 if (isVolatile && 342 LI->getParent()->getTerminator()->getNumSuccessors() != 1) 343 return nullptr; 344 } 345 346 // Okay, they are all the same operation. Create a new PHI node of the 347 // correct type, and PHI together all of the LHS's of the instructions. 348 PHINode *NewPN = PHINode::Create(FirstLI->getOperand(0)->getType(), 349 PN.getNumIncomingValues(), 350 PN.getName()+".in"); 351 352 Value *InVal = FirstLI->getOperand(0); 353 NewPN->addIncoming(InVal, PN.getIncomingBlock(0)); 354 355 // Add all operands to the new PHI. 356 for (unsigned i = 1, e = PN.getNumIncomingValues(); i != e; ++i) { 357 Value *NewInVal = cast<LoadInst>(PN.getIncomingValue(i))->getOperand(0); 358 if (NewInVal != InVal) 359 InVal = nullptr; 360 NewPN->addIncoming(NewInVal, PN.getIncomingBlock(i)); 361 } 362 363 Value *PhiVal; 364 if (InVal) { 365 // The new PHI unions all of the same values together. This is really 366 // common, so we handle it intelligently here for compile-time speed. 367 PhiVal = InVal; 368 delete NewPN; 369 } else { 370 InsertNewInstBefore(NewPN, PN); 371 PhiVal = NewPN; 372 } 373 374 // If this was a volatile load that we are merging, make sure to loop through 375 // and mark all the input loads as non-volatile. If we don't do this, we will 376 // insert a new volatile load and the old ones will not be deletable. 377 if (isVolatile) 378 for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i) 379 cast<LoadInst>(PN.getIncomingValue(i))->setVolatile(false); 380 381 LoadInst *NewLI = new LoadInst(PhiVal, "", isVolatile, LoadAlignment); 382 NewLI->setDebugLoc(FirstLI->getDebugLoc()); 383 return NewLI; 384 } 385 386 387 388 /// FoldPHIArgOpIntoPHI - If all operands to a PHI node are the same "unary" 389 /// operator and they all are only used by the PHI, PHI together their 390 /// inputs, and do the operation once, to the result of the PHI. 391 Instruction *InstCombiner::FoldPHIArgOpIntoPHI(PHINode &PN) { 392 Instruction *FirstInst = cast<Instruction>(PN.getIncomingValue(0)); 393 394 if (isa<GetElementPtrInst>(FirstInst)) 395 return FoldPHIArgGEPIntoPHI(PN); 396 if (isa<LoadInst>(FirstInst)) 397 return FoldPHIArgLoadIntoPHI(PN); 398 399 // Scan the instruction, looking for input operations that can be folded away. 400 // If all input operands to the phi are the same instruction (e.g. a cast from 401 // the same type or "+42") we can pull the operation through the PHI, reducing 402 // code size and simplifying code. 403 Constant *ConstantOp = nullptr; 404 Type *CastSrcTy = nullptr; 405 bool isNUW = false, isNSW = false, isExact = false; 406 407 if (isa<CastInst>(FirstInst)) { 408 CastSrcTy = FirstInst->getOperand(0)->getType(); 409 410 // Be careful about transforming integer PHIs. We don't want to pessimize 411 // the code by turning an i32 into an i1293. 412 if (PN.getType()->isIntegerTy() && CastSrcTy->isIntegerTy()) { 413 if (!ShouldChangeType(PN.getType(), CastSrcTy)) 414 return nullptr; 415 } 416 } else if (isa<BinaryOperator>(FirstInst) || isa<CmpInst>(FirstInst)) { 417 // Can fold binop, compare or shift here if the RHS is a constant, 418 // otherwise call FoldPHIArgBinOpIntoPHI. 419 ConstantOp = dyn_cast<Constant>(FirstInst->getOperand(1)); 420 if (!ConstantOp) 421 return FoldPHIArgBinOpIntoPHI(PN); 422 423 if (OverflowingBinaryOperator *BO = 424 dyn_cast<OverflowingBinaryOperator>(FirstInst)) { 425 isNUW = BO->hasNoUnsignedWrap(); 426 isNSW = BO->hasNoSignedWrap(); 427 } else if (PossiblyExactOperator *PEO = 428 dyn_cast<PossiblyExactOperator>(FirstInst)) 429 isExact = PEO->isExact(); 430 } else { 431 return nullptr; // Cannot fold this operation. 