1 //===- InstructionSimplify.cpp - Fold instruction operands ----------------===// 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 routines for folding instructions into simpler forms 11 // that do not require creating new instructions. This does constant folding 12 // ("add i32 1, 1" -> "2") but can also handle non-constant operands, either 13 // returning a constant ("and i32 %x, 0" -> "0") or an already existing value 14 // ("and i32 %x, %x" -> "%x"). All operands are assumed to have already been 15 // simplified: This is usually true and assuming it simplifies the logic (if 16 // they have not been simplified then results are correct but maybe suboptimal). 17 // 18 //===----------------------------------------------------------------------===// 19 20 #include "llvm/Analysis/InstructionSimplify.h" 21 #include "llvm/ADT/SetVector.h" 22 #include "llvm/ADT/Statistic.h" 23 #include "llvm/Analysis/AliasAnalysis.h" 24 #include "llvm/Analysis/CaptureTracking.h" 25 #include "llvm/Analysis/ConstantFolding.h" 26 #include "llvm/Analysis/MemoryBuiltins.h" 27 #include "llvm/Analysis/ValueTracking.h" 28 #include "llvm/Analysis/VectorUtils.h" 29 #include "llvm/IR/ConstantRange.h" 30 #include "llvm/IR/DataLayout.h" 31 #include "llvm/IR/Dominators.h" 32 #include "llvm/IR/GetElementPtrTypeIterator.h" 33 #include "llvm/IR/GlobalAlias.h" 34 #include "llvm/IR/Operator.h" 35 #include "llvm/IR/PatternMatch.h" 36 #include "llvm/IR/ValueHandle.h" 37 #include <algorithm> 38 using namespace llvm; 39 using namespace llvm::PatternMatch; 40 41 #define DEBUG_TYPE "instsimplify" 42 43 enum { RecursionLimit = 3 }; 44 45 STATISTIC(NumExpand, "Number of expansions"); 46 STATISTIC(NumReassoc, "Number of reassociations"); 47 48 namespace { 49 struct Query { 50 const DataLayout &DL; 51 const TargetLibraryInfo *TLI; 52 const DominatorTree *DT; 53 AssumptionCache *AC; 54 const Instruction *CxtI; 55 56 Query(const DataLayout &DL, const TargetLibraryInfo *tli, 57 const DominatorTree *dt, AssumptionCache *ac = nullptr, 58 const Instruction *cxti = nullptr) 59 : DL(DL), TLI(tli), DT(dt), AC(ac), CxtI(cxti) {} 60 }; 61 } // end anonymous namespace 62 63 static Value *SimplifyAndInst(Value *, Value *, const Query &, unsigned); 64 static Value *SimplifyBinOp(unsigned, Value *, Value *, const Query &, 65 unsigned); 66 static Value *SimplifyFPBinOp(unsigned, Value *, Value *, const FastMathFlags &, 67 const Query &, unsigned); 68 static Value *SimplifyCmpInst(unsigned, Value *, Value *, const Query &, 69 unsigned); 70 static Value *SimplifyOrInst(Value *, Value *, const Query &, unsigned); 71 static Value *SimplifyXorInst(Value *, Value *, const Query &, unsigned); 72 static Value *SimplifyTruncInst(Value *, Type *, const Query &, unsigned); 73 74 /// For a boolean type, or a vector of boolean type, return false, or 75 /// a vector with every element false, as appropriate for the type. 76 static Constant *getFalse(Type *Ty) { 77 assert(Ty->getScalarType()->isIntegerTy(1) && 78 "Expected i1 type or a vector of i1!"); 79 return Constant::getNullValue(Ty); 80 } 81 82 /// For a boolean type, or a vector of boolean type, return true, or 83 /// a vector with every element true, as appropriate for the type. 84 static Constant *getTrue(Type *Ty) { 85 assert(Ty->getScalarType()->isIntegerTy(1) && 86 "Expected i1 type or a vector of i1!"); 87 return Constant::getAllOnesValue(Ty); 88 } 89 90 /// isSameCompare - Is V equivalent to the comparison "LHS Pred RHS"? 91 static bool isSameCompare(Value *V, CmpInst::Predicate Pred, Value *LHS, 92 Value *RHS) { 93 CmpInst *Cmp = dyn_cast<CmpInst>(V); 94 if (!Cmp) 95 return false; 96 CmpInst::Predicate CPred = Cmp->getPredicate(); 97 Value *CLHS = Cmp->getOperand(0), *CRHS = Cmp->getOperand(1); 98 if (CPred == Pred && CLHS == LHS && CRHS == RHS) 99 return true; 100 return CPred == CmpInst::getSwappedPredicate(Pred) && CLHS == RHS && 101 CRHS == LHS; 102 } 103 104 /// Does the given value dominate the specified phi node? 105 static bool ValueDominatesPHI(Value *V, PHINode *P, const DominatorTree *DT) { 106 Instruction *I = dyn_cast<Instruction>(V); 107 if (!I) 108 // Arguments and constants dominate all instructions. 109 return true; 110 111 // If we are processing instructions (and/or basic blocks) that have not been 112 // fully added to a function, the parent nodes may still be null. Simply 113 // return the conservative answer in these cases. 114 if (!I->getParent() || !P->getParent() || !I->getParent()->getParent()) 115 return false; 116 117 // If we have a DominatorTree then do a precise test. 118 if (DT) { 119 if (!DT->isReachableFromEntry(P->getParent())) 120 return true; 121 if (!DT->isReachableFromEntry(I->getParent())) 122 return false; 123 return DT->dominates(I, P); 124 } 125 126 // Otherwise, if the instruction is in the entry block and is not an invoke, 127 // then it obviously dominates all phi nodes. 128 if (I->getParent() == &I->getParent()->getParent()->getEntryBlock() && 129 !isa<InvokeInst>(I)) 130 return true; 131 132 return false; 133 } 134 135 /// Simplify "A op (B op' C)" by distributing op over op', turning it into 136 /// "(A op B) op' (A op C)". Here "op" is given by Opcode and "op'" is 137 /// given by OpcodeToExpand, while "A" corresponds to LHS and "B op' C" to RHS. 138 /// Also performs the transform "(A op' B) op C" -> "(A op C) op' (B op C)". 139 /// Returns the simplified value, or null if no simplification was performed. 140 static Value *ExpandBinOp(unsigned Opcode, Value *LHS, Value *RHS, 141 unsigned OpcToExpand, const Query &Q, 142 unsigned MaxRecurse) { 143 Instruction::BinaryOps OpcodeToExpand = (Instruction::BinaryOps)OpcToExpand; 144 // Recursion is always used, so bail out at once if we already hit the limit. 145 if (!MaxRecurse--) 146 return nullptr; 147 148 // Check whether the expression has the form "(A op' B) op C". 149 if (BinaryOperator *Op0 = dyn_cast<BinaryOperator>(LHS)) 150 if (Op0->getOpcode() == OpcodeToExpand) { 151 // It does! Try turning it into "(A op C) op' (B op C)". 152 Value *A = Op0->getOperand(0), *B = Op0->getOperand(1), *C = RHS; 153 // Do "A op C" and "B op C" both simplify? 154 if (Value *L = SimplifyBinOp(Opcode, A, C, Q, MaxRecurse)) 155 if (Value *R = SimplifyBinOp(Opcode, B, C, Q, MaxRecurse)) { 156 // They do! Return "L op' R" if it simplifies or is already available. 157 // If "L op' R" equals "A op' B" then "L op' R" is just the LHS. 158 if ((L == A && R == B) || (Instruction::isCommutative(OpcodeToExpand) 159 && L == B && R == A)) { 160 ++NumExpand; 161 return LHS; 162 } 163 // Otherwise return "L op' R" if it simplifies. 164 if (Value *V = SimplifyBinOp(OpcodeToExpand, L, R, Q, MaxRecurse)) { 165 ++NumExpand; 166 return V; 167 } 168 } 169 } 170 171 // Check whether the expression has the form "A op (B op' C)". 172 if (BinaryOperator *Op1 = dyn_cast<BinaryOperator>(RHS)) 173 if (Op1->getOpcode() == OpcodeToExpand) { 174 // It does! Try turning it into "(A op B) op' (A op C)". 175 Value *A = LHS, *B = Op1->getOperand(0), *C = Op1->getOperand(1); 176 // Do "A op B" and "A op C" both simplify? 177 if (Value *L = SimplifyBinOp(Opcode, A, B, Q, MaxRecurse)) 178 if (Value *R = SimplifyBinOp(Opcode, A, C, Q, MaxRecurse)) { 179 // They do! Return "L op' R" if it simplifies or is already available. 180 // If "L op' R" equals "B op' C" then "L op' R" is just the RHS. 181 if ((L == B && R == C) || (Instruction::isCommutative(OpcodeToExpand) 182 && L == C && R == B)) { 183 ++NumExpand; 184 return RHS; 185 } 186 // Otherwise return "L op' R" if it simplifies. 187 if (Value *V = SimplifyBinOp(OpcodeToExpand, L, R, Q, MaxRecurse)) { 188 ++NumExpand; 189 return V; 190 } 191 } 192 } 193 194 return nullptr; 195 } 196 197 /// Generic simplifications for associative binary operations. 198 /// Returns the simpler value, or null if none was found. 199 static Value *SimplifyAssociativeBinOp(unsigned Opc, Value *LHS, Value *RHS, 200 const Query &Q, unsigned MaxRecurse) { 201 Instruction::BinaryOps Opcode = (Instruction::BinaryOps)Opc; 202 assert(Instruction::isAssociative(Opcode) && "Not an associative operation!"); 203 204 // Recursion is always used, so bail out at once if we already hit the limit. 205 if (!MaxRecurse--) 206 return nullptr; 207 208 BinaryOperator *Op0 = dyn_cast<BinaryOperator>(LHS); 209 BinaryOperator *Op1 = dyn_cast<BinaryOperator>(RHS); 210 211 // Transform: "(A op B) op C" ==> "A op (B op C)" if it simplifies completely. 212 if (Op0 && Op0->getOpcode() == Opcode) { 213 Value *A = Op0->getOperand(0); 214 Value *B = Op0->getOperand(1); 215 Value *C = RHS; 216 217 // Does "B op C" simplify? 218 if (Value *V = SimplifyBinOp(Opcode, B, C, Q, MaxRecurse)) { 219 // It does! Return "A op V" if it simplifies or is already available. 220 // If V equals B then "A op V" is just the LHS. 221 if (V == B) return LHS; 222 // Otherwise return "A op V" if it simplifies. 223 if (Value *W = SimplifyBinOp(Opcode, A, V, Q, MaxRecurse)) { 224 ++NumReassoc; 225 return W; 226 } 227 } 228 } 229 230 // Transform: "A op (B op C)" ==> "(A op B) op C" if it simplifies completely. 231 if (Op1 && Op1->getOpcode() == Opcode) { 232 Value *A = LHS; 233 Value *B = Op1->getOperand(0); 234 Value *C = Op1->getOperand(1); 235 236 // Does "A op B" simplify? 237 if (Value *V = SimplifyBinOp(Opcode, A, B, Q, MaxRecurse)) { 238 // It does! Return "V op C" if it simplifies or is already available. 239 // If V equals B then "V op C" is just the RHS. 240 if (V == B) return RHS; 241 // Otherwise return "V op C" if it simplifies. 242 if (Value *W = SimplifyBinOp(Opcode, V, C, Q, MaxRecurse)) { 243 ++NumReassoc; 244 return W; 245 } 246 } 247 } 248 249 // The remaining transforms require commutativity as well as associativity. 250 if (!Instruction::isCommutative(Opcode)) 251 return nullptr; 252 253 // Transform: "(A op B) op C" ==> "(C op A) op B" if it simplifies completely. 254 if (Op0 && Op0->getOpcode() == Opcode) { 255 Value *A = Op0->getOperand(0); 256 Value *B = Op0->getOperand(1); 257 Value *C = RHS; 258 259 // Does "C op A" simplify? 260 if (Value *V = SimplifyBinOp(Opcode, C, A, Q, MaxRecurse)) { 261 // It does! Return "V op B" if it simplifies or is already available. 262 // If V equals A then "V op B" is just the LHS. 263 if (V == A) return LHS; 264 // Otherwise return "V op B" if it simplifies. 265 if (Value *W = SimplifyBinOp(Opcode, V, B, Q, MaxRecurse)) { 266 ++NumReassoc; 267 return W; 268 } 269 } 270 } 271 272 // Transform: "A op (B op C)" ==> "B op (C op A)" if it simplifies completely. 273 if (Op1 && Op1->getOpcode() == Opcode) { 274 Value *A = LHS; 275 Value *B = Op1->getOperand(0); 276 Value *C = Op1->getOperand(1); 277 278 // Does "C op A" simplify? 279 if (Value *V = SimplifyBinOp(Opcode, C, A, Q, MaxRecurse)) { 280 // It does! Return "B op V" if it simplifies or is already available. 281 // If V equals C then "B op V" is just the RHS. 282 if (V == C) return RHS; 283 // Otherwise return "B op V" if it simplifies. 284 if (Value *W = SimplifyBinOp(Opcode, B, V, Q, MaxRecurse)) { 285 ++NumReassoc; 286 return W; 287 } 288 } 289 } 290 291 return nullptr; 292 } 293 294 /// In the case of a binary operation with a select instruction as an operand, 295 /// try to simplify the binop by seeing whether evaluating it on both branches 296 /// of the select results in the same value. Returns the common value if so, 297 /// otherwise returns null. 298 static Value *ThreadBinOpOverSelect(unsigned Opcode, Value *LHS, Value *RHS, 299 const Query &Q, unsigned MaxRecurse) { 300 // Recursion is always used, so bail out at once if we already hit the limit. 301 if (!MaxRecurse--) 302 return nullptr; 303 304 SelectInst *SI; 305 if (isa<SelectInst>(LHS)) { 306 SI = cast<SelectInst>(LHS); 307 } else { 308 assert(isa<SelectInst>(RHS) && "No select instruction operand!"); 309 SI = cast<SelectInst>(RHS); 310 } 311 312 // Evaluate the BinOp on the true and false branches of the select. 313 Value *TV; 314 Value *FV; 315 if (SI == LHS) { 316 TV = SimplifyBinOp(Opcode, SI->getTrueValue(), RHS, Q, MaxRecurse); 317 FV = SimplifyBinOp(Opcode, SI->getFalseValue(), RHS, Q, MaxRecurse); 318 } else { 319 TV = SimplifyBinOp(Opcode, LHS, SI->getTrueValue(), Q, MaxRecurse); 320 FV = SimplifyBinOp(Opcode, LHS, SI->getFalseValue(), Q, MaxRecurse); 321 } 322 323 // If they simplified to the same value, then return the common value. 324 // If they both failed to simplify then return null. 325 if (TV == FV) 326 return TV; 327 328 // If one branch simplified to undef, return the other one. 329 if (TV && isa<UndefValue>(TV)) 330 return FV; 331 if (FV && isa<UndefValue>(FV)) 332 return TV; 333 334 // If applying the operation did not change the true and false select values, 335 // then the result of the binop is the select itself. 336 if (TV == SI->getTrueValue() && FV == SI->getFalseValue()) 337 return SI; 338 339 // If one branch simplified and the other did not, and the simplified 340 // value is equal to the unsimplified one, return the simplified value. 341 // For example, select (cond, X, X & Z) & Z -> X & Z. 342 if ((FV && !TV) || (TV && !FV)) { 343 // Check that the simplified value has the form "X op Y" where "op" is the 344 // same as the original operation. 345 Instruction *Simplified = dyn_cast<Instruction>(FV ? FV : TV); 346 if (Simplified && Simplified->getOpcode() == Opcode) { 347 // The value that didn't simplify is "UnsimplifiedLHS op UnsimplifiedRHS". 348 // We already know that "op" is the same as for the simplified value. See 349 // if the operands match too. If so, return the simplified value. 350 Value *UnsimplifiedBranch = FV ? SI->getTrueValue() : SI->getFalseValue(); 351 Value *UnsimplifiedLHS = SI == LHS ? UnsimplifiedBranch : LHS; 352 Value *UnsimplifiedRHS = SI == LHS ? RHS : UnsimplifiedBranch; 353 if (Simplified->getOperand(0) == UnsimplifiedLHS && 354 Simplified->getOperand(1) == UnsimplifiedRHS) 355 return Simplified; 356 if (Simplified->isCommutative() && 357 Simplified->getOperand(1) == UnsimplifiedLHS && 358 Simplified->getOperand(0) == UnsimplifiedRHS) 359 return Simplified; 360 } 361 } 362 363 return nullptr; 364 } 365 366 /// In the case of a comparison with a select instruction, try to simplify the 367 /// comparison by seeing whether both branches of the select result in the same 368 /// value. Returns the common value if so, otherwise returns null. 369 static Value *ThreadCmpOverSelect(CmpInst::Predicate Pred, Value *LHS, 370 Value *RHS, const Query &Q, 371 unsigned MaxRecurse) { 372 // Recursion is always used, so bail out at once if we already hit the limit. 373 if (!MaxRecurse--) 374 return nullptr; 375 376 // Make sure the select is on the LHS. 377 if (!isa<SelectInst>(LHS)) { 378 std::swap(LHS, RHS); 379 Pred = CmpInst::getSwappedPredicate(Pred); 380 } 381 assert(isa<SelectInst>(LHS) && "Not comparing with a select instruction!"); 382 SelectInst *SI = cast<SelectInst>(LHS); 383 Value *Cond = SI->getCondition(); 384 Value *TV = SI->getTrueValue(); 385 Value *FV = SI->getFalseValue(); 386 387 // Now that we have "cmp select(Cond, TV, FV), RHS", analyse it. 388 // Does "cmp TV, RHS" simplify? 389 Value *TCmp = SimplifyCmpInst(Pred, TV, RHS, Q, MaxRecurse); 390 if (TCmp == Cond) { 391 // It not only simplified, it simplified to the select condition. Replace 392 // it with 'true'. 393 TCmp = getTrue(Cond->getType()); 394 } else if (!TCmp) { 395 // It didn't simplify. However if "cmp TV, RHS" is equal to the select 396 // condition then we can replace it with 'true'. Otherwise give up. 397 if (!isSameCompare(Cond, Pred, TV, RHS)) 398 return nullptr; 399 TCmp = getTrue(Cond->getType()); 400 } 401 402 // Does "cmp FV, RHS" simplify? 403 Value *FCmp = SimplifyCmpInst(Pred, FV, RHS, Q, MaxRecurse); 404 if (FCmp == Cond) { 405 // It not only simplified, it simplified to the select condition. Replace 406 // it with 'false'. 407 FCmp = getFalse(Cond->getType()); 408 } else if (!FCmp) { 409 // It didn't simplify. However if "cmp FV, RHS" is equal to the select 410 // condition then we can replace it with 'false'. Otherwise give up. 411 if (!isSameCompare(Cond, Pred, FV, RHS)) 412 return nullptr; 413 FCmp = getFalse(Cond->getType()); 414 } 415 416 // If both sides simplified to the same value, then use it as the result of 417 // the original comparison. 418 if (TCmp == FCmp) 419 return TCmp; 420 421 // The remaining cases only make sense if the select condition has the same 422 // type as the result of the comparison, so bail out if this is not so. 423 if (Cond->getType()->isVectorTy() != RHS->getType()->isVectorTy()) 424 return nullptr; 425 // If the false value simplified to false, then the result of the compare 426 // is equal to "Cond && TCmp". This also catches the case when the false 427 // value simplified to false and the true value to true, returning "Cond". 428 if (match(FCmp, m_Zero())) 429 if (Value *V = SimplifyAndInst(Cond, TCmp, Q, MaxRecurse)) 430 return V; 431 // If the true value simplified to true, then the result of the compare 432 // is equal to "Cond || FCmp". 433 if (match(TCmp, m_One())) 434 if (Value *V = SimplifyOrInst(Cond, FCmp, Q, MaxRecurse)) 435 return V; 436 // Finally, if the false value simplified to true and the true value to 437 // false, then the result of the compare is equal to "!Cond". 438 if (match(FCmp, m_One()) && match(TCmp, m_Zero())) 439 if (Value *V = 440 SimplifyXorInst(Cond, Constant::getAllOnesValue(Cond->getType()), 441 Q, MaxRecurse)) 442 return V; 443 444 return nullptr; 445 } 446 447 /// In the case of a binary operation with an operand that is a PHI instruction, 448 /// try to simplify the binop by seeing whether evaluating it on the incoming 449 /// phi values yields the same result for every value. If so returns the common 450 /// value, otherwise returns null. 451 static Value *ThreadBinOpOverPHI(unsigned Opcode, Value *LHS, Value *RHS, 452 const Query &Q, unsigned MaxRecurse) { 453 // Recursion is always used, so bail out at once if we already hit the limit. 454 if (!MaxRecurse--) 455 return nullptr; 456 457 PHINode *PI; 458 if (isa<PHINode>(LHS)) { 459 PI = cast<PHINode>(LHS); 460 // Bail out if RHS and the phi may be mutually interdependent due to a loop. 461 if (!ValueDominatesPHI(RHS, PI, Q.DT)) 462 return nullptr; 463 } else { 464 assert(isa<PHINode>(RHS) && "No PHI instruction operand!"); 465 PI = cast<PHINode>(RHS); 466 // Bail out if LHS and the phi may be mutually interdependent due to a loop. 467 if (!ValueDominatesPHI(LHS, PI, Q.DT)) 468 return nullptr; 469 } 470 471 // Evaluate the BinOp on the incoming phi values. 472 Value *CommonValue = nullptr; 473 for (Value *Incoming : PI->incoming_values()) { 474 // If the incoming value is the phi node itself, it can safely be skipped. 475 if (Incoming == PI) continue; 476 Value *V = PI == LHS ? 477 SimplifyBinOp(Opcode, Incoming, RHS, Q, MaxRecurse) : 478 SimplifyBinOp(Opcode, LHS, Incoming, Q, MaxRecurse); 479 // If the operation failed to simplify, or simplified to a different value 480 // to previously, then give up. 481 if (!V || (CommonValue && V != CommonValue)) 482 return nullptr; 483 CommonValue = V; 484 } 485 486 return CommonValue; 487 } 488 489 /// In the case of a comparison with a PHI instruction, try to simplify the 490 /// comparison by seeing whether comparing with all of the incoming phi values 491 /// yields the same result every time. If so returns the common result, 492 /// otherwise returns null. 493 static Value *ThreadCmpOverPHI(CmpInst::Predicate Pred, Value *LHS, Value *RHS, 494 const Query &Q, unsigned MaxRecurse) { 495 // Recursion is always used, so bail out at once if we already hit the limit. 496 if (!MaxRecurse--) 497 return nullptr; 498 499 // Make sure the phi is on the LHS. 500 if (!isa<PHINode>(LHS)) { 501 std::swap(LHS, RHS); 502 Pred = CmpInst::getSwappedPredicate(Pred); 503 } 504 assert(isa<PHINode>(LHS) && "Not comparing with a phi instruction!"); 505 PHINode *PI = cast<PHINode>(LHS); 506 507 // Bail out if RHS and the phi may be mutually interdependent due to a loop. 508 if (!ValueDominatesPHI(RHS, PI, Q.DT)) 509 return nullptr; 510 511 // Evaluate the BinOp on the incoming phi values. 512 Value *CommonValue = nullptr; 513 for (Value *Incoming : PI->incoming_values()) { 514 // If the incoming value is the phi node itself, it can safely be skipped. 