InstCombineCalls.cpp revision a9b2313c13a1bc8cbae751da03a9049ecaf0f918
1//===- InstCombineCalls.cpp -----------------------------------------------===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file implements the visitCall and visitInvoke functions. 11// 12//===----------------------------------------------------------------------===// 13 14#include "InstCombine.h" 15#include "llvm/IntrinsicInst.h" 16#include "llvm/Support/CallSite.h" 17#include "llvm/Target/TargetData.h" 18#include "llvm/Analysis/MemoryBuiltins.h" 19#include "llvm/Transforms/Utils/BuildLibCalls.h" 20using namespace llvm; 21 22/// getPromotedType - Return the specified type promoted as it would be to pass 23/// though a va_arg area. 24static const Type *getPromotedType(const Type *Ty) { 25 if (const IntegerType* ITy = dyn_cast<IntegerType>(Ty)) { 26 if (ITy->getBitWidth() < 32) 27 return Type::getInt32Ty(Ty->getContext()); 28 } 29 return Ty; 30} 31 32/// EnforceKnownAlignment - If the specified pointer points to an object that 33/// we control, modify the object's alignment to PrefAlign. This isn't 34/// often possible though. If alignment is important, a more reliable approach 35/// is to simply align all global variables and allocation instructions to 36/// their preferred alignment from the beginning. 37/// 38static unsigned EnforceKnownAlignment(Value *V, 39 unsigned Align, unsigned PrefAlign) { 40 41 User *U = dyn_cast<User>(V); 42 if (!U) return Align; 43 44 switch (Operator::getOpcode(U)) { 45 default: break; 46 case Instruction::BitCast: 47 return EnforceKnownAlignment(U->getOperand(0), Align, PrefAlign); 48 case Instruction::GetElementPtr: { 49 // If all indexes are zero, it is just the alignment of the base pointer. 50 bool AllZeroOperands = true; 51 for (User::op_iterator i = U->op_begin() + 1, e = U->op_end(); i != e; ++i) 52 if (!isa<Constant>(*i) || 53 !cast<Constant>(*i)->isNullValue()) { 54 AllZeroOperands = false; 55 break; 56 } 57 58 if (AllZeroOperands) { 59 // Treat this like a bitcast. 60 return EnforceKnownAlignment(U->getOperand(0), Align, PrefAlign); 61 } 62 break; 63 } 64 } 65 66 if (GlobalValue *GV = dyn_cast<GlobalValue>(V)) { 67 // If there is a large requested alignment and we can, bump up the alignment 68 // of the global. 69 if (!GV->isDeclaration()) { 70 if (GV->getAlignment() >= PrefAlign) 71 Align = GV->getAlignment(); 72 else { 73 GV->setAlignment(PrefAlign); 74 Align = PrefAlign; 75 } 76 } 77 } else if (AllocaInst *AI = dyn_cast<AllocaInst>(V)) { 78 // If there is a requested alignment and if this is an alloca, round up. 79 if (AI->getAlignment() >= PrefAlign) 80 Align = AI->getAlignment(); 81 else { 82 AI->setAlignment(PrefAlign); 83 Align = PrefAlign; 84 } 85 } 86 87 return Align; 88} 89 90/// GetOrEnforceKnownAlignment - If the specified pointer has an alignment that 91/// we can determine, return it, otherwise return 0. If PrefAlign is specified, 92/// and it is more than the alignment of the ultimate object, see if we can 93/// increase the alignment of the ultimate object, making this check succeed. 94unsigned InstCombiner::GetOrEnforceKnownAlignment(Value *V, 95 unsigned PrefAlign) { 96 unsigned BitWidth = TD ? TD->getTypeSizeInBits(V->getType()) : 97 sizeof(PrefAlign) * CHAR_BIT; 98 APInt Mask = APInt::getAllOnesValue(BitWidth); 99 APInt KnownZero(BitWidth, 0), KnownOne(BitWidth, 0); 100 ComputeMaskedBits(V, Mask, KnownZero, KnownOne); 101 unsigned TrailZ = KnownZero.countTrailingOnes(); 102 unsigned Align = 1u << std::min(BitWidth - 1, TrailZ); 103 104 if (PrefAlign > Align) 105 Align = EnforceKnownAlignment(V, Align, PrefAlign); 106 107 // We don't need to make any adjustment. 108 return Align; 109} 110 111Instruction *InstCombiner::SimplifyMemTransfer(MemIntrinsic *MI) { 112 unsigned DstAlign = GetOrEnforceKnownAlignment(MI->getOperand(1)); 113 unsigned SrcAlign = GetOrEnforceKnownAlignment(MI->getOperand(2)); 114 unsigned MinAlign = std::min(DstAlign, SrcAlign); 115 unsigned CopyAlign = MI->getAlignment(); 116 117 if (CopyAlign < MinAlign) { 118 MI->setAlignment(ConstantInt::get(MI->getAlignmentType(), 119 MinAlign, false)); 120 return MI; 121 } 122 123 // If MemCpyInst length is 1/2/4/8 bytes then replace memcpy with 124 // load/store. 125 ConstantInt *MemOpLength = dyn_cast<ConstantInt>(MI->getOperand(3)); 126 if (MemOpLength == 0) return 0; 127 128 // Source and destination pointer types are always "i8*" for intrinsic. See 129 // if the size is something we can handle with a single primitive load/store. 130 // A single load+store correctly handles overlapping memory in the memmove 131 // case. 132 unsigned Size = MemOpLength->getZExtValue(); 133 if (Size == 0) return MI; // Delete this mem transfer. 134 135 if (Size > 8 || (Size&(Size-1))) 136 return 0; // If not 1/2/4/8 bytes, exit. 137 138 // Use an integer load+store unless we can find something better. 139 unsigned SrcAddrSp = 140 cast<PointerType>(MI->getOperand(2)->getType())->getAddressSpace(); 141 unsigned DstAddrSp = 142 cast<PointerType>(MI->getOperand(1)->getType())->getAddressSpace(); 143 144 const IntegerType* IntType = IntegerType::get(MI->getContext(), Size<<3); 145 Type *NewSrcPtrTy = PointerType::get(IntType, SrcAddrSp); 146 Type *NewDstPtrTy = PointerType::get(IntType, DstAddrSp); 147 148 // Memcpy forces the use of i8* for the source and destination. That means 149 // that if you're using memcpy to move one double around, you'll get a cast 150 // from double* to i8*. We'd much rather use a double load+store rather than 151 // an i64 load+store, here because this improves the odds that the source or 152 // dest address will be promotable. See if we can find a better type than the 153 // integer datatype. 