432 } 433 434 // Check to see if all arguments are the same operation. 435 for (unsigned i = 1, e = PN.getNumIncomingValues(); i != e; ++i) { 436 Instruction *I = dyn_cast<Instruction>(PN.getIncomingValue(i)); 437 if (!I || !I->hasOneUse() || !I->isSameOperationAs(FirstInst)) 438 return nullptr; 439 if (CastSrcTy) { 440 if (I->getOperand(0)->getType() != CastSrcTy) 441 return nullptr; // Cast operation must match. 442 } else if (I->getOperand(1) != ConstantOp) { 443 return nullptr; 444 } 445 446 if (isNUW) 447 isNUW = cast<OverflowingBinaryOperator>(I)->hasNoUnsignedWrap(); 448 if (isNSW) 449 isNSW = cast<OverflowingBinaryOperator>(I)->hasNoSignedWrap(); 450 if (isExact) 451 isExact = cast<PossiblyExactOperator>(I)->isExact(); 452 } 453 454 // Okay, they are all the same operation. Create a new PHI node of the 455 // correct type, and PHI together all of the LHS's of the instructions. 456 PHINode *NewPN = PHINode::Create(FirstInst->getOperand(0)->getType(), 457 PN.getNumIncomingValues(), 458 PN.getName()+".in"); 459 460 Value *InVal = FirstInst->getOperand(0); 461 NewPN->addIncoming(InVal, PN.getIncomingBlock(0)); 462 463 // Add all operands to the new PHI. 464 for (unsigned i = 1, e = PN.getNumIncomingValues(); i != e; ++i) { 465 Value *NewInVal = cast<Instruction>(PN.getIncomingValue(i))->getOperand(0); 466 if (NewInVal != InVal) 467 InVal = nullptr; 468 NewPN->addIncoming(NewInVal, PN.getIncomingBlock(i)); 469 } 470 471 Value *PhiVal; 472 if (InVal) { 473 // The new PHI unions all of the same values together. This is really 474 // common, so we handle it intelligently here for compile-time speed. 475 PhiVal = InVal; 476 delete NewPN; 477 } else { 478 InsertNewInstBefore(NewPN, PN); 479 PhiVal = NewPN; 480 } 481 482 // Insert and return the new operation. 483 if (CastInst *FirstCI = dyn_cast<CastInst>(FirstInst)) { 484 CastInst *NewCI = CastInst::Create(FirstCI->getOpcode(), PhiVal, 485 PN.getType()); 486 NewCI->setDebugLoc(FirstInst->getDebugLoc()); 487 return NewCI; 488 } 489 490 if (BinaryOperator *BinOp = dyn_cast<BinaryOperator>(FirstInst)) { 491 BinOp = BinaryOperator::Create(BinOp->getOpcode(), PhiVal, ConstantOp); 492 if (isNUW) BinOp->setHasNoUnsignedWrap(); 493 if (isNSW) BinOp->setHasNoSignedWrap(); 494 if (isExact) BinOp->setIsExact(); 495 BinOp->setDebugLoc(FirstInst->getDebugLoc()); 496 return BinOp; 497 } 498 499 CmpInst *CIOp = cast<CmpInst>(FirstInst); 500 CmpInst *NewCI = CmpInst::Create(CIOp->getOpcode(), CIOp->getPredicate(), 501 PhiVal, ConstantOp); 502 NewCI->setDebugLoc(FirstInst->getDebugLoc()); 503 return NewCI; 504 } 505 506 /// DeadPHICycle - Return true if this PHI node is only used by a PHI node cycle 507 /// that is dead. 508 static bool DeadPHICycle(PHINode *PN, 509 SmallPtrSet<PHINode*, 16> &PotentiallyDeadPHIs) { 510 if (PN->use_empty()) return true; 511 if (!PN->hasOneUse()) return false; 512 513 // Remember this node, and if we find the cycle, return. 514 if (!PotentiallyDeadPHIs.insert(PN)) 515 return true; 516 517 // Don't scan crazily complex things. 518 if (PotentiallyDeadPHIs.size() == 16) 519 return false; 520 521 if (PHINode *PU = dyn_cast<PHINode>(PN->user_back())) 522 return DeadPHICycle(PU, PotentiallyDeadPHIs); 523 524 return false; 525 } 526 527 /// PHIsEqualValue - Return true if this phi node is always equal to 528 /// NonPhiInVal. This happens with mutually cyclic phi nodes like: 529 /// z = some value; x = phi (y, z); y = phi (x, z) 530 static bool PHIsEqualValue(PHINode *PN, Value *NonPhiInVal, 531 SmallPtrSet<PHINode*, 16> &ValueEqualPHIs) { 532 // See if we already saw this PHI node. 533 if (!ValueEqualPHIs.insert(PN)) 534 return true; 535 536 // Don't scan crazily complex things. 