515 if (Incoming == PI) continue; 516 Value *V = SimplifyCmpInst(Pred, Incoming, RHS, Q, MaxRecurse); 517 // If the operation failed to simplify, or simplified to a different value 518 // to previously, then give up. 519 if (!V || (CommonValue && V != CommonValue)) 520 return nullptr; 521 CommonValue = V; 522 } 523 524 return CommonValue; 525 } 526 527 /// Given operands for an Add, see if we can fold the result. 528 /// If not, this returns null. 529 static Value *SimplifyAddInst(Value *Op0, Value *Op1, bool isNSW, bool isNUW, 530 const Query &Q, unsigned MaxRecurse) { 531 if (Constant *CLHS = dyn_cast<Constant>(Op0)) { 532 if (Constant *CRHS = dyn_cast<Constant>(Op1)) 533 return ConstantFoldBinaryOpOperands(Instruction::Add, CLHS, CRHS, Q.DL); 534 535 // Canonicalize the constant to the RHS. 536 std::swap(Op0, Op1); 537 } 538 539 // X + undef -> undef 540 if (match(Op1, m_Undef())) 541 return Op1; 542 543 // X + 0 -> X 544 if (match(Op1, m_Zero())) 545 return Op0; 546 547 // X + (Y - X) -> Y 548 // (Y - X) + X -> Y 549 // Eg: X + -X -> 0 550 Value *Y = nullptr; 551 if (match(Op1, m_Sub(m_Value(Y), m_Specific(Op0))) || 552 match(Op0, m_Sub(m_Value(Y), m_Specific(Op1)))) 553 return Y; 554 555 // X + ~X -> -1 since ~X = -X-1 556 if (match(Op0, m_Not(m_Specific(Op1))) || 557 match(Op1, m_Not(m_Specific(Op0)))) 558 return Constant::getAllOnesValue(Op0->getType()); 559 560 /// i1 add -> xor. 561 if (MaxRecurse && Op0->getType()->isIntegerTy(1)) 562 if (Value *V = SimplifyXorInst(Op0, Op1, Q, MaxRecurse-1)) 563 return V; 564 565 // Try some generic simplifications for associative operations. 566 if (Value *V = SimplifyAssociativeBinOp(Instruction::Add, Op0, Op1, Q, 567 MaxRecurse)) 568 return V; 569 570 // Threading Add over selects and phi nodes is pointless, so don't bother. 571 // Threading over the select in "A + select(cond, B, C)" means evaluating 572 // "A+B" and "A+C" and seeing if they are equal; but they are equal if and 573 // only if B and C are equal. If B and C are equal then (since we assume 574 // that operands have already been simplified) "select(cond, B, C)" should 575 // have been simplified to the common value of B and C already. Analysing 576 // "A+B" and "A+C" thus gains nothing, but costs compile time. Similarly 577 // for threading over phi nodes. 578 579 return nullptr; 580 } 581 582 Value *llvm::SimplifyAddInst(Value *Op0, Value *Op1, bool isNSW, bool isNUW, 583 const DataLayout &DL, const TargetLibraryInfo *TLI, 584 const DominatorTree *DT, AssumptionCache *AC, 585 const Instruction *CxtI) { 586 return ::SimplifyAddInst(Op0, Op1, isNSW, isNUW, Query(DL, TLI, DT, AC, CxtI), 587 RecursionLimit); 588 } 589 590 /// \brief Compute the base pointer and cumulative constant offsets for V. 591 /// 592 /// This strips all constant offsets off of V, leaving it the base pointer, and 593 /// accumulates the total constant offset applied in the returned constant. It 594 /// returns 0 if V is not a pointer, and returns the constant '0' if there are 595 /// no constant offsets applied. 596 /// 597 /// This is very similar to GetPointerBaseWithConstantOffset except it doesn't 598 /// follow non-inbounds geps. This allows it to remain usable for icmp ult/etc. 599 /// folding. 600 static Constant *stripAndComputeConstantOffsets(const DataLayout &DL, Value *&V, 601 bool AllowNonInbounds = false) { 602 assert(V->getType()->getScalarType()->isPointerTy()); 603 604 Type *IntPtrTy = DL.getIntPtrType(V->getType())->getScalarType(); 605 APInt Offset = APInt::getNullValue(IntPtrTy->getIntegerBitWidth()); 606 607 // Even though we don't look through PHI nodes, we could be called on an 608 // instruction in an unreachable block, which may be on a cycle. 609 SmallPtrSet<Value *, 4> Visited; 610 Visited.insert(V); 611 do { 612 if (GEPOperator *GEP = dyn_cast<GEPOperator>(V)) { 613 if ((!AllowNonInbounds && !GEP->isInBounds()) || 614 !GEP->accumulateConstantOffset(DL, Offset)) 615 break; 616 V = GEP->getPointerOperand(); 617 } else if (Operator::getOpcode(V) == Instruction::BitCast) { 618 V = cast<Operator>(V)->getOperand(0); 619 } else if (GlobalAlias *GA = dyn_cast<GlobalAlias>(V)) { 620 if (GA->isInterposable()) 621 break; 622 V = GA->getAliasee(); 623 } else { 624 if (auto CS = CallSite(V)) 625 if (Value *RV = CS.getReturnedArgOperand()) { 626 V = RV; 627 continue; 628 } 629 break; 630 } 631 assert(V->getType()->getScalarType()->isPointerTy() && 632 "Unexpected operand type!"); 633 } while (Visited.insert(V).second); 634 635 Constant *OffsetIntPtr = ConstantInt::get(IntPtrTy, Offset); 636 if (V->getType()->isVectorTy()) 637 return ConstantVector::getSplat(V->getType()->getVectorNumElements(), 638 OffsetIntPtr); 639 return OffsetIntPtr; 640 } 641 642 /// \brief Compute the constant difference between two pointer values. 643 /// If the difference is not a constant, returns zero. 644 static Constant *computePointerDifference(const DataLayout &DL, Value *LHS, 645 Value *RHS) { 646 Constant *LHSOffset = stripAndComputeConstantOffsets(DL, LHS); 647 Constant *RHSOffset = stripAndComputeConstantOffsets(DL, RHS); 648 649 // If LHS and RHS are not related via constant offsets to the same base 650 // value, there is nothing we can do here. 651 if (LHS != RHS) 652 return nullptr; 653 654 // Otherwise, the difference of LHS - RHS can be computed as: 655 // LHS - RHS 656 // = (LHSOffset + Base) - (RHSOffset + Base) 657 // = LHSOffset - RHSOffset 658 return ConstantExpr::getSub(LHSOffset, RHSOffset); 659 } 660 661 /// Given operands for a Sub, see if we can fold the result. 662 /// If not, this returns null. 663 static Value *SimplifySubInst(Value *Op0, Value *Op1, bool isNSW, bool isNUW, 664 const Query &Q, unsigned MaxRecurse) { 665 if (Constant *CLHS = dyn_cast<Constant>(Op0)) 666 if (Constant *CRHS = dyn_cast<Constant>(Op1)) 667 return ConstantFoldBinaryOpOperands(Instruction::Sub, CLHS, CRHS, Q.DL); 668 669 // X - undef -> undef 670 // undef - X -> undef 671 if (match(Op0, m_Undef()) || match(Op1, m_Undef())) 672 return UndefValue::get(Op0->getType()); 673 674 // X - 0 -> X 675 if (match(Op1, m_Zero())) 676 return Op0; 677 678 // X - X -> 0 679 if (Op0 == Op1) 680 return Constant::getNullValue(Op0->getType()); 681 682 // 0 - X -> 0 if the sub is NUW. 683 if (isNUW && match(Op0, m_Zero())) 684 return Op0; 685 686 // (X + Y) - Z -> X + (Y - Z) or Y + (X - Z) if everything simplifies. 687 // For example, (X + Y) - Y -> X; (Y + X) - Y -> X 688 Value *X = nullptr, *Y = nullptr, *Z = Op1; 689 if (MaxRecurse && match(Op0, m_Add(m_Value(X), m_Value(Y)))) { // (X + Y) - Z 690 // See if "V === Y - Z" simplifies. 691 if (Value *V = SimplifyBinOp(Instruction::Sub, Y, Z, Q, MaxRecurse-1)) 692 // It does! Now see if "X + V" simplifies. 693 if (Value *W = SimplifyBinOp(Instruction::Add, X, V, Q, MaxRecurse-1)) { 694 // It does, we successfully reassociated! 695 ++NumReassoc; 696 return W; 697 } 698 // See if "V === X - Z" simplifies. 699 if (Value *V = SimplifyBinOp(Instruction::Sub, X, Z, Q, MaxRecurse-1)) 700 // It does! Now see if "Y + V" simplifies. 701 if (Value *W = SimplifyBinOp(Instruction::Add, Y, V, Q, MaxRecurse-1)) { 702 // It does, we successfully reassociated! 703 ++NumReassoc; 704 return W; 705 } 706 } 707 708 // X - (Y + Z) -> (X - Y) - Z or (X - Z) - Y if everything simplifies. 709 // For example, X - (X + 1) -> -1 710 X = Op0; 711 if (MaxRecurse && match(Op1, m_Add(m_Value(Y), m_Value(Z)))) { // X - (Y + Z) 712 // See if "V === X - Y" simplifies. 713 if (Value *V = SimplifyBinOp(Instruction::Sub, X, Y, Q, MaxRecurse-1)) 714 // It does! Now see if "V - Z" simplifies. 715 if (Value *W = SimplifyBinOp(Instruction::Sub, V, Z, Q, MaxRecurse-1)) { 716 // It does, we successfully reassociated! 717 ++NumReassoc; 718 return W; 719 } 720 // See if "V === X - Z" simplifies. 721 if (Value *V = SimplifyBinOp(Instruction::Sub, X, Z, Q, MaxRecurse-1)) 722 // It does! Now see if "V - Y" simplifies. 723 if (Value *W = SimplifyBinOp(Instruction::Sub, V, Y, Q, MaxRecurse-1)) { 724 // It does, we successfully reassociated! 725 ++NumReassoc; 726 return W; 727 } 728 } 729 730 // Z - (X - Y) -> (Z - X) + Y if everything simplifies. 731 // For example, X - (X - Y) -> Y. 732 Z = Op0; 733 if (MaxRecurse && match(Op1, m_Sub(m_Value(X), m_Value(Y)))) // Z - (X - Y) 734 // See if "V === Z - X" simplifies. 735 if (Value *V = SimplifyBinOp(Instruction::Sub, Z, X, Q, MaxRecurse-1)) 736 // It does! Now see if "V + Y" simplifies. 737 if (Value *W = SimplifyBinOp(Instruction::Add, V, Y, Q, MaxRecurse-1)) { 738 // It does, we successfully reassociated! 739 ++NumReassoc; 740 return W; 741 } 742 743 // trunc(X) - trunc(Y) -> trunc(X - Y) if everything simplifies. 744 if (MaxRecurse && match(Op0, m_Trunc(m_Value(X))) && 745 match(Op1, m_Trunc(m_Value(Y)))) 746 if (X->getType() == Y->getType()) 747 // See if "V === X - Y" simplifies. 748 if (Value *V = SimplifyBinOp(Instruction::Sub, X, Y, Q, MaxRecurse-1)) 749 // It does! Now see if "trunc V" simplifies. 750 if (Value *W = SimplifyTruncInst(V, Op0->getType(), Q, MaxRecurse-1)) 751 // It does, return the simplified "trunc V". 752 return W; 753 754 // Variations on GEP(base, I, ...) - GEP(base, i, ...) -> GEP(null, I-i, ...). 755 if (match(Op0, m_PtrToInt(m_Value(X))) && 756 match(Op1, m_PtrToInt(m_Value(Y)))) 757 if (Constant *Result = computePointerDifference(Q.DL, X, Y)) 758 return ConstantExpr::getIntegerCast(Result, Op0->getType(), true); 759 760 // i1 sub -> xor. 761 if (MaxRecurse && Op0->getType()->isIntegerTy(1)) 762 if (Value *V = SimplifyXorInst(Op0, Op1, Q, MaxRecurse-1)) 763 return V; 764 765 // Threading Sub over selects and phi nodes is pointless, so don't bother. 766 // Threading over the select in "A - select(cond, B, C)" means evaluating 767 // "A-B" and "A-C" and seeing if they are equal; but they are equal if and 768 // only if B and C are equal. If B and C are equal then (since we assume 769 // that operands have already been simplified) "select(cond, B, C)" should 770 // have been simplified to the common value of B and C already. Analysing 771 // "A-B" and "A-C" thus gains nothing, but costs compile time. Similarly 772 // for threading over phi nodes. 773 774 return nullptr; 775 } 776 777 Value *llvm::SimplifySubInst(Value *Op0, Value *Op1, bool isNSW, bool isNUW, 778 const DataLayout &DL, const TargetLibraryInfo *TLI, 779 const DominatorTree *DT, AssumptionCache *AC, 780 const Instruction *CxtI) { 781 return ::SimplifySubInst(Op0, Op1, isNSW, isNUW, Query(DL, TLI, DT, AC, CxtI), 782 RecursionLimit); 783 } 784 785 /// Given operands for an FAdd, see if we can fold the result. If not, this 786 /// returns null. 787 static Value *SimplifyFAddInst(Value *Op0, Value *Op1, FastMathFlags FMF, 788 const Query &Q, unsigned MaxRecurse) { 789 if (Constant *CLHS = dyn_cast<Constant>(Op0)) { 790 if (Constant *CRHS = dyn_cast<Constant>(Op1)) 791 return ConstantFoldBinaryOpOperands(Instruction::FAdd, CLHS, CRHS, Q.DL); 792 793 // Canonicalize the constant to the RHS. 794 std::swap(Op0, Op1); 795 } 796 797 // fadd X, -0 ==> X 798 if (match(Op1, m_NegZero())) 799 return Op0; 800 801 // fadd X, 0 ==> X, when we know X is not -0 802 if (match(Op1, m_Zero()) && 803 (FMF.noSignedZeros() || CannotBeNegativeZero(Op0, Q.TLI))) 804 return Op0; 805 806 // fadd [nnan ninf] X, (fsub [nnan ninf] 0, X) ==> 0 807 // where nnan and ninf have to occur at least once somewhere in this 808 // expression 809 Value *SubOp = nullptr; 810 if (match(Op1, m_FSub(m_AnyZero(), m_Specific(Op0)))) 811 SubOp = Op1; 812 else if (match(Op0, m_FSub(m_AnyZero(), m_Specific(Op1)))) 813 SubOp = Op0; 814 if (SubOp) { 815 Instruction *FSub = cast<Instruction>(SubOp); 816 if ((FMF.noNaNs() || FSub->hasNoNaNs()) && 817 (FMF.noInfs() || FSub->hasNoInfs())) 818 return Constant::getNullValue(Op0->getType()); 819 } 820 821 return nullptr; 822 } 823 824 /// Given operands for an FSub, see if we can fold the result. If not, this 825 /// returns null. 826 static Value *SimplifyFSubInst(Value *Op0, Value *Op1, FastMathFlags FMF, 827 const Query &Q, unsigned MaxRecurse) { 828 if (Constant *CLHS = dyn_cast<Constant>(Op0)) { 829 if (Constant *CRHS = dyn_cast<Constant>(Op1)) 830 return ConstantFoldBinaryOpOperands(Instruction::FSub, CLHS, CRHS, Q.DL); 831 } 832 833 // fsub X, 0 ==> X 834 if (match(Op1, m_Zero())) 835 return Op0; 836 837 // fsub X, -0 ==> X, when we know X is not -0 838 if (match(Op1, m_NegZero()) && 839 (FMF.noSignedZeros() || CannotBeNegativeZero(Op0, Q.TLI))) 840 return Op0; 841 842 // fsub -0.0, (fsub -0.0, X) ==> X 843 Value *X; 844 if (match(Op0, m_NegZero()) && match(Op1, m_FSub(m_NegZero(), m_Value(X)))) 845 return X; 846 847 // fsub 0.0, (fsub 0.0, X) ==> X if signed zeros are ignored. 848 if (FMF.noSignedZeros() && match(Op0, m_AnyZero()) && 849 match(Op1, m_FSub(m_AnyZero(), m_Value(X)))) 850 return X; 851 852 // fsub nnan x, x ==> 0.0 853 if (FMF.noNaNs() && Op0 == Op1) 854 return Constant::getNullValue(Op0->getType()); 855 856 return nullptr; 857 } 858 859 /// Given the operands for an FMul, see if we can fold the result 860 static Value *SimplifyFMulInst(Value *Op0, Value *Op1, 861 FastMathFlags FMF, 862 const Query &Q, 863 unsigned MaxRecurse) { 864 if (Constant *CLHS = dyn_cast<Constant>(Op0)) { 865 if (Constant *CRHS = dyn_cast<Constant>(Op1)) 866 return ConstantFoldBinaryOpOperands(Instruction::FMul, CLHS, CRHS, Q.DL); 867 868 // Canonicalize the constant to the RHS. 869 std::swap(Op0, Op1); 870 } 871 872 // fmul X, 1.0 ==> X 873 if (match(Op1, m_FPOne())) 874 return Op0; 875 876 // fmul nnan nsz X, 0 ==> 0 877 if (FMF.noNaNs() && FMF.noSignedZeros() && match(Op1, m_AnyZero())) 878 return Op1; 879 880 return nullptr; 881 } 882 883 /// Given operands for a Mul, see if we can fold the result. 884 /// If not, this returns null. 885 static Value *SimplifyMulInst(Value *Op0, Value *Op1, const Query &Q, 886 unsigned MaxRecurse) { 887 if (Constant *CLHS = dyn_cast<Constant>(Op0)) { 888 if (Constant *CRHS = dyn_cast<Constant>(Op1)) 889 return ConstantFoldBinaryOpOperands(Instruction::Mul, CLHS, CRHS, Q.DL); 890 891 // Canonicalize the constant to the RHS. 892 std::swap(Op0, Op1); 893 } 894 895 // X * undef -> 0 896 if (match(Op1, m_Undef())) 897 return Constant::getNullValue(Op0->getType()); 898 899 // X * 0 -> 0 900 if (match(Op1, m_Zero())) 901 return Op1; 902 903 // X * 1 -> X 904 if (match(Op1, m_One())) 905 return Op0; 906 907 // (X / Y) * Y -> X if the division is exact. 908 Value *X = nullptr; 909 if (match(Op0, m_Exact(m_IDiv(m_Value(X), m_Specific(Op1)))) || // (X / Y) * Y 910 match(Op1, m_Exact(m_IDiv(m_Value(X), m_Specific(Op0))))) // Y * (X / Y) 911 return X; 912 913 // i1 mul -> and. 914 if (MaxRecurse && Op0->getType()->isIntegerTy(1)) 915 if (Value *V = SimplifyAndInst(Op0, Op1, Q, MaxRecurse-1)) 916 return V; 917 918 // Try some generic simplifications for associative operations. 919 if (Value *V = SimplifyAssociativeBinOp(Instruction::Mul, Op0, Op1, Q, 920 MaxRecurse)) 921 return V; 922 923 // Mul distributes over Add. Try some generic simplifications based on this. 924 if (Value *V = ExpandBinOp(Instruction::Mul, Op0, Op1, Instruction::Add, 925 Q, MaxRecurse)) 926 return V; 927 928 // If the operation is with the result of a select instruction, check whether 929 // operating on either branch of the select always yields the same value. 930 if (isa<SelectInst>(Op0) || isa<SelectInst>(Op1)) 931 if (Value *V = ThreadBinOpOverSelect(Instruction::Mul, Op0, Op1, Q, 932 MaxRecurse)) 933 return V; 934 935 // If the operation is with the result of a phi instruction, check whether 936 // operating on all incoming values of the phi always yields the same value. 937 if (isa<PHINode>(Op0) || isa<PHINode>(Op1)) 938 if (Value *V = ThreadBinOpOverPHI(Instruction::Mul, Op0, Op1, Q, 939 MaxRecurse)) 940 return V; 941 942 return nullptr; 943 } 944 945 Value *llvm::SimplifyFAddInst(Value *Op0, Value *Op1, FastMathFlags FMF, 946 const DataLayout &DL, 947 const TargetLibraryInfo *TLI, 948 const DominatorTree *DT, AssumptionCache *AC, 949 const Instruction *CxtI) { 950 return ::SimplifyFAddInst(Op0, Op1, FMF, Query(DL, TLI, DT, AC, CxtI), 951 RecursionLimit); 952 } 953 954 Value *llvm::SimplifyFSubInst(Value *Op0, Value *Op1, FastMathFlags FMF, 955 const DataLayout &DL, 956 const TargetLibraryInfo *TLI, 957 const DominatorTree *DT, AssumptionCache *AC, 958 const Instruction *CxtI) { 959 return ::SimplifyFSubInst(Op0, Op1, FMF, Query(DL, TLI, DT, AC, CxtI), 960 RecursionLimit); 961 } 962 963 Value *llvm::SimplifyFMulInst(Value *Op0, Value *Op1, FastMathFlags FMF, 964 const DataLayout &DL, 965 const TargetLibraryInfo *TLI, 966 const DominatorTree *DT, AssumptionCache *AC, 967 const Instruction *CxtI) { 968 return ::SimplifyFMulInst(Op0, Op1, FMF, Query(DL, TLI, DT, AC, CxtI), 969 RecursionLimit); 970 } 971 972 Value *llvm::SimplifyMulInst(Value *Op0, Value *Op1, const DataLayout &DL, 973 const TargetLibraryInfo *TLI, 974 const DominatorTree *DT, AssumptionCache *AC, 975 const Instruction *CxtI) { 976 return ::SimplifyMulInst(Op0, Op1, Query(DL, TLI, DT, AC, CxtI), 977 RecursionLimit); 978 } 979 980 /// Given operands for an SDiv or UDiv, see if we can fold the result. 981 /// If not, this returns null. 982 static Value *SimplifyDiv(Instruction::BinaryOps Opcode, Value *Op0, Value *Op1, 983 const Query &Q, unsigned MaxRecurse) { 984 if (Constant *C0 = dyn_cast<Constant>(Op0)) 985 if (Constant *C1 = dyn_cast<Constant>(Op1)) 986 return ConstantFoldBinaryOpOperands(Opcode, C0, C1, Q.DL); 987 988 bool isSigned = Opcode == Instruction::SDiv; 989 990 // X / undef -> undef 991 if (match(Op1, m_Undef())) 992 return Op1; 993 994 // X / 0 -> undef, we don't need to preserve faults! 995 if (match(Op1, m_Zero())) 996 return UndefValue::get(Op1->getType()); 997 998 // undef / X -> 0 999 if (match(Op0, m_Undef())) 1000 return Constant::getNullValue(Op0->getType()); 1001 1002 // 0 / X -> 0, we don't need to preserve faults! 1003 if (match(Op0, m_Zero())) 1004 return Op0; 1005 1006 // X / 1 -> X 1007 if (match(Op1, m_One())) 1008 return Op0; 1009 1010 if (Op0->getType()->isIntegerTy(1)) 1011 // It can't be division by zero, hence it must be division by one. 1012 return Op0; 1013 1014 // X / X -> 1 1015 if (Op0 == Op1) 1016 return ConstantInt::get(Op0->getType(), 1); 1017 1018 // (X * Y) / Y -> X if the multiplication does not overflow. 1019 Value *X = nullptr, *Y = nullptr; 1020 if (match(Op0, m_Mul(m_Value(X), m_Value(Y))) && (X == Op1 || Y == Op1)) { 1021 if (Y != Op1) std::swap(X, Y); // Ensure expression is (X * Y) / Y, Y = Op1 1022 OverflowingBinaryOperator *Mul = cast<OverflowingBinaryOperator>(Op0); 1023 // If the Mul knows it does not overflow, then we are good to go. 1024 if ((isSigned && Mul->hasNoSignedWrap()) || 1025 (!isSigned && Mul->hasNoUnsignedWrap())) 1026 return X; 1027 // If X has the form X = A / Y then X * Y cannot overflow. 1028 if (BinaryOperator *Div = dyn_cast<BinaryOperator>(X)) 1029 if (Div->getOpcode() == Opcode && Div->getOperand(1) == Y) 1030 return X; 1031 } 1032 1033 // (X rem Y) / Y -> 0 1034 if ((isSigned && match(Op0, m_SRem(m_Value(), m_Specific(Op1)))) || 1035 (!isSigned && match(Op0, m_URem(m_Value(), m_Specific(Op1))))) 1036 return Constant::getNullValue(Op0->getType()); 1037 1038 // (X /u C1) /u C2 -> 0 if C1 * C2 overflow 1039 ConstantInt *C1, *C2; 1040 if (!isSigned && match(Op0, m_UDiv(m_Value(X), m_ConstantInt(C1))) && 1041 match(Op1, m_ConstantInt(C2))) { 1042 bool Overflow; 1043 C1->getValue().umul_ov(C2->getValue(), Overflow); 1044 if (Overflow) 1045 return Constant::getNullValue(Op0->getType()); 1046 } 1047 1048 // If the operation is with the result of a select instruction, check whether 1049 // operating on either branch of the select always yields the same value. 1050 if (isa<SelectInst>(Op0) || isa<SelectInst>(Op1)) 1051 if (Value *V = ThreadBinOpOverSelect(Opcode, Op0, Op1, Q, MaxRecurse)) 1052 return V; 1053 1054 // If the operation is with the result of a phi instruction, check whether 1055 // operating on all incoming values of the phi always yields the same value. 1056 if (isa<PHINode>(Op0) || isa<PHINode>(Op1)) 1057 if (Value *V = ThreadBinOpOverPHI(Opcode, Op0, Op1, Q, MaxRecurse)) 1058 return V; 1059 1060 return nullptr; 1061 } 1062 1063 /// Given operands for an SDiv, see if we can fold the result. 