154 Value *StrippedDest = MI->getOperand(1)->stripPointerCasts(); 155 if (StrippedDest != MI->getOperand(1)) { 156 const Type *SrcETy = cast<PointerType>(StrippedDest->getType()) 157 ->getElementType(); 158 if (TD && SrcETy->isSized() && TD->getTypeStoreSize(SrcETy) == Size) { 159 // The SrcETy might be something like {{{double}}} or [1 x double]. Rip 160 // down through these levels if so. 161 while (!SrcETy->isSingleValueType()) { 162 if (const StructType *STy = dyn_cast<StructType>(SrcETy)) { 163 if (STy->getNumElements() == 1) 164 SrcETy = STy->getElementType(0); 165 else 166 break; 167 } else if (const ArrayType *ATy = dyn_cast<ArrayType>(SrcETy)) { 168 if (ATy->getNumElements() == 1) 169 SrcETy = ATy->getElementType(); 170 else 171 break; 172 } else 173 break; 174 } 175 176 if (SrcETy->isSingleValueType()) { 177 NewSrcPtrTy = PointerType::get(SrcETy, SrcAddrSp); 178 NewDstPtrTy = PointerType::get(SrcETy, DstAddrSp); 179 } 180 } 181 } 182 183 184 // If the memcpy/memmove provides better alignment info than we can 185 // infer, use it. 186 SrcAlign = std::max(SrcAlign, CopyAlign); 187 DstAlign = std::max(DstAlign, CopyAlign); 188 189 Value *Src = Builder->CreateBitCast(MI->getOperand(2), NewSrcPtrTy); 190 Value *Dest = Builder->CreateBitCast(MI->getOperand(1), NewDstPtrTy); 191 Instruction *L = new LoadInst(Src, "tmp", MI->isVolatile(), SrcAlign); 192 InsertNewInstBefore(L, *MI); 193 InsertNewInstBefore(new StoreInst(L, Dest, MI->isVolatile(), DstAlign), 194 *MI); 195 196 // Set the size of the copy to 0, it will be deleted on the next iteration. 197 MI->setOperand(3, Constant::getNullValue(MemOpLength->getType())); 198 return MI; 199} 200 201Instruction *InstCombiner::SimplifyMemSet(MemSetInst *MI) { 202 unsigned Alignment = GetOrEnforceKnownAlignment(MI->getDest()); 203 if (MI->getAlignment() < Alignment) { 204 MI->setAlignment(ConstantInt::get(MI->getAlignmentType(), 205 Alignment, false)); 206 return MI; 207 } 208 209 // Extract the length and alignment and fill if they are constant. 210 ConstantInt *LenC = dyn_cast<ConstantInt>(MI->getLength()); 211 ConstantInt *FillC = dyn_cast<ConstantInt>(MI->getValue()); 212 if (!LenC || !FillC || !FillC->getType()->isIntegerTy(8)) 213 return 0; 214 uint64_t Len = LenC->getZExtValue(); 215 Alignment = MI->getAlignment(); 216 217 // If the length is zero, this is a no-op 218 if (Len == 0) return MI; // memset(d,c,0,a) -> noop 219 220 // memset(s,c,n) -> store s, c (for n=1,2,4,8) 221 if (Len <= 8 && isPowerOf2_32((uint32_t)Len)) { 222 const Type *ITy = IntegerType::get(MI->getContext(), Len*8); // n=1 -> i8. 223 224 Value *Dest = MI->getDest(); 225 Dest = Builder->CreateBitCast(Dest, PointerType::getUnqual(ITy)); 226 227 // Alignment 0 is identity for alignment 1 for memset, but not store. 228 if (Alignment == 0) Alignment = 1; 229 230 // Extract the fill value and store. 231 uint64_t Fill = FillC->getZExtValue()*0x0101010101010101ULL; 232 InsertNewInstBefore(new StoreInst(ConstantInt::get(ITy, Fill), 233 Dest, false, Alignment), *MI); 234 235 // Set the size of the copy to 0, it will be deleted on the next iteration. 236 MI->setLength(Constant::getNullValue(LenC->getType())); 237 return MI; 238 } 239 240 return 0; 241} 242 243/// visitCallInst - CallInst simplification. This mostly only handles folding 244/// of intrinsic instructions. For normal calls, it allows visitCallSite to do 245/// the heavy lifting. 246/// 247Instruction *InstCombiner::visitCallInst(CallInst &CI) { 248 if (isFreeCall(&CI)) 249 return visitFree(CI); 250 251 // If the caller function is nounwind, mark the call as nounwind, even if the 252 // callee isn't. 253 if (CI.getParent()->getParent()->doesNotThrow() && 254 !CI.doesNotThrow()) { 255 CI.setDoesNotThrow(); 256 return &CI; 257 } 258 259 IntrinsicInst *II = dyn_cast<IntrinsicInst>(&CI); 260 if (!II) return visitCallSite(&CI); 261 262 // Intrinsics cannot occur in an invoke, so handle them here instead of in 263 // visitCallSite. 264 if (MemIntrinsic *MI = dyn_cast<MemIntrinsic>(II)) { 265 bool Changed = false; 266 267 // memmove/cpy/set of zero bytes is a noop. 268 if (Constant *NumBytes = dyn_cast<Constant>(MI->getLength())) { 269 if (NumBytes->isNullValue()) return EraseInstFromFunction(CI); 270 271 if (ConstantInt *CI = dyn_cast<ConstantInt>(NumBytes)) 272 if (CI->getZExtValue() == 1) { 273 // Replace the instruction with just byte operations. We would 274 // transform other cases to loads/stores, but we don't know if 275 // alignment is sufficient. 276 } 277 } 278 279 // If we have a memmove and the source operation is a constant global, 280 // then the source and dest pointers can't alias, so we can change this 281 // into a call to memcpy. 282 if (MemMoveInst *MMI = dyn_cast<MemMoveInst>(MI)) { 283 if (GlobalVariable *GVSrc = dyn_cast<GlobalVariable>(MMI->getSource())) 284 if (GVSrc->isConstant()) { 285 Module *M = CI.getParent()->getParent()->getParent(); 286 Intrinsic::ID MemCpyID = Intrinsic::memcpy; 287 const Type *Tys[3] = { CI.getOperand(1)->getType(), 288 CI.getOperand(2)->getType(), 289 CI.getOperand(3)->getType() }; 290 CI.setCalledFunction( 291 Intrinsic::getDeclaration(M, MemCpyID, Tys, 3)); 292 Changed = true; 293 } 294 } 295 296 if (MemTransferInst *MTI = dyn_cast<MemTransferInst>(MI)) { 297 // memmove(x,x,size) -> noop. 298 if (MTI->getSource() == MTI->getDest()) 299 return EraseInstFromFunction(CI); 300 } 301 302 // If we can determine a pointer alignment that is bigger than currently 303 // set, update the alignment. 304 if (isa<MemTransferInst>(MI)) { 305 if (Instruction *I = SimplifyMemTransfer(MI)) 306 return I; 307 } else if (MemSetInst *MSI = dyn_cast<MemSetInst>(MI)) { 308 if (Instruction *I = SimplifyMemSet(MSI)) 309 return I; 310 } 311 312 if (Changed) return II; 313 } 314 315 switch (II->getIntrinsicID()) { 316 default: break; 317 case Intrinsic::objectsize: { 318 // We need target data for just about everything so depend on it. 319 if (!TD) break; 320 321 const Type *ReturnTy = CI.getType(); 322 bool Min = (cast<ConstantInt>(II->getOperand(2))->getZExtValue() == 1); 323 324 // Get to the real allocated thing and offset as fast as possible. 325 Value *Op1 = II->getOperand(1)->stripPointerCasts(); 326 327 // If we've stripped down to a single global variable that we 328 // can know the size of then just return that. 329 if (GlobalVariable *GV = dyn_cast<GlobalVariable>(Op1)) { 330 if (GV->hasDefinitiveInitializer()) { 331 Constant *C = GV->getInitializer(); 332 uint64_t GlobalSize = TD->getTypeAllocSize(C->getType()); 333 return ReplaceInstUsesWith(CI, ConstantInt::get(ReturnTy, GlobalSize)); 334 } else { 335 // Can't determine size of the GV. 336 Constant *RetVal = ConstantInt::get(ReturnTy, Min ? 