537 if (ValueEqualPHIs.size() == 16) 538 return false; 539 540 // Scan the operands to see if they are either phi nodes or are equal to 541 // the value. 542 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) { 543 Value *Op = PN->getIncomingValue(i); 544 if (PHINode *OpPN = dyn_cast<PHINode>(Op)) { 545 if (!PHIsEqualValue(OpPN, NonPhiInVal, ValueEqualPHIs)) 546 return false; 547 } else if (Op != NonPhiInVal) 548 return false; 549 } 550 551 return true; 552 } 553 554 555 namespace { 556 struct PHIUsageRecord { 557 unsigned PHIId; // The ID # of the PHI (something determinstic to sort on) 558 unsigned Shift; // The amount shifted. 559 Instruction *Inst; // The trunc instruction. 560 561 PHIUsageRecord(unsigned pn, unsigned Sh, Instruction *User) 562 : PHIId(pn), Shift(Sh), Inst(User) {} 563 564 bool operator<(const PHIUsageRecord &RHS) const { 565 if (PHIId < RHS.PHIId) return true; 566 if (PHIId > RHS.PHIId) return false; 567 if (Shift < RHS.Shift) return true; 568 if (Shift > RHS.Shift) return false; 569 return Inst->getType()->getPrimitiveSizeInBits() < 570 RHS.Inst->getType()->getPrimitiveSizeInBits(); 571 } 572 }; 573 574 struct LoweredPHIRecord { 575 PHINode *PN; // The PHI that was lowered. 576 unsigned Shift; // The amount shifted. 577 unsigned Width; // The width extracted. 578 579 LoweredPHIRecord(PHINode *pn, unsigned Sh, Type *Ty) 580 : PN(pn), Shift(Sh), Width(Ty->getPrimitiveSizeInBits()) {} 581 582 // Ctor form used by DenseMap. 583 LoweredPHIRecord(PHINode *pn, unsigned Sh) 584 : PN(pn), Shift(Sh), Width(0) {} 585 }; 586 } 587 588 namespace llvm { 589 template<> 590 struct DenseMapInfo<LoweredPHIRecord> { 591 static inline LoweredPHIRecord getEmptyKey() { 592 return LoweredPHIRecord(nullptr, 0); 593 } 594 static inline LoweredPHIRecord getTombstoneKey() { 595 return LoweredPHIRecord(nullptr, 1); 596 } 597 static unsigned getHashValue(const LoweredPHIRecord &Val) { 598 return DenseMapInfo<PHINode*>::getHashValue(Val.PN) ^ (Val.Shift>>3) ^ 599 (Val.Width>>3); 600 } 601 static bool isEqual(const LoweredPHIRecord &LHS, 602 const LoweredPHIRecord &RHS) { 603 return LHS.PN == RHS.PN && LHS.Shift == RHS.Shift && 604 LHS.Width == RHS.Width; 605 } 606 }; 607 } 608 609 610 /// SliceUpIllegalIntegerPHI - This is an integer PHI and we know that it has an 611 /// illegal type: see if it is only used by trunc or trunc(lshr) operations. If 612 /// so, we split the PHI into the various pieces being extracted. This sort of 613 /// thing is introduced when SROA promotes an aggregate to large integer values. 614 /// 615 /// TODO: The user of the trunc may be an bitcast to float/double/vector or an 616 /// inttoptr. We should produce new PHIs in the right type. 617 /// 618 Instruction *InstCombiner::SliceUpIllegalIntegerPHI(PHINode &FirstPhi) { 619 // PHIUsers - Keep track of all of the truncated values extracted from a set 620 // of PHIs, along with their offset. These are the things we want to rewrite. 621 SmallVector<PHIUsageRecord, 16> PHIUsers; 622 623 // PHIs are often mutually cyclic, so we keep track of a whole set of PHI 624 // nodes which are extracted from. PHIsToSlice is a set we use to avoid 625 // revisiting PHIs, PHIsInspected is a ordered list of PHIs that we need to 626 // check the uses of (to ensure they are all extracts). 