1064 /// If not, this returns null. 1065 static Value *SimplifySDivInst(Value *Op0, Value *Op1, const Query &Q, 1066 unsigned MaxRecurse) { 1067 if (Value *V = SimplifyDiv(Instruction::SDiv, Op0, Op1, Q, MaxRecurse)) 1068 return V; 1069 1070 return nullptr; 1071 } 1072 1073 Value *llvm::SimplifySDivInst(Value *Op0, Value *Op1, const DataLayout &DL, 1074 const TargetLibraryInfo *TLI, 1075 const DominatorTree *DT, AssumptionCache *AC, 1076 const Instruction *CxtI) { 1077 return ::SimplifySDivInst(Op0, Op1, Query(DL, TLI, DT, AC, CxtI), 1078 RecursionLimit); 1079 } 1080 1081 /// Given operands for a UDiv, see if we can fold the result. 1082 /// If not, this returns null. 1083 static Value *SimplifyUDivInst(Value *Op0, Value *Op1, const Query &Q, 1084 unsigned MaxRecurse) { 1085 if (Value *V = SimplifyDiv(Instruction::UDiv, Op0, Op1, Q, MaxRecurse)) 1086 return V; 1087 1088 return nullptr; 1089 } 1090 1091 Value *llvm::SimplifyUDivInst(Value *Op0, Value *Op1, const DataLayout &DL, 1092 const TargetLibraryInfo *TLI, 1093 const DominatorTree *DT, AssumptionCache *AC, 1094 const Instruction *CxtI) { 1095 return ::SimplifyUDivInst(Op0, Op1, Query(DL, TLI, DT, AC, CxtI), 1096 RecursionLimit); 1097 } 1098 1099 static Value *SimplifyFDivInst(Value *Op0, Value *Op1, FastMathFlags FMF, 1100 const Query &Q, unsigned) { 1101 // undef / X -> undef (the undef could be a snan). 1102 if (match(Op0, m_Undef())) 1103 return Op0; 1104 1105 // X / undef -> undef 1106 if (match(Op1, m_Undef())) 1107 return Op1; 1108 1109 // 0 / X -> 0 1110 // Requires that NaNs are off (X could be zero) and signed zeroes are 1111 // ignored (X could be positive or negative, so the output sign is unknown). 1112 if (FMF.noNaNs() && FMF.noSignedZeros() && match(Op0, m_AnyZero())) 1113 return Op0; 1114 1115 if (FMF.noNaNs()) { 1116 // X / X -> 1.0 is legal when NaNs are ignored. 1117 if (Op0 == Op1) 1118 return ConstantFP::get(Op0->getType(), 1.0); 1119 1120 // -X / X -> -1.0 and 1121 // X / -X -> -1.0 are legal when NaNs are ignored. 1122 // We can ignore signed zeros because +-0.0/+-0.0 is NaN and ignored. 1123 if ((BinaryOperator::isFNeg(Op0, /*IgnoreZeroSign=*/true) && 1124 BinaryOperator::getFNegArgument(Op0) == Op1) || 1125 (BinaryOperator::isFNeg(Op1, /*IgnoreZeroSign=*/true) && 1126 BinaryOperator::getFNegArgument(Op1) == Op0)) 1127 return ConstantFP::get(Op0->getType(), -1.0); 1128 } 1129 1130 return nullptr; 1131 } 1132 1133 Value *llvm::SimplifyFDivInst(Value *Op0, Value *Op1, FastMathFlags FMF, 1134 const DataLayout &DL, 1135 const TargetLibraryInfo *TLI, 1136 const DominatorTree *DT, AssumptionCache *AC, 1137 const Instruction *CxtI) { 1138 return ::SimplifyFDivInst(Op0, Op1, FMF, Query(DL, TLI, DT, AC, CxtI), 1139 RecursionLimit); 1140 } 1141 1142 /// Given operands for an SRem or URem, see if we can fold the result. 1143 /// If not, this returns null. 1144 static Value *SimplifyRem(Instruction::BinaryOps Opcode, Value *Op0, Value *Op1, 1145 const Query &Q, unsigned MaxRecurse) { 1146 if (Constant *C0 = dyn_cast<Constant>(Op0)) 1147 if (Constant *C1 = dyn_cast<Constant>(Op1)) 1148 return ConstantFoldBinaryOpOperands(Opcode, C0, C1, Q.DL); 1149 1150 // X % undef -> undef 1151 if (match(Op1, m_Undef())) 1152 return Op1; 1153 1154 // undef % X -> 0 1155 if (match(Op0, m_Undef())) 1156 return Constant::getNullValue(Op0->getType()); 1157 1158 // 0 % X -> 0, we don't need to preserve faults! 1159 if (match(Op0, m_Zero())) 1160 return Op0; 1161 1162 // X % 0 -> undef, we don't need to preserve faults! 1163 if (match(Op1, m_Zero())) 1164 return UndefValue::get(Op0->getType()); 1165 1166 // X % 1 -> 0 1167 if (match(Op1, m_One())) 1168 return Constant::getNullValue(Op0->getType()); 1169 1170 if (Op0->getType()->isIntegerTy(1)) 1171 // It can't be remainder by zero, hence it must be remainder by one. 1172 return Constant::getNullValue(Op0->getType()); 1173 1174 // X % X -> 0 1175 if (Op0 == Op1) 1176 return Constant::getNullValue(Op0->getType()); 1177 1178 // (X % Y) % Y -> X % Y 1179 if ((Opcode == Instruction::SRem && 1180 match(Op0, m_SRem(m_Value(), m_Specific(Op1)))) || 1181 (Opcode == Instruction::URem && 1182 match(Op0, m_URem(m_Value(), m_Specific(Op1))))) 1183 return Op0; 1184 1185 // If the operation is with the result of a select instruction, check whether 1186 // operating on either branch of the select always yields the same value. 1187 if (isa<SelectInst>(Op0) || isa<SelectInst>(Op1)) 1188 if (Value *V = ThreadBinOpOverSelect(Opcode, Op0, Op1, Q, MaxRecurse)) 1189 return V; 1190 1191 // If the operation is with the result of a phi instruction, check whether 1192 // operating on all incoming values of the phi always yields the same value. 1193 if (isa<PHINode>(Op0) || isa<PHINode>(Op1)) 1194 if (Value *V = ThreadBinOpOverPHI(Opcode, Op0, Op1, Q, MaxRecurse)) 1195 return V; 1196 1197 return nullptr; 1198 } 1199 1200 /// Given operands for an SRem, see if we can fold the result. 1201 /// If not, this returns null. 1202 static Value *SimplifySRemInst(Value *Op0, Value *Op1, const Query &Q, 1203 unsigned MaxRecurse) { 1204 if (Value *V = SimplifyRem(Instruction::SRem, Op0, Op1, Q, MaxRecurse)) 1205 return V; 1206 1207 return nullptr; 1208 } 1209 1210 Value *llvm::SimplifySRemInst(Value *Op0, Value *Op1, const DataLayout &DL, 1211 const TargetLibraryInfo *TLI, 1212 const DominatorTree *DT, AssumptionCache *AC, 1213 const Instruction *CxtI) { 1214 return ::SimplifySRemInst(Op0, Op1, Query(DL, TLI, DT, AC, CxtI), 1215 RecursionLimit); 1216 } 1217 1218 /// Given operands for a URem, see if we can fold the result. 1219 /// If not, this returns null. 1220 static Value *SimplifyURemInst(Value *Op0, Value *Op1, const Query &Q, 1221 unsigned MaxRecurse) { 1222 if (Value *V = SimplifyRem(Instruction::URem, Op0, Op1, Q, MaxRecurse)) 1223 return V; 1224 1225 return nullptr; 1226 } 1227 1228 Value *llvm::SimplifyURemInst(Value *Op0, Value *Op1, const DataLayout &DL, 1229 const TargetLibraryInfo *TLI, 1230 const DominatorTree *DT, AssumptionCache *AC, 1231 const Instruction *CxtI) { 1232 return ::SimplifyURemInst(Op0, Op1, Query(DL, TLI, DT, AC, CxtI), 1233 RecursionLimit); 1234 } 1235 1236 static Value *SimplifyFRemInst(Value *Op0, Value *Op1, FastMathFlags FMF, 1237 const Query &, unsigned) { 1238 // undef % X -> undef (the undef could be a snan). 1239 if (match(Op0, m_Undef())) 1240 return Op0; 1241 1242 // X % undef -> undef 1243 if (match(Op1, m_Undef())) 1244 return Op1; 1245 1246 // 0 % X -> 0 1247 // Requires that NaNs are off (X could be zero) and signed zeroes are 1248 // ignored (X could be positive or negative, so the output sign is unknown). 1249 if (FMF.noNaNs() && FMF.noSignedZeros() && match(Op0, m_AnyZero())) 1250 return Op0; 1251 1252 return nullptr; 1253 } 1254 1255 Value *llvm::SimplifyFRemInst(Value *Op0, Value *Op1, FastMathFlags FMF, 1256 const DataLayout &DL, 1257 const TargetLibraryInfo *TLI, 1258 const DominatorTree *DT, AssumptionCache *AC, 1259 const Instruction *CxtI) { 1260 return ::SimplifyFRemInst(Op0, Op1, FMF, Query(DL, TLI, DT, AC, CxtI), 1261 RecursionLimit); 1262 } 1263 1264 /// Returns true if a shift by \c Amount always yields undef. 1265 static bool isUndefShift(Value *Amount) { 1266 Constant *C = dyn_cast<Constant>(Amount); 1267 if (!C) 1268 return false; 1269 1270 // X shift by undef -> undef because it may shift by the bitwidth. 1271 if (isa<UndefValue>(C)) 1272 return true; 1273 1274 // Shifting by the bitwidth or more is undefined. 1275 if (ConstantInt *CI = dyn_cast<ConstantInt>(C)) 1276 if (CI->getValue().getLimitedValue() >= 1277 CI->getType()->getScalarSizeInBits()) 1278 return true; 1279 1280 // If all lanes of a vector shift are undefined the whole shift is. 1281 if (isa<ConstantVector>(C) || isa<ConstantDataVector>(C)) { 1282 for (unsigned I = 0, E = C->getType()->getVectorNumElements(); I != E; ++I) 1283 if (!isUndefShift(C->getAggregateElement(I))) 1284 return false; 1285 return true; 1286 } 1287 1288 return false; 1289 } 1290 1291 /// Given operands for an Shl, LShr or AShr, see if we can fold the result. 1292 /// If not, this returns null. 1293 static Value *SimplifyShift(unsigned Opcode, Value *Op0, Value *Op1, 1294 const Query &Q, unsigned MaxRecurse) { 1295 if (Constant *C0 = dyn_cast<Constant>(Op0)) 1296 if (Constant *C1 = dyn_cast<Constant>(Op1)) 1297 return ConstantFoldBinaryOpOperands(Opcode, C0, C1, Q.DL); 1298 1299 // 0 shift by X -> 0 1300 if (match(Op0, m_Zero())) 1301 return Op0; 1302 1303 // X shift by 0 -> X 1304 if (match(Op1, m_Zero())) 1305 return Op0; 1306 1307 // Fold undefined shifts. 1308 if (isUndefShift(Op1)) 1309 return UndefValue::get(Op0->getType()); 1310 1311 // If the operation is with the result of a select instruction, check whether 1312 // operating on either branch of the select always yields the same value. 1313 if (isa<SelectInst>(Op0) || isa<SelectInst>(Op1)) 1314 if (Value *V = ThreadBinOpOverSelect(Opcode, Op0, Op1, Q, MaxRecurse)) 1315 return V; 1316 1317 // If the operation is with the result of a phi instruction, check whether 1318 // operating on all incoming values of the phi always yields the same value. 1319 if (isa<PHINode>(Op0) || isa<PHINode>(Op1)) 1320 if (Value *V = ThreadBinOpOverPHI(Opcode, Op0, Op1, Q, MaxRecurse)) 1321 return V; 1322 1323 // If any bits in the shift amount make that value greater than or equal to 1324 // the number of bits in the type, the shift is undefined. 1325 unsigned BitWidth = Op1->getType()->getScalarSizeInBits(); 1326 APInt KnownZero(BitWidth, 0); 1327 APInt KnownOne(BitWidth, 0); 1328 computeKnownBits(Op1, KnownZero, KnownOne, Q.DL, 0, Q.AC, Q.CxtI, Q.DT); 1329 if (KnownOne.getLimitedValue() >= BitWidth) 1330 return UndefValue::get(Op0->getType()); 1331 1332 // If all valid bits in the shift amount are known zero, the first operand is 1333 // unchanged. 1334 unsigned NumValidShiftBits = Log2_32_Ceil(BitWidth); 1335 APInt ShiftAmountMask = APInt::getLowBitsSet(BitWidth, NumValidShiftBits); 1336 if ((KnownZero & ShiftAmountMask) == ShiftAmountMask) 1337 return Op0; 1338 1339 return nullptr; 1340 } 1341 1342 /// \brief Given operands for an Shl, LShr or AShr, see if we can 1343 /// fold the result. If not, this returns null. 1344 static Value *SimplifyRightShift(unsigned Opcode, Value *Op0, Value *Op1, 1345 bool isExact, const Query &Q, 1346 unsigned MaxRecurse) { 1347 if (Value *V = SimplifyShift(Opcode, Op0, Op1, Q, MaxRecurse)) 1348 return V; 1349 1350 // X >> X -> 0 1351 if (Op0 == Op1) 1352 return Constant::getNullValue(Op0->getType()); 1353 1354 // undef >> X -> 0 1355 // undef >> X -> undef (if it's exact) 1356 if (match(Op0, m_Undef())) 1357 return isExact ? Op0 : Constant::getNullValue(Op0->getType()); 1358 1359 // The low bit cannot be shifted out of an exact shift if it is set. 1360 if (isExact) { 1361 unsigned BitWidth = Op0->getType()->getScalarSizeInBits(); 1362 APInt Op0KnownZero(BitWidth, 0); 1363 APInt Op0KnownOne(BitWidth, 0); 1364 computeKnownBits(Op0, Op0KnownZero, Op0KnownOne, Q.DL, /*Depth=*/0, Q.AC, 1365 Q.CxtI, Q.DT); 1366 if (Op0KnownOne[0]) 1367 return Op0; 1368 } 1369 1370 return nullptr; 1371 } 1372 1373 /// Given operands for an Shl, see if we can fold the result. 1374 /// If not, this returns null. 1375 static Value *SimplifyShlInst(Value *Op0, Value *Op1, bool isNSW, bool isNUW, 1376 const Query &Q, unsigned MaxRecurse) { 1377 if (Value *V = SimplifyShift(Instruction::Shl, Op0, Op1, Q, MaxRecurse)) 1378 return V; 1379 1380 // undef << X -> 0 1381 // undef << X -> undef if (if it's NSW/NUW) 1382 if (match(Op0, m_Undef())) 1383 return isNSW || isNUW ? Op0 : Constant::getNullValue(Op0->getType()); 1384 1385 // (X >> A) << A -> X 1386 Value *X; 1387 if (match(Op0, m_Exact(m_Shr(m_Value(X), m_Specific(Op1))))) 1388 return X; 1389 return nullptr; 1390 } 1391 1392 Value *llvm::SimplifyShlInst(Value *Op0, Value *Op1, bool isNSW, bool isNUW, 1393 const DataLayout &DL, const TargetLibraryInfo *TLI, 1394 const DominatorTree *DT, AssumptionCache *AC, 1395 const Instruction *CxtI) { 1396 return ::SimplifyShlInst(Op0, Op1, isNSW, isNUW, Query(DL, TLI, DT, AC, CxtI), 1397 RecursionLimit); 1398 } 1399 1400 /// Given operands for an LShr, see if we can fold the result. 1401 /// If not, this returns null. 1402 static Value *SimplifyLShrInst(Value *Op0, Value *Op1, bool isExact, 1403 const Query &Q, unsigned MaxRecurse) { 1404 if (Value *V = SimplifyRightShift(Instruction::LShr, Op0, Op1, isExact, Q, 1405 MaxRecurse)) 1406 return V; 1407 1408 // (X << A) >> A -> X 1409 Value *X; 1410 if (match(Op0, m_NUWShl(m_Value(X), m_Specific(Op1)))) 1411 return X; 1412 1413 return nullptr; 1414 } 1415 1416 Value *llvm::SimplifyLShrInst(Value *Op0, Value *Op1, bool isExact, 1417 const DataLayout &DL, 1418 const TargetLibraryInfo *TLI, 1419 const DominatorTree *DT, AssumptionCache *AC, 1420 const Instruction *CxtI) { 1421 return ::SimplifyLShrInst(Op0, Op1, isExact, Query(DL, TLI, DT, AC, CxtI), 1422 RecursionLimit); 1423 } 1424 1425 /// Given operands for an AShr, see if we can fold the result. 1426 /// If not, this returns null. 1427 static Value *SimplifyAShrInst(Value *Op0, Value *Op1, bool isExact, 1428 const Query &Q, unsigned MaxRecurse) { 1429 if (Value *V = SimplifyRightShift(Instruction::AShr, Op0, Op1, isExact, Q, 1430 MaxRecurse)) 1431 return V; 1432 1433 // all ones >>a X -> all ones 1434 if (match(Op0, m_AllOnes())) 1435 return Op0; 1436 1437 // (X << A) >> A -> X 1438 Value *X; 1439 if (match(Op0, m_NSWShl(m_Value(X), m_Specific(Op1)))) 1440 return X; 1441 1442 // Arithmetic shifting an all-sign-bit value is a no-op. 1443 unsigned NumSignBits = ComputeNumSignBits(Op0, Q.DL, 0, Q.AC, Q.CxtI, Q.DT); 1444 if (NumSignBits == Op0->getType()->getScalarSizeInBits()) 1445 return Op0; 1446 1447 return nullptr; 1448 } 1449 1450 Value *llvm::SimplifyAShrInst(Value *Op0, Value *Op1, bool isExact, 1451 const DataLayout &DL, 1452 const TargetLibraryInfo *TLI, 1453 const DominatorTree *DT, AssumptionCache *AC, 1454 const Instruction *CxtI) { 1455 return ::SimplifyAShrInst(Op0, Op1, isExact, Query(DL, TLI, DT, AC, CxtI), 1456 RecursionLimit); 1457 } 1458 1459 static Value *simplifyUnsignedRangeCheck(ICmpInst *ZeroICmp, 1460 ICmpInst *UnsignedICmp, bool IsAnd) { 1461 Value *X, *Y; 1462 1463 ICmpInst::Predicate EqPred; 1464 if (!match(ZeroICmp, m_ICmp(EqPred, m_Value(Y), m_Zero())) || 1465 !ICmpInst::isEquality(EqPred)) 1466 return nullptr; 1467 1468 ICmpInst::Predicate UnsignedPred; 1469 if (match(UnsignedICmp, m_ICmp(UnsignedPred, m_Value(X), m_Specific(Y))) && 1470 ICmpInst::isUnsigned(UnsignedPred)) 1471 ; 1472 else if (match(UnsignedICmp, 1473 m_ICmp(UnsignedPred, m_Value(Y), m_Specific(X))) && 1474 ICmpInst::isUnsigned(UnsignedPred)) 1475 UnsignedPred = ICmpInst::getSwappedPredicate(UnsignedPred); 1476 else 1477 return nullptr; 1478 1479 // X < Y && Y != 0 --> X < Y 1480 // X < Y || Y != 0 --> Y != 0 1481 if (UnsignedPred == ICmpInst::ICMP_ULT && EqPred == ICmpInst::ICMP_NE) 1482 return IsAnd ? UnsignedICmp : ZeroICmp; 1483 1484 // X >= Y || Y != 0 --> true 1485 // X >= Y || Y == 0 --> X >= Y 1486 if (UnsignedPred == ICmpInst::ICMP_UGE && !IsAnd) { 1487 if (EqPred == ICmpInst::ICMP_NE) 1488 return getTrue(UnsignedICmp->getType()); 1489 return UnsignedICmp; 1490 } 1491 1492 // X < Y && Y == 0 --> false 1493 if (UnsignedPred == ICmpInst::ICMP_ULT && EqPred == ICmpInst::ICMP_EQ && 1494 IsAnd) 1495 return getFalse(UnsignedICmp->getType()); 1496 1497 return nullptr; 1498 } 1499 1500 static Value *SimplifyAndOfICmps(ICmpInst *Op0, ICmpInst *Op1) { 1501 Type *ITy = Op0->getType(); 1502 ICmpInst::Predicate Pred0, Pred1; 1503 ConstantInt *CI1, *CI2; 1504 Value *V; 1505 1506 if (Value *X = simplifyUnsignedRangeCheck(Op0, Op1, /*IsAnd=*/true)) 1507 return X; 1508 1509 // Look for this pattern: (icmp V, C0) & (icmp V, C1)). 1510 const APInt *C0, *C1; 1511 if (match(Op0, m_ICmp(Pred0, m_Value(V), m_APInt(C0))) && 1512 match(Op1, m_ICmp(Pred1, m_Specific(V), m_APInt(C1)))) { 1513 // Make a constant range that's the intersection of the two icmp ranges. 1514 // If the intersection is empty, we know that the result is false. 1515 auto Range0 = ConstantRange::makeAllowedICmpRegion(Pred0, *C0); 1516 auto Range1 = ConstantRange::makeAllowedICmpRegion(Pred1, *C1); 1517 if (Range0.intersectWith(Range1).isEmptySet()) 1518 return getFalse(ITy); 1519 } 1520 1521 if (!match(Op0, m_ICmp(Pred0, m_Add(m_Value(V), m_ConstantInt(CI1)), 1522 m_ConstantInt(CI2)))) 1523 return nullptr; 1524 1525 if (!match(Op1, m_ICmp(Pred1, m_Specific(V), m_Specific(CI1)))) 1526 return nullptr; 1527 1528 auto *AddInst = cast<BinaryOperator>(Op0->getOperand(0)); 1529 bool isNSW = AddInst->hasNoSignedWrap(); 1530 bool isNUW = AddInst->hasNoUnsignedWrap(); 1531 1532 const APInt &CI1V = CI1->getValue(); 1533 const APInt &CI2V = CI2->getValue(); 1534 const APInt Delta = CI2V - CI1V; 1535 if (CI1V.isStrictlyPositive()) { 1536 if (Delta == 2) { 1537 if (Pred0 == ICmpInst::ICMP_ULT && Pred1 == ICmpInst::ICMP_SGT) 1538 return getFalse(ITy); 1539 if (Pred0 == ICmpInst::ICMP_SLT && Pred1 == ICmpInst::ICMP_SGT && isNSW) 1540 return getFalse(ITy); 1541 } 1542 if (Delta == 1) { 1543 if (Pred0 == ICmpInst::ICMP_ULE && Pred1 == ICmpInst::ICMP_SGT) 1544 return getFalse(ITy); 1545 if (Pred0 == ICmpInst::ICMP_SLE && Pred1 == ICmpInst::ICMP_SGT && isNSW) 1546 return getFalse(ITy); 1547 } 1548 } 1549 if (CI1V.getBoolValue() && isNUW) { 1550 if (Delta == 2) 1551 if (Pred0 == ICmpInst::ICMP_ULT && Pred1 == ICmpInst::ICMP_UGT) 1552 return getFalse(ITy); 1553 if (Delta == 1) 1554 if (Pred0 == ICmpInst::ICMP_ULE && Pred1 == ICmpInst::ICMP_UGT) 1555 return getFalse(ITy); 1556 } 1557 1558 return nullptr; 1559 } 1560 1561 /// Given operands for an And, see if we can fold the result. 1562 /// If not, this returns null. 1563 static Value *SimplifyAndInst(Value *Op0, Value *Op1, const Query &Q, 1564 unsigned MaxRecurse) { 1565 if (Constant *CLHS = dyn_cast<Constant>(Op0)) { 1566 if (Constant *CRHS = dyn_cast<Constant>(Op1)) 1567 return ConstantFoldBinaryOpOperands(Instruction::And, CLHS, CRHS, Q.DL); 1568 1569 // Canonicalize the constant to the RHS. 1570 std::swap(Op0, Op1); 1571 } 1572 1573 // X & undef -> 0 1574 if (match(Op1, m_Undef())) 1575 return Constant::getNullValue(Op0->getType()); 1576 1577 // X & X = X 1578 if (Op0 == Op1) 1579 return Op0; 1580 1581 // X & 0 = 0 1582 if (match(Op1, m_Zero())) 1583 return Op1; 1584 1585 // X & -1 = X 1586 if (match(Op1, m_AllOnes())) 1587 return Op0; 1588 1589 // A & ~A = ~A & A = 0 1590 if (match(Op0, m_Not(m_Specific(Op1))) || 1591 match(Op1, m_Not(m_Specific(Op0)))) 1592 return Constant::getNullValue(Op0->getType()); 1593 1594 // (A | ?) & A = A 1595 Value *A = nullptr, *B = nullptr; 1596 if (match(Op0, m_Or(m_Value(A), m_Value(B))) && 1597 (A == Op1 || B == Op1)) 1598 return Op1; 1599 1600 // A & (A | ?) = A 1601 if (match(Op1, m_Or(m_Value(A), m_Value(B))) && 1602 (A == Op0 || B == Op0)) 1603 return Op0; 1604 1605 // A & (-A) = A if A is a power of two or zero. 1606 if (match(Op0, m_Neg(m_Specific(Op1))) || 1607 match(Op1, m_Neg(m_Specific(Op0)))) { 1608 if (isKnownToBeAPowerOfTwo(Op0, Q.DL, /*OrZero*/ true, 0, Q.AC, Q.CxtI, 1609 Q.DT)) 1610 return Op0; 1611 if (isKnownToBeAPowerOfTwo(Op1, Q.DL, /*OrZero*/ true, 0, Q.AC, Q.CxtI, 1612 Q.DT)) 1613 return Op1; 1614 } 1615 1616 if (auto *ICILHS = dyn_cast<ICmpInst>(Op0)) { 1617 if (auto *ICIRHS = dyn_cast<ICmpInst>(Op1)) { 1618 if (Value *V = SimplifyAndOfICmps(ICILHS, ICIRHS)) 1619 return V; 1620 if (Value *V = SimplifyAndOfICmps(ICIRHS, ICILHS)) 1621 return V; 1622 } 1623 } 1624 1625 // The compares may be hidden behind casts. Look through those and try the 1626 // same folds as above. 1627 auto *Cast0 = dyn_cast<CastInst>(Op0); 1628 auto *Cast1 = dyn_cast<CastInst>(Op1); 1629 if (Cast0 && Cast1 && Cast0->getOpcode() == Cast1->getOpcode() && 1630 Cast0->getSrcTy() == Cast1->getSrcTy()) { 1631 auto *Cmp0 = dyn_cast<ICmpInst>(Cast0->getOperand(0)); 1632 auto *Cmp1 = dyn_cast<ICmpInst>(Cast1->getOperand(0)); 1633 if (Cmp0 && Cmp1) { 1634 Instruction::CastOps CastOpc = Cast0->getOpcode(); 1635 Type *ResultType = Cast0->getType(); 1636 if (auto *V = dyn_cast_or_null<Constant>(SimplifyAndOfICmps(Cmp0, Cmp1))) 1637 return ConstantExpr::getCast(CastOpc, V, ResultType); 1638 if (auto *V = dyn_cast_or_null<Constant>(SimplifyAndOfICmps(Cmp1, Cmp0))) 1639 return ConstantExpr::getCast(CastOpc, V, ResultType); 1640 } 1641 } 1642 1643 // Try some generic simplifications for associative operations. 