lib/Transforms/InstCombine/InstCombineVectorOps.cpp

   1 //===- InstCombineVectorOps.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 instcombine for ExtractElement, InsertElement and
  11 // ShuffleVector.
  12 //
  13 //===----------------------------------------------------------------------===//
  14
  15 #include "InstCombine.h"
  16 using namespace llvm;
  17
  18 /// CheapToScalarize - Return true if the value is cheaper to scalarize than it
  19 /// is to leave as a vector operation.
  20 static bool CheapToScalarize(Value *V, bool isConstant) {
  21   if (isa<ConstantAggregateZero>(V))
  22     return true;
  23   if (ConstantVector *C = dyn_cast<ConstantVector>(V)) {
  24     if (isConstant) return true;
  25     // If all elts are the same, we can extract.
  26     Constant *Op0 = C->getOperand(0);
  27     for (unsigned i = 1; i < C->getNumOperands(); ++i)
  28       if (C->getOperand(i) != Op0)
  29         return false;
  30     return true;
  31   }
  32   Instruction *I = dyn_cast<Instruction>(V);
  33   if (!I) return false;
  34
  35   // Insert element gets simplified to the inserted element or is deleted if
  36   // this is constant idx extract element and its a constant idx insertelt.
  37   if (I->getOpcode() == Instruction::InsertElement && isConstant &&
  38       isa<ConstantInt>(I->getOperand(2)))
  39     return true;
  40   if (I->getOpcode() == Instruction::Load && I->hasOneUse())
  41     return true;
  42   if (BinaryOperator *BO = dyn_cast<BinaryOperator>(I))
  43     if (BO->hasOneUse() &&
  44         (CheapToScalarize(BO->getOperand(0), isConstant) ||
  45          CheapToScalarize(BO->getOperand(1), isConstant)))
  46       return true;
  47   if (CmpInst *CI = dyn_cast<CmpInst>(I))
  48     if (CI->hasOneUse() &&
  49         (CheapToScalarize(CI->getOperand(0), isConstant) ||
  50          CheapToScalarize(CI->getOperand(1), isConstant)))
  51       return true;
  52
  53   return false;
  54 }
  55
  56 /// getShuffleMask - Read and decode a shufflevector mask.
  57 /// Turn undef elements into negative values.
  58 static std::vector<int> getShuffleMask(const ShuffleVectorInst *SVI) {
  59   unsigned NElts = SVI->getType()->getNumElements();
  60   if (isa<ConstantAggregateZero>(SVI->getOperand(2)))
  61     return std::vector<int>(NElts, 0);
  62   if (isa<UndefValue>(SVI->getOperand(2)))
  63     return std::vector<int>(NElts, -1);
  64
  65   std::vector<int> Result;
  66   const ConstantVector *CP = cast<ConstantVector>(SVI->getOperand(2));
  67   for (User::const_op_iterator i = CP->op_begin(), e = CP->op_end(); i!=e; ++i)
  68     if (isa<UndefValue>(*i))
  69       Result.push_back(-1);  // undef
  70     else
  71       Result.push_back(cast<ConstantInt>(*i)->getZExtValue());
  72   return Result;
  73 }
  74
  75 /// FindScalarElement - Given a vector and an element number, see if the scalar
  76 /// value is already around as a register, for example if it were inserted then
  77 /// extracted from the vector.
  78 static Value *FindScalarElement(Value *V, unsigned EltNo) {
  79   assert(V->getType()->isVectorTy() && "Not looking at a vector?");
  80   const VectorType *PTy = cast<VectorType>(V->getType());
  81   unsigned Width = PTy->getNumElements();
  82   if (EltNo >= Width)  // Out of range access.
  83     return UndefValue::get(PTy->getElementType());
  84
  85   if (isa<UndefValue>(V))
  86     return UndefValue::get(PTy->getElementType());
  87   if (isa<ConstantAggregateZero>(V))
  88     return Constant::getNullValue(PTy->getElementType());
  89   if (ConstantVector *CP = dyn_cast<ConstantVector>(V))
  90     return CP->getOperand(EltNo);
  91
  92   if (InsertElementInst *III = dyn_cast<InsertElementInst>(V)) {
  93     // If this is an insert to a variable element, we don't know what it is.
  94     if (!isa<ConstantInt>(III->getOperand(2)))
  95       return 0;
  96     unsigned IIElt = cast<ConstantInt>(III->getOperand(2))->getZExtValue();
  97
  98     // If this is an insert to the element we are looking for, return the
  99     // inserted value.
 100     if (EltNo == IIElt)
 101       return III->getOperand(1);
 102
 103     // Otherwise, the insertelement doesn't modify the value, recurse on its
 104     // vector input.