0 : -1ULL); 337 return ReplaceInstUsesWith(CI, RetVal); 338 } 339 } else if (AllocaInst *AI = dyn_cast<AllocaInst>(Op1)) { 340 // Get alloca size. 341 if (AI->getAllocatedType()->isSized()) { 342 uint64_t AllocaSize = TD->getTypeAllocSize(AI->getAllocatedType()); 343 if (AI->isArrayAllocation()) { 344 const ConstantInt *C = dyn_cast<ConstantInt>(AI->getArraySize()); 345 if (!C) break; 346 AllocaSize *= C->getZExtValue(); 347 } 348 return ReplaceInstUsesWith(CI, ConstantInt::get(ReturnTy, AllocaSize)); 349 } 350 } else if (CallInst *MI = extractMallocCall(Op1)) { 351 const Type* MallocType = getMallocAllocatedType(MI); 352 // Get alloca size. 353 if (MallocType && MallocType->isSized()) { 354 if (Value *NElems = getMallocArraySize(MI, TD, true)) { 355 if (ConstantInt *NElements = dyn_cast<ConstantInt>(NElems)) 356 return ReplaceInstUsesWith(CI, ConstantInt::get(ReturnTy, 357 (NElements->getZExtValue() * TD->getTypeAllocSize(MallocType)))); 358 } 359 } 360 } else if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Op1)) { 361 // Only handle constant GEPs here. 362 if (CE->getOpcode() != Instruction::GetElementPtr) break; 363 GEPOperator *GEP = cast<GEPOperator>(CE); 364 365 // Make sure we're not a constant offset from an external 366 // global. 367 Value *Operand = GEP->getPointerOperand(); 368 Operand = Operand->stripPointerCasts(); 369 if (GlobalVariable *GV = dyn_cast<GlobalVariable>(Operand)) 370 if (!GV->hasDefinitiveInitializer()) break; 371 372 // Get what we're pointing to and its size. 373 const PointerType *BaseType = 374 cast<PointerType>(Operand->getType()); 375 uint64_t Size = TD->getTypeAllocSize(BaseType->getElementType()); 376 377 // Get the current byte offset into the thing. Use the original 378 // operand in case we're looking through a bitcast. 379 SmallVector<Value*, 8> Ops(CE->op_begin()+1, CE->op_end()); 380 const PointerType *OffsetType = 381 cast<PointerType>(GEP->getPointerOperand()->getType()); 382 uint64_t Offset = TD->getIndexedOffset(OffsetType, &Ops[0], Ops.size()); 383 384 if (Size < Offset) { 385 // Out of bound reference? Negative index normalized to large 386 // index? Just return "I don't know". 387 Constant *RetVal = ConstantInt::get(ReturnTy, Min ? 0 : -1ULL); 388 return ReplaceInstUsesWith(CI, RetVal); 389 } 390 391 Constant *RetVal = ConstantInt::get(ReturnTy, Size-Offset); 392 return ReplaceInstUsesWith(CI, RetVal); 393 394 } 395 396 // Do not return "I don't know" here. Later optimization passes could 397 // make it possible to evaluate objectsize to a constant. 398 break; 399 } 400 case Intrinsic::bswap: 401 // bswap(bswap(x)) -> x 402 if (IntrinsicInst *Operand = dyn_cast<IntrinsicInst>(II->getOperand(1))) 403 if (Operand->getIntrinsicID() == Intrinsic::bswap) 404 return ReplaceInstUsesWith(CI, Operand->getOperand(1)); 405 406 // bswap(trunc(bswap(x))) -> trunc(lshr(x, c)) 407 if (TruncInst *TI = dyn_cast<TruncInst>(II->getOperand(1))) { 408 if (IntrinsicInst *Operand = dyn_cast<IntrinsicInst>(TI->getOperand(0))) 409 if (Operand->getIntrinsicID() == Intrinsic::bswap) { 410 unsigned C = Operand->getType()->getPrimitiveSizeInBits() - 411 TI->getType()->getPrimitiveSizeInBits(); 412 Value *CV = ConstantInt::get(Operand->getType(), C); 413 Value *V = Builder->CreateLShr(Operand->getOperand(1), CV); 414 return new TruncInst(V, TI->getType()); 415 } 416 } 417 418 break; 419 case Intrinsic::powi: 420 if (ConstantInt *Power = dyn_cast<ConstantInt>(II->getOperand(2))) { 421 // powi(x, 0) -> 1.0 422 if (Power->isZero()) 423 return ReplaceInstUsesWith(CI, ConstantFP::get(CI.getType(), 1.0)); 424 // powi(x, 1) -> x 425 if (Power->isOne()) 426 return ReplaceInstUsesWith(CI, II->getOperand(1)); 427 // powi(x, -1) -> 1/x 428 if (Power->isAllOnesValue()) 429 return BinaryOperator::CreateFDiv(ConstantFP::get(CI.getType(), 1.0), 430 II->getOperand(1)); 431 } 432 break; 433 case Intrinsic::cttz: { 434 // If all bits below the first known one are known zero, 435 // this value is constant. 436 const IntegerType *IT = cast<IntegerType>(II->getOperand(1)->getType()); 437 uint32_t BitWidth = IT->getBitWidth(); 438 APInt KnownZero(BitWidth, 0); 439 APInt KnownOne(BitWidth, 0); 440 ComputeMaskedBits(II->getOperand(1), APInt::getAllOnesValue(BitWidth), 441 KnownZero, KnownOne); 442 unsigned TrailingZeros = KnownOne.countTrailingZeros(); 443 APInt Mask(APInt::getLowBitsSet(BitWidth, TrailingZeros)); 444 if ((Mask & KnownZero) == Mask) 445 return ReplaceInstUsesWith(CI, ConstantInt::get(IT, 446 APInt(BitWidth, TrailingZeros))); 447 448 } 449 break; 450 case Intrinsic::ctlz: { 451 // If all bits above the first known one are known zero, 452 // this value is constant. 453 const IntegerType *IT = cast<IntegerType>(II->getOperand(1)->getType()); 454 uint32_t BitWidth = IT->getBitWidth(); 455 APInt KnownZero(BitWidth, 0); 456 APInt KnownOne(BitWidth, 0); 457 ComputeMaskedBits(II->getOperand(1), APInt::getAllOnesValue(BitWidth), 458 KnownZero, KnownOne); 459 unsigned LeadingZeros = KnownOne.countLeadingZeros(); 460 APInt Mask(APInt::getHighBitsSet(BitWidth, LeadingZeros)); 461 if ((Mask & KnownZero) == Mask) 462 return ReplaceInstUsesWith(CI, ConstantInt::get(IT, 463 APInt(BitWidth, LeadingZeros))); 464 465 } 466 break; 467 case Intrinsic::uadd_with_overflow: { 468 Value *LHS = II->getOperand(1), *RHS = II->getOperand(2); 469 const IntegerType *IT = cast<IntegerType>(II->getOperand(1)->getType()); 470 uint32_t BitWidth = IT->getBitWidth(); 471 APInt Mask = APInt::getSignBit(BitWidth); 472 APInt LHSKnownZero(BitWidth, 0); 473 APInt LHSKnownOne(BitWidth, 0); 474 ComputeMaskedBits(LHS, Mask, LHSKnownZero, LHSKnownOne); 475 bool LHSKnownNegative = LHSKnownOne[BitWidth - 1]; 476 bool LHSKnownPositive = LHSKnownZero[BitWidth - 1]; 477 478 if (LHSKnownNegative || LHSKnownPositive) { 479 APInt RHSKnownZero(BitWidth, 0); 480 APInt RHSKnownOne(BitWidth, 0); 481 ComputeMaskedBits(RHS, Mask, RHSKnownZero, RHSKnownOne); 482 bool RHSKnownNegative = RHSKnownOne[BitWidth - 1]; 483 bool RHSKnownPositive = RHSKnownZero[BitWidth - 1]; 484 if (LHSKnownNegative && RHSKnownNegative) { 485 // The sign bit is set in both cases: this MUST overflow. 