627 SmallVector<PHINode*, 8> PHIsToSlice; 628 SmallPtrSet<PHINode*, 8> PHIsInspected; 629 630 PHIsToSlice.push_back(&FirstPhi); 631 PHIsInspected.insert(&FirstPhi); 632 633 for (unsigned PHIId = 0; PHIId != PHIsToSlice.size(); ++PHIId) { 634 PHINode *PN = PHIsToSlice[PHIId]; 635 636 // Scan the input list of the PHI. If any input is an invoke, and if the 637 // input is defined in the predecessor, then we won't be split the critical 638 // edge which is required to insert a truncate. Because of this, we have to 639 // bail out. 640 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) { 641 InvokeInst *II = dyn_cast<InvokeInst>(PN->getIncomingValue(i)); 642 if (!II) continue; 643 if (II->getParent() != PN->getIncomingBlock(i)) 644 continue; 645 646 // If we have a phi, and if it's directly in the predecessor, then we have 647 // a critical edge where we need to put the truncate. Since we can't 648 // split the edge in instcombine, we have to bail out. 649 return nullptr; 650 } 651 652 for (User *U : PN->users()) { 653 Instruction *UserI = cast<Instruction>(U); 654 655 // If the user is a PHI, inspect its uses recursively. 656 if (PHINode *UserPN = dyn_cast<PHINode>(UserI)) { 657 if (PHIsInspected.insert(UserPN)) 658 PHIsToSlice.push_back(UserPN); 659 continue; 660 } 661 662 // Truncates are always ok. 663 if (isa<TruncInst>(UserI)) { 664 PHIUsers.push_back(PHIUsageRecord(PHIId, 0, UserI)); 665 continue; 666 } 667 668 // Otherwise it must be a lshr which can only be used by one trunc. 669 if (UserI->getOpcode() != Instruction::LShr || 670 !UserI->hasOneUse() || !isa<TruncInst>(UserI->user_back()) || 671 !isa<ConstantInt>(UserI->getOperand(1))) 672 return nullptr; 673 674 unsigned Shift = cast<ConstantInt>(UserI->getOperand(1))->getZExtValue(); 675 PHIUsers.push_back(PHIUsageRecord(PHIId, Shift, UserI->user_back())); 676 } 677 } 678 679 // If we have no users, they must be all self uses, just nuke the PHI. 680 if (PHIUsers.empty()) 681 return ReplaceInstUsesWith(FirstPhi, UndefValue::get(FirstPhi.getType())); 682 683 // If this phi node is transformable, create new PHIs for all the pieces 684 // extracted out of it. First, sort the users by their offset and size. 685 array_pod_sort(PHIUsers.begin(), PHIUsers.end()); 686 687 DEBUG(dbgs() << "SLICING UP PHI: " << FirstPhi << '\n'; 688 for (unsigned i = 1, e = PHIsToSlice.size(); i != e; ++i) 689 dbgs() << "AND USER PHI #" << i << ": " << *PHIsToSlice[i] << '\n'; 690 ); 691 692 // PredValues - This is a temporary used when rewriting PHI nodes. It is 693 // hoisted out here to avoid construction/destruction thrashing. 694 DenseMap<BasicBlock*, Value*> PredValues; 695 696 // ExtractedVals - Each new PHI we introduce is saved here so we don't 697 // introduce redundant PHIs. 698 DenseMap<LoweredPHIRecord, PHINode*> ExtractedVals; 699 700 for (unsigned UserI = 0, UserE = PHIUsers.size(); UserI != UserE; ++UserI) { 701 unsigned PHIId = PHIUsers[UserI].PHIId; 702 PHINode *PN = PHIsToSlice[PHIId]; 703 unsigned Offset = PHIUsers[UserI].Shift; 704 Type *Ty = PHIUsers[UserI].Inst->getType(); 705 706 PHINode *EltPHI; 707 708 // If we've already lowered a user like this, reuse the previously lowered 709 // value. 710 if ((EltPHI = ExtractedVals[LoweredPHIRecord(PN, Offset, Ty)]) == nullptr) { 711 712 // Otherwise, Create the new PHI node for this user. 713 EltPHI = PHINode::Create(Ty, PN->getNumIncomingValues(), 714 PN->getName()+".off"+Twine(Offset), PN); 715 assert(EltPHI->getType() != PN->getType() && 716 "Truncate didn't shrink phi?"); 717 718 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) { 719 BasicBlock *Pred = PN->getIncomingBlock(i); 720 Value *&PredVal = PredValues[Pred]; 721 722 // If we already have a value for this predecessor, reuse it. 723 if (PredVal) { 724 EltPHI->addIncoming(PredVal, Pred); 725 continue; 726 } 727 728 // Handle the PHI self-reuse case. 729 Value *InVal = PN->getIncomingValue(i); 730 if (InVal == PN) { 731 PredVal = EltPHI; 732 EltPHI->addIncoming(PredVal, Pred); 733 continue; 734 } 735 736 if (PHINode *InPHI = dyn_cast<PHINode>(PN)) { 737 // If the incoming value was a PHI, and if it was one of the PHIs we 738 // already rewrote it, just use the lowered value. 