1644 if (Value *V = SimplifyAssociativeBinOp(Instruction::And, Op0, Op1, Q, 1645 MaxRecurse)) 1646 return V; 1647 1648 // And distributes over Or. Try some generic simplifications based on this. 1649 if (Value *V = ExpandBinOp(Instruction::And, Op0, Op1, Instruction::Or, 1650 Q, MaxRecurse)) 1651 return V; 1652 1653 // And distributes over Xor. Try some generic simplifications based on this. 1654 if (Value *V = ExpandBinOp(Instruction::And, Op0, Op1, Instruction::Xor, 1655 Q, MaxRecurse)) 1656 return V; 1657 1658 // If the operation is with the result of a select instruction, check whether 1659 // operating on either branch of the select always yields the same value. 1660 if (isa<SelectInst>(Op0) || isa<SelectInst>(Op1)) 1661 if (Value *V = ThreadBinOpOverSelect(Instruction::And, Op0, Op1, Q, 1662 MaxRecurse)) 1663 return V; 1664 1665 // If the operation is with the result of a phi instruction, check whether 1666 // operating on all incoming values of the phi always yields the same value. 1667 if (isa<PHINode>(Op0) || isa<PHINode>(Op1)) 1668 if (Value *V = ThreadBinOpOverPHI(Instruction::And, Op0, Op1, Q, 1669 MaxRecurse)) 1670 return V; 1671 1672 return nullptr; 1673 } 1674 1675 Value *llvm::SimplifyAndInst(Value *Op0, Value *Op1, const DataLayout &DL, 1676 const TargetLibraryInfo *TLI, 1677 const DominatorTree *DT, AssumptionCache *AC, 1678 const Instruction *CxtI) { 1679 return ::SimplifyAndInst(Op0, Op1, Query(DL, TLI, DT, AC, CxtI), 1680 RecursionLimit); 1681 } 1682 1683 /// Simplify (or (icmp ...) (icmp ...)) to true when we can tell that the union 1684 /// contains all possible values. 1685 static Value *SimplifyOrOfICmps(ICmpInst *Op0, ICmpInst *Op1) { 1686 ICmpInst::Predicate Pred0, Pred1; 1687 ConstantInt *CI1, *CI2; 1688 Value *V; 1689 1690 if (Value *X = simplifyUnsignedRangeCheck(Op0, Op1, /*IsAnd=*/false)) 1691 return X; 1692 1693 if (!match(Op0, m_ICmp(Pred0, m_Add(m_Value(V), m_ConstantInt(CI1)), 1694 m_ConstantInt(CI2)))) 1695 return nullptr; 1696 1697 if (!match(Op1, m_ICmp(Pred1, m_Specific(V), m_Specific(CI1)))) 1698 return nullptr; 1699 1700 Type *ITy = Op0->getType(); 1701 1702 auto *AddInst = cast<BinaryOperator>(Op0->getOperand(0)); 1703 bool isNSW = AddInst->hasNoSignedWrap(); 1704 bool isNUW = AddInst->hasNoUnsignedWrap(); 1705 1706 const APInt &CI1V = CI1->getValue(); 1707 const APInt &CI2V = CI2->getValue(); 1708 const APInt Delta = CI2V - CI1V; 1709 if (CI1V.isStrictlyPositive()) { 1710 if (Delta == 2) { 1711 if (Pred0 == ICmpInst::ICMP_UGE && Pred1 == ICmpInst::ICMP_SLE) 1712 return getTrue(ITy); 1713 if (Pred0 == ICmpInst::ICMP_SGE && Pred1 == ICmpInst::ICMP_SLE && isNSW) 1714 return getTrue(ITy); 1715 } 1716 if (Delta == 1) { 1717 if (Pred0 == ICmpInst::ICMP_UGT && Pred1 == ICmpInst::ICMP_SLE) 1718 return getTrue(ITy); 1719 if (Pred0 == ICmpInst::ICMP_SGT && Pred1 == ICmpInst::ICMP_SLE && isNSW) 1720 return getTrue(ITy); 1721 } 1722 } 1723 if (CI1V.getBoolValue() && isNUW) { 1724 if (Delta == 2) 1725 if (Pred0 == ICmpInst::ICMP_UGE && Pred1 == ICmpInst::ICMP_ULE) 1726 return getTrue(ITy); 1727 if (Delta == 1) 1728 if (Pred0 == ICmpInst::ICMP_UGT && Pred1 == ICmpInst::ICMP_ULE) 1729 return getTrue(ITy); 1730 } 1731 1732 return nullptr; 1733 } 1734 1735 /// Given operands for an Or, see if we can fold the result. 1736 /// If not, this returns null. 1737 static Value *SimplifyOrInst(Value *Op0, Value *Op1, const Query &Q, 1738 unsigned MaxRecurse) { 1739 if (Constant *CLHS = dyn_cast<Constant>(Op0)) { 1740 if (Constant *CRHS = dyn_cast<Constant>(Op1)) 1741 return ConstantFoldBinaryOpOperands(Instruction::Or, CLHS, CRHS, Q.DL); 1742 1743 // Canonicalize the constant to the RHS. 1744 std::swap(Op0, Op1); 1745 } 1746 1747 // X | undef -> -1 1748 if (match(Op1, m_Undef())) 1749 return Constant::getAllOnesValue(Op0->getType()); 1750 1751 // X | X = X 1752 if (Op0 == Op1) 1753 return Op0; 1754 1755 // X | 0 = X 1756 if (match(Op1, m_Zero())) 1757 return Op0; 1758 1759 // X | -1 = -1 1760 if (match(Op1, m_AllOnes())) 1761 return Op1; 1762 1763 // A | ~A = ~A | A = -1 1764 if (match(Op0, m_Not(m_Specific(Op1))) || 1765 match(Op1, m_Not(m_Specific(Op0)))) 1766 return Constant::getAllOnesValue(Op0->getType()); 1767 1768 // (A & ?) | A = A 1769 Value *A = nullptr, *B = nullptr; 1770 if (match(Op0, m_And(m_Value(A), m_Value(B))) && 1771 (A == Op1 || B == Op1)) 1772 return Op1; 1773 1774 // A | (A & ?) = A 1775 if (match(Op1, m_And(m_Value(A), m_Value(B))) && 1776 (A == Op0 || B == Op0)) 1777 return Op0; 1778 1779 // ~(A & ?) | A = -1 1780 if (match(Op0, m_Not(m_And(m_Value(A), m_Value(B)))) && 1781 (A == Op1 || B == Op1)) 1782 return Constant::getAllOnesValue(Op1->getType()); 1783 1784 // A | ~(A & ?) = -1 1785 if (match(Op1, m_Not(m_And(m_Value(A), m_Value(B)))) && 1786 (A == Op0 || B == Op0)) 1787 return Constant::getAllOnesValue(Op0->getType()); 1788 1789 if (auto *ICILHS = dyn_cast<ICmpInst>(Op0)) { 1790 if (auto *ICIRHS = dyn_cast<ICmpInst>(Op1)) { 1791 if (Value *V = SimplifyOrOfICmps(ICILHS, ICIRHS)) 1792 return V; 1793 if (Value *V = SimplifyOrOfICmps(ICIRHS, ICILHS)) 1794 return V; 1795 } 1796 } 1797 1798 // Try some generic simplifications for associative operations. 1799 if (Value *V = SimplifyAssociativeBinOp(Instruction::Or, Op0, Op1, Q, 1800 MaxRecurse)) 1801 return V; 1802 1803 // Or distributes over And. Try some generic simplifications based on this. 1804 if (Value *V = ExpandBinOp(Instruction::Or, Op0, Op1, Instruction::And, Q, 1805 MaxRecurse)) 1806 return V; 1807 1808 // If the operation is with the result of a select instruction, check whether 1809 // operating on either branch of the select always yields the same value. 1810 if (isa<SelectInst>(Op0) || isa<SelectInst>(Op1)) 1811 if (Value *V = ThreadBinOpOverSelect(Instruction::Or, Op0, Op1, Q, 1812 MaxRecurse)) 1813 return V; 1814 1815 // (A & C)|(B & D) 1816 Value *C = nullptr, *D = nullptr; 1817 if (match(Op0, m_And(m_Value(A), m_Value(C))) && 1818 match(Op1, m_And(m_Value(B), m_Value(D)))) { 1819 ConstantInt *C1 = dyn_cast<ConstantInt>(C); 1820 ConstantInt *C2 = dyn_cast<ConstantInt>(D); 1821 if (C1 && C2 && (C1->getValue() == ~C2->getValue())) { 1822 // (A & C1)|(B & C2) 1823 // If we have: ((V + N) & C1) | (V & C2) 1824 // .. and C2 = ~C1 and C2 is 0+1+ and (N & C2) == 0 1825 // replace with V+N. 1826 Value *V1, *V2; 1827 if ((C2->getValue() & (C2->getValue() + 1)) == 0 && // C2 == 0+1+ 1828 match(A, m_Add(m_Value(V1), m_Value(V2)))) { 1829 // Add commutes, try both ways. 1830 if (V1 == B && 1831 MaskedValueIsZero(V2, C2->getValue(), Q.DL, 0, Q.AC, Q.CxtI, Q.DT)) 1832 return A; 1833 if (V2 == B && 1834 MaskedValueIsZero(V1, C2->getValue(), Q.DL, 0, Q.AC, Q.CxtI, Q.DT)) 1835 return A; 1836 } 1837 // Or commutes, try both ways. 1838 if ((C1->getValue() & (C1->getValue() + 1)) == 0 && 1839 match(B, m_Add(m_Value(V1), m_Value(V2)))) { 1840 // Add commutes, try both ways. 1841 if (V1 == A && 1842 MaskedValueIsZero(V2, C1->getValue(), Q.DL, 0, Q.AC, Q.CxtI, Q.DT)) 1843 return B; 1844 if (V2 == A && 1845 MaskedValueIsZero(V1, C1->getValue(), Q.DL, 0, Q.AC, Q.CxtI, Q.DT)) 1846 return B; 1847 } 1848 } 1849 } 1850 1851 // If the operation is with the result of a phi instruction, check whether 1852 // operating on all incoming values of the phi always yields the same value. 1853 if (isa<PHINode>(Op0) || isa<PHINode>(Op1)) 1854 if (Value *V = ThreadBinOpOverPHI(Instruction::Or, Op0, Op1, Q, MaxRecurse)) 1855 return V; 1856 1857 return nullptr; 1858 } 1859 1860 Value *llvm::SimplifyOrInst(Value *Op0, Value *Op1, const DataLayout &DL, 1861 const TargetLibraryInfo *TLI, 1862 const DominatorTree *DT, AssumptionCache *AC, 1863 const Instruction *CxtI) { 1864 return ::SimplifyOrInst(Op0, Op1, Query(DL, TLI, DT, AC, CxtI), 1865 RecursionLimit); 1866 } 1867 1868 /// Given operands for a Xor, see if we can fold the result. 1869 /// If not, this returns null. 1870 static Value *SimplifyXorInst(Value *Op0, Value *Op1, const Query &Q, 1871 unsigned MaxRecurse) { 1872 if (Constant *CLHS = dyn_cast<Constant>(Op0)) { 1873 if (Constant *CRHS = dyn_cast<Constant>(Op1)) 1874 return ConstantFoldBinaryOpOperands(Instruction::Xor, CLHS, CRHS, Q.DL); 1875 1876 // Canonicalize the constant to the RHS. 1877 std::swap(Op0, Op1); 1878 } 1879 1880 // A ^ undef -> undef 1881 if (match(Op1, m_Undef())) 1882 return Op1; 1883 1884 // A ^ 0 = A 1885 if (match(Op1, m_Zero())) 1886 return Op0; 1887 1888 // A ^ A = 0 1889 if (Op0 == Op1) 1890 return Constant::getNullValue(Op0->getType()); 1891 1892 // A ^ ~A = ~A ^ A = -1 1893 if (match(Op0, m_Not(m_Specific(Op1))) || 1894 match(Op1, m_Not(m_Specific(Op0)))) 1895 return Constant::getAllOnesValue(Op0->getType()); 1896 1897 // Try some generic simplifications for associative operations. 1898 if (Value *V = SimplifyAssociativeBinOp(Instruction::Xor, Op0, Op1, Q, 1899 MaxRecurse)) 1900 return V; 1901 1902 // Threading Xor over selects and phi nodes is pointless, so don't bother. 1903 // Threading over the select in "A ^ select(cond, B, C)" means evaluating 1904 // "A^B" and "A^C" and seeing if they are equal; but they are equal if and 1905 // only if B and C are equal. If B and C are equal then (since we assume 1906 // that operands have already been simplified) "select(cond, B, C)" should 1907 // have been simplified to the common value of B and C already. Analysing 1908 // "A^B" and "A^C" thus gains nothing, but costs compile time. Similarly 1909 // for threading over phi nodes. 1910 1911 return nullptr; 1912 } 1913 1914 Value *llvm::SimplifyXorInst(Value *Op0, Value *Op1, const DataLayout &DL, 1915 const TargetLibraryInfo *TLI, 1916 const DominatorTree *DT, AssumptionCache *AC, 1917 const Instruction *CxtI) { 1918 return ::SimplifyXorInst(Op0, Op1, Query(DL, TLI, DT, AC, CxtI), 1919 RecursionLimit); 1920 } 1921 1922 static Type *GetCompareTy(Value *Op) { 1923 return CmpInst::makeCmpResultType(Op->getType()); 1924 } 1925 1926 /// Rummage around inside V looking for something equivalent to the comparison 1927 /// "LHS Pred RHS". Return such a value if found, otherwise return null. 1928 /// Helper function for analyzing max/min idioms. 1929 static Value *ExtractEquivalentCondition(Value *V, CmpInst::Predicate Pred, 1930 Value *LHS, Value *RHS) { 1931 SelectInst *SI = dyn_cast<SelectInst>(V); 1932 if (!SI) 1933 return nullptr; 1934 CmpInst *Cmp = dyn_cast<CmpInst>(SI->getCondition()); 1935 if (!Cmp) 1936 return nullptr; 1937 Value *CmpLHS = Cmp->getOperand(0), *CmpRHS = Cmp->getOperand(1); 1938 if (Pred == Cmp->getPredicate() && LHS == CmpLHS && RHS == CmpRHS) 1939 return Cmp; 1940 if (Pred == CmpInst::getSwappedPredicate(Cmp->getPredicate()) && 1941 LHS == CmpRHS && RHS == CmpLHS) 1942 return Cmp; 1943 return nullptr; 1944 } 1945 1946 // A significant optimization not implemented here is assuming that alloca 1947 // addresses are not equal to incoming argument values. They don't *alias*, 1948 // as we say, but that doesn't mean they aren't equal, so we take a 1949 // conservative approach. 1950 // 1951 // This is inspired in part by C++11 5.10p1: 1952 // "Two pointers of the same type compare equal if and only if they are both 1953 // null, both point to the same function, or both represent the same 1954 // address." 1955 // 1956 // This is pretty permissive. 1957 // 1958 // It's also partly due to C11 6.5.9p6: 1959 // "Two pointers compare equal if and only if both are null pointers, both are 1960 // pointers to the same object (including a pointer to an object and a 1961 // subobject at its beginning) or function, both are pointers to one past the 1962 // last element of the same array object, or one is a pointer to one past the 1963 // end of one array object and the other is a pointer to the start of a 1964 // different array object that happens to immediately follow the first array 1965 // object in the address space.) 1966 // 1967 // C11's version is more restrictive, however there's no reason why an argument 1968 // couldn't be a one-past-the-end value for a stack object in the caller and be 1969 // equal to the beginning of a stack object in the callee. 1970 // 1971 // If the C and C++ standards are ever made sufficiently restrictive in this 1972 // area, it may be possible to update LLVM's semantics accordingly and reinstate 1973 // this optimization. 1974 static Constant * 1975 computePointerICmp(const DataLayout &DL, const TargetLibraryInfo *TLI, 1976 const DominatorTree *DT, CmpInst::Predicate Pred, 1977 const Instruction *CxtI, Value *LHS, Value *RHS) { 1978 // First, skip past any trivial no-ops. 1979 LHS = LHS->stripPointerCasts(); 1980 RHS = RHS->stripPointerCasts(); 1981 1982 // A non-null pointer is not equal to a null pointer. 1983 if (llvm::isKnownNonNull(LHS) && isa<ConstantPointerNull>(RHS) && 1984 (Pred == CmpInst::ICMP_EQ || Pred == CmpInst::ICMP_NE)) 1985 return ConstantInt::get(GetCompareTy(LHS), 1986 !CmpInst::isTrueWhenEqual(Pred)); 1987 1988 // We can only fold certain predicates on pointer comparisons. 1989 switch (Pred) { 1990 default: 1991 return nullptr; 1992 1993 // Equality comaprisons are easy to fold. 1994 case CmpInst::ICMP_EQ: 1995 case CmpInst::ICMP_NE: 1996 break; 1997 1998 // We can only handle unsigned relational comparisons because 'inbounds' on 1999 // a GEP only protects against unsigned wrapping. 2000 case CmpInst::ICMP_UGT: 2001 case CmpInst::ICMP_UGE: 2002 case CmpInst::ICMP_ULT: 2003 case CmpInst::ICMP_ULE: 2004 // However, we have to switch them to their signed variants to handle 2005 // negative indices from the base pointer. 2006 Pred = ICmpInst::getSignedPredicate(Pred); 2007 break; 2008 } 2009 2010 // Strip off any constant offsets so that we can reason about them. 2011 // It's tempting to use getUnderlyingObject or even just stripInBoundsOffsets 2012 // here and compare base addresses like AliasAnalysis does, however there are 2013 // numerous hazards. AliasAnalysis and its utilities rely on special rules 2014 // governing loads and stores which don't apply to icmps. Also, AliasAnalysis 2015 // doesn't need to guarantee pointer inequality when it says NoAlias. 2016 Constant *LHSOffset = stripAndComputeConstantOffsets(DL, LHS); 2017 Constant *RHSOffset = stripAndComputeConstantOffsets(DL, RHS); 2018 2019 // If LHS and RHS are related via constant offsets to the same base 2020 // value, we can replace it with an icmp which just compares the offsets. 2021 if (LHS == RHS) 2022 return ConstantExpr::getICmp(Pred, LHSOffset, RHSOffset); 2023 2024 // Various optimizations for (in)equality comparisons. 2025 if (Pred == CmpInst::ICMP_EQ || Pred == CmpInst::ICMP_NE) { 2026 // Different non-empty allocations that exist at the same time have 2027 // different addresses (if the program can tell). Global variables always 2028 // exist, so they always exist during the lifetime of each other and all 2029 // allocas. Two different allocas usually have different addresses... 2030 // 2031 // However, if there's an @llvm.stackrestore dynamically in between two 2032 // allocas, they may have the same address. It's tempting to reduce the 2033 // scope of the problem by only looking at *static* allocas here. That would 2034 // cover the majority of allocas while significantly reducing the likelihood 2035 // of having an @llvm.stackrestore pop up in the middle. However, it's not 2036 // actually impossible for an @llvm.stackrestore to pop up in the middle of 2037 // an entry block. Also, if we have a block that's not attached to a 2038 // function, we can't tell if it's "static" under the current definition. 2039 // Theoretically, this problem could be fixed by creating a new kind of 2040 // instruction kind specifically for static allocas. Such a new instruction 2041 // could be required to be at the top of the entry block, thus preventing it 2042 // from being subject to a @llvm.stackrestore. Instcombine could even 2043 // convert regular allocas into these special allocas. It'd be nifty. 2044 // However, until then, this problem remains open. 2045 // 2046 // So, we'll assume that two non-empty allocas have different addresses 2047 // for now. 2048 // 2049 // With all that, if the offsets are within the bounds of their allocations 2050 // (and not one-past-the-end! so we can't use inbounds!), and their 2051 // allocations aren't the same, the pointers are not equal. 2052 // 2053 // Note that it's not necessary to check for LHS being a global variable 2054 // address, due to canonicalization and constant folding. 2055 if (isa<AllocaInst>(LHS) && 2056 (isa<AllocaInst>(RHS) || isa<GlobalVariable>(RHS))) { 2057 ConstantInt *LHSOffsetCI = dyn_cast<ConstantInt>(LHSOffset); 2058 ConstantInt *RHSOffsetCI = dyn_cast<ConstantInt>(RHSOffset); 2059 uint64_t LHSSize, RHSSize; 2060 if (LHSOffsetCI && RHSOffsetCI && 2061 getObjectSize(LHS, LHSSize, DL, TLI) && 2062 getObjectSize(RHS, RHSSize, DL, TLI)) { 2063 const APInt &LHSOffsetValue = LHSOffsetCI->getValue(); 2064 const APInt &RHSOffsetValue = RHSOffsetCI->getValue(); 2065 if (!LHSOffsetValue.isNegative() && 2066 !RHSOffsetValue.isNegative() && 2067 LHSOffsetValue.ult(LHSSize) && 2068 RHSOffsetValue.ult(RHSSize)) { 2069 return ConstantInt::get(GetCompareTy(LHS), 2070 !CmpInst::isTrueWhenEqual(Pred)); 2071 } 2072 } 2073 2074 // Repeat the above check but this time without depending on DataLayout 2075 // or being able to compute a precise size. 2076 if (!cast<PointerType>(LHS->getType())->isEmptyTy() && 2077 !cast<PointerType>(RHS->getType())->isEmptyTy() && 2078 LHSOffset->isNullValue() && 2079 RHSOffset->isNullValue()) 2080 return ConstantInt::get(GetCompareTy(LHS), 2081 !CmpInst::isTrueWhenEqual(Pred)); 2082 } 2083 2084 // Even if an non-inbounds GEP occurs along the path we can still optimize 2085 // equality comparisons concerning the result. We avoid walking the whole 2086 // chain again by starting where the last calls to 2087 // stripAndComputeConstantOffsets left off and accumulate the offsets. 2088 Constant *LHSNoBound = stripAndComputeConstantOffsets(DL, LHS, true); 2089 Constant *RHSNoBound = stripAndComputeConstantOffsets(DL, RHS, true); 2090 if (LHS == RHS) 2091 return ConstantExpr::getICmp(Pred, 2092 ConstantExpr::getAdd(LHSOffset, LHSNoBound), 2093 ConstantExpr::getAdd(RHSOffset, RHSNoBound)); 2094 2095 // If one side of the equality comparison must come from a noalias call 2096 // (meaning a system memory allocation function), and the other side must 2097 // come from a pointer that cannot overlap with dynamically-allocated 2098 // memory within the lifetime of the current function (allocas, byval 2099 // arguments, globals), then determine the comparison result here. 2100 SmallVector<Value *, 8> LHSUObjs, RHSUObjs; 2101 GetUnderlyingObjects(LHS, LHSUObjs, DL); 2102 GetUnderlyingObjects(RHS, RHSUObjs, DL); 2103 2104 // Is the set of underlying objects all noalias calls? 2105 auto IsNAC = [](SmallVectorImpl<Value *> &Objects) { 2106 return std::all_of(Objects.begin(), Objects.end(), isNoAliasCall); 2107 }; 2108 2109 // Is the set of underlying objects all things which must be disjoint from 2110 // noalias calls. For allocas, we consider only static ones (dynamic 2111 // allocas might be transformed into calls to malloc not simultaneously 2112 // live with the compared-to allocation). For globals, we exclude symbols 2113 // that might be resolve lazily to symbols in another dynamically-loaded 2114 // library (and, thus, could be malloc'ed by the implementation). 2115 auto IsAllocDisjoint = [](SmallVectorImpl<Value *> &Objects) { 2116 return std::all_of(Objects.begin(), Objects.end(), [](Value *V) { 2117 if (const AllocaInst *AI = dyn_cast<AllocaInst>(V)) 2118 return AI->getParent() && AI->getFunction() && AI->isStaticAlloca(); 2119 if (const GlobalValue *GV = dyn_cast<GlobalValue>(V)) 2120 return (GV->hasLocalLinkage() || GV->hasHiddenVisibility() || 2121 GV->hasProtectedVisibility() || GV->hasGlobalUnnamedAddr()) && 2122 !GV->isThreadLocal(); 2123 if (const Argument *A = dyn_cast<Argument>(V)) 2124 return A->hasByValAttr(); 2125 return false; 2126 }); 2127 }; 2128 2129 if ((IsNAC(LHSUObjs) && IsAllocDisjoint(RHSUObjs)) || 2130 (IsNAC(RHSUObjs) && IsAllocDisjoint(LHSUObjs))) 2131 return ConstantInt::get(GetCompareTy(LHS), 2132 !CmpInst::isTrueWhenEqual(Pred)); 2133 2134 // Fold comparisons for non-escaping pointer even if the allocation call 2135 // cannot be elided. We cannot fold malloc comparison to null. Also, the 2136 // dynamic allocation call could be either of the operands. 