486 // Create a simple add instruction, and insert it into the struct. 487 Instruction *Add = BinaryOperator::CreateAdd(LHS, RHS, "", &CI); 488 Worklist.Add(Add); 489 Constant *V[] = { 490 UndefValue::get(LHS->getType()),ConstantInt::getTrue(II->getContext()) 491 }; 492 Constant *Struct = ConstantStruct::get(II->getContext(), V, 2, false); 493 return InsertValueInst::Create(Struct, Add, 0); 494 } 495 496 if (LHSKnownPositive && RHSKnownPositive) { 497 // The sign bit is clear in both cases: this CANNOT overflow. 498 // Create a simple add instruction, and insert it into the struct. 499 Instruction *Add = BinaryOperator::CreateNUWAdd(LHS, RHS, "", &CI); 500 Worklist.Add(Add); 501 Constant *V[] = { 502 UndefValue::get(LHS->getType()), 503 ConstantInt::getFalse(II->getContext()) 504 }; 505 Constant *Struct = ConstantStruct::get(II->getContext(), V, 2, false); 506 return InsertValueInst::Create(Struct, Add, 0); 507 } 508 } 509 } 510 // FALL THROUGH uadd into sadd 511 case Intrinsic::sadd_with_overflow: 512 // Canonicalize constants into the RHS. 513 if (isa<Constant>(II->getOperand(1)) && 514 !isa<Constant>(II->getOperand(2))) { 515 Value *LHS = II->getOperand(1); 516 II->setOperand(1, II->getOperand(2)); 517 II->setOperand(2, LHS); 518 return II; 519 } 520 521 // X + undef -> undef 522 if (isa<UndefValue>(II->getOperand(2))) 523 return ReplaceInstUsesWith(CI, UndefValue::get(II->getType())); 524 525 if (ConstantInt *RHS = dyn_cast<ConstantInt>(II->getOperand(2))) { 526 // X + 0 -> {X, false} 527 if (RHS->isZero()) { 528 Constant *V[] = { 529 UndefValue::get(II->getCalledValue()->getType()), 530 ConstantInt::getFalse(II->getContext()) 531 }; 532 Constant *Struct = ConstantStruct::get(II->getContext(), V, 2, false); 533 return InsertValueInst::Create(Struct, II->getOperand(1), 0); 534 } 535 } 536 break; 537 case Intrinsic::usub_with_overflow: 538 case Intrinsic::ssub_with_overflow: 539 // undef - X -> undef 540 // X - undef -> undef 541 if (isa<UndefValue>(II->getOperand(1)) || 542 isa<UndefValue>(II->getOperand(2))) 543 return ReplaceInstUsesWith(CI, UndefValue::get(II->getType())); 544 545 if (ConstantInt *RHS = dyn_cast<ConstantInt>(II->getOperand(2))) { 546 // X - 0 -> {X, false} 547 if (RHS->isZero()) { 548 Constant *V[] = { 549 UndefValue::get(II->getOperand(1)->getType()), 550 ConstantInt::getFalse(II->getContext()) 551 }; 552 Constant *Struct = ConstantStruct::get(II->getContext(), V, 2, false); 553 return InsertValueInst::Create(Struct, II->getOperand(1), 0); 554 } 555 } 556 break; 557 case Intrinsic::umul_with_overflow: 558 case Intrinsic::smul_with_overflow: 559 // Canonicalize constants into the RHS. 560 if (isa<Constant>(II->getOperand(1)) && 561 !isa<Constant>(II->getOperand(2))) { 562 Value *LHS = II->getOperand(1); 563 II->setOperand(1, II->getOperand(2)); 564 II->setOperand(2, LHS); 565 return II; 566 } 567 568 // X * undef -> undef 569 if (isa<UndefValue>(II->getOperand(2))) 570 return ReplaceInstUsesWith(CI, UndefValue::get(II->getType())); 571 572 if (ConstantInt *RHSI = dyn_cast<ConstantInt>(II->getOperand(2))) { 573 // X*0 -> {0, false} 574 if (RHSI->isZero()) 575 return ReplaceInstUsesWith(CI, Constant::getNullValue(II->getType())); 576 577 // X * 1 -> {X, false} 578 if (RHSI->equalsInt(1)) { 579 Constant *V[] = { 580 UndefValue::get(II->getOperand(1)->getType()), 581 ConstantInt::getFalse(II->getContext()) 582 }; 583 Constant *Struct = ConstantStruct::get(II->getContext(), V, 2, false); 584 return InsertValueInst::Create(Struct, II->getOperand(1), 0); 585 } 586 } 587 break; 588 case Intrinsic::ppc_altivec_lvx: 589 case Intrinsic::ppc_altivec_lvxl: 590 case Intrinsic::x86_sse_loadu_ps: 591 case Intrinsic::x86_sse2_loadu_pd: 592 case Intrinsic::x86_sse2_loadu_dq: 593 // Turn PPC lvx -> load if the pointer is known aligned. 594 // Turn X86 loadups -> load if the pointer is known aligned. 595 if (GetOrEnforceKnownAlignment(II->getOperand(1), 16) >= 16) { 596 Value *Ptr = Builder->CreateBitCast(II->getOperand(1), 597 PointerType::getUnqual(II->getType())); 598 return new LoadInst(Ptr); 599 } 600 break; 601 case Intrinsic::ppc_altivec_stvx: 602 case Intrinsic::ppc_altivec_stvxl: 603 // Turn stvx -> store if the pointer is known aligned. 604 if (GetOrEnforceKnownAlignment(II->getOperand(2), 16) >= 16) { 605 const Type *OpPtrTy = 606 PointerType::getUnqual(II->getOperand(1)->getType()); 607 Value *Ptr = Builder->CreateBitCast(II->getOperand(2), OpPtrTy); 608 return new StoreInst(II->getOperand(1), Ptr); 609 } 610 break; 611 case Intrinsic::x86_sse_storeu_ps: 612 case Intrinsic::x86_sse2_storeu_pd: 613 case Intrinsic::x86_sse2_storeu_dq: 614 // Turn X86 storeu -> store if the pointer is known aligned. 615 if (GetOrEnforceKnownAlignment(II->getOperand(1), 16) >= 16) { 616 const Type *OpPtrTy = 617 PointerType::getUnqual(II->getOperand(2)->getType()); 618 Value *Ptr = Builder->CreateBitCast(II->getOperand(1), OpPtrTy); 619 return new StoreInst(II->getOperand(2), Ptr); 620 } 621 break; 622 623 case Intrinsic::x86_sse_cvttss2si: { 624 // These intrinsics only demands the 0th element of its input vector. If 625 // we can simplify the input based on that, do so now. 626 unsigned VWidth = 627 cast<VectorType>(II->getOperand(1)->getType())->getNumElements(); 628 APInt DemandedElts(VWidth, 1); 629 APInt UndefElts(VWidth, 0); 630 if (Value *V = SimplifyDemandedVectorElts(II->getOperand(1), DemandedElts, 631 UndefElts)) { 632 II->setOperand(1, V); 633 return II; 634 } 635 break; 636 } 637 638 case Intrinsic::ppc_altivec_vperm: 639 // Turn vperm(V1,V2,mask) -> shuffle(V1,V2,mask) if mask is a constant. 640 if (ConstantVector *Mask = dyn_cast<ConstantVector>(II->getOperand(3))) { 641 assert(Mask->getNumOperands() == 16 && "Bad type for intrinsic!"); 642 643 // Check that all of the elements are integer constants or undefs. 