739 if (Value *Res = ExtractedVals[LoweredPHIRecord(InPHI, Offset, Ty)]) { 740 PredVal = Res; 741 EltPHI->addIncoming(PredVal, Pred); 742 continue; 743 } 744 } 745 746 // Otherwise, do an extract in the predecessor. 747 Builder->SetInsertPoint(Pred, Pred->getTerminator()); 748 Value *Res = InVal; 749 if (Offset) 750 Res = Builder->CreateLShr(Res, ConstantInt::get(InVal->getType(), 751 Offset), "extract"); 752 Res = Builder->CreateTrunc(Res, Ty, "extract.t"); 753 PredVal = Res; 754 EltPHI->addIncoming(Res, Pred); 755 756 // If the incoming value was a PHI, and if it was one of the PHIs we are 757 // rewriting, we will ultimately delete the code we inserted. This 758 // means we need to revisit that PHI to make sure we extract out the 759 // needed piece. 760 if (PHINode *OldInVal = dyn_cast<PHINode>(PN->getIncomingValue(i))) 761 if (PHIsInspected.count(OldInVal)) { 762 unsigned RefPHIId = std::find(PHIsToSlice.begin(),PHIsToSlice.end(), 763 OldInVal)-PHIsToSlice.begin(); 764 PHIUsers.push_back(PHIUsageRecord(RefPHIId, Offset, 765 cast<Instruction>(Res))); 766 ++UserE; 767 } 768 } 769 PredValues.clear(); 770 771 DEBUG(dbgs() << " Made element PHI for offset " << Offset << ": " 772 << *EltPHI << '\n'); 773 ExtractedVals[LoweredPHIRecord(PN, Offset, Ty)] = EltPHI; 774 } 775 776 // Replace the use of this piece with the PHI node. 777 ReplaceInstUsesWith(*PHIUsers[UserI].Inst, EltPHI); 778 } 779 780 // Replace all the remaining uses of the PHI nodes (self uses and the lshrs) 781 // with undefs. 782 Value *Undef = UndefValue::get(FirstPhi.getType()); 783 for (unsigned i = 1, e = PHIsToSlice.size(); i != e; ++i) 784 ReplaceInstUsesWith(*PHIsToSlice[i], Undef); 785 return ReplaceInstUsesWith(FirstPhi, Undef); 786 } 787 788 // PHINode simplification 789 // 790 Instruction *InstCombiner::visitPHINode(PHINode &PN) { 791 if (Value *V = SimplifyInstruction(&PN, DL, TLI)) 792 return ReplaceInstUsesWith(PN, V); 793 794 // If all PHI operands are the same operation, pull them through the PHI, 795 // reducing code size. 796 if (isa<Instruction>(PN.getIncomingValue(0)) && 797 isa<Instruction>(PN.getIncomingValue(1)) && 798 cast<Instruction>(PN.getIncomingValue(0))->getOpcode() == 799 cast<Instruction>(PN.getIncomingValue(1))->getOpcode() && 800 // FIXME: The hasOneUse check will fail for PHIs that use the value more 801 // than themselves more than once. 802 PN.getIncomingValue(0)->hasOneUse()) 803 if (Instruction *Result = FoldPHIArgOpIntoPHI(PN)) 804 return Result; 805 806 // If this is a trivial cycle in the PHI node graph, remove it. Basically, if 807 // this PHI only has a single use (a PHI), and if that PHI only has one use (a 808 // PHI)... break the cycle. 