2137 Value *MI = nullptr; 2138 if (isAllocLikeFn(LHS, TLI) && llvm::isKnownNonNullAt(RHS, CxtI, DT)) 2139 MI = LHS; 2140 else if (isAllocLikeFn(RHS, TLI) && llvm::isKnownNonNullAt(LHS, CxtI, DT)) 2141 MI = RHS; 2142 // FIXME: We should also fold the compare when the pointer escapes, but the 2143 // compare dominates the pointer escape 2144 if (MI && !PointerMayBeCaptured(MI, true, true)) 2145 return ConstantInt::get(GetCompareTy(LHS), 2146 CmpInst::isFalseWhenEqual(Pred)); 2147 } 2148 2149 // Otherwise, fail. 2150 return nullptr; 2151 } 2152 2153 /// Given operands for an ICmpInst, see if we can fold the result. 2154 /// If not, this returns null. 2155 static Value *SimplifyICmpInst(unsigned Predicate, Value *LHS, Value *RHS, 2156 const Query &Q, unsigned MaxRecurse) { 2157 CmpInst::Predicate Pred = (CmpInst::Predicate)Predicate; 2158 assert(CmpInst::isIntPredicate(Pred) && "Not an integer compare!"); 2159 2160 if (Constant *CLHS = dyn_cast<Constant>(LHS)) { 2161 if (Constant *CRHS = dyn_cast<Constant>(RHS)) 2162 return ConstantFoldCompareInstOperands(Pred, CLHS, CRHS, Q.DL, Q.TLI); 2163 2164 // If we have a constant, make sure it is on the RHS. 2165 std::swap(LHS, RHS); 2166 Pred = CmpInst::getSwappedPredicate(Pred); 2167 } 2168 2169 Type *ITy = GetCompareTy(LHS); // The return type. 2170 Type *OpTy = LHS->getType(); // The operand type. 2171 2172 // icmp X, X -> true/false 2173 // X icmp undef -> true/false. For example, icmp ugt %X, undef -> false 2174 // because X could be 0. 2175 if (LHS == RHS || isa<UndefValue>(RHS)) 2176 return ConstantInt::get(ITy, CmpInst::isTrueWhenEqual(Pred)); 2177 2178 // Special case logic when the operands have i1 type. 2179 if (OpTy->getScalarType()->isIntegerTy(1)) { 2180 switch (Pred) { 2181 default: break; 2182 case ICmpInst::ICMP_EQ: 2183 // X == 1 -> X 2184 if (match(RHS, m_One())) 2185 return LHS; 2186 break; 2187 case ICmpInst::ICMP_NE: 2188 // X != 0 -> X 2189 if (match(RHS, m_Zero())) 2190 return LHS; 2191 break; 2192 case ICmpInst::ICMP_UGT: 2193 // X >u 0 -> X 2194 if (match(RHS, m_Zero())) 2195 return LHS; 2196 break; 2197 case ICmpInst::ICMP_UGE: { 2198 // X >=u 1 -> X 2199 if (match(RHS, m_One())) 2200 return LHS; 2201 if (isImpliedCondition(RHS, LHS, Q.DL).getValueOr(false)) 2202 return getTrue(ITy); 2203 break; 2204 } 2205 case ICmpInst::ICMP_SGE: { 2206 /// For signed comparison, the values for an i1 are 0 and -1 2207 /// respectively. This maps into a truth table of: 2208 /// LHS | RHS | LHS >=s RHS | LHS implies RHS 2209 /// 0 | 0 | 1 (0 >= 0) | 1 2210 /// 0 | 1 | 1 (0 >= -1) | 1 2211 /// 1 | 0 | 0 (-1 >= 0) | 0 2212 /// 1 | 1 | 1 (-1 >= -1) | 1 2213 if (isImpliedCondition(LHS, RHS, Q.DL).getValueOr(false)) 2214 return getTrue(ITy); 2215 break; 2216 } 2217 case ICmpInst::ICMP_SLT: 2218 // X <s 0 -> X 2219 if (match(RHS, m_Zero())) 2220 return LHS; 2221 break; 2222 case ICmpInst::ICMP_SLE: 2223 // X <=s -1 -> X 2224 if (match(RHS, m_One())) 2225 return LHS; 2226 break; 2227 case ICmpInst::ICMP_ULE: { 2228 if (isImpliedCondition(LHS, RHS, Q.DL).getValueOr(false)) 2229 return getTrue(ITy); 2230 break; 2231 } 2232 } 2233 } 2234 2235 // If we are comparing with zero then try hard since this is a common case. 2236 if (match(RHS, m_Zero())) { 2237 bool LHSKnownNonNegative, LHSKnownNegative; 2238 switch (Pred) { 2239 default: llvm_unreachable("Unknown ICmp predicate!"); 2240 case ICmpInst::ICMP_ULT: 2241 return getFalse(ITy); 2242 case ICmpInst::ICMP_UGE: 2243 return getTrue(ITy); 2244 case ICmpInst::ICMP_EQ: 2245 case ICmpInst::ICMP_ULE: 2246 if (isKnownNonZero(LHS, Q.DL, 0, Q.AC, Q.CxtI, Q.DT)) 2247 return getFalse(ITy); 2248 break; 2249 case ICmpInst::ICMP_NE: 2250 case ICmpInst::ICMP_UGT: 2251 if (isKnownNonZero(LHS, Q.DL, 0, Q.AC, Q.CxtI, Q.DT)) 2252 return getTrue(ITy); 2253 break; 2254 case ICmpInst::ICMP_SLT: 2255 ComputeSignBit(LHS, LHSKnownNonNegative, LHSKnownNegative, Q.DL, 0, Q.AC, 2256 Q.CxtI, Q.DT); 2257 if (LHSKnownNegative) 2258 return getTrue(ITy); 2259 if (LHSKnownNonNegative) 2260 return getFalse(ITy); 2261 break; 2262 case ICmpInst::ICMP_SLE: 2263 ComputeSignBit(LHS, LHSKnownNonNegative, LHSKnownNegative, Q.DL, 0, Q.AC, 2264 Q.CxtI, Q.DT); 2265 if (LHSKnownNegative) 2266 return getTrue(ITy); 2267 if (LHSKnownNonNegative && 2268 isKnownNonZero(LHS, Q.DL, 0, Q.AC, Q.CxtI, Q.DT)) 2269 return getFalse(ITy); 2270 break; 2271 case ICmpInst::ICMP_SGE: 2272 ComputeSignBit(LHS, LHSKnownNonNegative, LHSKnownNegative, Q.DL, 0, Q.AC, 2273 Q.CxtI, Q.DT); 2274 if (LHSKnownNegative) 2275 return getFalse(ITy); 2276 if (LHSKnownNonNegative) 2277 return getTrue(ITy); 2278 break; 2279 case ICmpInst::ICMP_SGT: 2280 ComputeSignBit(LHS, LHSKnownNonNegative, LHSKnownNegative, Q.DL, 0, Q.AC, 2281 Q.CxtI, Q.DT); 2282 if (LHSKnownNegative) 2283 return getFalse(ITy); 2284 if (LHSKnownNonNegative && 2285 isKnownNonZero(LHS, Q.DL, 0, Q.AC, Q.CxtI, Q.DT)) 2286 return getTrue(ITy); 2287 break; 2288 } 2289 } 2290 2291 // See if we are doing a comparison with a constant integer. 2292 if (ConstantInt *CI = dyn_cast<ConstantInt>(RHS)) { 2293 // Rule out tautological comparisons (eg., ult 0 or uge 0). 2294 ConstantRange RHS_CR = ICmpInst::makeConstantRange(Pred, CI->getValue()); 2295 if (RHS_CR.isEmptySet()) 2296 return ConstantInt::getFalse(CI->getContext()); 2297 if (RHS_CR.isFullSet()) 2298 return ConstantInt::getTrue(CI->getContext()); 2299 2300 // Many binary operators with constant RHS have easy to compute constant 2301 // range. Use them to check whether the comparison is a tautology. 2302 unsigned Width = CI->getBitWidth(); 2303 APInt Lower = APInt(Width, 0); 2304 APInt Upper = APInt(Width, 0); 2305 ConstantInt *CI2; 2306 if (match(LHS, m_URem(m_Value(), m_ConstantInt(CI2)))) { 2307 // 'urem x, CI2' produces [0, CI2). 2308 Upper = CI2->getValue(); 2309 } else if (match(LHS, m_SRem(m_Value(), m_ConstantInt(CI2)))) { 2310 // 'srem x, CI2' produces (-|CI2|, |CI2|). 2311 Upper = CI2->getValue().abs(); 2312 Lower = (-Upper) + 1; 2313 } else if (match(LHS, m_UDiv(m_ConstantInt(CI2), m_Value()))) { 2314 // 'udiv CI2, x' produces [0, CI2]. 2315 Upper = CI2->getValue() + 1; 2316 } else if (match(LHS, m_UDiv(m_Value(), m_ConstantInt(CI2)))) { 2317 // 'udiv x, CI2' produces [0, UINT_MAX / CI2]. 2318 APInt NegOne = APInt::getAllOnesValue(Width); 2319 if (!CI2->isZero()) 2320 Upper = NegOne.udiv(CI2->getValue()) + 1; 2321 } else if (match(LHS, m_SDiv(m_ConstantInt(CI2), m_Value()))) { 2322 if (CI2->isMinSignedValue()) { 2323 // 'sdiv INT_MIN, x' produces [INT_MIN, INT_MIN / -2]. 2324 Lower = CI2->getValue(); 2325 Upper = Lower.lshr(1) + 1; 2326 } else { 2327 // 'sdiv CI2, x' produces [-|CI2|, |CI2|]. 2328 Upper = CI2->getValue().abs() + 1; 2329 Lower = (-Upper) + 1; 2330 } 2331 } else if (match(LHS, m_SDiv(m_Value(), m_ConstantInt(CI2)))) { 2332 APInt IntMin = APInt::getSignedMinValue(Width); 2333 APInt IntMax = APInt::getSignedMaxValue(Width); 2334 const APInt &Val = CI2->getValue(); 2335 if (Val.isAllOnesValue()) { 2336 // 'sdiv x, -1' produces [INT_MIN + 1, INT_MAX] 2337 // where CI2 != -1 and CI2 != 0 and CI2 != 1 2338 Lower = IntMin + 1; 2339 Upper = IntMax + 1; 2340 } else if (Val.countLeadingZeros() < Width - 1) { 2341 // 'sdiv x, CI2' produces [INT_MIN / CI2, INT_MAX / CI2] 2342 // where CI2 != -1 and CI2 != 0 and CI2 != 1 2343 Lower = IntMin.sdiv(Val); 2344 Upper = IntMax.sdiv(Val); 2345 if (Lower.sgt(Upper)) 2346 std::swap(Lower, Upper); 2347 Upper = Upper + 1; 2348 assert(Upper != Lower && "Upper part of range has wrapped!"); 2349 } 2350 } else if (match(LHS, m_NUWShl(m_ConstantInt(CI2), m_Value()))) { 2351 // 'shl nuw CI2, x' produces [CI2, CI2 << CLZ(CI2)] 2352 Lower = CI2->getValue(); 2353 Upper = Lower.shl(Lower.countLeadingZeros()) + 1; 2354 } else if (match(LHS, m_NSWShl(m_ConstantInt(CI2), m_Value()))) { 2355 if (CI2->isNegative()) { 2356 // 'shl nsw CI2, x' produces [CI2 << CLO(CI2)-1, CI2] 2357 unsigned ShiftAmount = CI2->getValue().countLeadingOnes() - 1; 2358 Lower = CI2->getValue().shl(ShiftAmount); 2359 Upper = CI2->getValue() + 1; 2360 } else { 2361 // 'shl nsw CI2, x' produces [CI2, CI2 << CLZ(CI2)-1] 2362 unsigned ShiftAmount = CI2->getValue().countLeadingZeros() - 1; 2363 Lower = CI2->getValue(); 2364 Upper = CI2->getValue().shl(ShiftAmount) + 1; 2365 } 2366 } else if (match(LHS, m_LShr(m_Value(), m_ConstantInt(CI2)))) { 2367 // 'lshr x, CI2' produces [0, UINT_MAX >> CI2]. 2368 APInt NegOne = APInt::getAllOnesValue(Width); 2369 if (CI2->getValue().ult(Width)) 2370 Upper = NegOne.lshr(CI2->getValue()) + 1; 2371 } else if (match(LHS, m_LShr(m_ConstantInt(CI2), m_Value()))) { 2372 // 'lshr CI2, x' produces [CI2 >> (Width-1), CI2]. 2373 unsigned ShiftAmount = Width - 1; 2374 if (!CI2->isZero() && cast<BinaryOperator>(LHS)->isExact()) 2375 ShiftAmount = CI2->getValue().countTrailingZeros(); 2376 Lower = CI2->getValue().lshr(ShiftAmount); 2377 Upper = CI2->getValue() + 1; 2378 } else if (match(LHS, m_AShr(m_Value(), m_ConstantInt(CI2)))) { 2379 // 'ashr x, CI2' produces [INT_MIN >> CI2, INT_MAX >> CI2]. 2380 APInt IntMin = APInt::getSignedMinValue(Width); 2381 APInt IntMax = APInt::getSignedMaxValue(Width); 2382 if (CI2->getValue().ult(Width)) { 2383 Lower = IntMin.ashr(CI2->getValue()); 2384 Upper = IntMax.ashr(CI2->getValue()) + 1; 2385 } 2386 } else if (match(LHS, m_AShr(m_ConstantInt(CI2), m_Value()))) { 2387 unsigned ShiftAmount = Width - 1; 2388 if (!CI2->isZero() && cast<BinaryOperator>(LHS)->isExact()) 2389 ShiftAmount = CI2->getValue().countTrailingZeros(); 2390 if (CI2->isNegative()) { 2391 // 'ashr CI2, x' produces [CI2, CI2 >> (Width-1)] 2392 Lower = CI2->getValue(); 2393 Upper = CI2->getValue().ashr(ShiftAmount) + 1; 2394 } else { 2395 // 'ashr CI2, x' produces [CI2 >> (Width-1), CI2] 2396 Lower = CI2->getValue().ashr(ShiftAmount); 2397 Upper = CI2->getValue() + 1; 2398 } 2399 } else if (match(LHS, m_Or(m_Value(), m_ConstantInt(CI2)))) { 2400 // 'or x, CI2' produces [CI2, UINT_MAX]. 2401 Lower = CI2->getValue(); 2402 } else if (match(LHS, m_And(m_Value(), m_ConstantInt(CI2)))) { 2403 // 'and x, CI2' produces [0, CI2]. 2404 Upper = CI2->getValue() + 1; 2405 } else if (match(LHS, m_NUWAdd(m_Value(), m_ConstantInt(CI2)))) { 2406 // 'add nuw x, CI2' produces [CI2, UINT_MAX]. 2407 Lower = CI2->getValue(); 2408 } 2409 2410 ConstantRange LHS_CR = Lower != Upper ? ConstantRange(Lower, Upper) 2411 : ConstantRange(Width, true); 2412 2413 if (auto *I = dyn_cast<Instruction>(LHS)) 2414 if (auto *Ranges = I->getMetadata(LLVMContext::MD_range)) 2415 LHS_CR = LHS_CR.intersectWith(getConstantRangeFromMetadata(*Ranges)); 2416 2417 if (!LHS_CR.isFullSet()) { 2418 if (RHS_CR.contains(LHS_CR)) 2419 return ConstantInt::getTrue(RHS->getContext()); 2420 if (RHS_CR.inverse().contains(LHS_CR)) 2421 return ConstantInt::getFalse(RHS->getContext()); 2422 } 2423 } 2424 2425 // If both operands have range metadata, use the metadata 2426 // to simplify the comparison. 2427 if (isa<Instruction>(RHS) && isa<Instruction>(LHS)) { 2428 auto RHS_Instr = dyn_cast<Instruction>(RHS); 2429 auto LHS_Instr = dyn_cast<Instruction>(LHS); 2430 2431 if (RHS_Instr->getMetadata(LLVMContext::MD_range) && 2432 LHS_Instr->getMetadata(LLVMContext::MD_range)) { 2433 auto RHS_CR = getConstantRangeFromMetadata( 2434 *RHS_Instr->getMetadata(LLVMContext::MD_range)); 2435 auto LHS_CR = getConstantRangeFromMetadata( 2436 *LHS_Instr->getMetadata(LLVMContext::MD_range)); 2437 2438 auto Satisfied_CR = ConstantRange::makeSatisfyingICmpRegion(Pred, RHS_CR); 2439 if (Satisfied_CR.contains(LHS_CR)) 2440 return ConstantInt::getTrue(RHS->getContext()); 2441 2442 auto InversedSatisfied_CR = ConstantRange::makeSatisfyingICmpRegion( 2443 CmpInst::getInversePredicate(Pred), RHS_CR); 2444 if (InversedSatisfied_CR.contains(LHS_CR)) 2445 return ConstantInt::getFalse(RHS->getContext()); 2446 } 2447 } 2448 2449 // Compare of cast, for example (zext X) != 0 -> X != 0 2450 if (isa<CastInst>(LHS) && (isa<Constant>(RHS) || isa<CastInst>(RHS))) { 2451 Instruction *LI = cast<CastInst>(LHS); 2452 Value *SrcOp = LI->getOperand(0); 2453 Type *SrcTy = SrcOp->getType(); 2454 Type *DstTy = LI->getType(); 2455 2456 // Turn icmp (ptrtoint x), (ptrtoint/constant) into a compare of the input 2457 // if the integer type is the same size as the pointer type. 2458 if (MaxRecurse && isa<PtrToIntInst>(LI) && 2459 Q.DL.getTypeSizeInBits(SrcTy) == DstTy->getPrimitiveSizeInBits()) { 2460 if (Constant *RHSC = dyn_cast<Constant>(RHS)) { 2461 // Transfer the cast to the constant. 2462 if (Value *V = SimplifyICmpInst(Pred, SrcOp, 2463 ConstantExpr::getIntToPtr(RHSC, SrcTy), 2464 Q, MaxRecurse-1)) 2465 return V; 2466 } else if (PtrToIntInst *RI = dyn_cast<PtrToIntInst>(RHS)) { 2467 if (RI->getOperand(0)->getType() == SrcTy) 2468 // Compare without the cast. 2469 if (Value *V = SimplifyICmpInst(Pred, SrcOp, RI->getOperand(0), 2470 Q, MaxRecurse-1)) 2471 return V; 2472 } 2473 } 2474 2475 if (isa<ZExtInst>(LHS)) { 2476 // Turn icmp (zext X), (zext Y) into a compare of X and Y if they have the 2477 // same type. 2478 if (ZExtInst *RI = dyn_cast<ZExtInst>(RHS)) { 2479 if (MaxRecurse && SrcTy == RI->getOperand(0)->getType()) 2480 // Compare X and Y. Note that signed predicates become unsigned. 2481 if (Value *V = SimplifyICmpInst(ICmpInst::getUnsignedPredicate(Pred), 2482 SrcOp, RI->getOperand(0), Q, 2483 MaxRecurse-1)) 2484 return V; 2485 } 2486 // Turn icmp (zext X), Cst into a compare of X and Cst if Cst is extended 2487 // too. If not, then try to deduce the result of the comparison. 2488 else if (ConstantInt *CI = dyn_cast<ConstantInt>(RHS)) { 2489 // Compute the constant that would happen if we truncated to SrcTy then 2490 // reextended to DstTy. 2491 Constant *Trunc = ConstantExpr::getTrunc(CI, SrcTy); 2492 Constant *RExt = ConstantExpr::getCast(CastInst::ZExt, Trunc, DstTy); 2493 2494 // If the re-extended constant didn't change then this is effectively 2495 // also a case of comparing two zero-extended values. 2496 if (RExt == CI && MaxRecurse) 2497 if (Value *V = SimplifyICmpInst(ICmpInst::getUnsignedPredicate(Pred), 2498 SrcOp, Trunc, Q, MaxRecurse-1)) 2499 return V; 2500 2501 // Otherwise the upper bits of LHS are zero while RHS has a non-zero bit 2502 // there. Use this to work out the result of the comparison. 2503 if (RExt != CI) { 2504 switch (Pred) { 2505 default: llvm_unreachable("Unknown ICmp predicate!"); 2506 // LHS <u RHS. 2507 case ICmpInst::ICMP_EQ: 2508 case ICmpInst::ICMP_UGT: 2509 case ICmpInst::ICMP_UGE: 2510 return ConstantInt::getFalse(CI->getContext()); 2511 2512 case ICmpInst::ICMP_NE: 2513 case ICmpInst::ICMP_ULT: 2514 case ICmpInst::ICMP_ULE: 2515 return ConstantInt::getTrue(CI->getContext()); 2516 2517 // LHS is non-negative. If RHS is negative then LHS >s LHS. If RHS 2518 // is non-negative then LHS <s RHS. 2519 case ICmpInst::ICMP_SGT: 2520 case ICmpInst::ICMP_SGE: 2521 return CI->getValue().isNegative() ? 2522 ConstantInt::getTrue(CI->getContext()) : 2523 ConstantInt::getFalse(CI->getContext()); 2524 2525 case ICmpInst::ICMP_SLT: 2526 case ICmpInst::ICMP_SLE: 2527 return CI->getValue().isNegative() ? 2528 ConstantInt::getFalse(CI->getContext()) : 2529 ConstantInt::getTrue(CI->getContext()); 2530 } 2531 } 2532 } 2533 } 2534 2535 if (isa<SExtInst>(LHS)) { 2536 // Turn icmp (sext X), (sext Y) into a compare of X and Y if they have the 2537 // same type. 2538 if (SExtInst *RI = dyn_cast<SExtInst>(RHS)) { 2539 if (MaxRecurse && SrcTy == RI->getOperand(0)->getType()) 2540 // Compare X and Y. Note that the predicate does not change. 2541 if (Value *V = SimplifyICmpInst(Pred, SrcOp, RI->getOperand(0), 2542 Q, MaxRecurse-1)) 2543 return V; 2544 } 2545 // Turn icmp (sext X), Cst into a compare of X and Cst if Cst is extended 2546 // too. If not, then try to deduce the result of the comparison. 2547 else if (ConstantInt *CI = dyn_cast<ConstantInt>(RHS)) { 2548 // Compute the constant that would happen if we truncated to SrcTy then 2549 // reextended to DstTy. 2550 Constant *Trunc = ConstantExpr::getTrunc(CI, SrcTy); 2551 Constant *RExt = ConstantExpr::getCast(CastInst::SExt, Trunc, DstTy); 2552 2553 // If the re-extended constant didn't change then this is effectively 2554 // also a case of comparing two sign-extended values. 2555 if (RExt == CI && MaxRecurse) 2556 if (Value *V = SimplifyICmpInst(Pred, SrcOp, Trunc, Q, MaxRecurse-1)) 2557 return V; 2558 2559 // Otherwise the upper bits of LHS are all equal, while RHS has varying 2560 // bits there. Use this to work out the result of the comparison. 2561 if (RExt != CI) { 2562 switch (Pred) { 2563 default: llvm_unreachable("Unknown ICmp predicate!"); 2564 case ICmpInst::ICMP_EQ: 2565 return ConstantInt::getFalse(CI->getContext()); 2566 case ICmpInst::ICMP_NE: 2567 return ConstantInt::getTrue(CI->getContext()); 2568 2569 // If RHS is non-negative then LHS <s RHS. If RHS is negative then 2570 // LHS >s RHS. 2571 case ICmpInst::ICMP_SGT: 2572 case ICmpInst::ICMP_SGE: 2573 return CI->getValue().isNegative() ? 2574 ConstantInt::getTrue(CI->getContext()) : 2575 ConstantInt::getFalse(CI->getContext()); 2576 case ICmpInst::ICMP_SLT: 2577 case ICmpInst::ICMP_SLE: 2578 return CI->getValue().isNegative() ? 2579 ConstantInt::getFalse(CI->getContext()) : 2580 ConstantInt::getTrue(CI->getContext()); 2581 2582 // If LHS is non-negative then LHS <u RHS. If LHS is negative then 2583 // LHS >u RHS. 2584 case ICmpInst::ICMP_UGT: 2585 case ICmpInst::ICMP_UGE: 2586 // Comparison is true iff the LHS <s 0. 2587 if (MaxRecurse) 2588 if (Value *V = SimplifyICmpInst(ICmpInst::ICMP_SLT, SrcOp, 2589 Constant::getNullValue(SrcTy), 2590 Q, MaxRecurse-1)) 2591 return V; 2592 break; 2593 case ICmpInst::ICMP_ULT: 2594 case ICmpInst::ICMP_ULE: 2595 // Comparison is true iff the LHS >=s 0. 2596 if (MaxRecurse) 2597 if (Value *V = SimplifyICmpInst(ICmpInst::ICMP_SGE, SrcOp, 2598 Constant::getNullValue(SrcTy), 2599 Q, MaxRecurse-1)) 2600 return V; 2601 break; 2602 } 2603 } 2604 } 2605 } 2606 } 2607 2608 // icmp eq|ne X, Y -> false|true if X != Y 2609 if ((Pred == ICmpInst::ICMP_EQ || Pred == ICmpInst::ICMP_NE) && 2610 isKnownNonEqual(LHS, RHS, Q.DL, Q.AC, Q.CxtI, Q.DT)) { 2611 LLVMContext &Ctx = LHS->getType()->getContext(); 2612 return Pred == ICmpInst::ICMP_NE ? 2613 ConstantInt::getTrue(Ctx) : ConstantInt::getFalse(Ctx); 2614 } 2615 2616 // Special logic for binary operators. 2617 BinaryOperator *LBO = dyn_cast<BinaryOperator>(LHS); 2618 BinaryOperator *RBO = dyn_cast<BinaryOperator>(RHS); 2619 if (MaxRecurse && (LBO || RBO)) { 2620 // Analyze the case when either LHS or RHS is an add instruction. 2621 Value *A = nullptr, *B = nullptr, *C = nullptr, *D = nullptr; 2622 // LHS = A + B (or A and B are null); RHS = C + D (or C and D are null). 2623 bool NoLHSWrapProblem = false, NoRHSWrapProblem = false; 2624 if (LBO && LBO->getOpcode() == Instruction::Add) { 2625 A = LBO->getOperand(0); B = LBO->getOperand(1); 2626 NoLHSWrapProblem = ICmpInst::isEquality(Pred) || 2627 (CmpInst::isUnsigned(Pred) && LBO->hasNoUnsignedWrap()) || 2628 (CmpInst::isSigned(Pred) && LBO->hasNoSignedWrap()); 2629 } 2630 if (RBO && RBO->getOpcode() == Instruction::Add) { 2631 C = RBO->getOperand(0); D = RBO->getOperand(1); 2632 NoRHSWrapProblem = ICmpInst::isEquality(Pred) || 2633 (CmpInst::isUnsigned(Pred) && RBO->hasNoUnsignedWrap()) || 2634 (CmpInst::isSigned(Pred) && RBO->hasNoSignedWrap()); 2635 } 2636 2637 // icmp (X+Y), X -> icmp Y, 0 for equalities or if there is no overflow. 2638 if ((A == RHS || B == RHS) && NoLHSWrapProblem) 2639 if (Value *V = SimplifyICmpInst(Pred, A == RHS ? B : A, 2640 Constant::getNullValue(RHS->getType()), 2641 Q, MaxRecurse-1)) 2642 return V; 2643 2644 // icmp X, (X+Y) -> icmp 0, Y for equalities or if there is no overflow. 2645 if ((C == LHS || D == LHS) && NoRHSWrapProblem) 2646 if (Value *V = SimplifyICmpInst(Pred, 2647 Constant::getNullValue(LHS->getType()), 2648 C == LHS ? D : C, Q, MaxRecurse-1)) 2649 return V; 2650 2651 // icmp (X+Y), (X+Z) -> icmp Y,Z for equalities or if there is no overflow. 2652 if (A && C && (A == C || A == D || B == C || B == D) && 2653 NoLHSWrapProblem && NoRHSWrapProblem) { 2654 // Determine Y and Z in the form icmp (X+Y), (X+Z). 