644 bool AllEltsOk = true; 645 for (unsigned i = 0; i != 16; ++i) { 646 if (!isa<ConstantInt>(Mask->getOperand(i)) && 647 !isa<UndefValue>(Mask->getOperand(i))) { 648 AllEltsOk = false; 649 break; 650 } 651 } 652 653 if (AllEltsOk) { 654 // Cast the input vectors to byte vectors. 655 Value *Op0 = Builder->CreateBitCast(II->getOperand(1), Mask->getType()); 656 Value *Op1 = Builder->CreateBitCast(II->getOperand(2), Mask->getType()); 657 Value *Result = UndefValue::get(Op0->getType()); 658 659 // Only extract each element once. 660 Value *ExtractedElts[32]; 661 memset(ExtractedElts, 0, sizeof(ExtractedElts)); 662 663 for (unsigned i = 0; i != 16; ++i) { 664 if (isa<UndefValue>(Mask->getOperand(i))) 665 continue; 666 unsigned Idx=cast<ConstantInt>(Mask->getOperand(i))->getZExtValue(); 667 Idx &= 31; // Match the hardware behavior. 668 669 if (ExtractedElts[Idx] == 0) { 670 ExtractedElts[Idx] = 671 Builder->CreateExtractElement(Idx < 16 ? Op0 : Op1, 672 ConstantInt::get(Type::getInt32Ty(II->getContext()), 673 Idx&15, false), "tmp"); 674 } 675 676 // Insert this value into the result vector. 677 Result = Builder->CreateInsertElement(Result, ExtractedElts[Idx], 678 ConstantInt::get(Type::getInt32Ty(II->getContext()), 679 i, false), "tmp"); 680 } 681 return CastInst::Create(Instruction::BitCast, Result, CI.getType()); 682 } 683 } 684 break; 685 686 case Intrinsic::stackrestore: { 687 // If the save is right next to the restore, remove the restore. This can 688 // happen when variable allocas are DCE'd. 689 if (IntrinsicInst *SS = dyn_cast<IntrinsicInst>(II->getOperand(1))) { 690 if (SS->getIntrinsicID() == Intrinsic::stacksave) { 691 BasicBlock::iterator BI = SS; 692 if (&*++BI == II) 693 return EraseInstFromFunction(CI); 694 } 695 } 696 697 // Scan down this block to see if there is another stack restore in the 698 // same block without an intervening call/alloca. 699 BasicBlock::iterator BI = II; 700 TerminatorInst *TI = II->getParent()->getTerminator(); 701 bool CannotRemove = false; 702 for (++BI; &*BI != TI; ++BI) { 703 if (isa<AllocaInst>(BI) || isMalloc(BI)) { 704 CannotRemove = true; 705 break; 706 } 707 if (CallInst *BCI = dyn_cast<CallInst>(BI)) { 708 if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(BCI)) { 709 // If there is a stackrestore below this one, remove this one. 710 if (II->getIntrinsicID() == Intrinsic::stackrestore) 711 return EraseInstFromFunction(CI); 712 // Otherwise, ignore the intrinsic. 713 } else { 714 // If we found a non-intrinsic call, we can't remove the stack 715 // restore. 716 CannotRemove = true; 717 break; 718 } 719 } 720 } 721 722 // If the stack restore is in a return/unwind block and if there are no 723 // allocas or calls between the restore and the return, nuke the restore. 724 if (!CannotRemove && (isa<ReturnInst>(TI) || isa<UnwindInst>(TI))) 725 return EraseInstFromFunction(CI); 726 break; 727 } 728 } 729 730 return visitCallSite(II); 731} 732 733// InvokeInst simplification 734// 735Instruction *InstCombiner::visitInvokeInst(InvokeInst &II) { 736 return visitCallSite(&II); 737} 738 739/// isSafeToEliminateVarargsCast - If this cast does not affect the value 740/// passed through the varargs area, we can eliminate the use of the cast. 741static bool isSafeToEliminateVarargsCast(const CallSite CS, 742 const CastInst * const CI, 743 const TargetData * const TD, 744 const int ix) { 745 if (!CI->isLosslessCast()) 746 return false; 747 748 // The size of ByVal arguments is derived from the type, so we 749 // can't change to a type with a different size. If the size were 750 // passed explicitly we could avoid this check. 751 if (!CS.paramHasAttr(ix, Attribute::ByVal)) 752 return true; 753 754 const Type* SrcTy = 755 cast<PointerType>(CI->getOperand(0)->getType())->getElementType(); 756 const Type* DstTy = cast<PointerType>(CI->getType())->getElementType(); 757 if (!SrcTy->isSized() || !DstTy->isSized()) 758 return false; 759 if (!TD || TD->getTypeAllocSize(SrcTy) != TD->getTypeAllocSize(DstTy)) 760 return false; 761 return true; 762} 763 764namespace { 765class InstCombineFortifiedLibCalls : public SimplifyFortifiedLibCalls { 766 InstCombiner *IC; 767protected: 768 void replaceCall(Value *With) { 769 NewInstruction = IC->ReplaceInstUsesWith(*CI, With); 770 } 771 bool isFoldable(unsigned SizeCIOp, unsigned SizeArgOp, bool isString) const { 772 if (ConstantInt *SizeCI = dyn_cast<ConstantInt>(CI->getOperand(SizeCIOp))) { 773 if (SizeCI->isAllOnesValue()) 774 return true; 775 if (isString) 776 return SizeCI->getZExtValue() >= 777 GetStringLength(CI->getOperand(SizeArgOp)); 778 if (ConstantInt *Arg = dyn_cast<ConstantInt>(CI->getOperand(SizeArgOp))) 779 return SizeCI->getZExtValue() >= Arg->getZExtValue(); 780 } 781 return false; 782 } 783public: 784 InstCombineFortifiedLibCalls(InstCombiner *IC) : IC(IC), NewInstruction(0) { } 785 Instruction *NewInstruction; 786}; 787} // end anonymous namespace 788 789// Try to fold some different type of calls here. 790// Currently we're only working with the checking functions, memcpy_chk, 791// mempcpy_chk, memmove_chk, memset_chk, strcpy_chk, stpcpy_chk, strncpy_chk, 792// strcat_chk and strncat_chk. 793Instruction *InstCombiner::tryOptimizeCall(CallInst *CI, const TargetData *TD) { 794 if (CI->getCalledFunction() == 0) return 0; 795 796 InstCombineFortifiedLibCalls Simplifier(this); 797 Simplifier.fold(CI, TD); 798 return Simplifier.NewInstruction; 799} 800 801// visitCallSite - Improvements for call and invoke instructions. 802// 803Instruction *InstCombiner::visitCallSite(CallSite CS) { 804 bool Changed = false; 805 806 // If the callee is a constexpr cast of a function, attempt to move the cast 807 // to the arguments of the call/invoke. 808 if (transformConstExprCastCall(CS)) return 0; 809 810 Value *Callee = CS.getCalledValue(); 811 812 if (Function *CalleeF = dyn_cast<Function>(Callee)) 813 // If the call and callee calling conventions don't match, this call must 814 // be unreachable, as the call is undefined. 