1 //===- ConstantFold.cpp - LLVM constant folder ----------------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // This file implements folding of constants for LLVM. This implements the 10 // (internal) ConstantFold.h interface, which is used by the 11 // ConstantExpr::get* methods to automatically fold constants when possible. 12 // 13 // The current constant folding implementation is implemented in two pieces: the 14 // pieces that don't need DataLayout, and the pieces that do. This is to avoid 15 // a dependence in IR on Target. 16 // 17 //===----------------------------------------------------------------------===// 18 19 #include "llvm/IR/ConstantFold.h" 20 #include "llvm/ADT/APSInt.h" 21 #include "llvm/ADT/SmallVector.h" 22 #include "llvm/IR/Constants.h" 23 #include "llvm/IR/DerivedTypes.h" 24 #include "llvm/IR/Function.h" 25 #include "llvm/IR/GetElementPtrTypeIterator.h" 26 #include "llvm/IR/GlobalAlias.h" 27 #include "llvm/IR/GlobalVariable.h" 28 #include "llvm/IR/Instructions.h" 29 #include "llvm/IR/Module.h" 30 #include "llvm/IR/Operator.h" 31 #include "llvm/IR/PatternMatch.h" 32 #include "llvm/Support/ErrorHandling.h" 33 using namespace llvm; 34 using namespace llvm::PatternMatch; 35 36 //===----------------------------------------------------------------------===// 37 // ConstantFold*Instruction Implementations 38 //===----------------------------------------------------------------------===// 39 40 /// This function determines which opcode to use to fold two constant cast 41 /// expressions together. It uses CastInst::isEliminableCastPair to determine 42 /// the opcode. Consequently its just a wrapper around that function. 43 /// Determine if it is valid to fold a cast of a cast 44 static unsigned 45 foldConstantCastPair( 46 unsigned opc, ///< opcode of the second cast constant expression 47 ConstantExpr *Op, ///< the first cast constant expression 48 Type *DstTy ///< destination type of the first cast 49 ) { 50 assert(Op && Op->isCast() && "Can't fold cast of cast without a cast!"); 51 assert(DstTy && DstTy->isFirstClassType() && "Invalid cast destination type"); 52 assert(CastInst::isCast(opc) && "Invalid cast opcode"); 53 54 // The types and opcodes for the two Cast constant expressions 55 Type *SrcTy = Op->getOperand(0)->getType(); 56 Type *MidTy = Op->getType(); 57 Instruction::CastOps firstOp = Instruction::CastOps(Op->getOpcode()); 58 Instruction::CastOps secondOp = Instruction::CastOps(opc); 59 60 // Assume that pointers are never more than 64 bits wide, and only use this 61 // for the middle type. Otherwise we could end up folding away illegal 62 // bitcasts between address spaces with different sizes. 63 IntegerType *FakeIntPtrTy = Type::getInt64Ty(DstTy->getContext()); 64 65 // Let CastInst::isEliminableCastPair do the heavy lifting. 66 return CastInst::isEliminableCastPair(firstOp, secondOp, SrcTy, MidTy, DstTy, 67 nullptr, FakeIntPtrTy, nullptr); 68 } 69 70 static Constant *FoldBitCast(Constant *V, Type *DestTy) { 71 Type *SrcTy = V->getType(); 72 if (SrcTy == DestTy) 73 return V; // no-op cast 74 75 // Handle casts from one vector constant to another. We know that the src 76 // and dest type have the same size (otherwise its an illegal cast). 77 if (VectorType *DestPTy = dyn_cast<VectorType>(DestTy)) { 78 if (V->isAllOnesValue()) 79 return Constant::getAllOnesValue(DestTy); 80 81 // Canonicalize scalar-to-vector bitcasts into vector-to-vector bitcasts 82 // This allows for other simplifications (although some of them 83 // can only be handled by Analysis/ConstantFolding.cpp). 84 if (isa<ConstantInt>(V) || isa<ConstantFP>(V)) 85 return ConstantExpr::getBitCast(ConstantVector::get(V), DestPTy); 86 return nullptr; 87 } 88 89 // Handle integral constant input. 90 if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) { 91 // See note below regarding the PPC_FP128 restriction. 92 if (DestTy->isFloatingPointTy() && !DestTy->isPPC_FP128Ty()) 93 return ConstantFP::get(DestTy->getContext(), 94 APFloat(DestTy->getFltSemantics(), 95 CI->getValue())); 96 97 // Otherwise, can't fold this (vector?) 98 return nullptr; 99 } 100 101 // Handle ConstantFP input: FP -> Integral. 102 if (ConstantFP *FP = dyn_cast<ConstantFP>(V)) { 103 // PPC_FP128 is really the sum of two consecutive doubles, where the first 104 // double is always stored first in memory, regardless of the target 105 // endianness. The memory layout of i128, however, depends on the target 106 // endianness, and so we can't fold this without target endianness 107 // information. This should instead be handled by 108 // Analysis/ConstantFolding.cpp 109 if (FP->getType()->isPPC_FP128Ty()) 110 return nullptr; 111 112 // Make sure dest type is compatible with the folded integer constant. 113 if (!DestTy->isIntegerTy()) 114 return nullptr; 115 116 return ConstantInt::get(FP->getContext(), 117 FP->getValueAPF().bitcastToAPInt()); 118 } 119 120 return nullptr; 121 } 122 123 124 /// V is an integer constant which only has a subset of its bytes used. 125 /// The bytes used are indicated by ByteStart (which is the first byte used, 126 /// counting from the least significant byte) and ByteSize, which is the number 127 /// of bytes used. 128 /// 129 /// This function analyzes the specified constant to see if the specified byte 130 /// range can be returned as a simplified constant. If so, the constant is 131 /// returned, otherwise null is returned. 132 static Constant *ExtractConstantBytes(Constant *C, unsigned ByteStart, 133 unsigned ByteSize) { 134 assert(C->getType()->isIntegerTy() && 135 (cast<IntegerType>(C->getType())->getBitWidth() & 7) == 0 && 136 "Non-byte sized integer input"); 137 [[maybe_unused]] unsigned CSize = cast<IntegerType>(C->getType())->getBitWidth()/8; 138 assert(ByteSize && "Must be accessing some piece"); 139 assert(ByteStart+ByteSize <= CSize && "Extracting invalid piece from input"); 140 assert(ByteSize != CSize && "Should not extract everything"); 141 142 // Constant Integers are simple. 143 if (ConstantInt *CI = dyn_cast<ConstantInt>(C)) { 144 APInt V = CI->getValue(); 145 if (ByteStart) 146 V.lshrInPlace(ByteStart*8); 147 V = V.trunc(ByteSize*8); 148 return ConstantInt::get(CI->getContext(), V); 149 } 150 151 // In the input is a constant expr, we might be able to recursively simplify. 152 // If not, we definitely can't do anything. 153 ConstantExpr *CE = dyn_cast<ConstantExpr>(C); 154 if (!CE) return nullptr; 155 156 switch (CE->getOpcode()) { 157 default: return nullptr; 158 case Instruction::Shl: { 159 ConstantInt *Amt = dyn_cast<ConstantInt>(CE->getOperand(1)); 160 if (!Amt) 161 return nullptr; 162 APInt ShAmt = Amt->getValue(); 163 // Cannot analyze non-byte shifts. 164 if ((ShAmt & 7) != 0) 165 return nullptr; 166 ShAmt.lshrInPlace(3); 167 168 // If the extract is known to be all zeros, return zero. 169 if (ShAmt.uge(ByteStart + ByteSize)) 170 return Constant::getNullValue( 171 IntegerType::get(CE->getContext(), ByteSize * 8)); 172 // If the extract is known to be fully in the input, extract it. 173 if (ShAmt.ule(ByteStart)) 174 return ExtractConstantBytes(CE->getOperand(0), 175 ByteStart - ShAmt.getZExtValue(), ByteSize); 176 177 // TODO: Handle the 'partially zero' case. 178 return nullptr; 179 } 180 } 181 } 182 183 static Constant *foldMaybeUndesirableCast(unsigned opc, Constant *V, 184 Type *DestTy) { 185 return ConstantExpr::isDesirableCastOp(opc) 186 ? ConstantExpr::getCast(opc, V, DestTy) 187 : ConstantFoldCastInstruction(opc, V, DestTy); 188 } 189 190 Constant *llvm::ConstantFoldCastInstruction(unsigned opc, Constant *V, 191 Type *DestTy) { 192 if (isa<PoisonValue>(V)) 193 return PoisonValue::get(DestTy); 194 195 if (isa<UndefValue>(V)) { 196 // zext(undef) = 0, because the top bits will be zero. 197 // sext(undef) = 0, because the top bits will all be the same. 198 // [us]itofp(undef) = 0, because the result value is bounded. 199 if (opc == Instruction::ZExt || opc == Instruction::SExt || 200 opc == Instruction::UIToFP || opc == Instruction::SIToFP) 201 return Constant::getNullValue(DestTy); 202 return UndefValue::get(DestTy); 203 } 204 205 if (V->isNullValue() && !DestTy->isX86_MMXTy() && !DestTy->isX86_AMXTy() && 206 opc != Instruction::AddrSpaceCast) 207 return Constant::getNullValue(DestTy); 208 209 // If the cast operand is a constant expression, there's a few things we can 210 // do to try to simplify it. 211 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V)) { 212 if (CE->isCast()) { 213 // Try hard to fold cast of cast because they are often eliminable. 