2655 Value *Y, *Z; 2656 if (A == C) { 2657 // C + B == C + D -> B == D 2658 Y = B; 2659 Z = D; 2660 } else if (A == D) { 2661 // D + B == C + D -> B == C 2662 Y = B; 2663 Z = C; 2664 } else if (B == C) { 2665 // A + C == C + D -> A == D 2666 Y = A; 2667 Z = D; 2668 } else { 2669 assert(B == D); 2670 // A + D == C + D -> A == C 2671 Y = A; 2672 Z = C; 2673 } 2674 if (Value *V = SimplifyICmpInst(Pred, Y, Z, Q, MaxRecurse-1)) 2675 return V; 2676 } 2677 } 2678 2679 { 2680 Value *Y = nullptr; 2681 // icmp pred (or X, Y), X 2682 if (LBO && match(LBO, m_c_Or(m_Value(Y), m_Specific(RHS)))) { 2683 if (Pred == ICmpInst::ICMP_ULT) 2684 return getFalse(ITy); 2685 if (Pred == ICmpInst::ICMP_UGE) 2686 return getTrue(ITy); 2687 2688 if (Pred == ICmpInst::ICMP_SLT || Pred == ICmpInst::ICMP_SGE) { 2689 bool RHSKnownNonNegative, RHSKnownNegative; 2690 bool YKnownNonNegative, YKnownNegative; 2691 ComputeSignBit(RHS, RHSKnownNonNegative, RHSKnownNegative, Q.DL, 0, 2692 Q.AC, Q.CxtI, Q.DT); 2693 ComputeSignBit(Y, YKnownNonNegative, YKnownNegative, Q.DL, 0, Q.AC, 2694 Q.CxtI, Q.DT); 2695 if (RHSKnownNonNegative && YKnownNegative) 2696 return Pred == ICmpInst::ICMP_SLT ? getTrue(ITy) : getFalse(ITy); 2697 if (RHSKnownNegative || YKnownNonNegative) 2698 return Pred == ICmpInst::ICMP_SLT ? getFalse(ITy) : getTrue(ITy); 2699 } 2700 } 2701 // icmp pred X, (or X, Y) 2702 if (RBO && match(RBO, m_c_Or(m_Value(Y), m_Specific(LHS)))) { 2703 if (Pred == ICmpInst::ICMP_ULE) 2704 return getTrue(ITy); 2705 if (Pred == ICmpInst::ICMP_UGT) 2706 return getFalse(ITy); 2707 2708 if (Pred == ICmpInst::ICMP_SGT || Pred == ICmpInst::ICMP_SLE) { 2709 bool LHSKnownNonNegative, LHSKnownNegative; 2710 bool YKnownNonNegative, YKnownNegative; 2711 ComputeSignBit(LHS, LHSKnownNonNegative, LHSKnownNegative, Q.DL, 0, 2712 Q.AC, Q.CxtI, Q.DT); 2713 ComputeSignBit(Y, YKnownNonNegative, YKnownNegative, Q.DL, 0, Q.AC, 2714 Q.CxtI, Q.DT); 2715 if (LHSKnownNonNegative && YKnownNegative) 2716 return Pred == ICmpInst::ICMP_SGT ? getTrue(ITy) : getFalse(ITy); 2717 if (LHSKnownNegative || YKnownNonNegative) 2718 return Pred == ICmpInst::ICMP_SGT ? getFalse(ITy) : getTrue(ITy); 2719 } 2720 } 2721 } 2722 2723 // icmp pred (and X, Y), X 2724 if (LBO && match(LBO, m_CombineOr(m_And(m_Value(), m_Specific(RHS)), 2725 m_And(m_Specific(RHS), m_Value())))) { 2726 if (Pred == ICmpInst::ICMP_UGT) 2727 return getFalse(ITy); 2728 if (Pred == ICmpInst::ICMP_ULE) 2729 return getTrue(ITy); 2730 } 2731 // icmp pred X, (and X, Y) 2732 if (RBO && match(RBO, m_CombineOr(m_And(m_Value(), m_Specific(LHS)), 2733 m_And(m_Specific(LHS), m_Value())))) { 2734 if (Pred == ICmpInst::ICMP_UGE) 2735 return getTrue(ITy); 2736 if (Pred == ICmpInst::ICMP_ULT) 2737 return getFalse(ITy); 2738 } 2739 2740 // 0 - (zext X) pred C 2741 if (!CmpInst::isUnsigned(Pred) && match(LHS, m_Neg(m_ZExt(m_Value())))) { 2742 if (ConstantInt *RHSC = dyn_cast<ConstantInt>(RHS)) { 2743 if (RHSC->getValue().isStrictlyPositive()) { 2744 if (Pred == ICmpInst::ICMP_SLT) 2745 return ConstantInt::getTrue(RHSC->getContext()); 2746 if (Pred == ICmpInst::ICMP_SGE) 2747 return ConstantInt::getFalse(RHSC->getContext()); 2748 if (Pred == ICmpInst::ICMP_EQ) 2749 return ConstantInt::getFalse(RHSC->getContext()); 2750 if (Pred == ICmpInst::ICMP_NE) 2751 return ConstantInt::getTrue(RHSC->getContext()); 2752 } 2753 if (RHSC->getValue().isNonNegative()) { 2754 if (Pred == ICmpInst::ICMP_SLE) 2755 return ConstantInt::getTrue(RHSC->getContext()); 2756 if (Pred == ICmpInst::ICMP_SGT) 2757 return ConstantInt::getFalse(RHSC->getContext()); 2758 } 2759 } 2760 } 2761 2762 // icmp pred (urem X, Y), Y 2763 if (LBO && match(LBO, m_URem(m_Value(), m_Specific(RHS)))) { 2764 bool KnownNonNegative, KnownNegative; 2765 switch (Pred) { 2766 default: 2767 break; 2768 case ICmpInst::ICMP_SGT: 2769 case ICmpInst::ICMP_SGE: 2770 ComputeSignBit(RHS, KnownNonNegative, KnownNegative, Q.DL, 0, Q.AC, 2771 Q.CxtI, Q.DT); 2772 if (!KnownNonNegative) 2773 break; 2774 // fall-through 2775 case ICmpInst::ICMP_EQ: 2776 case ICmpInst::ICMP_UGT: 2777 case ICmpInst::ICMP_UGE: 2778 return getFalse(ITy); 2779 case ICmpInst::ICMP_SLT: 2780 case ICmpInst::ICMP_SLE: 2781 ComputeSignBit(RHS, KnownNonNegative, KnownNegative, Q.DL, 0, Q.AC, 2782 Q.CxtI, Q.DT); 2783 if (!KnownNonNegative) 2784 break; 2785 // fall-through 2786 case ICmpInst::ICMP_NE: 2787 case ICmpInst::ICMP_ULT: 2788 case ICmpInst::ICMP_ULE: 2789 return getTrue(ITy); 2790 } 2791 } 2792 2793 // icmp pred X, (urem Y, X) 2794 if (RBO && match(RBO, m_URem(m_Value(), m_Specific(LHS)))) { 2795 bool KnownNonNegative, KnownNegative; 2796 switch (Pred) { 2797 default: 2798 break; 2799 case ICmpInst::ICMP_SGT: 2800 case ICmpInst::ICMP_SGE: 2801 ComputeSignBit(LHS, KnownNonNegative, KnownNegative, Q.DL, 0, Q.AC, 2802 Q.CxtI, Q.DT); 2803 if (!KnownNonNegative) 2804 break; 2805 // fall-through 2806 case ICmpInst::ICMP_NE: 2807 case ICmpInst::ICMP_UGT: 2808 case ICmpInst::ICMP_UGE: 2809 return getTrue(ITy); 2810 case ICmpInst::ICMP_SLT: 2811 case ICmpInst::ICMP_SLE: 2812 ComputeSignBit(LHS, KnownNonNegative, KnownNegative, Q.DL, 0, Q.AC, 2813 Q.CxtI, Q.DT); 2814 if (!KnownNonNegative) 2815 break; 2816 // fall-through 2817 case ICmpInst::ICMP_EQ: 2818 case ICmpInst::ICMP_ULT: 2819 case ICmpInst::ICMP_ULE: 2820 return getFalse(ITy); 2821 } 2822 } 2823 2824 // x >> y <=u x 2825 // x udiv y <=u x. 2826 if (LBO && (match(LBO, m_LShr(m_Specific(RHS), m_Value())) || 2827 match(LBO, m_UDiv(m_Specific(RHS), m_Value())))) { 2828 // icmp pred (X op Y), X 2829 if (Pred == ICmpInst::ICMP_UGT) 2830 return getFalse(ITy); 2831 if (Pred == ICmpInst::ICMP_ULE) 2832 return getTrue(ITy); 2833 } 2834 2835 // handle: 2836 // CI2 << X == CI 2837 // CI2 << X != CI 2838 // 2839 // where CI2 is a power of 2 and CI isn't 2840 if (auto *CI = dyn_cast<ConstantInt>(RHS)) { 2841 const APInt *CI2Val, *CIVal = &CI->getValue(); 2842 if (LBO && match(LBO, m_Shl(m_APInt(CI2Val), m_Value())) && 2843 CI2Val->isPowerOf2()) { 2844 if (!CIVal->isPowerOf2()) { 2845 // CI2 << X can equal zero in some circumstances, 2846 // this simplification is unsafe if CI is zero. 2847 // 2848 // We know it is safe if: 2849 // - The shift is nsw, we can't shift out the one bit. 2850 // - The shift is nuw, we can't shift out the one bit. 2851 // - CI2 is one 2852 // - CI isn't zero 2853 if (LBO->hasNoSignedWrap() || LBO->hasNoUnsignedWrap() || 2854 *CI2Val == 1 || !CI->isZero()) { 2855 if (Pred == ICmpInst::ICMP_EQ) 2856 return ConstantInt::getFalse(RHS->getContext()); 2857 if (Pred == ICmpInst::ICMP_NE) 2858 return ConstantInt::getTrue(RHS->getContext()); 2859 } 2860 } 2861 if (CIVal->isSignBit() && *CI2Val == 1) { 2862 if (Pred == ICmpInst::ICMP_UGT) 2863 return ConstantInt::getFalse(RHS->getContext()); 2864 if (Pred == ICmpInst::ICMP_ULE) 2865 return ConstantInt::getTrue(RHS->getContext()); 2866 } 2867 } 2868 } 2869 2870 if (MaxRecurse && LBO && RBO && LBO->getOpcode() == RBO->getOpcode() && 2871 LBO->getOperand(1) == RBO->getOperand(1)) { 2872 switch (LBO->getOpcode()) { 2873 default: break; 2874 case Instruction::UDiv: 2875 case Instruction::LShr: 2876 if (ICmpInst::isSigned(Pred)) 2877 break; 2878 // fall-through 2879 case Instruction::SDiv: 2880 case Instruction::AShr: 2881 if (!LBO->isExact() || !RBO->isExact()) 2882 break; 2883 if (Value *V = SimplifyICmpInst(Pred, LBO->getOperand(0), 2884 RBO->getOperand(0), Q, MaxRecurse-1)) 2885 return V; 2886 break; 2887 case Instruction::Shl: { 2888 bool NUW = LBO->hasNoUnsignedWrap() && RBO->hasNoUnsignedWrap(); 2889 bool NSW = LBO->hasNoSignedWrap() && RBO->hasNoSignedWrap(); 2890 if (!NUW && !NSW) 2891 break; 2892 if (!NSW && ICmpInst::isSigned(Pred)) 2893 break; 2894 if (Value *V = SimplifyICmpInst(Pred, LBO->getOperand(0), 2895 RBO->getOperand(0), Q, MaxRecurse-1)) 2896 return V; 2897 break; 2898 } 2899 } 2900 } 2901 2902 // Simplify comparisons involving max/min. 2903 Value *A, *B; 2904 CmpInst::Predicate P = CmpInst::BAD_ICMP_PREDICATE; 2905 CmpInst::Predicate EqP; // Chosen so that "A == max/min(A,B)" iff "A EqP B". 2906 2907 // Signed variants on "max(a,b)>=a -> true". 2908 if (match(LHS, m_SMax(m_Value(A), m_Value(B))) && (A == RHS || B == RHS)) { 2909 if (A != RHS) std::swap(A, B); // smax(A, B) pred A. 2910 EqP = CmpInst::ICMP_SGE; // "A == smax(A, B)" iff "A sge B". 2911 // We analyze this as smax(A, B) pred A. 2912 P = Pred; 2913 } else if (match(RHS, m_SMax(m_Value(A), m_Value(B))) && 2914 (A == LHS || B == LHS)) { 2915 if (A != LHS) std::swap(A, B); // A pred smax(A, B). 2916 EqP = CmpInst::ICMP_SGE; // "A == smax(A, B)" iff "A sge B". 2917 // We analyze this as smax(A, B) swapped-pred A. 2918 P = CmpInst::getSwappedPredicate(Pred); 2919 } else if (match(LHS, m_SMin(m_Value(A), m_Value(B))) && 2920 (A == RHS || B == RHS)) { 2921 if (A != RHS) std::swap(A, B); // smin(A, B) pred A. 2922 EqP = CmpInst::ICMP_SLE; // "A == smin(A, B)" iff "A sle B". 2923 // We analyze this as smax(-A, -B) swapped-pred -A. 2924 // Note that we do not need to actually form -A or -B thanks to EqP. 2925 P = CmpInst::getSwappedPredicate(Pred); 2926 } else if (match(RHS, m_SMin(m_Value(A), m_Value(B))) && 2927 (A == LHS || B == LHS)) { 2928 if (A != LHS) std::swap(A, B); // A pred smin(A, B). 2929 EqP = CmpInst::ICMP_SLE; // "A == smin(A, B)" iff "A sle B". 2930 // We analyze this as smax(-A, -B) pred -A. 2931 // Note that we do not need to actually form -A or -B thanks to EqP. 2932 P = Pred; 2933 } 2934 if (P != CmpInst::BAD_ICMP_PREDICATE) { 2935 // Cases correspond to "max(A, B) p A". 2936 switch (P) { 2937 default: 2938 break; 2939 case CmpInst::ICMP_EQ: 2940 case CmpInst::ICMP_SLE: 2941 // Equivalent to "A EqP B". This may be the same as the condition tested 2942 // in the max/min; if so, we can just return that. 2943 if (Value *V = ExtractEquivalentCondition(LHS, EqP, A, B)) 2944 return V; 2945 if (Value *V = ExtractEquivalentCondition(RHS, EqP, A, B)) 2946 return V; 2947 // Otherwise, see if "A EqP B" simplifies. 2948 if (MaxRecurse) 2949 if (Value *V = SimplifyICmpInst(EqP, A, B, Q, MaxRecurse-1)) 2950 return V; 2951 break; 2952 case CmpInst::ICMP_NE: 2953 case CmpInst::ICMP_SGT: { 2954 CmpInst::Predicate InvEqP = CmpInst::getInversePredicate(EqP); 2955 // Equivalent to "A InvEqP B". This may be the same as the condition 2956 // tested in the max/min; if so, we can just return that. 2957 if (Value *V = ExtractEquivalentCondition(LHS, InvEqP, A, B)) 2958 return V; 2959 if (Value *V = ExtractEquivalentCondition(RHS, InvEqP, A, B)) 2960 return V; 2961 // Otherwise, see if "A InvEqP B" simplifies. 2962 if (MaxRecurse) 2963 if (Value *V = SimplifyICmpInst(InvEqP, A, B, Q, MaxRecurse-1)) 2964 return V; 2965 break; 2966 } 2967 case CmpInst::ICMP_SGE: 2968 // Always true. 2969 return getTrue(ITy); 2970 case CmpInst::ICMP_SLT: 2971 // Always false. 2972 return getFalse(ITy); 2973 } 2974 } 2975 2976 // Unsigned variants on "max(a,b)>=a -> true". 2977 P = CmpInst::BAD_ICMP_PREDICATE; 2978 if (match(LHS, m_UMax(m_Value(A), m_Value(B))) && (A == RHS || B == RHS)) { 2979 if (A != RHS) std::swap(A, B); // umax(A, B) pred A. 2980 EqP = CmpInst::ICMP_UGE; // "A == umax(A, B)" iff "A uge B". 2981 // We analyze this as umax(A, B) pred A. 2982 P = Pred; 2983 } else if (match(RHS, m_UMax(m_Value(A), m_Value(B))) && 2984 (A == LHS || B == LHS)) { 2985 if (A != LHS) std::swap(A, B); // A pred umax(A, B). 2986 EqP = CmpInst::ICMP_UGE; // "A == umax(A, B)" iff "A uge B". 2987 // We analyze this as umax(A, B) swapped-pred A. 2988 P = CmpInst::getSwappedPredicate(Pred); 2989 } else if (match(LHS, m_UMin(m_Value(A), m_Value(B))) && 2990 (A == RHS || B == RHS)) { 2991 if (A != RHS) std::swap(A, B); // umin(A, B) pred A. 2992 EqP = CmpInst::ICMP_ULE; // "A == umin(A, B)" iff "A ule B". 2993 // We analyze this as umax(-A, -B) swapped-pred -A. 2994 // Note that we do not need to actually form -A or -B thanks to EqP. 2995 P = CmpInst::getSwappedPredicate(Pred); 2996 } else if (match(RHS, m_UMin(m_Value(A), m_Value(B))) && 2997 (A == LHS || B == LHS)) { 2998 if (A != LHS) std::swap(A, B); // A pred umin(A, B). 2999 EqP = CmpInst::ICMP_ULE; // "A == umin(A, B)" iff "A ule B". 3000 // We analyze this as umax(-A, -B) pred -A. 3001 // Note that we do not need to actually form -A or -B thanks to EqP. 3002 P = Pred; 3003 } 3004 if (P != CmpInst::BAD_ICMP_PREDICATE) { 3005 // Cases correspond to "max(A, B) p A". 3006 switch (P) { 3007 default: 3008 break; 3009 case CmpInst::ICMP_EQ: 3010 case CmpInst::ICMP_ULE: 3011 // Equivalent to "A EqP B". This may be the same as the condition tested 3012 // in the max/min; if so, we can just return that. 3013 if (Value *V = ExtractEquivalentCondition(LHS, EqP, A, B)) 3014 return V; 3015 if (Value *V = ExtractEquivalentCondition(RHS, EqP, A, B)) 3016 return V; 3017 // Otherwise, see if "A EqP B" simplifies. 3018 if (MaxRecurse) 3019 if (Value *V = SimplifyICmpInst(EqP, A, B, Q, MaxRecurse-1)) 3020 return V; 3021 break; 3022 case CmpInst::ICMP_NE: 3023 case CmpInst::ICMP_UGT: { 3024 CmpInst::Predicate InvEqP = CmpInst::getInversePredicate(EqP); 3025 // Equivalent to "A InvEqP B". This may be the same as the condition 3026 // tested in the max/min; if so, we can just return that. 3027 if (Value *V = ExtractEquivalentCondition(LHS, InvEqP, A, B)) 3028 return V; 3029 if (Value *V = ExtractEquivalentCondition(RHS, InvEqP, A, B)) 3030 return V; 3031 // Otherwise, see if "A InvEqP B" simplifies. 3032 if (MaxRecurse) 3033 if (Value *V = SimplifyICmpInst(InvEqP, A, B, Q, MaxRecurse-1)) 3034 return V; 3035 break; 3036 } 3037 case CmpInst::ICMP_UGE: 3038 // Always true. 3039 return getTrue(ITy); 3040 case CmpInst::ICMP_ULT: 3041 // Always false. 3042 return getFalse(ITy); 3043 } 3044 } 3045 3046 // Variants on "max(x,y) >= min(x,z)". 3047 Value *C, *D; 3048 if (match(LHS, m_SMax(m_Value(A), m_Value(B))) && 3049 match(RHS, m_SMin(m_Value(C), m_Value(D))) && 3050 (A == C || A == D || B == C || B == D)) { 3051 // max(x, ?) pred min(x, ?). 3052 if (Pred == CmpInst::ICMP_SGE) 3053 // Always true. 3054 return getTrue(ITy); 3055 if (Pred == CmpInst::ICMP_SLT) 3056 // Always false. 3057 return getFalse(ITy); 3058 } else if (match(LHS, m_SMin(m_Value(A), m_Value(B))) && 3059 match(RHS, m_SMax(m_Value(C), m_Value(D))) && 3060 (A == C || A == D || B == C || B == D)) { 3061 // min(x, ?) pred max(x, ?). 3062 if (Pred == CmpInst::ICMP_SLE) 3063 // Always true. 3064 return getTrue(ITy); 3065 if (Pred == CmpInst::ICMP_SGT) 3066 // Always false. 3067 return getFalse(ITy); 3068 } else if (match(LHS, m_UMax(m_Value(A), m_Value(B))) && 3069 match(RHS, m_UMin(m_Value(C), m_Value(D))) && 3070 (A == C || A == D || B == C || B == D)) { 3071 // max(x, ?) pred min(x, ?). 3072 if (Pred == CmpInst::ICMP_UGE) 3073 // Always true. 3074 return getTrue(ITy); 3075 if (Pred == CmpInst::ICMP_ULT) 3076 // Always false. 3077 return getFalse(ITy); 3078 } else if (match(LHS, m_UMin(m_Value(A), m_Value(B))) && 3079 match(RHS, m_UMax(m_Value(C), m_Value(D))) && 3080 (A == C || A == D || B == C || B == D)) { 3081 // min(x, ?) pred max(x, ?). 3082 if (Pred == CmpInst::ICMP_ULE) 3083 // Always true. 3084 return getTrue(ITy); 3085 if (Pred == CmpInst::ICMP_UGT) 3086 // Always false. 3087 return getFalse(ITy); 3088 } 3089 3090 // Simplify comparisons of related pointers using a powerful, recursive 3091 // GEP-walk when we have target data available.. 3092 if (LHS->getType()->isPointerTy()) 3093 if (auto *C = computePointerICmp(Q.DL, Q.TLI, Q.DT, Pred, Q.CxtI, LHS, RHS)) 3094 return C; 3095 3096 if (GetElementPtrInst *GLHS = dyn_cast<GetElementPtrInst>(LHS)) { 3097 if (GEPOperator *GRHS = dyn_cast<GEPOperator>(RHS)) { 3098 if (GLHS->getPointerOperand() == GRHS->getPointerOperand() && 3099 GLHS->hasAllConstantIndices() && GRHS->hasAllConstantIndices() && 3100 (ICmpInst::isEquality(Pred) || 3101 (GLHS->isInBounds() && GRHS->isInBounds() && 3102 Pred == ICmpInst::getSignedPredicate(Pred)))) { 3103 // The bases are equal and the indices are constant. Build a constant 3104 // expression GEP with the same indices and a null base pointer to see 3105 // what constant folding can make out of it. 3106 Constant *Null = Constant::getNullValue(GLHS->getPointerOperandType()); 3107 SmallVector<Value *, 4> IndicesLHS(GLHS->idx_begin(), GLHS->idx_end()); 3108 Constant *NewLHS = ConstantExpr::getGetElementPtr( 3109 GLHS->getSourceElementType(), Null, IndicesLHS); 3110 3111 SmallVector<Value *, 4> IndicesRHS(GRHS->idx_begin(), GRHS->idx_end()); 3112 Constant *NewRHS = ConstantExpr::getGetElementPtr( 3113 GLHS->getSourceElementType(), Null, IndicesRHS); 3114 return ConstantExpr::getICmp(Pred, NewLHS, NewRHS); 3115 } 3116 } 3117 } 3118 3119 // If a bit is known to be zero for A and known to be one for B, 3120 // then A and B cannot be equal. 3121 if (ICmpInst::isEquality(Pred)) { 3122 if (ConstantInt *CI = dyn_cast<ConstantInt>(RHS)) { 3123 uint32_t BitWidth = CI->getBitWidth(); 3124 APInt LHSKnownZero(BitWidth, 0); 3125 APInt LHSKnownOne(BitWidth, 0); 3126 computeKnownBits(LHS, LHSKnownZero, LHSKnownOne, Q.DL, /*Depth=*/0, Q.AC, 3127 Q.CxtI, Q.DT); 3128 const APInt &RHSVal = CI->getValue(); 3129 if (((LHSKnownZero & RHSVal) != 0) || ((LHSKnownOne & ~RHSVal) != 0)) 3130 return Pred == ICmpInst::ICMP_EQ 3131 ? ConstantInt::getFalse(CI->getContext()) 3132 : ConstantInt::getTrue(CI->getContext()); 3133 } 3134 } 3135 3136 // If the comparison is with the result of a select instruction, check whether 3137 // comparing with either branch of the select always yields the same value. 3138 if (isa<SelectInst>(LHS) || isa<SelectInst>(RHS)) 3139 if (Value *V = ThreadCmpOverSelect(Pred, LHS, RHS, Q, MaxRecurse)) 3140 return V; 3141 3142 // If the comparison is with the result of a phi instruction, check whether 3143 // doing the compare with each incoming phi value yields a common result. 3144 if (isa<PHINode>(LHS) || isa<PHINode>(RHS)) 3145 if (Value *V = ThreadCmpOverPHI(Pred, LHS, RHS, Q, MaxRecurse)) 3146 return V; 3147 3148 return nullptr; 3149 } 3150 3151 Value *llvm::SimplifyICmpInst(unsigned Predicate, Value *LHS, Value *RHS, 3152 const DataLayout &DL, 3153 const TargetLibraryInfo *TLI, 3154 const DominatorTree *DT, AssumptionCache *AC, 3155 const Instruction *CxtI) { 3156 return ::SimplifyICmpInst(Predicate, LHS, RHS, Query(DL, TLI, DT, AC, CxtI), 3157 RecursionLimit); 3158 } 3159 3160 /// Given operands for an FCmpInst, see if we can fold the result. 3161 /// If not, this returns null. 3162 static Value *SimplifyFCmpInst(unsigned Predicate, Value *LHS, Value *RHS, 3163 FastMathFlags FMF, const Query &Q, 3164 unsigned MaxRecurse) { 3165 CmpInst::Predicate Pred = (CmpInst::Predicate)Predicate; 3166 assert(CmpInst::isFPPredicate(Pred) && "Not an FP compare!"); 3167 3168 if (Constant *CLHS = dyn_cast<Constant>(LHS)) { 3169 if (Constant *CRHS = dyn_cast<Constant>(RHS)) 3170 return ConstantFoldCompareInstOperands(Pred, CLHS, CRHS, Q.DL, Q.TLI); 3171 3172 // If we have a constant, make sure it is on the RHS. 3173 std::swap(LHS, RHS); 3174 Pred = CmpInst::getSwappedPredicate(Pred); 3175 } 3176 3177 // Fold trivial predicates. 3178 if (Pred == FCmpInst::FCMP_FALSE) 3179 return ConstantInt::get(GetCompareTy(LHS), 0); 3180 if (Pred == FCmpInst::FCMP_TRUE) 3181 return ConstantInt::get(GetCompareTy(LHS), 1); 3182 3183 // UNO/ORD predicates can be trivially folded if NaNs are ignored. 3184 if (FMF.noNaNs()) { 3185 if (Pred == FCmpInst::FCMP_UNO) 3186 return ConstantInt::get(GetCompareTy(LHS), 0); 3187 if (Pred == FCmpInst::FCMP_ORD) 3188 return ConstantInt::get(GetCompareTy(LHS), 1); 3189 } 3190 3191 // fcmp pred x, undef and fcmp pred undef, x 3192 // fold to true if unordered, false if ordered 3193 if (isa<UndefValue>(LHS) || isa<UndefValue>(RHS)) { 3194 // Choosing NaN for the undef will always make unordered comparison succeed 3195 // and ordered comparison fail. 3196 return ConstantInt::get(GetCompareTy(LHS), CmpInst::isUnordered(Pred)); 3197 } 3198 3199 // fcmp x,x -> true/false. Not all compares are foldable. 