815 if (CalleeF->getCallingConv() != CS.getCallingConv() && 816 // Only do this for calls to a function with a body. A prototype may 817 // not actually end up matching the implementation's calling conv for a 818 // variety of reasons (e.g. it may be written in assembly). 819 !CalleeF->isDeclaration()) { 820 Instruction *OldCall = CS.getInstruction(); 821 new StoreInst(ConstantInt::getTrue(Callee->getContext()), 822 UndefValue::get(Type::getInt1PtrTy(Callee->getContext())), 823 OldCall); 824 // If OldCall dues not return void then replaceAllUsesWith undef. 825 // This allows ValueHandlers and custom metadata to adjust itself. 826 if (!OldCall->getType()->isVoidTy()) 827 OldCall->replaceAllUsesWith(UndefValue::get(OldCall->getType())); 828 if (isa<CallInst>(OldCall)) 829 return EraseInstFromFunction(*OldCall); 830 831 // We cannot remove an invoke, because it would change the CFG, just 832 // change the callee to a null pointer. 833 cast<InvokeInst>(OldCall)->setCalledFunction( 834 Constant::getNullValue(CalleeF->getType())); 835 return 0; 836 } 837 838 if (isa<ConstantPointerNull>(Callee) || isa<UndefValue>(Callee)) { 839 // This instruction is not reachable, just remove it. We insert a store to 840 // undef so that we know that this code is not reachable, despite the fact 841 // that we can't modify the CFG here. 842 new StoreInst(ConstantInt::getTrue(Callee->getContext()), 843 UndefValue::get(Type::getInt1PtrTy(Callee->getContext())), 844 CS.getInstruction()); 845 846 // If CS dues not return void then replaceAllUsesWith undef. 847 // This allows ValueHandlers and custom metadata to adjust itself. 848 if (!CS.getInstruction()->getType()->isVoidTy()) 849 CS.getInstruction()-> 850 replaceAllUsesWith(UndefValue::get(CS.getInstruction()->getType())); 851 852 if (InvokeInst *II = dyn_cast<InvokeInst>(CS.getInstruction())) { 853 // Don't break the CFG, insert a dummy cond branch. 854 BranchInst::Create(II->getNormalDest(), II->getUnwindDest(), 855 ConstantInt::getTrue(Callee->getContext()), II); 856 } 857 return EraseInstFromFunction(*CS.getInstruction()); 858 } 859 860 if (BitCastInst *BC = dyn_cast<BitCastInst>(Callee)) 861 if (IntrinsicInst *In = dyn_cast<IntrinsicInst>(BC->getOperand(0))) 862 if (In->getIntrinsicID() == Intrinsic::init_trampoline) 863 return transformCallThroughTrampoline(CS); 864 865 const PointerType *PTy = cast<PointerType>(Callee->getType()); 866 const FunctionType *FTy = cast<FunctionType>(PTy->getElementType()); 867 if (FTy->isVarArg()) { 868 int ix = FTy->getNumParams() + (isa<InvokeInst>(Callee) ? 3 : 1); 869 // See if we can optimize any arguments passed through the varargs area of 870 // the call. 871 for (CallSite::arg_iterator I = CS.arg_begin()+FTy->getNumParams(), 872 E = CS.arg_end(); I != E; ++I, ++ix) { 873 CastInst *CI = dyn_cast<CastInst>(*I); 874 if (CI && isSafeToEliminateVarargsCast(CS, CI, TD, ix)) { 875 *I = CI->getOperand(0); 876 Changed = true; 877 } 878 } 879 } 880 881 if (isa<InlineAsm>(Callee) && !CS.doesNotThrow()) { 882 // Inline asm calls cannot throw - mark them 'nounwind'. 883 CS.setDoesNotThrow(); 884 Changed = true; 885 } 886 887 // Try to optimize the call if possible, we require TargetData for most of 888 // this. None of these calls are seen as possibly dead so go ahead and 889 // delete the instruction now. 890 if (CallInst *CI = dyn_cast<CallInst>(CS.getInstruction())) { 891 Instruction *I = tryOptimizeCall(CI, TD); 892 // If we changed something return the result, etc. Otherwise let 893 // the fallthrough check. 894 if (I) return EraseInstFromFunction(*I); 895 } 896 897 return Changed ? CS.getInstruction() : 0; 898} 899 900// transformConstExprCastCall - If the callee is a constexpr cast of a function, 901// attempt to move the cast to the arguments of the call/invoke. 902// 903bool InstCombiner::transformConstExprCastCall(CallSite CS) { 904 if (!isa<ConstantExpr>(CS.getCalledValue())) return false; 905 ConstantExpr *CE = cast<ConstantExpr>(CS.getCalledValue()); 906 if (CE->getOpcode() != Instruction::BitCast || 907 !isa<Function>(CE->getOperand(0))) 908 return false; 909 Function *Callee = cast<Function>(CE->getOperand(0)); 910 Instruction *Caller = CS.getInstruction(); 911 const AttrListPtr &CallerPAL = CS.getAttributes(); 912 913 // Okay, this is a cast from a function to a different type. Unless doing so 914 // would cause a type conversion of one of our arguments, change this call to 915 // be a direct call with arguments casted to the appropriate types. 916 // 917 const FunctionType *FT = Callee->getFunctionType(); 918 const Type *OldRetTy = Caller->getType(); 919 const Type *NewRetTy = FT->getReturnType(); 920 921 if (NewRetTy->isStructTy()) 922 return false; // TODO: Handle multiple return values. 923 924 // Check to see if we are changing the return type... 925 if (OldRetTy != NewRetTy) { 926 if (Callee->isDeclaration() && 927 // Conversion is ok if changing from one pointer type to another or from 928 // a pointer to an integer of the same size. 929 !((OldRetTy->isPointerTy() || !TD || 930 OldRetTy == TD->getIntPtrType(Caller->getContext())) && 931 (NewRetTy->isPointerTy() || !TD || 932 NewRetTy == TD->getIntPtrType(Caller->getContext())))) 933 return false; // Cannot transform this return value. 934 935 if (!Caller->use_empty() && 936 // void -> non-void is handled specially 937 !NewRetTy->isVoidTy() && !CastInst::isCastable(NewRetTy, OldRetTy)) 938 return false; // Cannot transform this return value. 939 940 if (!CallerPAL.isEmpty() && !Caller->use_empty()) { 941 Attributes RAttrs = CallerPAL.getRetAttributes(); 942 if (RAttrs & Attribute::typeIncompatible(NewRetTy)) 943 return false; // Attribute not compatible with transformed value. 