214 if (unsigned newOpc = foldConstantCastPair(opc, CE, DestTy)) 215 return foldMaybeUndesirableCast(newOpc, CE->getOperand(0), DestTy); 216 } 217 } 218 219 // If the cast operand is a constant vector, perform the cast by 220 // operating on each element. In the cast of bitcasts, the element 221 // count may be mismatched; don't attempt to handle that here. 222 if ((isa<ConstantVector>(V) || isa<ConstantDataVector>(V)) && 223 DestTy->isVectorTy() && 224 cast<FixedVectorType>(DestTy)->getNumElements() == 225 cast<FixedVectorType>(V->getType())->getNumElements()) { 226 VectorType *DestVecTy = cast<VectorType>(DestTy); 227 Type *DstEltTy = DestVecTy->getElementType(); 228 // Fast path for splatted constants. 229 if (Constant *Splat = V->getSplatValue()) { 230 Constant *Res = foldMaybeUndesirableCast(opc, Splat, DstEltTy); 231 if (!Res) 232 return nullptr; 233 return ConstantVector::getSplat( 234 cast<VectorType>(DestTy)->getElementCount(), Res); 235 } 236 SmallVector<Constant *, 16> res; 237 Type *Ty = IntegerType::get(V->getContext(), 32); 238 for (unsigned i = 0, 239 e = cast<FixedVectorType>(V->getType())->getNumElements(); 240 i != e; ++i) { 241 Constant *C = ConstantExpr::getExtractElement(V, ConstantInt::get(Ty, i)); 242 Constant *Casted = foldMaybeUndesirableCast(opc, C, DstEltTy); 243 if (!Casted) 244 return nullptr; 245 res.push_back(Casted); 246 } 247 return ConstantVector::get(res); 248 } 249 250 // We actually have to do a cast now. Perform the cast according to the 251 // opcode specified. 252 switch (opc) { 253 default: 254 llvm_unreachable("Failed to cast constant expression"); 255 case Instruction::FPTrunc: 256 case Instruction::FPExt: 257 if (ConstantFP *FPC = dyn_cast<ConstantFP>(V)) { 258 bool ignored; 259 APFloat Val = FPC->getValueAPF(); 260 Val.convert(DestTy->getFltSemantics(), APFloat::rmNearestTiesToEven, 261 &ignored); 262 return ConstantFP::get(V->getContext(), Val); 263 } 264 return nullptr; // Can't fold. 265 case Instruction::FPToUI: 266 case Instruction::FPToSI: 267 if (ConstantFP *FPC = dyn_cast<ConstantFP>(V)) { 268 const APFloat &V = FPC->getValueAPF(); 269 bool ignored; 270 uint32_t DestBitWidth = cast<IntegerType>(DestTy)->getBitWidth(); 271 APSInt IntVal(DestBitWidth, opc == Instruction::FPToUI); 272 if (APFloat::opInvalidOp == 273 V.convertToInteger(IntVal, APFloat::rmTowardZero, &ignored)) { 274 // Undefined behavior invoked - the destination type can't represent 275 // the input constant. 276 return PoisonValue::get(DestTy); 277 } 278 return ConstantInt::get(FPC->getContext(), IntVal); 279 } 280 return nullptr; // Can't fold. 281 case Instruction::UIToFP: 282 case Instruction::SIToFP: 283 if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) { 284 const APInt &api = CI->getValue(); 285 APFloat apf(DestTy->getFltSemantics(), 286 APInt::getZero(DestTy->getPrimitiveSizeInBits())); 287 apf.convertFromAPInt(api, opc==Instruction::SIToFP, 288 APFloat::rmNearestTiesToEven); 289 return ConstantFP::get(V->getContext(), apf); 290 } 291 return nullptr; 292 case Instruction::ZExt: 293 if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) { 294 uint32_t BitWidth = cast<IntegerType>(DestTy)->getBitWidth(); 295 return ConstantInt::get(V->getContext(), 296 CI->getValue().zext(BitWidth)); 297 } 298 return nullptr; 299 case Instruction::SExt: 300 if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) { 301 uint32_t BitWidth = cast<IntegerType>(DestTy)->getBitWidth(); 302 return ConstantInt::get(V->getContext(), 303 CI->getValue().sext(BitWidth)); 304 } 305 return nullptr; 306 case Instruction::Trunc: { 307 if (V->getType()->isVectorTy()) 308 return nullptr; 309 310 uint32_t DestBitWidth = cast<IntegerType>(DestTy)->getBitWidth(); 311 if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) { 312 return ConstantInt::get(V->getContext(), 313 CI->getValue().trunc(DestBitWidth)); 314 } 315 316 // The input must be a constantexpr. See if we can simplify this based on 317 // the bytes we are demanding. Only do this if the source and dest are an 318 // even multiple of a byte. 319 if ((DestBitWidth & 7) == 0 && 320 (cast<IntegerType>(V->getType())->getBitWidth() & 7) == 0) 321 if (Constant *Res = ExtractConstantBytes(V, 0, DestBitWidth / 8)) 322 return Res; 323 324 return nullptr; 325 } 326 case Instruction::BitCast: 327 return FoldBitCast(V, DestTy); 328 case Instruction::AddrSpaceCast: 329 case Instruction::IntToPtr: 330 case Instruction::PtrToInt: 331 return nullptr; 332 } 333 } 334 335 Constant *llvm::ConstantFoldSelectInstruction(Constant *Cond, 336 Constant *V1, Constant *V2) { 337 // Check for i1 and vector true/false conditions. 338 if (Cond->isNullValue()) return V2; 339 if (Cond->isAllOnesValue()) return V1; 340 341 // If the condition is a vector constant, fold the result elementwise. 342 if (ConstantVector *CondV = dyn_cast<ConstantVector>(Cond)) { 343 auto *V1VTy = CondV->getType(); 344 SmallVector<Constant*, 16> Result; 345 Type *Ty = IntegerType::get(CondV->getContext(), 32); 346 for (unsigned i = 0, e = V1VTy->getNumElements(); i != e; ++i) { 347 Constant *V; 348 Constant *V1Element = ConstantExpr::getExtractElement(V1, 349 ConstantInt::get(Ty, i)); 350 Constant *V2Element = ConstantExpr::getExtractElement(V2, 351 ConstantInt::get(Ty, i)); 352 auto *Cond = cast<Constant>(CondV->getOperand(i)); 353 if (isa<PoisonValue>(Cond)) { 354 V = PoisonValue::get(V1Element->getType()); 355 } else if (V1Element == V2Element) { 356 V = V1Element; 357 } else if (isa<UndefValue>(Cond)) { 358 V = isa<UndefValue>(V1Element) ? V1Element : V2Element; 359 } else { 360 if (!isa<ConstantInt>(Cond)) break; 361 V = Cond->isNullValue() ? V2Element : V1Element; 362 } 363 Result.push_back(V); 364 } 365 366 // If we were able to build the vector, return it. 367 if (Result.size() == V1VTy->getNumElements()) 368 return ConstantVector::get(Result); 369 } 370 371 if (isa<PoisonValue>(Cond)) 372 return PoisonValue::get(V1->getType()); 373 374 if (isa<UndefValue>(Cond)) { 375 if (isa<UndefValue>(V1)) return V1; 376 return V2; 377 } 378 379 if (V1 == V2) return V1; 380 381 if (isa<PoisonValue>(V1)) 382 return V2; 383 if (isa<PoisonValue>(V2)) 384 return V1; 385 386 // If the true or false value is undef, we can fold to the other value as 387 // long as the other value isn't poison. 388 auto NotPoison = [](Constant *C) { 389 if (isa<PoisonValue>(C)) 390 return false; 391 392 // TODO: We can analyze ConstExpr by opcode to determine if there is any 393 // possibility of poison. 394 if (isa<ConstantExpr>(C)) 395 return false; 396 397 if (isa<ConstantInt>(C) || isa<GlobalVariable>(C) || isa<ConstantFP>(C) || 398 isa<ConstantPointerNull>(C) || isa<Function>(C)) 399 return true; 400 401 if (C->getType()->isVectorTy()) 402 return !C->containsPoisonElement() && !C->containsConstantExpression(); 403 404 // TODO: Recursively analyze aggregates or other constants. 405 return false; 406 }; 407 if (isa<UndefValue>(V1) && NotPoison(V2)) return V2; 408 if (isa<UndefValue>(V2) && NotPoison(V1)) return V1; 409 410 return nullptr; 411 } 412 413 Constant *llvm::ConstantFoldExtractElementInstruction(Constant *Val, 414 Constant *Idx) { 415 auto *ValVTy = cast<VectorType>(Val->getType()); 416 417 // extractelt poison, C -> poison 418 // extractelt C, undef -> poison 419 if (isa<PoisonValue>(Val) || isa<UndefValue>(Idx)) 420 return PoisonValue::get(ValVTy->getElementType()); 421 422 // extractelt undef, C -> undef 423 if (isa<UndefValue>(Val)) 424 return UndefValue::get(ValVTy->getElementType()); 425 426 auto *CIdx = dyn_cast<ConstantInt>(Idx); 427 if (!CIdx) 428 return nullptr; 429 430 if (auto *ValFVTy = dyn_cast<FixedVectorType>(Val->getType())) { 431 // ee({w,x,y,z}, wrong_value) -> poison 432 if (CIdx->uge(ValFVTy->getNumElements())) 433 return PoisonValue::get(ValFVTy->getElementType()); 434 } 435 436 // ee (gep (ptr, idx0, ...), idx) -> gep (ee (ptr, idx), ee (idx0, idx), ...) 