3200 if (LHS == RHS) { 3201 if (CmpInst::isTrueWhenEqual(Pred)) 3202 return ConstantInt::get(GetCompareTy(LHS), 1); 3203 if (CmpInst::isFalseWhenEqual(Pred)) 3204 return ConstantInt::get(GetCompareTy(LHS), 0); 3205 } 3206 3207 // Handle fcmp with constant RHS 3208 const ConstantFP *CFP = nullptr; 3209 if (const auto *RHSC = dyn_cast<Constant>(RHS)) { 3210 if (RHS->getType()->isVectorTy()) 3211 CFP = dyn_cast_or_null<ConstantFP>(RHSC->getSplatValue()); 3212 else 3213 CFP = dyn_cast<ConstantFP>(RHSC); 3214 } 3215 if (CFP) { 3216 // If the constant is a nan, see if we can fold the comparison based on it. 3217 if (CFP->getValueAPF().isNaN()) { 3218 if (FCmpInst::isOrdered(Pred)) // True "if ordered and foo" 3219 return ConstantInt::getFalse(CFP->getContext()); 3220 assert(FCmpInst::isUnordered(Pred) && 3221 "Comparison must be either ordered or unordered!"); 3222 // True if unordered. 3223 return ConstantInt::get(GetCompareTy(LHS), 1); 3224 } 3225 // Check whether the constant is an infinity. 3226 if (CFP->getValueAPF().isInfinity()) { 3227 if (CFP->getValueAPF().isNegative()) { 3228 switch (Pred) { 3229 case FCmpInst::FCMP_OLT: 3230 // No value is ordered and less than negative infinity. 3231 return ConstantInt::get(GetCompareTy(LHS), 0); 3232 case FCmpInst::FCMP_UGE: 3233 // All values are unordered with or at least negative infinity. 3234 return ConstantInt::get(GetCompareTy(LHS), 1); 3235 default: 3236 break; 3237 } 3238 } else { 3239 switch (Pred) { 3240 case FCmpInst::FCMP_OGT: 3241 // No value is ordered and greater than infinity. 3242 return ConstantInt::get(GetCompareTy(LHS), 0); 3243 case FCmpInst::FCMP_ULE: 3244 // All values are unordered with and at most infinity. 3245 return ConstantInt::get(GetCompareTy(LHS), 1); 3246 default: 3247 break; 3248 } 3249 } 3250 } 3251 if (CFP->getValueAPF().isZero()) { 3252 switch (Pred) { 3253 case FCmpInst::FCMP_UGE: 3254 if (CannotBeOrderedLessThanZero(LHS, Q.TLI)) 3255 return ConstantInt::get(GetCompareTy(LHS), 1); 3256 break; 3257 case FCmpInst::FCMP_OLT: 3258 // X < 0 3259 if (CannotBeOrderedLessThanZero(LHS, Q.TLI)) 3260 return ConstantInt::get(GetCompareTy(LHS), 0); 3261 break; 3262 default: 3263 break; 3264 } 3265 } 3266 } 3267 3268 // If the comparison is with the result of a select instruction, check whether 3269 // comparing with either branch of the select always yields the same value. 3270 if (isa<SelectInst>(LHS) || isa<SelectInst>(RHS)) 3271 if (Value *V = ThreadCmpOverSelect(Pred, LHS, RHS, Q, MaxRecurse)) 3272 return V; 3273 3274 // If the comparison is with the result of a phi instruction, check whether 3275 // doing the compare with each incoming phi value yields a common result. 3276 if (isa<PHINode>(LHS) || isa<PHINode>(RHS)) 3277 if (Value *V = ThreadCmpOverPHI(Pred, LHS, RHS, Q, MaxRecurse)) 3278 return V; 3279 3280 return nullptr; 3281 } 3282 3283 Value *llvm::SimplifyFCmpInst(unsigned Predicate, Value *LHS, Value *RHS, 3284 FastMathFlags FMF, const DataLayout &DL, 3285 const TargetLibraryInfo *TLI, 3286 const DominatorTree *DT, AssumptionCache *AC, 3287 const Instruction *CxtI) { 3288 return ::SimplifyFCmpInst(Predicate, LHS, RHS, FMF, 3289 Query(DL, TLI, DT, AC, CxtI), RecursionLimit); 3290 } 3291 3292 /// See if V simplifies when its operand Op is replaced with RepOp. 3293 static const Value *SimplifyWithOpReplaced(Value *V, Value *Op, Value *RepOp, 3294 const Query &Q, 3295 unsigned MaxRecurse) { 3296 // Trivial replacement. 3297 if (V == Op) 3298 return RepOp; 3299 3300 auto *I = dyn_cast<Instruction>(V); 3301 if (!I) 3302 return nullptr; 3303 3304 // If this is a binary operator, try to simplify it with the replaced op. 3305 if (auto *B = dyn_cast<BinaryOperator>(I)) { 3306 // Consider: 3307 // %cmp = icmp eq i32 %x, 2147483647 3308 // %add = add nsw i32 %x, 1 3309 // %sel = select i1 %cmp, i32 -2147483648, i32 %add 3310 // 3311 // We can't replace %sel with %add unless we strip away the flags. 3312 if (isa<OverflowingBinaryOperator>(B)) 3313 if (B->hasNoSignedWrap() || B->hasNoUnsignedWrap()) 3314 return nullptr; 3315 if (isa<PossiblyExactOperator>(B)) 3316 if (B->isExact()) 3317 return nullptr; 3318 3319 if (MaxRecurse) { 3320 if (B->getOperand(0) == Op) 3321 return SimplifyBinOp(B->getOpcode(), RepOp, B->getOperand(1), Q, 3322 MaxRecurse - 1); 3323 if (B->getOperand(1) == Op) 3324 return SimplifyBinOp(B->getOpcode(), B->getOperand(0), RepOp, Q, 3325 MaxRecurse - 1); 3326 } 3327 } 3328 3329 // Same for CmpInsts. 3330 if (CmpInst *C = dyn_cast<CmpInst>(I)) { 3331 if (MaxRecurse) { 3332 if (C->getOperand(0) == Op) 3333 return SimplifyCmpInst(C->getPredicate(), RepOp, C->getOperand(1), Q, 3334 MaxRecurse - 1); 3335 if (C->getOperand(1) == Op) 3336 return SimplifyCmpInst(C->getPredicate(), C->getOperand(0), RepOp, Q, 3337 MaxRecurse - 1); 3338 } 3339 } 3340 3341 // TODO: We could hand off more cases to instsimplify here. 3342 3343 // If all operands are constant after substituting Op for RepOp then we can 3344 // constant fold the instruction. 3345 if (Constant *CRepOp = dyn_cast<Constant>(RepOp)) { 3346 // Build a list of all constant operands. 3347 SmallVector<Constant *, 8> ConstOps; 3348 for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i) { 3349 if (I->getOperand(i) == Op) 3350 ConstOps.push_back(CRepOp); 3351 else if (Constant *COp = dyn_cast<Constant>(I->getOperand(i))) 3352 ConstOps.push_back(COp); 3353 else 3354 break; 3355 } 3356 3357 // All operands were constants, fold it. 3358 if (ConstOps.size() == I->getNumOperands()) { 3359 if (CmpInst *C = dyn_cast<CmpInst>(I)) 3360 return ConstantFoldCompareInstOperands(C->getPredicate(), ConstOps[0], 3361 ConstOps[1], Q.DL, Q.TLI); 3362 3363 if (LoadInst *LI = dyn_cast<LoadInst>(I)) 3364 if (!LI->isVolatile()) 3365 return ConstantFoldLoadFromConstPtr(ConstOps[0], LI->getType(), Q.DL); 3366 3367 return ConstantFoldInstOperands(I, ConstOps, Q.DL, Q.TLI); 3368 } 3369 } 3370 3371 return nullptr; 3372 } 3373 3374 /// Given operands for a SelectInst, see if we can fold the result. 3375 /// If not, this returns null. 3376 static Value *SimplifySelectInst(Value *CondVal, Value *TrueVal, 3377 Value *FalseVal, const Query &Q, 3378 unsigned MaxRecurse) { 3379 // select true, X, Y -> X 3380 // select false, X, Y -> Y 3381 if (Constant *CB = dyn_cast<Constant>(CondVal)) { 3382 if (CB->isAllOnesValue()) 3383 return TrueVal; 3384 if (CB->isNullValue()) 3385 return FalseVal; 3386 } 3387 3388 // select C, X, X -> X 3389 if (TrueVal == FalseVal) 3390 return TrueVal; 3391 3392 if (isa<UndefValue>(CondVal)) { // select undef, X, Y -> X or Y 3393 if (isa<Constant>(TrueVal)) 3394 return TrueVal; 3395 return FalseVal; 3396 } 3397 if (isa<UndefValue>(TrueVal)) // select C, undef, X -> X 3398 return FalseVal; 3399 if (isa<UndefValue>(FalseVal)) // select C, X, undef -> X 3400 return TrueVal; 3401 3402 if (const auto *ICI = dyn_cast<ICmpInst>(CondVal)) { 3403 unsigned BitWidth = Q.DL.getTypeSizeInBits(TrueVal->getType()); 3404 ICmpInst::Predicate Pred = ICI->getPredicate(); 3405 Value *CmpLHS = ICI->getOperand(0); 3406 Value *CmpRHS = ICI->getOperand(1); 3407 APInt MinSignedValue = APInt::getSignBit(BitWidth); 3408 Value *X; 3409 const APInt *Y; 3410 bool TrueWhenUnset; 3411 bool IsBitTest = false; 3412 if (ICmpInst::isEquality(Pred) && 3413 match(CmpLHS, m_And(m_Value(X), m_APInt(Y))) && 3414 match(CmpRHS, m_Zero())) { 3415 IsBitTest = true; 3416 TrueWhenUnset = Pred == ICmpInst::ICMP_EQ; 3417 } else if (Pred == ICmpInst::ICMP_SLT && match(CmpRHS, m_Zero())) { 3418 X = CmpLHS; 3419 Y = &MinSignedValue; 3420 IsBitTest = true; 3421 TrueWhenUnset = false; 3422 } else if (Pred == ICmpInst::ICMP_SGT && match(CmpRHS, m_AllOnes())) { 3423 X = CmpLHS; 3424 Y = &MinSignedValue; 3425 IsBitTest = true; 3426 TrueWhenUnset = true; 3427 } 3428 if (IsBitTest) { 3429 const APInt *C; 3430 // (X & Y) == 0 ? X & ~Y : X --> X 3431 // (X & Y) != 0 ? X & ~Y : X --> X & ~Y 3432 if (FalseVal == X && match(TrueVal, m_And(m_Specific(X), m_APInt(C))) && 3433 *Y == ~*C) 3434 return TrueWhenUnset ? FalseVal : TrueVal; 3435 // (X & Y) == 0 ? X : X & ~Y --> X & ~Y 3436 // (X & Y) != 0 ? X : X & ~Y --> X 3437 if (TrueVal == X && match(FalseVal, m_And(m_Specific(X), m_APInt(C))) && 3438 *Y == ~*C) 3439 return TrueWhenUnset ? FalseVal : TrueVal; 3440 3441 if (Y->isPowerOf2()) { 3442 // (X & Y) == 0 ? X | Y : X --> X | Y 3443 // (X & Y) != 0 ? X | Y : X --> X 3444 if (FalseVal == X && match(TrueVal, m_Or(m_Specific(X), m_APInt(C))) && 3445 *Y == *C) 3446 return TrueWhenUnset ? TrueVal : FalseVal; 3447 // (X & Y) == 0 ? X : X | Y --> X 3448 // (X & Y) != 0 ? X : X | Y --> X | Y 3449 if (TrueVal == X && match(FalseVal, m_Or(m_Specific(X), m_APInt(C))) && 3450 *Y == *C) 3451 return TrueWhenUnset ? TrueVal : FalseVal; 3452 } 3453 } 3454 if (ICI->hasOneUse()) { 3455 const APInt *C; 3456 if (match(CmpRHS, m_APInt(C))) { 3457 // X < MIN ? T : F --> F 3458 if (Pred == ICmpInst::ICMP_SLT && C->isMinSignedValue()) 3459 return FalseVal; 3460 // X < MIN ? T : F --> F 3461 if (Pred == ICmpInst::ICMP_ULT && C->isMinValue()) 3462 return FalseVal; 3463 // X > MAX ? T : F --> F 3464 if (Pred == ICmpInst::ICMP_SGT && C->isMaxSignedValue()) 3465 return FalseVal; 3466 // X > MAX ? T : F --> F 3467 if (Pred == ICmpInst::ICMP_UGT && C->isMaxValue()) 3468 return FalseVal; 3469 } 3470 } 3471 3472 // If we have an equality comparison then we know the value in one of the 3473 // arms of the select. See if substituting this value into the arm and 3474 // simplifying the result yields the same value as the other arm. 3475 if (Pred == ICmpInst::ICMP_EQ) { 3476 if (SimplifyWithOpReplaced(FalseVal, CmpLHS, CmpRHS, Q, MaxRecurse) == 3477 TrueVal || 3478 SimplifyWithOpReplaced(FalseVal, CmpRHS, CmpLHS, Q, MaxRecurse) == 3479 TrueVal) 3480 return FalseVal; 3481 if (SimplifyWithOpReplaced(TrueVal, CmpLHS, CmpRHS, Q, MaxRecurse) == 3482 FalseVal || 3483 SimplifyWithOpReplaced(TrueVal, CmpRHS, CmpLHS, Q, MaxRecurse) == 3484 FalseVal) 3485 return FalseVal; 3486 } else if (Pred == ICmpInst::ICMP_NE) { 3487 if (SimplifyWithOpReplaced(TrueVal, CmpLHS, CmpRHS, Q, MaxRecurse) == 3488 FalseVal || 3489 SimplifyWithOpReplaced(TrueVal, CmpRHS, CmpLHS, Q, MaxRecurse) == 3490 FalseVal) 3491 return TrueVal; 3492 if (SimplifyWithOpReplaced(FalseVal, CmpLHS, CmpRHS, Q, MaxRecurse) == 3493 TrueVal || 3494 SimplifyWithOpReplaced(FalseVal, CmpRHS, CmpLHS, Q, MaxRecurse) == 3495 TrueVal) 3496 return TrueVal; 3497 } 3498 } 3499 3500 return nullptr; 3501 } 3502 3503 Value *llvm::SimplifySelectInst(Value *Cond, Value *TrueVal, Value *FalseVal, 3504 const DataLayout &DL, 3505 const TargetLibraryInfo *TLI, 3506 const DominatorTree *DT, AssumptionCache *AC, 3507 const Instruction *CxtI) { 3508 return ::SimplifySelectInst(Cond, TrueVal, FalseVal, 3509 Query(DL, TLI, DT, AC, CxtI), RecursionLimit); 3510 } 3511 3512 /// Given operands for an GetElementPtrInst, see if we can fold the result. 3513 /// If not, this returns null. 3514 static Value *SimplifyGEPInst(Type *SrcTy, ArrayRef<Value *> Ops, 3515 const Query &Q, unsigned) { 3516 // The type of the GEP pointer operand. 3517 unsigned AS = 3518 cast<PointerType>(Ops[0]->getType()->getScalarType())->getAddressSpace(); 3519 3520 // getelementptr P -> P. 3521 if (Ops.size() == 1) 3522 return Ops[0]; 3523 3524 // Compute the (pointer) type returned by the GEP instruction. 3525 Type *LastType = GetElementPtrInst::getIndexedType(SrcTy, Ops.slice(1)); 3526 Type *GEPTy = PointerType::get(LastType, AS); 3527 if (VectorType *VT = dyn_cast<VectorType>(Ops[0]->getType())) 3528 GEPTy = VectorType::get(GEPTy, VT->getNumElements()); 3529 3530 if (isa<UndefValue>(Ops[0])) 3531 return UndefValue::get(GEPTy); 3532 3533 if (Ops.size() == 2) { 3534 // getelementptr P, 0 -> P. 3535 if (match(Ops[1], m_Zero())) 3536 return Ops[0]; 3537 3538 Type *Ty = SrcTy; 3539 if (Ty->isSized()) { 3540 Value *P; 3541 uint64_t C; 3542 uint64_t TyAllocSize = Q.DL.getTypeAllocSize(Ty); 3543 // getelementptr P, N -> P if P points to a type of zero size. 3544 if (TyAllocSize == 0) 3545 return Ops[0]; 3546 3547 // The following transforms are only safe if the ptrtoint cast 3548 // doesn't truncate the pointers. 3549 if (Ops[1]->getType()->getScalarSizeInBits() == 3550 Q.DL.getPointerSizeInBits(AS)) { 3551 auto PtrToIntOrZero = [GEPTy](Value *P) -> Value * { 3552 if (match(P, m_Zero())) 3553 return Constant::getNullValue(GEPTy); 3554 Value *Temp; 3555 if (match(P, m_PtrToInt(m_Value(Temp)))) 3556 if (Temp->getType() == GEPTy) 3557 return Temp; 3558 return nullptr; 3559 }; 3560 3561 // getelementptr V, (sub P, V) -> P if P points to a type of size 1. 3562 if (TyAllocSize == 1 && 3563 match(Ops[1], m_Sub(m_Value(P), m_PtrToInt(m_Specific(Ops[0]))))) 3564 if (Value *R = PtrToIntOrZero(P)) 3565 return R; 3566 3567 // getelementptr V, (ashr (sub P, V), C) -> Q 3568 // if P points to a type of size 1 << C. 3569 if (match(Ops[1], 3570 m_AShr(m_Sub(m_Value(P), m_PtrToInt(m_Specific(Ops[0]))), 3571 m_ConstantInt(C))) && 3572 TyAllocSize == 1ULL << C) 3573 if (Value *R = PtrToIntOrZero(P)) 3574 return R; 3575 3576 // getelementptr V, (sdiv (sub P, V), C) -> Q 3577 // if P points to a type of size C. 3578 if (match(Ops[1], 3579 m_SDiv(m_Sub(m_Value(P), m_PtrToInt(m_Specific(Ops[0]))), 3580 m_SpecificInt(TyAllocSize)))) 3581 if (Value *R = PtrToIntOrZero(P)) 3582 return R; 3583 } 3584 } 3585 } 3586 3587 // Check to see if this is constant foldable. 3588 for (unsigned i = 0, e = Ops.size(); i != e; ++i) 3589 if (!isa<Constant>(Ops[i])) 3590 return nullptr; 3591 3592 return ConstantExpr::getGetElementPtr(SrcTy, cast<Constant>(Ops[0]), 3593 Ops.slice(1)); 3594 } 3595 3596 Value *llvm::SimplifyGEPInst(Type *SrcTy, ArrayRef<Value *> Ops, 3597 const DataLayout &DL, 3598 const TargetLibraryInfo *TLI, 3599 const DominatorTree *DT, AssumptionCache *AC, 3600 const Instruction *CxtI) { 3601 return ::SimplifyGEPInst(SrcTy, Ops, 3602 Query(DL, TLI, DT, AC, CxtI), RecursionLimit); 3603 } 3604 3605 /// Given operands for an InsertValueInst, see if we can fold the result. 3606 /// If not, this returns null. 3607 static Value *SimplifyInsertValueInst(Value *Agg, Value *Val, 3608 ArrayRef<unsigned> Idxs, const Query &Q, 3609 unsigned) { 3610 if (Constant *CAgg = dyn_cast<Constant>(Agg)) 3611 if (Constant *CVal = dyn_cast<Constant>(Val)) 3612 return ConstantFoldInsertValueInstruction(CAgg, CVal, Idxs); 3613 3614 // insertvalue x, undef, n -> x 3615 if (match(Val, m_Undef())) 3616 return Agg; 3617 3618 // insertvalue x, (extractvalue y, n), n 3619 if (ExtractValueInst *EV = dyn_cast<ExtractValueInst>(Val)) 3620 if (EV->getAggregateOperand()->getType() == Agg->getType() && 3621 EV->getIndices() == Idxs) { 3622 // insertvalue undef, (extractvalue y, n), n -> y 3623 if (match(Agg, m_Undef())) 3624 return EV->getAggregateOperand(); 3625 3626 // insertvalue y, (extractvalue y, n), n -> y 3627 if (Agg == EV->getAggregateOperand()) 3628 return Agg; 3629 } 3630 3631 return nullptr; 3632 } 3633 3634 Value *llvm::SimplifyInsertValueInst( 3635 Value *Agg, Value *Val, ArrayRef<unsigned> Idxs, const DataLayout &DL, 3636 const TargetLibraryInfo *TLI, const DominatorTree *DT, AssumptionCache *AC, 3637 const Instruction *CxtI) { 3638 return ::SimplifyInsertValueInst(Agg, Val, Idxs, Query(DL, TLI, DT, AC, CxtI), 3639 RecursionLimit); 3640 } 3641 3642 /// Given operands for an ExtractValueInst, see if we can fold the result. 3643 /// If not, this returns null. 3644 static Value *SimplifyExtractValueInst(Value *Agg, ArrayRef<unsigned> Idxs, 3645 const Query &, unsigned) { 3646 if (auto *CAgg = dyn_cast<Constant>(Agg)) 3647 return ConstantFoldExtractValueInstruction(CAgg, Idxs); 3648 3649 // extractvalue x, (insertvalue y, elt, n), n -> elt 3650 unsigned NumIdxs = Idxs.size(); 3651 for (auto *IVI = dyn_cast<InsertValueInst>(Agg); IVI != nullptr; 3652 IVI = dyn_cast<InsertValueInst>(IVI->getAggregateOperand())) { 3653 ArrayRef<unsigned> InsertValueIdxs = IVI->getIndices(); 3654 unsigned NumInsertValueIdxs = InsertValueIdxs.size(); 3655 unsigned NumCommonIdxs = std::min(NumInsertValueIdxs, NumIdxs); 3656 if (InsertValueIdxs.slice(0, NumCommonIdxs) == 3657 Idxs.slice(0, NumCommonIdxs)) { 3658 if (NumIdxs == NumInsertValueIdxs) 3659 return IVI->getInsertedValueOperand(); 3660 break; 3661 } 3662 } 3663 3664 return nullptr; 3665 } 3666 3667 Value *llvm::SimplifyExtractValueInst(Value *Agg, ArrayRef<unsigned> Idxs, 3668 const DataLayout &DL, 3669 const TargetLibraryInfo *TLI, 3670 const DominatorTree *DT, 3671 AssumptionCache *AC, 3672 const Instruction *CxtI) { 3673 return ::SimplifyExtractValueInst(Agg, Idxs, Query(DL, TLI, DT, AC, CxtI), 3674 RecursionLimit); 3675 } 3676 3677 /// Given operands for an ExtractElementInst, see if we can fold the result. 3678 /// If not, this returns null. 3679 static Value *SimplifyExtractElementInst(Value *Vec, Value *Idx, const Query &, 3680 unsigned) { 3681 if (auto *CVec = dyn_cast<Constant>(Vec)) { 3682 if (auto *CIdx = dyn_cast<Constant>(Idx)) 3683 return ConstantFoldExtractElementInstruction(CVec, CIdx); 3684 3685 // The index is not relevant if our vector is a splat. 3686 if (auto *Splat = CVec->getSplatValue()) 3687 return Splat; 3688 3689 if (isa<UndefValue>(Vec)) 3690 return UndefValue::get(Vec->getType()->getVectorElementType()); 3691 } 3692 3693 // If extracting a specified index from the vector, see if we can recursively 3694 // find a previously computed scalar that was inserted into the vector. 3695 if (auto *IdxC = dyn_cast<ConstantInt>(Idx)) 3696 if (Value *Elt = findScalarElement(Vec, IdxC->getZExtValue())) 3697 return Elt; 3698 3699 return nullptr; 3700 } 3701 3702 Value *llvm::SimplifyExtractElementInst( 3703 Value *Vec, Value *Idx, const DataLayout &DL, const TargetLibraryInfo *TLI, 3704 const DominatorTree *DT, AssumptionCache *AC, const Instruction *CxtI) { 3705 return ::SimplifyExtractElementInst(Vec, Idx, Query(DL, TLI, DT, AC, CxtI), 3706 RecursionLimit); 3707 } 3708 3709 /// See if we can fold the given phi. If not, returns null. 3710 static Value *SimplifyPHINode(PHINode *PN, const Query &Q) { 3711 // If all of the PHI's incoming values are the same then replace the PHI node 3712 // with the common value. 3713 Value *CommonValue = nullptr; 3714 bool HasUndefInput = false; 3715 for (Value *Incoming : PN->incoming_values()) { 3716 // If the incoming value is the phi node itself, it can safely be skipped. 3717 if (Incoming == PN) continue; 3718 if (isa<UndefValue>(Incoming)) { 3719 // Remember that we saw an undef value, but otherwise ignore them. 3720 HasUndefInput = true; 3721 continue; 3722 } 3723 if (CommonValue && Incoming != CommonValue) 3724 return nullptr; // Not the same, bail out. 3725 CommonValue = Incoming; 3726 } 3727 3728 // If CommonValue is null then all of the incoming values were either undef or 3729 // equal to the phi node itself. 3730 if (!CommonValue) 3731 return UndefValue::get(PN->getType()); 3732 3733 // If we have a PHI node like phi(X, undef, X), where X is defined by some 3734 // instruction, we cannot return X as the result of the PHI node unless it 3735 // dominates the PHI block. 3736 if (HasUndefInput) 3737 return ValueDominatesPHI(CommonValue, PN, Q.DT) ? CommonValue : nullptr; 3738 3739 return CommonValue; 3740 } 3741 3742 static Value *SimplifyTruncInst(Value *Op, Type *Ty, const Query &Q, unsigned) { 3743 if (Constant *C = dyn_cast<Constant>(Op)) 3744 return ConstantFoldCastOperand(Instruction::Trunc, C, Ty, Q.DL); 3745 3746 return nullptr; 3747 } 3748 3749 Value *llvm::SimplifyTruncInst(Value *Op, Type *Ty, const DataLayout &DL, 3750 const TargetLibraryInfo *TLI, 3751 const DominatorTree *DT, AssumptionCache *AC, 3752 const Instruction *CxtI) { 3753 return ::SimplifyTruncInst(Op, Ty, Query(DL, TLI, DT, AC, CxtI), 3754 RecursionLimit); 3755 } 3756 3757 //=== Helper functions for higher up the class hierarchy. 3758 3759 /// Given operands for a BinaryOperator, see if we can fold the result. 