944 } 945 946 // If the callsite is an invoke instruction, and the return value is used by 947 // a PHI node in a successor, we cannot change the return type of the call 948 // because there is no place to put the cast instruction (without breaking 949 // the critical edge). Bail out in this case. 950 if (!Caller->use_empty()) 951 if (InvokeInst *II = dyn_cast<InvokeInst>(Caller)) 952 for (Value::use_iterator UI = II->use_begin(), E = II->use_end(); 953 UI != E; ++UI) 954 if (PHINode *PN = dyn_cast<PHINode>(*UI)) 955 if (PN->getParent() == II->getNormalDest() || 956 PN->getParent() == II->getUnwindDest()) 957 return false; 958 } 959 960 unsigned NumActualArgs = unsigned(CS.arg_end()-CS.arg_begin()); 961 unsigned NumCommonArgs = std::min(FT->getNumParams(), NumActualArgs); 962 963 CallSite::arg_iterator AI = CS.arg_begin(); 964 for (unsigned i = 0, e = NumCommonArgs; i != e; ++i, ++AI) { 965 const Type *ParamTy = FT->getParamType(i); 966 const Type *ActTy = (*AI)->getType(); 967 968 if (!CastInst::isCastable(ActTy, ParamTy)) 969 return false; // Cannot transform this parameter value. 970 971 if (CallerPAL.getParamAttributes(i + 1) 972 & Attribute::typeIncompatible(ParamTy)) 973 return false; // Attribute not compatible with transformed value. 974 975 // Converting from one pointer type to another or between a pointer and an 976 // integer of the same size is safe even if we do not have a body. 977 bool isConvertible = ActTy == ParamTy || 978 (TD && ((ParamTy->isPointerTy() || 979 ParamTy == TD->getIntPtrType(Caller->getContext())) && 980 (ActTy->isPointerTy() || 981 ActTy == TD->getIntPtrType(Caller->getContext())))); 982 if (Callee->isDeclaration() && !isConvertible) return false; 983 } 984 985 if (FT->getNumParams() < NumActualArgs && !FT->isVarArg() && 986 Callee->isDeclaration()) 987 return false; // Do not delete arguments unless we have a function body. 988 989 if (FT->getNumParams() < NumActualArgs && FT->isVarArg() && 990 !CallerPAL.isEmpty()) 991 // In this case we have more arguments than the new function type, but we 992 // won't be dropping them. Check that these extra arguments have attributes 993 // that are compatible with being a vararg call argument. 994 for (unsigned i = CallerPAL.getNumSlots(); i; --i) { 995 if (CallerPAL.getSlot(i - 1).Index <= FT->getNumParams()) 996 break; 997 Attributes PAttrs = CallerPAL.getSlot(i - 1).Attrs; 998 if (PAttrs & Attribute::VarArgsIncompatible) 999 return false; 1000 } 1001 1002 // Okay, we decided that this is a safe thing to do: go ahead and start 1003 // inserting cast instructions as necessary... 1004 std::vector<Value*> Args; 1005 Args.reserve(NumActualArgs); 1006 SmallVector<AttributeWithIndex, 8> attrVec; 1007 attrVec.reserve(NumCommonArgs); 1008 1009 // Get any return attributes. 1010 Attributes RAttrs = CallerPAL.getRetAttributes(); 1011 1012 // If the return value is not being used, the type may not be compatible 1013 // with the existing attributes. Wipe out any problematic attributes. 1014 RAttrs &= ~Attribute::typeIncompatible(NewRetTy); 1015 1016 // Add the new return attributes. 1017 if (RAttrs) 1018 attrVec.push_back(AttributeWithIndex::get(0, RAttrs)); 1019 1020 AI = CS.arg_begin(); 1021 for (unsigned i = 0; i != NumCommonArgs; ++i, ++AI) { 1022 const Type *ParamTy = FT->getParamType(i); 1023 if ((*AI)->getType() == ParamTy) { 1024 Args.push_back(*AI); 1025 } else { 1026 Instruction::CastOps opcode = CastInst::getCastOpcode(*AI, 1027 false, ParamTy, false); 1028 Args.push_back(Builder->CreateCast(opcode, *AI, ParamTy, "tmp")); 1029 } 1030 1031 // Add any parameter attributes. 1032 if (Attributes PAttrs = CallerPAL.getParamAttributes(i + 1)) 1033 attrVec.push_back(AttributeWithIndex::get(i + 1, PAttrs)); 1034 } 1035 1036 // If the function takes more arguments than the call was taking, add them 1037 // now. 1038 for (unsigned i = NumCommonArgs; i != FT->getNumParams(); ++i) 1039 Args.push_back(Constant::getNullValue(FT->getParamType(i))); 1040 1041 // If we are removing arguments to the function, emit an obnoxious warning. 1042 if (FT->getNumParams() < NumActualArgs) { 1043 if (!FT->isVarArg()) { 1044 errs() << "WARNING: While resolving call to function '" 1045 << Callee->getName() << "' arguments were dropped!\n"; 1046 } else { 1047 // Add all of the arguments in their promoted form to the arg list. 1048 for (unsigned i = FT->getNumParams(); i != NumActualArgs; ++i, ++AI) { 1049 const Type *PTy = getPromotedType((*AI)->getType()); 1050 if (PTy != (*AI)->getType()) { 1051 // Must promote to pass through va_arg area! 1052 Instruction::CastOps opcode = 1053 CastInst::getCastOpcode(*AI, false, PTy, false); 1054 Args.push_back(Builder->CreateCast(opcode, *AI, PTy, "tmp")); 1055 } else { 1056 Args.push_back(*AI); 1057 } 1058 1059 // Add any parameter attributes. 1060 if (Attributes PAttrs = CallerPAL.getParamAttributes(i + 1)) 1061 attrVec.push_back(AttributeWithIndex::get(i + 1, PAttrs)); 1062 } 1063 } 1064 } 1065 1066 if (Attributes FnAttrs = CallerPAL.getFnAttributes()) 1067 attrVec.push_back(AttributeWithIndex::get(~0, FnAttrs)); 1068 1069 if (NewRetTy->isVoidTy()) 1070 Caller->setName(""); // Void type should not have a name. 1071 1072 const AttrListPtr &NewCallerPAL = AttrListPtr::get(attrVec.begin(), 1073 attrVec.end()); 1074 1075 Instruction *NC; 1076 if (InvokeInst *II = dyn_cast<InvokeInst>(Caller)) { 1077 NC = InvokeInst::Create(Callee, II->getNormalDest(), II->getUnwindDest(), 1078 Args.begin(), Args.end(), 1079 Caller->getName(), Caller); 1080 cast<InvokeInst>(NC)->setCallingConv(II->getCallingConv()); 1081 cast<InvokeInst>(NC)->setAttributes(NewCallerPAL); 1082 } else { 1083 NC = CallInst::Create(Callee, Args.begin(), Args.end(), 1084 Caller->getName(), Caller); 1085 CallInst *CI = cast<CallInst>(Caller); 1086 if (CI->isTailCall()) 1087 cast<CallInst>(NC)->setTailCall(); 1088 cast<CallInst>(NC)->setCallingConv(CI->getCallingConv()); 1089 cast<CallInst>(NC)->setAttributes(NewCallerPAL); 1090 } 1091 1092 // Insert a cast of the return type as necessary. 1093 Value *NV = NC; 1094 if (OldRetTy != NV->getType() && !Caller->use_empty()) { 1095 if (!NV->getType()->isVoidTy()) { 1096 Instruction::CastOps opcode = CastInst::getCastOpcode(NC, false, 1097 OldRetTy, false); 1098 NV = NC = CastInst::Create(opcode, NC, OldRetTy, "tmp"); 1099 1100 // If this is an invoke instruction, we should insert it after the first 1101 // non-phi, instruction in the normal successor block. 