437 if (auto *CE = dyn_cast<ConstantExpr>(Val)) { 438 if (auto *GEP = dyn_cast<GEPOperator>(CE)) { 439 SmallVector<Constant *, 8> Ops; 440 Ops.reserve(CE->getNumOperands()); 441 for (unsigned i = 0, e = CE->getNumOperands(); i != e; ++i) { 442 Constant *Op = CE->getOperand(i); 443 if (Op->getType()->isVectorTy()) { 444 Constant *ScalarOp = ConstantExpr::getExtractElement(Op, Idx); 445 if (!ScalarOp) 446 return nullptr; 447 Ops.push_back(ScalarOp); 448 } else 449 Ops.push_back(Op); 450 } 451 return CE->getWithOperands(Ops, ValVTy->getElementType(), false, 452 GEP->getSourceElementType()); 453 } else if (CE->getOpcode() == Instruction::InsertElement) { 454 if (const auto *IEIdx = dyn_cast<ConstantInt>(CE->getOperand(2))) { 455 if (APSInt::isSameValue(APSInt(IEIdx->getValue()), 456 APSInt(CIdx->getValue()))) { 457 return CE->getOperand(1); 458 } else { 459 return ConstantExpr::getExtractElement(CE->getOperand(0), CIdx); 460 } 461 } 462 } 463 } 464 465 if (Constant *C = Val->getAggregateElement(CIdx)) 466 return C; 467 468 // Lane < Splat minimum vector width => extractelt Splat(x), Lane -> x 469 if (CIdx->getValue().ult(ValVTy->getElementCount().getKnownMinValue())) { 470 if (Constant *SplatVal = Val->getSplatValue()) 471 return SplatVal; 472 } 473 474 return nullptr; 475 } 476 477 Constant *llvm::ConstantFoldInsertElementInstruction(Constant *Val, 478 Constant *Elt, 479 Constant *Idx) { 480 if (isa<UndefValue>(Idx)) 481 return PoisonValue::get(Val->getType()); 482 483 // Inserting null into all zeros is still all zeros. 484 // TODO: This is true for undef and poison splats too. 485 if (isa<ConstantAggregateZero>(Val) && Elt->isNullValue()) 486 return Val; 487 488 ConstantInt *CIdx = dyn_cast<ConstantInt>(Idx); 489 if (!CIdx) return nullptr; 490 491 // Do not iterate on scalable vector. The num of elements is unknown at 492 // compile-time. 493 if (isa<ScalableVectorType>(Val->getType())) 494 return nullptr; 495 496 auto *ValTy = cast<FixedVectorType>(Val->getType()); 497 498 unsigned NumElts = ValTy->getNumElements(); 499 if (CIdx->uge(NumElts)) 500 return PoisonValue::get(Val->getType()); 501 502 SmallVector<Constant*, 16> Result; 503 Result.reserve(NumElts); 504 auto *Ty = Type::getInt32Ty(Val->getContext()); 505 uint64_t IdxVal = CIdx->getZExtValue(); 506 for (unsigned i = 0; i != NumElts; ++i) { 507 if (i == IdxVal) { 508 Result.push_back(Elt); 509 continue; 510 } 511 512 Constant *C = ConstantExpr::getExtractElement(Val, ConstantInt::get(Ty, i)); 513 Result.push_back(C); 514 } 515 516 return ConstantVector::get(Result); 517 } 518 519 Constant *llvm::ConstantFoldShuffleVectorInstruction(Constant *V1, Constant *V2, 520 ArrayRef<int> Mask) { 521 auto *V1VTy = cast<VectorType>(V1->getType()); 522 unsigned MaskNumElts = Mask.size(); 523 auto MaskEltCount = 524 ElementCount::get(MaskNumElts, isa<ScalableVectorType>(V1VTy)); 525 Type *EltTy = V1VTy->getElementType(); 526 527 // Poison shuffle mask -> poison value. 528 if (all_of(Mask, [](int Elt) { return Elt == PoisonMaskElem; })) { 529 return PoisonValue::get(VectorType::get(EltTy, MaskEltCount)); 530 } 531 532 // If the mask is all zeros this is a splat, no need to go through all 533 // elements. 534 if (all_of(Mask, [](int Elt) { return Elt == 0; })) { 535 Type *Ty = IntegerType::get(V1->getContext(), 32); 536 Constant *Elt = 537 ConstantExpr::getExtractElement(V1, ConstantInt::get(Ty, 0)); 538 539 if (Elt->isNullValue()) { 540 auto *VTy = VectorType::get(EltTy, MaskEltCount); 541 return ConstantAggregateZero::get(VTy); 542 } else if (!MaskEltCount.isScalable()) 543 return ConstantVector::getSplat(MaskEltCount, Elt); 544 } 545 // Do not iterate on scalable vector. The num of elements is unknown at 546 // compile-time. 547 if (isa<ScalableVectorType>(V1VTy)) 548 return nullptr; 549 550 unsigned SrcNumElts = V1VTy->getElementCount().getKnownMinValue(); 551 552 // Loop over the shuffle mask, evaluating each element. 553 SmallVector<Constant*, 32> Result; 554 for (unsigned i = 0; i != MaskNumElts; ++i) { 555 int Elt = Mask[i]; 556 if (Elt == -1) { 557 Result.push_back(UndefValue::get(EltTy)); 558 continue; 559 } 560 Constant *InElt; 561 if (unsigned(Elt) >= SrcNumElts*2) 562 InElt = UndefValue::get(EltTy); 563 else if (unsigned(Elt) >= SrcNumElts) { 564 Type *Ty = IntegerType::get(V2->getContext(), 32); 565 InElt = 566 ConstantExpr::getExtractElement(V2, 567 ConstantInt::get(Ty, Elt - SrcNumElts)); 568 } else { 569 Type *Ty = IntegerType::get(V1->getContext(), 32); 570 InElt = ConstantExpr::getExtractElement(V1, ConstantInt::get(Ty, Elt)); 571 } 572 Result.push_back(InElt); 573 } 574 575 return ConstantVector::get(Result); 576 } 577 578 Constant *llvm::ConstantFoldExtractValueInstruction(Constant *Agg, 579 ArrayRef<unsigned> Idxs) { 580 // Base case: no indices, so return the entire value. 581 if (Idxs.empty()) 582 return Agg; 583 584 if (Constant *C = Agg->getAggregateElement(Idxs[0])) 585 return ConstantFoldExtractValueInstruction(C, Idxs.slice(1)); 586 587 return nullptr; 588 } 589 590 Constant *llvm::ConstantFoldInsertValueInstruction(Constant *Agg, 591 Constant *Val, 592 ArrayRef<unsigned> Idxs) { 593 // Base case: no indices, so replace the entire value. 594 if (Idxs.empty()) 595 return Val; 596 597 unsigned NumElts; 598 if (StructType *ST = dyn_cast<StructType>(Agg->getType())) 599 NumElts = ST->getNumElements(); 600 else 601 NumElts = cast<ArrayType>(Agg->getType())->getNumElements(); 602 603 SmallVector<Constant*, 32> Result; 604 for (unsigned i = 0; i != NumElts; ++i) { 605 Constant *C = Agg->getAggregateElement(i); 606 if (!C) return nullptr; 607 608 if (Idxs[0] == i) 609 C = ConstantFoldInsertValueInstruction(C, Val, Idxs.slice(1)); 610 611 Result.push_back(C); 612 } 613 614 if (StructType *ST = dyn_cast<StructType>(Agg->getType())) 615 return ConstantStruct::get(ST, Result); 616 return ConstantArray::get(cast<ArrayType>(Agg->getType()), Result); 617 } 618 619 Constant *llvm::ConstantFoldUnaryInstruction(unsigned Opcode, Constant *C) { 620 assert(Instruction::isUnaryOp(Opcode) && "Non-unary instruction detected"); 621 622 // Handle scalar UndefValue and scalable vector UndefValue. Fixed-length 623 // vectors are always evaluated per element. 624 bool IsScalableVector = isa<ScalableVectorType>(C->getType()); 625 bool HasScalarUndefOrScalableVectorUndef = 626 (!C->getType()->isVectorTy() || IsScalableVector) && isa<UndefValue>(C); 627 628 if (HasScalarUndefOrScalableVectorUndef) { 629 switch (static_cast<Instruction::UnaryOps>(Opcode)) { 630 case Instruction::FNeg: 631 return C; // -undef -> undef 632 case Instruction::UnaryOpsEnd: 633 llvm_unreachable("Invalid UnaryOp"); 634 } 635 } 636 637 // Constant should not be UndefValue, unless these are vector constants. 638 assert(!HasScalarUndefOrScalableVectorUndef && "Unexpected UndefValue"); 639 // We only have FP UnaryOps right now. 640 assert(!isa<ConstantInt>(C) && "Unexpected Integer UnaryOp"); 641 642 if (ConstantFP *CFP = dyn_cast<ConstantFP>(C)) { 643 const APFloat &CV = CFP->getValueAPF(); 644 switch (Opcode) { 645 default: 646 break; 647 case Instruction::FNeg: 648 return ConstantFP::get(C->getContext(), neg(CV)); 649 } 650 } else if (auto *VTy = dyn_cast<FixedVectorType>(C->getType())) { 651 652 Type *Ty = IntegerType::get(VTy->getContext(), 32); 653 // Fast path for splatted constants. 654 if (Constant *Splat = C->getSplatValue()) 655 if (Constant *Elt = ConstantFoldUnaryInstruction(Opcode, Splat)) 656 return ConstantVector::getSplat(VTy->getElementCount(), Elt); 657 658 // Fold each element and create a vector constant from those constants. 659 SmallVector<Constant *, 16> Result; 660 for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i) { 661 Constant *ExtractIdx = ConstantInt::get(Ty, i); 662 Constant *Elt = ConstantExpr::getExtractElement(C, ExtractIdx); 663 Constant *Res = ConstantFoldUnaryInstruction(Opcode, Elt); 664 if (!Res) 665 return nullptr; 666 Result.push_back(Res); 667 } 668 669 return ConstantVector::get(Result); 670 } 671 672 // We don't know how to fold this. 673 return nullptr; 674 } 675 676 Constant *llvm::ConstantFoldBinaryInstruction(unsigned Opcode, Constant *C1, 677 Constant *C2) { 678 assert(Instruction::isBinaryOp(Opcode) && "Non-binary instruction detected"); 679 680 // Simplify BinOps with their identity values first. They are no-ops and we 681 // can always return the other value, including undef or poison values. 