3760 /// If not, this returns null. 3761 static Value *SimplifyBinOp(unsigned Opcode, Value *LHS, Value *RHS, 3762 const Query &Q, unsigned MaxRecurse) { 3763 switch (Opcode) { 3764 case Instruction::Add: 3765 return SimplifyAddInst(LHS, RHS, /*isNSW*/false, /*isNUW*/false, 3766 Q, MaxRecurse); 3767 case Instruction::FAdd: 3768 return SimplifyFAddInst(LHS, RHS, FastMathFlags(), Q, MaxRecurse); 3769 3770 case Instruction::Sub: 3771 return SimplifySubInst(LHS, RHS, /*isNSW*/false, /*isNUW*/false, 3772 Q, MaxRecurse); 3773 case Instruction::FSub: 3774 return SimplifyFSubInst(LHS, RHS, FastMathFlags(), Q, MaxRecurse); 3775 3776 case Instruction::Mul: return SimplifyMulInst (LHS, RHS, Q, MaxRecurse); 3777 case Instruction::FMul: 3778 return SimplifyFMulInst (LHS, RHS, FastMathFlags(), Q, MaxRecurse); 3779 case Instruction::SDiv: return SimplifySDivInst(LHS, RHS, Q, MaxRecurse); 3780 case Instruction::UDiv: return SimplifyUDivInst(LHS, RHS, Q, MaxRecurse); 3781 case Instruction::FDiv: 3782 return SimplifyFDivInst(LHS, RHS, FastMathFlags(), Q, MaxRecurse); 3783 case Instruction::SRem: return SimplifySRemInst(LHS, RHS, Q, MaxRecurse); 3784 case Instruction::URem: return SimplifyURemInst(LHS, RHS, Q, MaxRecurse); 3785 case Instruction::FRem: 3786 return SimplifyFRemInst(LHS, RHS, FastMathFlags(), Q, MaxRecurse); 3787 case Instruction::Shl: 3788 return SimplifyShlInst(LHS, RHS, /*isNSW*/false, /*isNUW*/false, 3789 Q, MaxRecurse); 3790 case Instruction::LShr: 3791 return SimplifyLShrInst(LHS, RHS, /*isExact*/false, Q, MaxRecurse); 3792 case Instruction::AShr: 3793 return SimplifyAShrInst(LHS, RHS, /*isExact*/false, Q, MaxRecurse); 3794 case Instruction::And: return SimplifyAndInst(LHS, RHS, Q, MaxRecurse); 3795 case Instruction::Or: return SimplifyOrInst (LHS, RHS, Q, MaxRecurse); 3796 case Instruction::Xor: return SimplifyXorInst(LHS, RHS, Q, MaxRecurse); 3797 default: 3798 if (Constant *CLHS = dyn_cast<Constant>(LHS)) 3799 if (Constant *CRHS = dyn_cast<Constant>(RHS)) 3800 return ConstantFoldBinaryOpOperands(Opcode, CLHS, CRHS, Q.DL); 3801 3802 // If the operation is associative, try some generic simplifications. 3803 if (Instruction::isAssociative(Opcode)) 3804 if (Value *V = SimplifyAssociativeBinOp(Opcode, LHS, RHS, Q, MaxRecurse)) 3805 return V; 3806 3807 // If the operation is with the result of a select instruction check whether 3808 // operating on either branch of the select always yields the same value. 3809 if (isa<SelectInst>(LHS) || isa<SelectInst>(RHS)) 3810 if (Value *V = ThreadBinOpOverSelect(Opcode, LHS, RHS, Q, MaxRecurse)) 3811 return V; 3812 3813 // If the operation is with the result of a phi instruction, check whether 3814 // operating on all incoming values of the phi always yields the same value. 3815 if (isa<PHINode>(LHS) || isa<PHINode>(RHS)) 3816 if (Value *V = ThreadBinOpOverPHI(Opcode, LHS, RHS, Q, MaxRecurse)) 3817 return V; 3818 3819 return nullptr; 3820 } 3821 } 3822 3823 /// Given operands for a BinaryOperator, see if we can fold the result. 3824 /// If not, this returns null. 3825 /// In contrast to SimplifyBinOp, try to use FastMathFlag when folding the 3826 /// result. In case we don't need FastMathFlags, simply fall to SimplifyBinOp. 3827 static Value *SimplifyFPBinOp(unsigned Opcode, Value *LHS, Value *RHS, 3828 const FastMathFlags &FMF, const Query &Q, 3829 unsigned MaxRecurse) { 3830 switch (Opcode) { 3831 case Instruction::FAdd: 3832 return SimplifyFAddInst(LHS, RHS, FMF, Q, MaxRecurse); 3833 case Instruction::FSub: 3834 return SimplifyFSubInst(LHS, RHS, FMF, Q, MaxRecurse); 3835 case Instruction::FMul: 3836 return SimplifyFMulInst(LHS, RHS, FMF, Q, MaxRecurse); 3837 default: 3838 return SimplifyBinOp(Opcode, LHS, RHS, Q, MaxRecurse); 3839 } 3840 } 3841 3842 Value *llvm::SimplifyBinOp(unsigned Opcode, Value *LHS, Value *RHS, 3843 const DataLayout &DL, const TargetLibraryInfo *TLI, 3844 const DominatorTree *DT, AssumptionCache *AC, 3845 const Instruction *CxtI) { 3846 return ::SimplifyBinOp(Opcode, LHS, RHS, Query(DL, TLI, DT, AC, CxtI), 3847 RecursionLimit); 3848 } 3849 3850 Value *llvm::SimplifyFPBinOp(unsigned Opcode, Value *LHS, Value *RHS, 3851 const FastMathFlags &FMF, const DataLayout &DL, 3852 const TargetLibraryInfo *TLI, 3853 const DominatorTree *DT, AssumptionCache *AC, 3854 const Instruction *CxtI) { 3855 return ::SimplifyFPBinOp(Opcode, LHS, RHS, FMF, Query(DL, TLI, DT, AC, CxtI), 3856 RecursionLimit); 3857 } 3858 3859 /// Given operands for a CmpInst, see if we can fold the result. 3860 static Value *SimplifyCmpInst(unsigned Predicate, Value *LHS, Value *RHS, 3861 const Query &Q, unsigned MaxRecurse) { 3862 if (CmpInst::isIntPredicate((CmpInst::Predicate)Predicate)) 3863 return SimplifyICmpInst(Predicate, LHS, RHS, Q, MaxRecurse); 3864 return SimplifyFCmpInst(Predicate, LHS, RHS, FastMathFlags(), Q, MaxRecurse); 3865 } 3866 3867 Value *llvm::SimplifyCmpInst(unsigned Predicate, Value *LHS, Value *RHS, 3868 const DataLayout &DL, const TargetLibraryInfo *TLI, 3869 const DominatorTree *DT, AssumptionCache *AC, 3870 const Instruction *CxtI) { 3871 return ::SimplifyCmpInst(Predicate, LHS, RHS, Query(DL, TLI, DT, AC, CxtI), 3872 RecursionLimit); 3873 } 3874 3875 static bool IsIdempotent(Intrinsic::ID ID) { 3876 switch (ID) { 3877 default: return false; 3878 3879 // Unary idempotent: f(f(x)) = f(x) 3880 case Intrinsic::fabs: 3881 case Intrinsic::floor: 3882 case Intrinsic::ceil: 3883 case Intrinsic::trunc: 3884 case Intrinsic::rint: 3885 case Intrinsic::nearbyint: 3886 case Intrinsic::round: 3887 return true; 3888 } 3889 } 3890 3891 static Value *SimplifyRelativeLoad(Constant *Ptr, Constant *Offset, 3892 const DataLayout &DL) { 3893 GlobalValue *PtrSym; 3894 APInt PtrOffset; 3895 if (!IsConstantOffsetFromGlobal(Ptr, PtrSym, PtrOffset, DL)) 3896 return nullptr; 3897 3898 Type *Int8PtrTy = Type::getInt8PtrTy(Ptr->getContext()); 3899 Type *Int32Ty = Type::getInt32Ty(Ptr->getContext()); 3900 Type *Int32PtrTy = Int32Ty->getPointerTo(); 3901 Type *Int64Ty = Type::getInt64Ty(Ptr->getContext()); 3902 3903 auto *OffsetConstInt = dyn_cast<ConstantInt>(Offset); 3904 if (!OffsetConstInt || OffsetConstInt->getType()->getBitWidth() > 64) 3905 return nullptr; 3906 3907 uint64_t OffsetInt = OffsetConstInt->getSExtValue(); 3908 if (OffsetInt % 4 != 0) 3909 return nullptr; 3910 3911 Constant *C = ConstantExpr::getGetElementPtr( 3912 Int32Ty, ConstantExpr::getBitCast(Ptr, Int32PtrTy), 3913 ConstantInt::get(Int64Ty, OffsetInt / 4)); 3914 Constant *Loaded = ConstantFoldLoadFromConstPtr(C, Int32Ty, DL); 3915 if (!Loaded) 3916 return nullptr; 3917 3918 auto *LoadedCE = dyn_cast<ConstantExpr>(Loaded); 3919 if (!LoadedCE) 3920 return nullptr; 3921 3922 if (LoadedCE->getOpcode() == Instruction::Trunc) { 3923 LoadedCE = dyn_cast<ConstantExpr>(LoadedCE->getOperand(0)); 3924 if (!LoadedCE) 3925 return nullptr; 3926 } 3927 3928 if (LoadedCE->getOpcode() != Instruction::Sub) 3929 return nullptr; 3930 3931 auto *LoadedLHS = dyn_cast<ConstantExpr>(LoadedCE->getOperand(0)); 3932 if (!LoadedLHS || LoadedLHS->getOpcode() != Instruction::PtrToInt) 3933 return nullptr; 3934 auto *LoadedLHSPtr = LoadedLHS->getOperand(0); 3935 3936 Constant *LoadedRHS = LoadedCE->getOperand(1); 3937 GlobalValue *LoadedRHSSym; 3938 APInt LoadedRHSOffset; 3939 if (!IsConstantOffsetFromGlobal(LoadedRHS, LoadedRHSSym, LoadedRHSOffset, 3940 DL) || 3941 PtrSym != LoadedRHSSym || PtrOffset != LoadedRHSOffset) 3942 return nullptr; 3943 3944 return ConstantExpr::getBitCast(LoadedLHSPtr, Int8PtrTy); 3945 } 3946 3947 static bool maskIsAllZeroOrUndef(Value *Mask) { 3948 auto *ConstMask = dyn_cast<Constant>(Mask); 3949 if (!ConstMask) 3950 return false; 3951 if (ConstMask->isNullValue() || isa<UndefValue>(ConstMask)) 3952 return true; 3953 for (unsigned I = 0, E = ConstMask->getType()->getVectorNumElements(); I != E; 3954 ++I) { 3955 if (auto *MaskElt = ConstMask->getAggregateElement(I)) 3956 if (MaskElt->isNullValue() || isa<UndefValue>(MaskElt)) 3957 continue; 3958 return false; 3959 } 3960 return true; 3961 } 3962 3963 template <typename IterTy> 3964 static Value *SimplifyIntrinsic(Function *F, IterTy ArgBegin, IterTy ArgEnd, 3965 const Query &Q, unsigned MaxRecurse) { 3966 Intrinsic::ID IID = F->getIntrinsicID(); 3967 unsigned NumOperands = std::distance(ArgBegin, ArgEnd); 3968 Type *ReturnType = F->getReturnType(); 3969 3970 // Binary Ops 3971 if (NumOperands == 2) { 3972 Value *LHS = *ArgBegin; 3973 Value *RHS = *(ArgBegin + 1); 3974 if (IID == Intrinsic::usub_with_overflow || 3975 IID == Intrinsic::ssub_with_overflow) { 3976 // X - X -> { 0, false } 3977 if (LHS == RHS) 3978 return Constant::getNullValue(ReturnType); 3979 3980 // X - undef -> undef 3981 // undef - X -> undef 3982 if (isa<UndefValue>(LHS) || isa<UndefValue>(RHS)) 3983 return UndefValue::get(ReturnType); 3984 } 3985 3986 if (IID == Intrinsic::uadd_with_overflow || 3987 IID == Intrinsic::sadd_with_overflow) { 3988 // X + undef -> undef 3989 if (isa<UndefValue>(RHS)) 3990 return UndefValue::get(ReturnType); 3991 } 3992 3993 if (IID == Intrinsic::umul_with_overflow || 3994 IID == Intrinsic::smul_with_overflow) { 3995 // X * 0 -> { 0, false } 3996 if (match(RHS, m_Zero())) 3997 return Constant::getNullValue(ReturnType); 3998 3999 // X * undef -> { 0, false } 4000 if (match(RHS, m_Undef())) 4001 return Constant::getNullValue(ReturnType); 4002 } 4003 4004 if (IID == Intrinsic::load_relative && isa<Constant>(LHS) && 4005 isa<Constant>(RHS)) 4006 return SimplifyRelativeLoad(cast<Constant>(LHS), cast<Constant>(RHS), 4007 Q.DL); 4008 } 4009 4010 // Simplify calls to llvm.masked.load.* 4011 if (IID == Intrinsic::masked_load) { 4012 Value *MaskArg = ArgBegin[2]; 4013 Value *PassthruArg = ArgBegin[3]; 4014 // If the mask is all zeros or undef, the "passthru" argument is the result. 4015 if (maskIsAllZeroOrUndef(MaskArg)) 4016 return PassthruArg; 4017 } 4018 4019 // Perform idempotent optimizations 4020 if (!IsIdempotent(IID)) 4021 return nullptr; 4022 4023 // Unary Ops 4024 if (NumOperands == 1) 4025 if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(*ArgBegin)) 4026 if (II->getIntrinsicID() == IID) 4027 return II; 4028 4029 return nullptr; 4030 } 4031 4032 template <typename IterTy> 4033 static Value *SimplifyCall(Value *V, IterTy ArgBegin, IterTy ArgEnd, 4034 const Query &Q, unsigned MaxRecurse) { 4035 Type *Ty = V->getType(); 4036 if (PointerType *PTy = dyn_cast<PointerType>(Ty)) 4037 Ty = PTy->getElementType(); 4038 FunctionType *FTy = cast<FunctionType>(Ty); 4039 4040 // call undef -> undef 4041 // call null -> undef 4042 if (isa<UndefValue>(V) || isa<ConstantPointerNull>(V)) 4043 return UndefValue::get(FTy->getReturnType()); 4044 4045 Function *F = dyn_cast<Function>(V); 4046 if (!F) 4047 return nullptr; 4048 4049 if (F->isIntrinsic()) 4050 if (Value *Ret = SimplifyIntrinsic(F, ArgBegin, ArgEnd, Q, MaxRecurse)) 4051 return Ret; 4052 4053 if (!canConstantFoldCallTo(F)) 4054 return nullptr; 4055 4056 SmallVector<Constant *, 4> ConstantArgs; 4057 ConstantArgs.reserve(ArgEnd - ArgBegin); 4058 for (IterTy I = ArgBegin, E = ArgEnd; I != E; ++I) { 4059 Constant *C = dyn_cast<Constant>(*I); 4060 if (!C) 4061 return nullptr; 4062 ConstantArgs.push_back(C); 4063 } 4064 4065 return ConstantFoldCall(F, ConstantArgs, Q.TLI); 4066 } 4067 4068 Value *llvm::SimplifyCall(Value *V, User::op_iterator ArgBegin, 4069 User::op_iterator ArgEnd, const DataLayout &DL, 4070 const TargetLibraryInfo *TLI, const DominatorTree *DT, 4071 AssumptionCache *AC, const Instruction *CxtI) { 4072 return ::SimplifyCall(V, ArgBegin, ArgEnd, Query(DL, TLI, DT, AC, CxtI), 4073 RecursionLimit); 4074 } 4075 4076 Value *llvm::SimplifyCall(Value *V, ArrayRef<Value *> Args, 4077 const DataLayout &DL, const TargetLibraryInfo *TLI, 4078 const DominatorTree *DT, AssumptionCache *AC, 4079 const Instruction *CxtI) { 4080 return ::SimplifyCall(V, Args.begin(), Args.end(), 4081 Query(DL, TLI, DT, AC, CxtI), RecursionLimit); 4082 } 4083 4084 /// See if we can compute a simplified version of this instruction. 4085 /// If not, this returns null. 4086 Value *llvm::SimplifyInstruction(Instruction *I, const DataLayout &DL, 4087 const TargetLibraryInfo *TLI, 4088 const DominatorTree *DT, AssumptionCache *AC) { 4089 Value *Result; 4090 4091 switch (I->getOpcode()) { 4092 default: 4093 Result = ConstantFoldInstruction(I, DL, TLI); 4094 break; 4095 case Instruction::FAdd: 4096 Result = SimplifyFAddInst(I->getOperand(0), I->getOperand(1), 4097 I->getFastMathFlags(), DL, TLI, DT, AC, I); 4098 break; 4099 case Instruction::Add: 4100 Result = SimplifyAddInst(I->getOperand(0), I->getOperand(1), 4101 cast<BinaryOperator>(I)->hasNoSignedWrap(), 4102 cast<BinaryOperator>(I)->hasNoUnsignedWrap(), DL, 4103 TLI, DT, AC, I); 4104 break; 4105 case Instruction::FSub: 4106 Result = SimplifyFSubInst(I->getOperand(0), I->getOperand(1), 4107 I->getFastMathFlags(), DL, TLI, DT, AC, I); 4108 break; 4109 case Instruction::Sub: 4110 Result = SimplifySubInst(I->getOperand(0), I->getOperand(1), 4111 cast<BinaryOperator>(I)->hasNoSignedWrap(), 4112 cast<BinaryOperator>(I)->hasNoUnsignedWrap(), DL, 4113 TLI, DT, AC, I); 4114 break; 4115 case Instruction::FMul: 4116 Result = SimplifyFMulInst(I->getOperand(0), I->getOperand(1), 4117 I->getFastMathFlags(), DL, TLI, DT, AC, I); 4118 break; 4119 case Instruction::Mul: 4120 Result = 4121 SimplifyMulInst(I->getOperand(0), I->getOperand(1), DL, TLI, DT, AC, I); 4122 break; 4123 case Instruction::SDiv: 4124 Result = SimplifySDivInst(I->getOperand(0), I->getOperand(1), DL, TLI, DT, 4125 AC, I); 4126 break; 4127 case Instruction::UDiv: 4128 Result = SimplifyUDivInst(I->getOperand(0), I->getOperand(1), DL, TLI, DT, 4129 AC, I); 4130 break; 4131 case Instruction::FDiv: 4132 Result = SimplifyFDivInst(I->getOperand(0), I->getOperand(1), 4133 I->getFastMathFlags(), DL, TLI, DT, AC, I); 4134 break; 4135 case Instruction::SRem: 4136 Result = SimplifySRemInst(I->getOperand(0), I->getOperand(1), DL, TLI, DT, 4137 AC, I); 4138 break; 4139 case Instruction::URem: 4140 Result = SimplifyURemInst(I->getOperand(0), I->getOperand(1), DL, TLI, DT, 4141 AC, I); 4142 break; 4143 case Instruction::FRem: 4144 Result = SimplifyFRemInst(I->getOperand(0), I->getOperand(1), 4145 I->getFastMathFlags(), DL, TLI, DT, AC, I); 4146 break; 4147 case Instruction::Shl: 4148 Result = SimplifyShlInst(I->getOperand(0), I->getOperand(1), 4149 cast<BinaryOperator>(I)->hasNoSignedWrap(), 4150 cast<BinaryOperator>(I)->hasNoUnsignedWrap(), DL, 4151 TLI, DT, AC, I); 4152 break; 4153 case Instruction::LShr: 4154 Result = SimplifyLShrInst(I->getOperand(0), I->getOperand(1), 4155 cast<BinaryOperator>(I)->isExact(), DL, TLI, DT, 4156 AC, I); 4157 break; 4158 case Instruction::AShr: 4159 Result = SimplifyAShrInst(I->getOperand(0), I->getOperand(1), 4160 cast<BinaryOperator>(I)->isExact(), DL, TLI, DT, 4161 AC, I); 4162 break; 4163 case Instruction::And: 4164 Result = 4165 SimplifyAndInst(I->getOperand(0), I->getOperand(1), DL, TLI, DT, AC, I); 4166 break; 4167 case Instruction::Or: 4168 Result = 4169 SimplifyOrInst(I->getOperand(0), I->getOperand(1), DL, TLI, DT, AC, I); 4170 break; 4171 case Instruction::Xor: 4172 Result = 4173 SimplifyXorInst(I->getOperand(0), I->getOperand(1), DL, TLI, DT, AC, I); 4174 break; 4175 case Instruction::ICmp: 4176 Result = 4177 SimplifyICmpInst(cast<ICmpInst>(I)->getPredicate(), I->getOperand(0), 4178 I->getOperand(1), DL, TLI, DT, AC, I); 4179 break; 4180 case Instruction::FCmp: 4181 Result = SimplifyFCmpInst(cast<FCmpInst>(I)->getPredicate(), 4182 I->getOperand(0), I->getOperand(1), 4183 I->getFastMathFlags(), DL, TLI, DT, AC, I); 4184 break; 4185 case Instruction::Select: 4186 Result = SimplifySelectInst(I->getOperand(0), I->getOperand(1), 4187 I->getOperand(2), DL, TLI, DT, AC, I); 4188 break; 4189 case Instruction::GetElementPtr: { 4190 SmallVector<Value*, 8> Ops(I->op_begin(), I->op_end()); 4191 Result = SimplifyGEPInst(cast<GetElementPtrInst>(I)->getSourceElementType(), 4192 Ops, DL, TLI, DT, AC, I); 4193 break; 4194 } 4195 case Instruction::InsertValue: { 4196 InsertValueInst *IV = cast<InsertValueInst>(I); 4197 Result = SimplifyInsertValueInst(IV->getAggregateOperand(), 4198 IV->getInsertedValueOperand(), 4199 IV->getIndices(), DL, TLI, DT, AC, I); 4200 break; 4201 } 4202 case Instruction::ExtractValue: { 4203 auto *EVI = cast<ExtractValueInst>(I); 4204 Result = SimplifyExtractValueInst(EVI->getAggregateOperand(), 4205 EVI->getIndices(), DL, TLI, DT, AC, I); 4206 break; 4207 } 4208 case Instruction::ExtractElement: { 4209 auto *EEI = cast<ExtractElementInst>(I); 4210 Result = SimplifyExtractElementInst( 4211 EEI->getVectorOperand(), EEI->getIndexOperand(), DL, TLI, DT, AC, I); 4212 break; 4213 } 4214 case Instruction::PHI: 4215 Result = SimplifyPHINode(cast<PHINode>(I), Query(DL, TLI, DT, AC, I)); 4216 break; 4217 case Instruction::Call: { 4218 CallSite CS(cast<CallInst>(I)); 4219 Result = SimplifyCall(CS.getCalledValue(), CS.arg_begin(), CS.arg_end(), DL, 4220 TLI, DT, AC, I); 4221 break; 4222 } 4223 case Instruction::Trunc: 4224 Result = 4225 SimplifyTruncInst(I->getOperand(0), I->getType(), DL, TLI, DT, AC, I); 4226 break; 4227 } 4228 4229 // In general, it is possible for computeKnownBits to determine all bits in a 4230 // value even when the operands are not all constants. 4231 if (!Result && I->getType()->isIntegerTy()) { 4232 unsigned BitWidth = I->getType()->getScalarSizeInBits(); 4233 APInt KnownZero(BitWidth, 0); 4234 APInt KnownOne(BitWidth, 0); 4235 computeKnownBits(I, KnownZero, KnownOne, DL, /*Depth*/0, AC, I, DT); 4236 if ((KnownZero | KnownOne).isAllOnesValue()) 4237 Result = ConstantInt::get(I->getContext(), KnownOne); 4238 } 4239 4240 /// If called on unreachable code, the above logic may report that the 4241 /// instruction simplified to itself. Make life easier for users by 4242 /// detecting that case here, returning a safe value instead. 4243 return Result == I ? UndefValue::get(I->getType()) : Result; 4244 } 4245 4246 /// \brief Implementation of recursive simplification through an instruction's 4247 /// uses. 4248 /// 4249 /// This is the common implementation of the recursive simplification routines. 4250 /// If we have a pre-simplified value in 'SimpleV', that is forcibly used to 4251 /// replace the instruction 'I'. Otherwise, we simply add 'I' to the list of 4252 /// instructions to process and attempt to simplify it using 4253 /// InstructionSimplify. 4254 /// 4255 /// This routine returns 'true' only when *it* simplifies something. The passed 4256 /// in simplified value does not count toward this. 4257 static bool replaceAndRecursivelySimplifyImpl(Instruction *I, Value *SimpleV, 4258 const TargetLibraryInfo *TLI, 4259 const DominatorTree *DT, 4260 AssumptionCache *AC) { 4261 bool Simplified = false; 4262 SmallSetVector<Instruction *, 8> Worklist; 4263 const DataLayout &DL = I->getModule()->getDataLayout(); 4264 4265 // If we have an explicit value to collapse to, do that round of the 4266 // simplification loop by hand initially. 4267 if (SimpleV) { 4268 for (User *U : I->users()) 4269 if (U != I) 4270 Worklist.insert(cast<Instruction>(U)); 4271 4272 // Replace the instruction with its simplified value. 4273 I->replaceAllUsesWith(SimpleV); 4274 4275 // Gracefully handle edge cases where the instruction is not wired into any 4276 // parent block. 4277 if (I->getParent()) 4278 I->eraseFromParent(); 4279 } else { 4280 Worklist.insert(I); 4281 } 4282 4283 // Note that we must test the size on each iteration, the worklist can grow. 4284 for (unsigned Idx = 0; Idx != Worklist.size(); ++Idx) { 4285 I = Worklist[Idx]; 4286 4287 // See if this instruction simplifies. 4288 SimpleV = SimplifyInstruction(I, DL, TLI, DT, AC); 4289 if (!SimpleV) 4290 continue; 4291 4292 Simplified = true; 4293 4294 // Stash away all the uses of the old instruction so we can check them for 4295 // recursive simplifications after a RAUW. This is cheaper than checking all 4296 // uses of To on the recursive step in most cases. 4297 for (User *U : I->users()) 4298 Worklist.insert(cast<Instruction>(U)); 4299 4300 // Replace the instruction with its simplified value. 4301 I->replaceAllUsesWith(SimpleV); 4302 4303 // Gracefully handle edge cases where the instruction is not wired into any 4304 // parent block. 4305 if (I->getParent()) 4306 I->eraseFromParent(); 4307 } 4308 return Simplified; 4309 } 4310 4311 bool llvm::recursivelySimplifyInstruction(Instruction *I, 4312 const TargetLibraryInfo *TLI, 4313 const DominatorTree *DT, 4314 AssumptionCache *AC) { 4315 return replaceAndRecursivelySimplifyImpl(I, nullptr, TLI, DT, AC); 4316 } 4317 4318 bool llvm::replaceAndRecursivelySimplify(Instruction *I, Value *SimpleV, 4319 const TargetLibraryInfo *TLI, 4320 const DominatorTree *DT, 4321 AssumptionCache *AC) { 4322 assert(I != SimpleV && "replaceAndRecursivelySimplify(X,X) is not valid!"); 4323 assert(SimpleV && "Must provide a simplified value."); 4324 return replaceAndRecursivelySimplifyImpl(I, SimpleV, TLI, DT, AC); 4325 } 4326