1102 if (InvokeInst *II = dyn_cast<InvokeInst>(Caller)) { 1103 BasicBlock::iterator I = II->getNormalDest()->getFirstNonPHI(); 1104 InsertNewInstBefore(NC, *I); 1105 } else { 1106 // Otherwise, it's a call, just insert cast right after the call instr 1107 InsertNewInstBefore(NC, *Caller); 1108 } 1109 Worklist.AddUsersToWorkList(*Caller); 1110 } else { 1111 NV = UndefValue::get(Caller->getType()); 1112 } 1113 } 1114 1115 1116 if (!Caller->use_empty()) 1117 Caller->replaceAllUsesWith(NV); 1118 1119 EraseInstFromFunction(*Caller); 1120 return true; 1121} 1122 1123// transformCallThroughTrampoline - Turn a call to a function created by the 1124// init_trampoline intrinsic into a direct call to the underlying function. 1125// 1126Instruction *InstCombiner::transformCallThroughTrampoline(CallSite CS) { 1127 Value *Callee = CS.getCalledValue(); 1128 const PointerType *PTy = cast<PointerType>(Callee->getType()); 1129 const FunctionType *FTy = cast<FunctionType>(PTy->getElementType()); 1130 const AttrListPtr &Attrs = CS.getAttributes(); 1131 1132 // If the call already has the 'nest' attribute somewhere then give up - 1133 // otherwise 'nest' would occur twice after splicing in the chain. 1134 if (Attrs.hasAttrSomewhere(Attribute::Nest)) 1135 return 0; 1136 1137 IntrinsicInst *Tramp = 1138 cast<IntrinsicInst>(cast<BitCastInst>(Callee)->getOperand(0)); 1139 1140 Function *NestF = cast<Function>(Tramp->getOperand(2)->stripPointerCasts()); 1141 const PointerType *NestFPTy = cast<PointerType>(NestF->getType()); 1142 const FunctionType *NestFTy = cast<FunctionType>(NestFPTy->getElementType()); 1143 1144 const AttrListPtr &NestAttrs = NestF->getAttributes(); 1145 if (!NestAttrs.isEmpty()) { 1146 unsigned NestIdx = 1; 1147 const Type *NestTy = 0; 1148 Attributes NestAttr = Attribute::None; 1149 1150 // Look for a parameter marked with the 'nest' attribute. 1151 for (FunctionType::param_iterator I = NestFTy->param_begin(), 1152 E = NestFTy->param_end(); I != E; ++NestIdx, ++I) 1153 if (NestAttrs.paramHasAttr(NestIdx, Attribute::Nest)) { 1154 // Record the parameter type and any other attributes. 1155 NestTy = *I; 1156 NestAttr = NestAttrs.getParamAttributes(NestIdx); 1157 break; 1158 } 1159 1160 if (NestTy) { 1161 Instruction *Caller = CS.getInstruction(); 1162 std::vector<Value*> NewArgs; 1163 NewArgs.reserve(unsigned(CS.arg_end()-CS.arg_begin())+1); 1164 1165 SmallVector<AttributeWithIndex, 8> NewAttrs; 1166 NewAttrs.reserve(Attrs.getNumSlots() + 1); 1167 1168 // Insert the nest argument into the call argument list, which may 1169 // mean appending it. Likewise for attributes. 1170 1171 // Add any result attributes. 1172 if (Attributes Attr = Attrs.getRetAttributes()) 1173 NewAttrs.push_back(AttributeWithIndex::get(0, Attr)); 1174 1175 { 1176 unsigned Idx = 1; 1177 CallSite::arg_iterator I = CS.arg_begin(), E = CS.arg_end(); 1178 do { 1179 if (Idx == NestIdx) { 1180 // Add the chain argument and attributes. 1181 Value *NestVal = Tramp->getOperand(3); 1182 if (NestVal->getType() != NestTy) 1183 NestVal = new BitCastInst(NestVal, NestTy, "nest", Caller); 1184 NewArgs.push_back(NestVal); 1185 NewAttrs.push_back(AttributeWithIndex::get(NestIdx, NestAttr)); 1186 } 1187 1188 if (I == E) 1189 break; 1190 1191 // Add the original argument and attributes. 1192 NewArgs.push_back(*I); 1193 if (Attributes Attr = Attrs.getParamAttributes(Idx)) 1194 NewAttrs.push_back 1195 (AttributeWithIndex::get(Idx + (Idx >= NestIdx), Attr)); 1196 1197 ++Idx, ++I; 1198 } while (1); 1199 } 1200 1201 // Add any function attributes. 1202 if (Attributes Attr = Attrs.getFnAttributes()) 1203 NewAttrs.push_back(AttributeWithIndex::get(~0, Attr)); 1204 1205 // The trampoline may have been bitcast to a bogus type (FTy). 1206 // Handle this by synthesizing a new function type, equal to FTy 1207 // with the chain parameter inserted. 1208 1209 std::vector<const Type*> NewTypes; 1210 NewTypes.reserve(FTy->getNumParams()+1); 1211 1212 // Insert the chain's type into the list of parameter types, which may 1213 // mean appending it. 1214 { 1215 unsigned Idx = 1; 1216 FunctionType::param_iterator I = FTy->param_begin(), 1217 E = FTy->param_end(); 1218 1219 do { 1220 if (Idx == NestIdx) 1221 // Add the chain's type. 1222 NewTypes.push_back(NestTy); 1223 1224 if (I == E) 1225 break; 1226 1227 // Add the original type. 1228 NewTypes.push_back(*I); 1229 1230 ++Idx, ++I; 1231 } while (1); 1232 } 1233 1234 // Replace the trampoline call with a direct call. Let the generic 1235 // code sort out any function type mismatches. 1236 FunctionType *NewFTy = FunctionType::get(FTy->getReturnType(), NewTypes, 1237 FTy->isVarArg()); 1238 Constant *NewCallee = 1239 NestF->getType() == PointerType::getUnqual(NewFTy) ? 1240 NestF : ConstantExpr::getBitCast(NestF, 1241 PointerType::getUnqual(NewFTy)); 1242 const AttrListPtr &NewPAL = AttrListPtr::get(NewAttrs.begin(), 1243 NewAttrs.end()); 1244 1245 Instruction *NewCaller; 1246 if (InvokeInst *II = dyn_cast<InvokeInst>(Caller)) { 1247 NewCaller = InvokeInst::Create(NewCallee, 1248 II->getNormalDest(), II->getUnwindDest(), 1249 NewArgs.begin(), NewArgs.end(), 1250 Caller->getName(), Caller); 1251 cast<InvokeInst>(NewCaller)->setCallingConv(II->getCallingConv()); 1252 cast<InvokeInst>(NewCaller)->setAttributes(NewPAL); 1253 } else { 1254 NewCaller = CallInst::Create(NewCallee, NewArgs.begin(), NewArgs.end(), 1255 Caller->getName(), Caller); 1256 if (cast<CallInst>(Caller)->isTailCall()) 1257 cast<CallInst>(NewCaller)->setTailCall(); 1258 cast<CallInst>(NewCaller)-> 1259 setCallingConv(cast<CallInst>(Caller)->getCallingConv()); 1260 cast<CallInst>(NewCaller)->setAttributes(NewPAL); 1261 } 1262 if (!Caller->getType()->isVoidTy()) 1263 Caller->replaceAllUsesWith(NewCaller); 1264 Caller->eraseFromParent(); 1265 Worklist.Remove(Caller); 1266 return 0; 1267 } 1268 } 1269 1270 // Replace the trampoline call with a direct call. Since there is no 'nest' 1271 // parameter, there is no need to adjust the argument list. Let the generic 1272 // code sort out any function type mismatches. 1273 Constant *NewCallee = 1274 NestF->getType() == PTy ? NestF : 1275 ConstantExpr::getBitCast(NestF, PTy); 1276 CS.setCalledFunction(NewCallee); 1277 return CS.getInstruction(); 1278} 1279 1280