682 // FIXME: remove unnecessary duplicated identity patterns below. 683 // FIXME: Use AllowRHSConstant with getBinOpIdentity to handle additional ops, 684 // like X << 0 = X. 685 Constant *Identity = ConstantExpr::getBinOpIdentity(Opcode, C1->getType()); 686 if (Identity) { 687 if (C1 == Identity) 688 return C2; 689 if (C2 == Identity) 690 return C1; 691 } 692 693 // Binary operations propagate poison. 694 if (isa<PoisonValue>(C1) || isa<PoisonValue>(C2)) 695 return PoisonValue::get(C1->getType()); 696 697 // Handle scalar UndefValue and scalable vector UndefValue. Fixed-length 698 // vectors are always evaluated per element. 699 bool IsScalableVector = isa<ScalableVectorType>(C1->getType()); 700 bool HasScalarUndefOrScalableVectorUndef = 701 (!C1->getType()->isVectorTy() || IsScalableVector) && 702 (isa<UndefValue>(C1) || isa<UndefValue>(C2)); 703 if (HasScalarUndefOrScalableVectorUndef) { 704 switch (static_cast<Instruction::BinaryOps>(Opcode)) { 705 case Instruction::Xor: 706 if (isa<UndefValue>(C1) && isa<UndefValue>(C2)) 707 // Handle undef ^ undef -> 0 special case. This is a common 708 // idiom (misuse). 709 return Constant::getNullValue(C1->getType()); 710 [[fallthrough]]; 711 case Instruction::Add: 712 case Instruction::Sub: 713 return UndefValue::get(C1->getType()); 714 case Instruction::And: 715 if (isa<UndefValue>(C1) && isa<UndefValue>(C2)) // undef & undef -> undef 716 return C1; 717 return Constant::getNullValue(C1->getType()); // undef & X -> 0 718 case Instruction::Mul: { 719 // undef * undef -> undef 720 if (isa<UndefValue>(C1) && isa<UndefValue>(C2)) 721 return C1; 722 const APInt *CV; 723 // X * undef -> undef if X is odd 724 if (match(C1, m_APInt(CV)) || match(C2, m_APInt(CV))) 725 if ((*CV)[0]) 726 return UndefValue::get(C1->getType()); 727 728 // X * undef -> 0 otherwise 729 return Constant::getNullValue(C1->getType()); 730 } 731 case Instruction::SDiv: 732 case Instruction::UDiv: 733 // X / undef -> poison 734 // X / 0 -> poison 735 if (match(C2, m_CombineOr(m_Undef(), m_Zero()))) 736 return PoisonValue::get(C2->getType()); 737 // undef / 1 -> undef 738 if (match(C2, m_One())) 739 return C1; 740 // undef / X -> 0 otherwise 741 return Constant::getNullValue(C1->getType()); 742 case Instruction::URem: 743 case Instruction::SRem: 744 // X % undef -> poison 745 // X % 0 -> poison 746 if (match(C2, m_CombineOr(m_Undef(), m_Zero()))) 747 return PoisonValue::get(C2->getType()); 748 // undef % X -> 0 otherwise 749 return Constant::getNullValue(C1->getType()); 750 case Instruction::Or: // X | undef -> -1 751 if (isa<UndefValue>(C1) && isa<UndefValue>(C2)) // undef | undef -> undef 752 return C1; 753 return Constant::getAllOnesValue(C1->getType()); // undef | X -> ~0 754 case Instruction::LShr: 755 // X >>l undef -> poison 756 if (isa<UndefValue>(C2)) 757 return PoisonValue::get(C2->getType()); 758 // undef >>l 0 -> undef 759 if (match(C2, m_Zero())) 760 return C1; 761 // undef >>l X -> 0 762 return Constant::getNullValue(C1->getType()); 763 case Instruction::AShr: 764 // X >>a undef -> poison 765 if (isa<UndefValue>(C2)) 766 return PoisonValue::get(C2->getType()); 767 // undef >>a 0 -> undef 768 if (match(C2, m_Zero())) 769 return C1; 770 // TODO: undef >>a X -> poison if the shift is exact 771 // undef >>a X -> 0 772 return Constant::getNullValue(C1->getType()); 773 case Instruction::Shl: 774 // X << undef -> undef 775 if (isa<UndefValue>(C2)) 776 return PoisonValue::get(C2->getType()); 777 // undef << 0 -> undef 778 if (match(C2, m_Zero())) 779 return C1; 780 // undef << X -> 0 781 return Constant::getNullValue(C1->getType()); 782 case Instruction::FSub: 783 // -0.0 - undef --> undef (consistent with "fneg undef") 784 if (match(C1, m_NegZeroFP()) && isa<UndefValue>(C2)) 785 return C2; 786 [[fallthrough]]; 787 case Instruction::FAdd: 788 case Instruction::FMul: 789 case Instruction::FDiv: 790 case Instruction::FRem: 791 // [any flop] undef, undef -> undef 792 if (isa<UndefValue>(C1) && isa<UndefValue>(C2)) 793 return C1; 794 // [any flop] C, undef -> NaN 795 // [any flop] undef, C -> NaN 796 // We could potentially specialize NaN/Inf constants vs. 'normal' 797 // constants (possibly differently depending on opcode and operand). This 798 // would allow returning undef sometimes. But it is always safe to fold to 799 // NaN because we can choose the undef operand as NaN, and any FP opcode 800 // with a NaN operand will propagate NaN. 801 return ConstantFP::getNaN(C1->getType()); 802 case Instruction::BinaryOpsEnd: 803 llvm_unreachable("Invalid BinaryOp"); 804 } 805 } 806 807 // Neither constant should be UndefValue, unless these are vector constants. 808 assert((!HasScalarUndefOrScalableVectorUndef) && "Unexpected UndefValue"); 809 810 // Handle simplifications when the RHS is a constant int. 811 if (ConstantInt *CI2 = dyn_cast<ConstantInt>(C2)) { 812 switch (Opcode) { 813 case Instruction::Add: 814 if (CI2->isZero()) return C1; // X + 0 == X 815 break; 816 case Instruction::Sub: 817 if (CI2->isZero()) return C1; // X - 0 == X 818 break; 819 case Instruction::Mul: 820 if (CI2->isZero()) return C2; // X * 0 == 0 821 if (CI2->isOne()) 822 return C1; // X * 1 == X 823 break; 824 case Instruction::UDiv: 825 case Instruction::SDiv: 826 if (CI2->isOne()) 827 return C1; // X / 1 == X 828 if (CI2->isZero()) 829 return PoisonValue::get(CI2->getType()); // X / 0 == poison 830 break; 831 case Instruction::URem: 832 case Instruction::SRem: 833 if (CI2->isOne()) 834 return Constant::getNullValue(CI2->getType()); // X % 1 == 0 835 if (CI2->isZero()) 836 return PoisonValue::get(CI2->getType()); // X % 0 == poison 837 break; 838 case Instruction::And: 839 if (CI2->isZero()) return C2; // X & 0 == 0 840 if (CI2->isMinusOne()) 841 return C1; // X & -1 == X 842 843 if (ConstantExpr *CE1 = dyn_cast<ConstantExpr>(C1)) { 844 // If and'ing the address of a global with a constant, fold it. 845 if (CE1->getOpcode() == Instruction::PtrToInt && 846 isa<GlobalValue>(CE1->getOperand(0))) { 847 GlobalValue *GV = cast<GlobalValue>(CE1->getOperand(0)); 848 849 Align GVAlign; // defaults to 1 850 851 if (Module *TheModule = GV->getParent()) { 852 const DataLayout &DL = TheModule->getDataLayout(); 853 GVAlign = GV->getPointerAlignment(DL); 854 855 // If the function alignment is not specified then assume that it 856 // is 4. 857 // This is dangerous; on x86, the alignment of the pointer 858 // corresponds to the alignment of the function, but might be less 859 // than 4 if it isn't explicitly specified. 860 // However, a fix for this behaviour was reverted because it 861 // increased code size (see https://reviews.llvm.org/D55115) 862 // FIXME: This code should be deleted once existing targets have 863 // appropriate defaults 864 if (isa<Function>(GV) && !DL.getFunctionPtrAlign()) 865 GVAlign = Align(4); 866 } else if (isa<GlobalVariable>(GV)) { 867 GVAlign = cast<GlobalVariable>(GV)->getAlign().valueOrOne(); 868 } 869 870 if (GVAlign > 1) { 871 unsigned DstWidth = CI2->getBitWidth(); 872 unsigned SrcWidth = std::min(DstWidth, Log2(GVAlign)); 873 APInt BitsNotSet(APInt::getLowBitsSet(DstWidth, SrcWidth)); 874 875 // If checking bits we know are clear, return zero. 876 if ((CI2->getValue() & BitsNotSet) == CI2->getValue()) 877 return Constant::getNullValue(CI2->getType()); 878 } 879 } 880 } 881 break; 882 case Instruction::Or: 883 if (CI2->isZero()) return C1; // X | 0 == X 884 if (CI2->isMinusOne()) 885 return C2; // X | -1 == -1 886 break; 887 case Instruction::Xor: 888 if (CI2->isZero()) return C1; // X ^ 0 == X 889 890 if (ConstantExpr *CE1 = dyn_cast<ConstantExpr>(C1)) { 891 switch (CE1->getOpcode()) { 892 default: break; 893 case Instruction::ICmp: 894 case Instruction::FCmp: 895 // cmp pred ^ true -> cmp !pred 896 assert(CI2->isOne()); 897 CmpInst::Predicate pred = (CmpInst::Predicate)CE1->getPredicate(); 898 pred = CmpInst::getInversePredicate(pred); 899 return ConstantExpr::getCompare(pred, CE1->getOperand(0), 900 CE1->getOperand(1)); 901 } 902 } 903 break; 904 } 905 } else if (isa<ConstantInt>(C1)) { 906 // If C1 is a ConstantInt and C2 is not, swap the operands. 