809 if (PN.hasOneUse()) { 810 Instruction *PHIUser = cast<Instruction>(PN.user_back()); 811 if (PHINode *PU = dyn_cast<PHINode>(PHIUser)) { 812 SmallPtrSet<PHINode*, 16> PotentiallyDeadPHIs; 813 PotentiallyDeadPHIs.insert(&PN); 814 if (DeadPHICycle(PU, PotentiallyDeadPHIs)) 815 return ReplaceInstUsesWith(PN, UndefValue::get(PN.getType())); 816 } 817 818 // If this phi has a single use, and if that use just computes a value for 819 // the next iteration of a loop, delete the phi. This occurs with unused 820 // induction variables, e.g. "for (int j = 0; ; ++j);". Detecting this 821 // common case here is good because the only other things that catch this 822 // are induction variable analysis (sometimes) and ADCE, which is only run 823 // late. 824 if (PHIUser->hasOneUse() && 825 (isa<BinaryOperator>(PHIUser) || isa<GetElementPtrInst>(PHIUser)) && 826 PHIUser->user_back() == &PN) { 827 return ReplaceInstUsesWith(PN, UndefValue::get(PN.getType())); 828 } 829 } 830 831 // We sometimes end up with phi cycles that non-obviously end up being the 832 // same value, for example: 833 // z = some value; x = phi (y, z); y = phi (x, z) 834 // where the phi nodes don't necessarily need to be in the same block. Do a 835 // quick check to see if the PHI node only contains a single non-phi value, if 836 // so, scan to see if the phi cycle is actually equal to that value. 837 { 838 unsigned InValNo = 0, NumIncomingVals = PN.getNumIncomingValues(); 839 // Scan for the first non-phi operand. 840 while (InValNo != NumIncomingVals && 841 isa<PHINode>(PN.getIncomingValue(InValNo))) 842 ++InValNo; 843 844 if (InValNo != NumIncomingVals) { 845 Value *NonPhiInVal = PN.getIncomingValue(InValNo); 846 847 // Scan the rest of the operands to see if there are any conflicts, if so 848 // there is no need to recursively scan other phis. 849 for (++InValNo; InValNo != NumIncomingVals; ++InValNo) { 850 Value *OpVal = PN.getIncomingValue(InValNo); 851 if (OpVal != NonPhiInVal && !isa<PHINode>(OpVal)) 852 break; 853 } 854 855 // If we scanned over all operands, then we have one unique value plus 856 // phi values. Scan PHI nodes to see if they all merge in each other or 857 // the value. 858 if (InValNo == NumIncomingVals) { 859 SmallPtrSet<PHINode*, 16> ValueEqualPHIs; 860 if (PHIsEqualValue(&PN, NonPhiInVal, ValueEqualPHIs)) 861 return ReplaceInstUsesWith(PN, NonPhiInVal); 862 } 863 } 864 } 865 866 // If there are multiple PHIs, sort their operands so that they all list 867 // the blocks in the same order. This will help identical PHIs be eliminated 868 // by other passes. Other passes shouldn't depend on this for correctness 869 // however. 870 PHINode *FirstPN = cast<PHINode>(PN.getParent()->begin()); 871 if (&PN != FirstPN) 872 for (unsigned i = 0, e = FirstPN->getNumIncomingValues(); i != e; ++i) { 873 BasicBlock *BBA = PN.getIncomingBlock(i); 874 BasicBlock *BBB = FirstPN->getIncomingBlock(i); 875 if (BBA != BBB) { 876 Value *VA = PN.getIncomingValue(i); 877 unsigned j = PN.getBasicBlockIndex(BBB); 878 Value *VB = PN.getIncomingValue(j); 879 PN.setIncomingBlock(i, BBB); 880 PN.setIncomingValue(i, VB); 881 PN.setIncomingBlock(j, BBA); 882 PN.setIncomingValue(j, VA); 883 // NOTE: Instcombine normally would want us to "return &PN" if we 884 // modified any of the operands of an instruction. However, since we 885 // aren't adding or removing uses (just rearranging them) we don't do 886 // this in this case. 887 } 888 } 889 890 // If this is an integer PHI and we know that it has an illegal type, see if 891 // it is only used by trunc or trunc(lshr) operations. If so, we split the 892 // PHI into the various pieces being extracted. This sort of thing is 893 // introduced when SROA promotes an aggregate to a single large integer type. 894 if (PN.getType()->isIntegerTy() && DL && 895 !DL->isLegalInteger(PN.getType()->getPrimitiveSizeInBits())) 896 if (Instruction *Res = SliceUpIllegalIntegerPHI(PN)) 897 return Res; 898 899 return nullptr; 900 } 901