907 if (Instruction::isCommutative(Opcode)) 908 return ConstantExpr::isDesirableBinOp(Opcode) 909 ? ConstantExpr::get(Opcode, C2, C1) 910 : ConstantFoldBinaryInstruction(Opcode, C2, C1); 911 } 912 913 if (ConstantInt *CI1 = dyn_cast<ConstantInt>(C1)) { 914 if (ConstantInt *CI2 = dyn_cast<ConstantInt>(C2)) { 915 const APInt &C1V = CI1->getValue(); 916 const APInt &C2V = CI2->getValue(); 917 switch (Opcode) { 918 default: 919 break; 920 case Instruction::Add: 921 return ConstantInt::get(CI1->getContext(), C1V + C2V); 922 case Instruction::Sub: 923 return ConstantInt::get(CI1->getContext(), C1V - C2V); 924 case Instruction::Mul: 925 return ConstantInt::get(CI1->getContext(), C1V * C2V); 926 case Instruction::UDiv: 927 assert(!CI2->isZero() && "Div by zero handled above"); 928 return ConstantInt::get(CI1->getContext(), C1V.udiv(C2V)); 929 case Instruction::SDiv: 930 assert(!CI2->isZero() && "Div by zero handled above"); 931 if (C2V.isAllOnes() && C1V.isMinSignedValue()) 932 return PoisonValue::get(CI1->getType()); // MIN_INT / -1 -> poison 933 return ConstantInt::get(CI1->getContext(), C1V.sdiv(C2V)); 934 case Instruction::URem: 935 assert(!CI2->isZero() && "Div by zero handled above"); 936 return ConstantInt::get(CI1->getContext(), C1V.urem(C2V)); 937 case Instruction::SRem: 938 assert(!CI2->isZero() && "Div by zero handled above"); 939 if (C2V.isAllOnes() && C1V.isMinSignedValue()) 940 return PoisonValue::get(CI1->getType()); // MIN_INT % -1 -> poison 941 return ConstantInt::get(CI1->getContext(), C1V.srem(C2V)); 942 case Instruction::And: 943 return ConstantInt::get(CI1->getContext(), C1V & C2V); 944 case Instruction::Or: 945 return ConstantInt::get(CI1->getContext(), C1V | C2V); 946 case Instruction::Xor: 947 return ConstantInt::get(CI1->getContext(), C1V ^ C2V); 948 case Instruction::Shl: 949 if (C2V.ult(C1V.getBitWidth())) 950 return ConstantInt::get(CI1->getContext(), C1V.shl(C2V)); 951 return PoisonValue::get(C1->getType()); // too big shift is poison 952 case Instruction::LShr: 953 if (C2V.ult(C1V.getBitWidth())) 954 return ConstantInt::get(CI1->getContext(), C1V.lshr(C2V)); 955 return PoisonValue::get(C1->getType()); // too big shift is poison 956 case Instruction::AShr: 957 if (C2V.ult(C1V.getBitWidth())) 958 return ConstantInt::get(CI1->getContext(), C1V.ashr(C2V)); 959 return PoisonValue::get(C1->getType()); // too big shift is poison 960 } 961 } 962 963 switch (Opcode) { 964 case Instruction::SDiv: 965 case Instruction::UDiv: 966 case Instruction::URem: 967 case Instruction::SRem: 968 case Instruction::LShr: 969 case Instruction::AShr: 970 case Instruction::Shl: 971 if (CI1->isZero()) return C1; 972 break; 973 default: 974 break; 975 } 976 } else if (ConstantFP *CFP1 = dyn_cast<ConstantFP>(C1)) { 977 if (ConstantFP *CFP2 = dyn_cast<ConstantFP>(C2)) { 978 const APFloat &C1V = CFP1->getValueAPF(); 979 const APFloat &C2V = CFP2->getValueAPF(); 980 APFloat C3V = C1V; // copy for modification 981 switch (Opcode) { 982 default: 983 break; 984 case Instruction::FAdd: 985 (void)C3V.add(C2V, APFloat::rmNearestTiesToEven); 986 return ConstantFP::get(C1->getContext(), C3V); 987 case Instruction::FSub: 988 (void)C3V.subtract(C2V, APFloat::rmNearestTiesToEven); 989 return ConstantFP::get(C1->getContext(), C3V); 990 case Instruction::FMul: 991 (void)C3V.multiply(C2V, APFloat::rmNearestTiesToEven); 992 return ConstantFP::get(C1->getContext(), C3V); 993 case Instruction::FDiv: 994 (void)C3V.divide(C2V, APFloat::rmNearestTiesToEven); 995 return ConstantFP::get(C1->getContext(), C3V); 996 case Instruction::FRem: 997 (void)C3V.mod(C2V); 998 return ConstantFP::get(C1->getContext(), C3V); 999 } 1000 } 1001 } else if (auto *VTy = dyn_cast<VectorType>(C1->getType())) { 1002 // Fast path for splatted constants. 1003 if (Constant *C2Splat = C2->getSplatValue()) { 1004 if (Instruction::isIntDivRem(Opcode) && C2Splat->isNullValue()) 1005 return PoisonValue::get(VTy); 1006 if (Constant *C1Splat = C1->getSplatValue()) { 1007 Constant *Res = 1008 ConstantExpr::isDesirableBinOp(Opcode) 1009 ? ConstantExpr::get(Opcode, C1Splat, C2Splat) 1010 : ConstantFoldBinaryInstruction(Opcode, C1Splat, C2Splat); 1011 if (!Res) 1012 return nullptr; 1013 return ConstantVector::getSplat(VTy->getElementCount(), Res); 1014 } 1015 } 1016 1017 if (auto *FVTy = dyn_cast<FixedVectorType>(VTy)) { 1018 // Fold each element and create a vector constant from those constants. 1019 SmallVector<Constant*, 16> Result; 1020 Type *Ty = IntegerType::get(FVTy->getContext(), 32); 1021 for (unsigned i = 0, e = FVTy->getNumElements(); i != e; ++i) { 1022 Constant *ExtractIdx = ConstantInt::get(Ty, i); 1023 Constant *LHS = ConstantExpr::getExtractElement(C1, ExtractIdx); 1024 Constant *RHS = ConstantExpr::getExtractElement(C2, ExtractIdx); 1025 1026 // If any element of a divisor vector is zero, the whole op is poison. 1027 if (Instruction::isIntDivRem(Opcode) && RHS->isNullValue()) 1028 return PoisonValue::get(VTy); 1029 1030 Constant *Res = ConstantExpr::isDesirableBinOp(Opcode) 1031 ? ConstantExpr::get(Opcode, LHS, RHS) 1032 : ConstantFoldBinaryInstruction(Opcode, LHS, RHS); 1033 if (!Res) 1034 return nullptr; 1035 Result.push_back(Res); 1036 } 1037 1038 return ConstantVector::get(Result); 1039 } 1040 } 1041 1042 if (ConstantExpr *CE1 = dyn_cast<ConstantExpr>(C1)) { 1043 // There are many possible foldings we could do here. We should probably 1044 // at least fold add of a pointer with an integer into the appropriate 1045 // getelementptr. This will improve alias analysis a bit. 1046 1047 // Given ((a + b) + c), if (b + c) folds to something interesting, return 1048 // (a + (b + c)). 1049 if (Instruction::isAssociative(Opcode) && CE1->getOpcode() == Opcode) { 1050 Constant *T = ConstantExpr::get(Opcode, CE1->getOperand(1), C2); 1051 if (!isa<ConstantExpr>(T) || cast<ConstantExpr>(T)->getOpcode() != Opcode) 1052 return ConstantExpr::get(Opcode, CE1->getOperand(0), T); 1053 } 1054 } else if (isa<ConstantExpr>(C2)) { 1055 // If C2 is a constant expr and C1 isn't, flop them around and fold the 1056 // other way if possible. 1057 if (Instruction::isCommutative(Opcode)) 1058 return ConstantFoldBinaryInstruction(Opcode, C2, C1); 1059 } 1060 1061 // i1 can be simplified in many cases. 1062 if (C1->getType()->isIntegerTy(1)) { 1063 switch (Opcode) { 1064 case Instruction::Add: 1065 case Instruction::Sub: 1066 return ConstantExpr::getXor(C1, C2); 1067 case Instruction::Shl: 1068 case Instruction::LShr: 1069 case Instruction::AShr: 1070 // We can assume that C2 == 0. If it were one the result would be 1071 // undefined because the shift value is as large as the bitwidth. 1072 return C1; 1073 case Instruction::SDiv: 1074 case Instruction::UDiv: 1075 // We can assume that C2 == 1. If it were zero the result would be 1076 // undefined through division by zero. 1077 return C1; 1078 case Instruction::URem: 1079 case Instruction::SRem: 1080 // We can assume that C2 == 1. If it were zero the result would be 1081 // undefined through division by zero. 1082 return ConstantInt::getFalse(C1->getContext()); 1083 default: 1084 break; 1085 } 1086 } 1087 1088 // We don't know how to fold this. 1089 return nullptr; 1090 } 1091 1092 static ICmpInst::Predicate areGlobalsPotentiallyEqual(const GlobalValue *GV1, 1093 const GlobalValue *GV2) { 1094 auto isGlobalUnsafeForEquality = [](const GlobalValue *GV) { 1095 if (GV->isInterposable() || GV->hasGlobalUnnamedAddr()) 1096 return true; 1097 if (const auto *GVar = dyn_cast<GlobalVariable>(GV)) { 1098 Type *Ty = GVar->getValueType(); 1099 // A global with opaque type might end up being zero sized. 1100 if (!Ty->isSized()) 1101 return true; 1102 // A global with an empty type might lie at the address of any other 1103 // global. 1104 if (Ty->isEmptyTy()) 1105 return true; 1106 } 1107 return false; 1108 }; 1109 // Don't try to decide equality of aliases. 1110 if (!isa<GlobalAlias>(GV1) && !isa<GlobalAlias>(GV2)) 1111 if (!isGlobalUnsafeForEquality(GV1) && !isGlobalUnsafeForEquality(GV2)) 1112 return ICmpInst::ICMP_NE; 1113 return ICmpInst::BAD_ICMP_PREDICATE; 1114 } 1115 1116 /// This function determines if there is anything we can decide about the two 1117 /// constants provided. This doesn't need to handle simple things like integer 1118 /// comparisons, but should instead handle ConstantExprs and GlobalValues. 1119 /// If we can determine that the two constants have a particular relation to 1120 /// each other, we should return the corresponding ICmp predicate, otherwise 1121 /// return ICmpInst::BAD_ICMP_PREDICATE. 1122 static ICmpInst::Predicate evaluateICmpRelation(Constant *V1, Constant *V2) { 1123 assert(V1->getType() == V2->getType() && 1124 "Cannot compare different types of values!"); 1125 if (V1 == V2) return ICmpInst::ICMP_EQ; 1126 1127 // The following folds only apply to pointers. 1128 if (!V1->getType()->isPointerTy()) 1129 return ICmpInst::BAD_ICMP_PREDICATE; 1130 1131 // To simplify this code we canonicalize the relation so that the first 1132 // operand is always the most "complex" of the two. We consider simple 1133 // constants (like ConstantPointerNull) to be the simplest, followed by 1134 // BlockAddress, GlobalValues, and ConstantExpr's (the most complex). 1135 auto GetComplexity = [](Constant *V) { 1136 if (isa<ConstantExpr>(V)) 1137 return 3; 1138 if (isa<GlobalValue>(V)) 1139 return 2; 1140 if (isa<BlockAddress>(V)) 1141 return 1; 1142 return 0; 1143 }; 1144 if (GetComplexity(V1) < GetComplexity(V2)) { 1145 ICmpInst::Predicate SwappedRelation = evaluateICmpRelation(V2, V1); 1146 if (SwappedRelation != ICmpInst::BAD_ICMP_PREDICATE) 1147 return ICmpInst::getSwappedPredicate(SwappedRelation); 1148 return ICmpInst::BAD_ICMP_PREDICATE; 1149 } 1150 1151 if (const BlockAddress *BA = dyn_cast<BlockAddress>(V1)) { 1152 // Now we know that the RHS is a BlockAddress or simple constant. 1153 if (const BlockAddress *BA2 = dyn_cast<BlockAddress>(V2)) { 1154 // Block address in another function can't equal this one, but block 1155 // addresses in the current function might be the same if blocks are 1156 // empty. 1157 if (BA2->getFunction() != BA->getFunction()) 1158 return ICmpInst::ICMP_NE; 1159 } else if (isa<ConstantPointerNull>(V2)) { 1160 return ICmpInst::ICMP_NE; 1161 } 1162 } else if (const GlobalValue *GV = dyn_cast<GlobalValue>(V1)) { 1163 // Now we know that the RHS is a GlobalValue, BlockAddress or simple 1164 // constant. 1165 if (const GlobalValue *GV2 = dyn_cast<GlobalValue>(V2)) { 1166 return areGlobalsPotentiallyEqual(GV, GV2); 1167 } else if (isa<BlockAddress>(V2)) { 1168 return ICmpInst::ICMP_NE; // Globals never equal labels. 1169 } else if (isa<ConstantPointerNull>(V2)) { 1170 // GlobalVals can never be null unless they have external weak linkage. 1171 // We don't try to evaluate aliases here. 1172 // NOTE: We should not be doing this constant folding if null pointer 1173 // is considered valid for the function. But currently there is no way to 1174 // query it from the Constant type. 1175 if (!GV->hasExternalWeakLinkage() && !isa<GlobalAlias>(GV) && 1176 !NullPointerIsDefined(nullptr /* F */, 1177 GV->getType()->getAddressSpace())) 1178 return ICmpInst::ICMP_UGT; 1179 } 1180 } else { 1181 // Ok, the LHS is known to be a constantexpr. The RHS can be any of a 1182 // constantexpr, a global, block address, or a simple constant. 1183 ConstantExpr *CE1 = cast<ConstantExpr>(V1); 1184 Constant *CE1Op0 = CE1->getOperand(0); 1185 1186 switch (CE1->getOpcode()) { 1187 case Instruction::GetElementPtr: { 1188 GEPOperator *CE1GEP = cast<GEPOperator>(CE1); 1189 // Ok, since this is a getelementptr, we know that the constant has a 1190 // pointer type. Check the various cases. 1191 if (isa<ConstantPointerNull>(V2)) { 1192 // If we are comparing a GEP to a null pointer, check to see if the base 1193 // of the GEP equals the null pointer. 1194 if (const GlobalValue *GV = dyn_cast<GlobalValue>(CE1Op0)) { 1195 // If its not weak linkage, the GVal must have a non-zero address 1196 // so the result is greater-than 1197 if (!GV->hasExternalWeakLinkage() && CE1GEP->isInBounds()) 1198 return ICmpInst::ICMP_UGT; 1199 } 1200 } else if (const GlobalValue *GV2 = dyn_cast<GlobalValue>(V2)) { 1201 if (const GlobalValue *GV = dyn_cast<GlobalValue>(CE1Op0)) { 1202 if (GV != GV2) { 1203 if (CE1GEP->hasAllZeroIndices()) 1204 return areGlobalsPotentiallyEqual(GV, GV2); 1205 return ICmpInst::BAD_ICMP_PREDICATE; 1206 } 1207 } 1208 } else if (const auto *CE2GEP = dyn_cast<GEPOperator>(V2)) { 1209 // By far the most common case to handle is when the base pointers are 1210 // obviously to the same global. 1211 const Constant *CE2Op0 = cast<Constant>(CE2GEP->getPointerOperand()); 1212 if (isa<GlobalValue>(CE1Op0) && isa<GlobalValue>(CE2Op0)) { 1213 // Don't know relative ordering, but check for inequality. 1214 if (CE1Op0 != CE2Op0) { 1215 if (CE1GEP->hasAllZeroIndices() && CE2GEP->hasAllZeroIndices()) 1216 return areGlobalsPotentiallyEqual(cast<GlobalValue>(CE1Op0), 1217 cast<GlobalValue>(CE2Op0)); 1218 return ICmpInst::BAD_ICMP_PREDICATE; 1219 } 1220 } 1221 } 1222 break; 1223 } 1224 default: 1225 break; 1226 } 1227 } 1228 1229 return ICmpInst::BAD_ICMP_PREDICATE; 1230 } 1231 1232 Constant *llvm::ConstantFoldCompareInstruction(CmpInst::Predicate Predicate, 1233 Constant *C1, Constant *C2) { 1234 Type *ResultTy; 1235 if (VectorType *VT = dyn_cast<VectorType>(C1->getType())) 1236 ResultTy = VectorType::get(Type::getInt1Ty(C1->getContext()), 1237 VT->getElementCount()); 1238 else 1239 ResultTy = Type::getInt1Ty(C1->getContext()); 1240 1241 // Fold FCMP_FALSE/FCMP_TRUE unconditionally. 1242 if (Predicate == FCmpInst::FCMP_FALSE) 1243 return Constant::getNullValue(ResultTy); 1244 1245 if (Predicate == FCmpInst::FCMP_TRUE) 1246 return Constant::getAllOnesValue(ResultTy); 1247 1248 // Handle some degenerate cases first 1249 if (isa<PoisonValue>(C1) || isa<PoisonValue>(C2)) 1250 return PoisonValue::get(ResultTy); 1251 1252 if (isa<UndefValue>(C1) || isa<UndefValue>(C2)) { 1253 bool isIntegerPredicate = ICmpInst::isIntPredicate(Predicate); 1254 // For EQ and NE, we can always pick a value for the undef to make the 1255 // predicate pass or fail, so we can return undef. 1256 // Also, if both operands are undef, we can return undef for int comparison. 1257 if (ICmpInst::isEquality(Predicate) || (isIntegerPredicate && C1 == C2)) 1258 return UndefValue::get(ResultTy); 1259 1260 // Otherwise, for integer compare, pick the same value as the non-undef 1261 // operand, and fold it to true or false. 1262 if (isIntegerPredicate) 1263 return ConstantInt::get(ResultTy, CmpInst::isTrueWhenEqual(Predicate)); 1264 1265 // Choosing NaN for the undef will always make unordered comparison succeed 1266 // and ordered comparison fails. 1267 return ConstantInt::get(ResultTy, CmpInst::isUnordered(Predicate)); 1268 } 1269 1270 if (C2->isNullValue()) { 1271 // The caller is expected to commute the operands if the constant expression 1272 // is C2. 1273 // C1 >= 0 --> true 1274 if (Predicate == ICmpInst::ICMP_UGE) 1275 return Constant::getAllOnesValue(ResultTy); 1276 // C1 < 0 --> false 1277 if (Predicate == ICmpInst::ICMP_ULT) 1278 return Constant::getNullValue(ResultTy); 1279 } 1280 1281 // If the comparison is a comparison between two i1's, simplify it. 1282 if (C1->getType()->isIntegerTy(1)) { 1283 switch (Predicate) { 1284 case ICmpInst::ICMP_EQ: 1285 if (isa<ConstantInt>(C2)) 1286 return ConstantExpr::getXor(C1, ConstantExpr::getNot(C2)); 1287 return ConstantExpr::getXor(ConstantExpr::getNot(C1), C2); 1288 case ICmpInst::ICMP_NE: 1289 return ConstantExpr::getXor(C1, C2); 1290 default: 1291 break; 1292 } 1293 } 1294 1295 if (isa<ConstantInt>(C1) && isa<ConstantInt>(C2)) { 1296 const APInt &V1 = cast<ConstantInt>(C1)->getValue(); 1297 const APInt &V2 = cast<ConstantInt>(C2)->getValue(); 1298 return ConstantInt::get(ResultTy, ICmpInst::compare(V1, V2, Predicate)); 1299 } else if (isa<ConstantFP>(C1) && isa<ConstantFP>(C2)) { 1300 const APFloat &C1V = cast<ConstantFP>(C1)->getValueAPF(); 1301 const APFloat &C2V = cast<ConstantFP>(C2)->getValueAPF(); 1302 return ConstantInt::get(ResultTy, FCmpInst::compare(C1V, C2V, Predicate)); 1303 } else if (auto *C1VTy = dyn_cast<VectorType>(C1->getType())) { 1304 1305 // Fast path for splatted constants. 1306 if (Constant *C1Splat = C1->getSplatValue()) 1307 if (Constant *C2Splat = C2->getSplatValue()) 1308 return ConstantVector::getSplat( 1309 C1VTy->getElementCount(), 1310 ConstantExpr::getCompare(Predicate, C1Splat, C2Splat)); 1311 1312 // Do not iterate on scalable vector. The number of elements is unknown at 1313 // compile-time. 1314 if (isa<ScalableVectorType>(C1VTy)) 1315 return nullptr; 1316 1317 // If we can constant fold the comparison of each element, constant fold 1318 // the whole vector comparison. 1319 SmallVector<Constant*, 4> ResElts; 1320 Type *Ty = IntegerType::get(C1->getContext(), 32); 1321 // Compare the elements, producing an i1 result or constant expr. 1322 for (unsigned I = 0, E = C1VTy->getElementCount().getKnownMinValue(); 1323 I != E; ++I) { 1324 Constant *C1E = 1325 ConstantExpr::getExtractElement(C1, ConstantInt::get(Ty, I)); 1326 Constant *C2E = 1327 ConstantExpr::getExtractElement(C2, ConstantInt::get(Ty, I)); 1328 1329 ResElts.push_back(ConstantExpr::getCompare(Predicate, C1E, C2E)); 1330 } 1331 1332 return ConstantVector::get(ResElts); 1333 } 1334 1335 if (C1->getType()->isFPOrFPVectorTy()) { 1336 if (C1 == C2) { 1337 // We know that C1 == C2 || isUnordered(C1, C2). 1338 if (Predicate == FCmpInst::FCMP_ONE) 1339 return ConstantInt::getFalse(ResultTy); 1340 else if (Predicate == FCmpInst::FCMP_UEQ) 1341 return ConstantInt::getTrue(ResultTy); 1342 } 1343 } else { 1344 // Evaluate the relation between the two constants, per the predicate. 1345 int Result = -1; // -1 = unknown, 0 = known false, 1 = known true. 1346 switch (evaluateICmpRelation(C1, C2)) { 1347 default: llvm_unreachable("Unknown relational!"); 1348 case ICmpInst::BAD_ICMP_PREDICATE: 1349 break; // Couldn't determine anything about these constants. 1350 case ICmpInst::ICMP_EQ: // We know the constants are equal! 1351 // If we know the constants are equal, we can decide the result of this 1352 // computation precisely. 1353 Result = ICmpInst::isTrueWhenEqual(Predicate); 1354 break; 1355 case ICmpInst::ICMP_ULT: 1356 switch (Predicate) { 1357 case ICmpInst::ICMP_ULT: case ICmpInst::ICMP_NE: case ICmpInst::ICMP_ULE: 1358 Result = 1; break; 1359 case ICmpInst::ICMP_UGT: case ICmpInst::ICMP_EQ: case ICmpInst::ICMP_UGE: 1360 Result = 0; break; 1361 default: 1362 break; 1363 } 1364 break; 1365 case ICmpInst::ICMP_SLT: 1366 switch (Predicate) { 1367 case ICmpInst::ICMP_SLT: case ICmpInst::ICMP_NE: case ICmpInst::ICMP_SLE: 1368 Result = 1; break; 1369 case ICmpInst::ICMP_SGT: case ICmpInst::ICMP_EQ: case ICmpInst::ICMP_SGE: 1370 Result = 0; break; 1371 default: 1372 break; 1373 } 1374 break; 1375 case ICmpInst::ICMP_UGT: 1376 switch (Predicate) { 1377 case ICmpInst::ICMP_UGT: case ICmpInst::ICMP_NE: case ICmpInst::ICMP_UGE: 1378 Result = 1; break; 1379 case ICmpInst::ICMP_ULT: case ICmpInst::ICMP_EQ: case ICmpInst::ICMP_ULE: 1380 Result = 0; break; 1381 default: 1382 break; 1383 } 1384 break; 1385 case ICmpInst::ICMP_SGT: 1386 switch (Predicate) { 1387 case ICmpInst::ICMP_SGT: case ICmpInst::ICMP_NE: case ICmpInst::ICMP_SGE: 1388 Result = 1; break; 1389 case ICmpInst::ICMP_SLT: case ICmpInst::ICMP_EQ: case ICmpInst::ICMP_SLE: 1390 Result = 0; break; 1391 default: 1392 break; 1393 } 1394 break; 1395 case ICmpInst::ICMP_ULE: 1396 if (Predicate == ICmpInst::ICMP_UGT) 1397 Result = 0; 1398 if (Predicate == ICmpInst::ICMP_ULT || Predicate == ICmpInst::ICMP_ULE) 1399 Result = 1; 1400 break; 1401 case ICmpInst::ICMP_SLE: 1402 if (Predicate == ICmpInst::ICMP_SGT) 1403 Result = 0; 1404 if (Predicate == ICmpInst::ICMP_SLT || Predicate == ICmpInst::ICMP_SLE) 1405 Result = 1; 1406 break; 1407 case ICmpInst::ICMP_UGE: 1408 if (Predicate == ICmpInst::ICMP_ULT) 1409 Result = 0; 1410 if (Predicate == ICmpInst::ICMP_UGT || Predicate == ICmpInst::ICMP_UGE) 1411 Result = 1; 1412 break; 1413 case ICmpInst::ICMP_SGE: 1414 if (Predicate == ICmpInst::ICMP_SLT) 1415 Result = 0; 1416 if (Predicate == ICmpInst::ICMP_SGT || Predicate == ICmpInst::ICMP_SGE) 1417 Result = 1; 1418 break; 1419 case ICmpInst::ICMP_NE: 1420 if (Predicate == ICmpInst::ICMP_EQ) 1421 Result = 0; 1422 if (Predicate == ICmpInst::ICMP_NE) 1423 Result = 1; 1424 break; 1425 } 1426 1427 // If we evaluated the result, return it now. 1428 if (Result != -1) 1429 return ConstantInt::get(ResultTy, Result); 1430 1431 if ((!isa<ConstantExpr>(C1) && isa<ConstantExpr>(C2)) || 1432 (C1->isNullValue() && !C2->isNullValue())) { 1433 // If C2 is a constant expr and C1 isn't, flip them around and fold the 1434 // other way if possible. 1435 // Also, if C1 is null and C2 isn't, flip them around. 1436 Predicate = ICmpInst::getSwappedPredicate(Predicate); 1437 return ConstantExpr::getICmp(Predicate, C2, C1); 1438 } 1439 } 1440 return nullptr; 1441 } 1442 1443 /// Test whether the given sequence of *normalized* indices is "inbounds". 1444 template<typename IndexTy> 1445 static bool isInBoundsIndices(ArrayRef<IndexTy> Idxs) { 1446 // No indices means nothing that could be out of bounds. 1447 if (Idxs.empty()) return true; 1448 1449 // If the first index is zero, it's in bounds. 1450 if (cast<Constant>(Idxs[0])->isNullValue()) return true; 1451 1452 // If the first index is one and all the rest are zero, it's in bounds, 1453 // by the one-past-the-end rule. 1454 if (auto *CI = dyn_cast<ConstantInt>(Idxs[0])) { 1455 if (!CI->isOne()) 1456 return false; 1457 } else { 1458 auto *CV = cast<ConstantDataVector>(Idxs[0]); 1459 CI = dyn_cast_or_null<ConstantInt>(CV->getSplatValue()); 1460 if (!CI || !CI->isOne()) 1461 return false; 1462 } 1463 1464 for (unsigned i = 1, e = Idxs.size(); i != e; ++i) 1465 if (!cast<Constant>(Idxs[i])->isNullValue()) 1466 return false; 1467 return true; 1468 } 1469 1470 /// Test whether a given ConstantInt is in-range for a SequentialType. 1471 static bool isIndexInRangeOfArrayType(uint64_t NumElements, 1472 const ConstantInt *CI) { 1473 // We cannot bounds check the index if it doesn't fit in an int64_t. 1474 if (CI->getValue().getSignificantBits() > 64) 1475 return false; 1476 1477 // A negative index or an index past the end of our sequential type is 1478 // considered out-of-range. 1479 int64_t IndexVal = CI->getSExtValue(); 1480 if (IndexVal < 0 || (IndexVal != 0 && (uint64_t)IndexVal >= NumElements)) 1481 return false; 1482 1483 // Otherwise, it is in-range. 1484 return true; 1485 } 1486 1487 // Combine Indices - If the source pointer to this getelementptr instruction 1488 // is a getelementptr instruction, combine the indices of the two 1489 // getelementptr instructions into a single instruction. 1490 static Constant *foldGEPOfGEP(GEPOperator *GEP, Type *PointeeTy, bool InBounds, 1491 ArrayRef<Value *> Idxs) { 1492 if (PointeeTy != GEP->getResultElementType()) 1493 return nullptr; 1494 1495 Constant *Idx0 = cast<Constant>(Idxs[0]); 1496 if (Idx0->isNullValue()) { 1497 // Handle the simple case of a zero index. 1498 SmallVector<Value*, 16> NewIndices; 1499 NewIndices.reserve(Idxs.size() + GEP->getNumIndices()); 1500 NewIndices.append(GEP->idx_begin(), GEP->idx_end()); 1501 NewIndices.append(Idxs.begin() + 1, Idxs.end()); 1502 return ConstantExpr::getGetElementPtr( 1503 GEP->getSourceElementType(), cast<Constant>(GEP->getPointerOperand()), 1504 NewIndices, InBounds && GEP->isInBounds(), GEP->getInRangeIndex()); 1505 } 1506 1507 gep_type_iterator LastI = gep_type_end(GEP); 1508 for (gep_type_iterator I = gep_type_begin(GEP), E = gep_type_end(GEP); 1509 I != E; ++I) 1510 LastI = I; 1511 1512 // We can't combine GEPs if the last index is a struct type. 1513 if (!LastI.isSequential()) 1514 return nullptr; 1515 // We could perform the transform with non-constant index, but prefer leaving 1516 // it as GEP of GEP rather than GEP of add for now. 1517 ConstantInt *CI = dyn_cast<ConstantInt>(Idx0); 1518 if (!CI) 1519 return nullptr; 1520 1521 // TODO: This code may be extended to handle vectors as well. 1522 auto *LastIdx = cast<Constant>(GEP->getOperand(GEP->getNumOperands()-1)); 1523 Type *LastIdxTy = LastIdx->getType(); 1524 if (LastIdxTy->isVectorTy()) 1525 return nullptr; 1526 1527 SmallVector<Value*, 16> NewIndices; 1528 NewIndices.reserve(Idxs.size() + GEP->getNumIndices()); 1529 NewIndices.append(GEP->idx_begin(), GEP->idx_end() - 1); 1530 1531 // Add the last index of the source with the first index of the new GEP. 1532 // Make sure to handle the case when they are actually different types. 1533 if (LastIdxTy != Idx0->getType()) { 1534 unsigned CommonExtendedWidth = 1535 std::max(LastIdxTy->getIntegerBitWidth(), 1536 Idx0->getType()->getIntegerBitWidth()); 1537 CommonExtendedWidth = std::max(CommonExtendedWidth, 64U); 1538 1539 Type *CommonTy = 1540 Type::getIntNTy(LastIdxTy->getContext(), CommonExtendedWidth); 1541 if (Idx0->getType() != CommonTy) 1542 Idx0 = ConstantFoldCastInstruction(Instruction::SExt, Idx0, CommonTy); 1543 if (LastIdx->getType() != CommonTy) 1544 LastIdx = 1545 ConstantFoldCastInstruction(Instruction::SExt, LastIdx, CommonTy); 1546 if (!Idx0 || !LastIdx) 1547 return nullptr; 1548 } 1549 1550 NewIndices.push_back(ConstantExpr::get(Instruction::Add, Idx0, LastIdx)); 1551 NewIndices.append(Idxs.begin() + 1, Idxs.end()); 1552 1553 // The combined GEP normally inherits its index inrange attribute from 1554 // the inner GEP, but if the inner GEP's last index was adjusted by the 1555 // outer GEP, any inbounds attribute on that index is invalidated. 1556 std::optional<unsigned> IRIndex = GEP->getInRangeIndex(); 1557 if (IRIndex && *IRIndex == GEP->getNumIndices() - 1) 1558 IRIndex = std::nullopt; 1559 1560 return ConstantExpr::getGetElementPtr( 1561 GEP->getSourceElementType(), cast<Constant>(GEP->getPointerOperand()), 1562 NewIndices, InBounds && GEP->isInBounds(), IRIndex); 1563 } 1564 1565 Constant *llvm::ConstantFoldGetElementPtr(Type *PointeeTy, Constant *C, 1566 bool InBounds, 1567 std::optional<unsigned> InRangeIndex, 1568 ArrayRef<Value *> Idxs) { 1569 if (Idxs.empty()) return C; 1570 1571 Type *GEPTy = GetElementPtrInst::getGEPReturnType( 1572 C, ArrayRef((Value *const *)Idxs.data(), Idxs.size())); 1573 1574 if (isa<PoisonValue>(C)) 1575 return PoisonValue::get(GEPTy); 1576 1577 if (isa<UndefValue>(C)) 1578 // If inbounds, we can choose an out-of-bounds pointer as a base pointer. 1579 return InBounds ? PoisonValue::get(GEPTy) : UndefValue::get(GEPTy); 1580 1581 auto IsNoOp = [&]() { 1582 // Avoid losing inrange information. 1583 if (InRangeIndex) 1584 return false; 1585 1586 return all_of(Idxs, [](Value *Idx) { 1587 Constant *IdxC = cast<Constant>(Idx); 1588 return IdxC->isNullValue() || isa<UndefValue>(IdxC); 1589 }); 1590 }; 1591 if (IsNoOp()) 1592 return GEPTy->isVectorTy() && !C->getType()->isVectorTy() 1593 ? ConstantVector::getSplat( 1594 cast<VectorType>(GEPTy)->getElementCount(), C) 1595 : C; 1596 1597 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C)) 1598 if (auto *GEP = dyn_cast<GEPOperator>(CE)) 1599 if (Constant *C = foldGEPOfGEP(GEP, PointeeTy, InBounds, Idxs)) 1600 return C; 1601 1602 // Check to see if any array indices are not within the corresponding 1603 // notional array or vector bounds. If so, try to determine if they can be 1604 // factored out into preceding dimensions. 1605 SmallVector<Constant *, 8> NewIdxs; 1606 Type *Ty = PointeeTy; 1607 Type *Prev = C->getType(); 1608 auto GEPIter = gep_type_begin(PointeeTy, Idxs); 1609 bool Unknown = 1610 !isa<ConstantInt>(Idxs[0]) && !isa<ConstantDataVector>(Idxs[0]); 1611 for (unsigned i = 1, e = Idxs.size(); i != e; 1612 Prev = Ty, Ty = (++GEPIter).getIndexedType(), ++i) { 1613 if (!isa<ConstantInt>(Idxs[i]) && !isa<ConstantDataVector>(Idxs[i])) { 1614 // We don't know if it's in range or not. 1615 Unknown = true; 1616 continue; 1617 } 1618 if (!isa<ConstantInt>(Idxs[i - 1]) && !isa<ConstantDataVector>(Idxs[i - 1])) 1619 // Skip if the type of the previous index is not supported. 1620 continue; 1621 if (InRangeIndex && i == *InRangeIndex + 1) { 1622 // If an index is marked inrange, we cannot apply this canonicalization to 1623 // the following index, as that will cause the inrange index to point to 1624 // the wrong element. 1625 continue; 1626 } 1627 if (isa<StructType>(Ty)) { 1628 // The verify makes sure that GEPs into a struct are in range. 1629 continue; 1630 } 1631 if (isa<VectorType>(Ty)) { 1632 // There can be awkward padding in after a non-power of two vector. 1633 Unknown = true; 1634 continue; 1635 } 1636 auto *STy = cast<ArrayType>(Ty); 1637 if (ConstantInt *CI = dyn_cast<ConstantInt>(Idxs[i])) { 1638 if (isIndexInRangeOfArrayType(STy->getNumElements(), CI)) 1639 // It's in range, skip to the next index. 1640 continue; 1641 if (CI->isNegative()) { 1642 // It's out of range and negative, don't try to factor it. 1643 Unknown = true; 1644 continue; 1645 } 1646 } else { 1647 auto *CV = cast<ConstantDataVector>(Idxs[i]); 1648 bool InRange = true; 1649 for (unsigned I = 0, E = CV->getNumElements(); I != E; ++I) { 1650 auto *CI = cast<ConstantInt>(CV->getElementAsConstant(I)); 1651 InRange &= isIndexInRangeOfArrayType(STy->getNumElements(), CI); 1652 if (CI->isNegative()) { 1653 Unknown = true; 1654 break; 1655 } 1656 } 1657 if (InRange || Unknown) 1658 // It's in range, skip to the next index. 1659 // It's out of range and negative, don't try to factor it. 1660 continue; 1661 } 1662 if (isa<StructType>(Prev)) { 1663 // It's out of range, but the prior dimension is a struct 1664 // so we can't do anything about it. 1665 Unknown = true; 1666 continue; 1667 } 1668 1669 // Determine the number of elements in our sequential type. 1670 uint64_t NumElements = STy->getArrayNumElements(); 1671 if (!NumElements) { 1672 Unknown = true; 1673 continue; 1674 } 1675 1676 // It's out of range, but we can factor it into the prior 1677 // dimension. 1678 NewIdxs.resize(Idxs.size()); 1679 1680 // Expand the current index or the previous index to a vector from a scalar 1681 // if necessary. 1682 Constant *CurrIdx = cast<Constant>(Idxs[i]); 1683 auto *PrevIdx = 1684 NewIdxs[i - 1] ? NewIdxs[i - 1] : cast<Constant>(Idxs[i - 1]); 1685 bool IsCurrIdxVector = CurrIdx->getType()->isVectorTy(); 1686 bool IsPrevIdxVector = PrevIdx->getType()->isVectorTy(); 1687 bool UseVector = IsCurrIdxVector || IsPrevIdxVector; 1688 1689 if (!IsCurrIdxVector && IsPrevIdxVector) 1690 CurrIdx = ConstantDataVector::getSplat( 1691 cast<FixedVectorType>(PrevIdx->getType())->getNumElements(), CurrIdx); 1692 1693 if (!IsPrevIdxVector && IsCurrIdxVector) 1694 PrevIdx = ConstantDataVector::getSplat( 1695 cast<FixedVectorType>(CurrIdx->getType())->getNumElements(), PrevIdx); 1696 1697 Constant *Factor = 1698 ConstantInt::get(CurrIdx->getType()->getScalarType(), NumElements); 1699 if (UseVector) 1700 Factor = ConstantDataVector::getSplat( 1701 IsPrevIdxVector 1702 ? cast<FixedVectorType>(PrevIdx->getType())->getNumElements() 1703 : cast<FixedVectorType>(CurrIdx->getType())->getNumElements(), 1704 Factor); 1705 1706 NewIdxs[i] = 1707 ConstantFoldBinaryInstruction(Instruction::SRem, CurrIdx, Factor); 1708 1709 Constant *Div = 1710 ConstantFoldBinaryInstruction(Instruction::SDiv, CurrIdx, Factor); 1711 1712 // We're working on either ConstantInt or vectors of ConstantInt, 1713 // so these should always fold. 1714 assert(NewIdxs[i] != nullptr && Div != nullptr && "Should have folded"); 1715 1716 unsigned CommonExtendedWidth = 1717 std::max(PrevIdx->getType()->getScalarSizeInBits(), 1718 Div->getType()->getScalarSizeInBits()); 1719 CommonExtendedWidth = std::max(CommonExtendedWidth, 64U); 1720 1721 // Before adding, extend both operands to i64 to avoid 1722 // overflow trouble. 1723 Type *ExtendedTy = Type::getIntNTy(Div->getContext(), CommonExtendedWidth); 1724 if (UseVector) 1725 ExtendedTy = FixedVectorType::get( 1726 ExtendedTy, 1727 IsPrevIdxVector 1728 ? cast<FixedVectorType>(PrevIdx->getType())->getNumElements() 1729 : cast<FixedVectorType>(CurrIdx->getType())->getNumElements()); 1730 1731 if (!PrevIdx->getType()->isIntOrIntVectorTy(CommonExtendedWidth)) 1732 PrevIdx = 1733 ConstantFoldCastInstruction(Instruction::SExt, PrevIdx, ExtendedTy); 1734 1735 if (!Div->getType()->isIntOrIntVectorTy(CommonExtendedWidth)) 1736 Div = ConstantFoldCastInstruction(Instruction::SExt, Div, ExtendedTy); 1737 1738 assert(PrevIdx && Div && "Should have folded"); 1739 NewIdxs[i - 1] = ConstantExpr::getAdd(PrevIdx, Div); 1740 } 1741 1742 // If we did any factoring, start over with the adjusted indices. 1743 if (!NewIdxs.empty()) { 1744 for (unsigned i = 0, e = Idxs.size(); i != e; ++i) 1745 if (!NewIdxs[i]) NewIdxs[i] = cast<Constant>(Idxs[i]); 1746 return ConstantExpr::getGetElementPtr(PointeeTy, C, NewIdxs, InBounds, 1747 InRangeIndex); 1748 } 1749 1750 // If all indices are known integers and normalized, we can do a simple 1751 // check for the "inbounds" property. 1752 if (!Unknown && !InBounds) 1753 if (auto *GV = dyn_cast<GlobalVariable>(C)) 1754 if (!GV->hasExternalWeakLinkage() && GV->getValueType() == PointeeTy && 1755 isInBoundsIndices(Idxs)) 1756 return ConstantExpr::getGetElementPtr(PointeeTy, C, Idxs, 1757 /*InBounds=*/true, InRangeIndex); 1758 1759 return nullptr; 1760 } 1761