 105     return FindScalarElement(III->getOperand(0), EltNo);
 106   }
 107
 108   if (ShuffleVectorInst *SVI = dyn_cast<ShuffleVectorInst>(V)) {
 109     unsigned LHSWidth =
 110       cast<VectorType>(SVI->getOperand(0)->getType())->getNumElements();
 111     int InEl = getShuffleMask(SVI)[EltNo];
 112     if (InEl < 0)
 113       return UndefValue::get(PTy->getElementType());
 114     if (InEl < (int)LHSWidth)
 115       return FindScalarElement(SVI->getOperand(0), InEl);
 116     return FindScalarElement(SVI->getOperand(1), InEl - LHSWidth);
 117   }
 118
 119   // Otherwise, we don't know.
 120   return 0;
 121 }
 122
 123 Instruction *InstCombiner::visitExtractElementInst(ExtractElementInst &EI) {
 124   // If vector val is undef, replace extract with scalar undef.
 125   if (isa<UndefValue>(EI.getOperand(0)))
 126     return ReplaceInstUsesWith(EI, UndefValue::get(EI.getType()));
 127
 128   // If vector val is constant 0, replace extract with scalar 0.
 129   if (isa<ConstantAggregateZero>(EI.getOperand(0)))
 130     return ReplaceInstUsesWith(EI, Constant::getNullValue(EI.getType()));
 131
 132   if (ConstantVector *C = dyn_cast<ConstantVector>(EI.getOperand(0))) {
 133     // If vector val is constant with all elements the same, replace EI with
 134     // that element. When the elements are not identical, we cannot replace yet
 135     // (we do that below, but only when the index is constant).
 136     Constant *op0 = C->getOperand(0);
 137     for (unsigned i = 1; i != C->getNumOperands(); ++i)
 138       if (C->getOperand(i) != op0) {
 139         op0 = 0;
 140         break;
 141       }
 142     if (op0)
 143       return ReplaceInstUsesWith(EI, op0);
 144   }
 145
 146   // If extracting a specified index from the vector, see if we can recursively
 147   // find a previously computed scalar that was inserted into the vector.
 148   if (ConstantInt *IdxC = dyn_cast<ConstantInt>(EI.getOperand(1))) {
 149     unsigned IndexVal = IdxC->getZExtValue();
 150     unsigned VectorWidth = EI.getVectorOperandType()->getNumElements();
 151
 152     // If this is extracting an invalid index, turn this into undef, to avoid
 153     // crashing the code below.
 154     if (IndexVal >= VectorWidth)
 155       return ReplaceInstUsesWith(EI, UndefValue::get(EI.getType()));
 156
 157     // This instruction only demands the single element from the input vector.
 158     // If the input vector has a single use, simplify it based on this use
 159     // property.
 160     if (EI.getOperand(0)->hasOneUse() && VectorWidth != 1) {
 161       APInt UndefElts(VectorWidth, 0);
 162       APInt DemandedMask(VectorWidth, 0);
 163       DemandedMask.set(IndexVal);
 164       if (Value *V = SimplifyDemandedVectorElts(EI.getOperand(0),
 165                                                 DemandedMask, UndefElts)) {
 166         EI.setOperand(0, V);
 167         return &EI;
 168       }
 169     }
 170
 171     if (Value *Elt = FindScalarElement(EI.getOperand(0), IndexVal))
 172       return ReplaceInstUsesWith(EI, Elt);
 173
 174     // If the this extractelement is directly using a bitcast from a vector of
 175     // the same number of elements, see if we can find the source element from
 176     // it.  In this case, we will end up needing to bitcast the scalars.
 177     if (BitCastInst *BCI = dyn_cast<BitCastInst>(EI.getOperand(0))) {
 178       if (const VectorType *VT =
 179           dyn_cast<VectorType>(BCI->getOperand(0)->getType()))
 180         if (VT->getNumElements() == VectorWidth)
 181           if (Value *Elt = FindScalarElement(BCI->getOperand(0), IndexVal))
 182             return new BitCastInst(Elt, EI.getType());
 183     }
 184   }
 185
 186   if (Instruction *I = dyn_cast<Instruction>(EI.getOperand(0))) {
 187     // Push extractelement into predecessor operation if legal and
 188     // profitable to do so
 189     if (BinaryOperator *BO = dyn_cast<BinaryOperator>(I)) {
 190       if (I->hasOneUse() &&
 191           CheapToScalarize(BO, isa<ConstantInt>(EI.getOperand(1)))) {
 192         Value *newEI0 =
 193           Builder->CreateExtractElement(BO->getOperand(0), EI.getOperand(1),
 194                                         EI.getName()+".lhs");
 195         Value *newEI1 =
 196           Builder->CreateExtractElement(BO->getOperand(1), EI.getOperand(1),
 197                                         EI.getName()+".rhs");
 198         return BinaryOperator::Create(BO->getOpcode(), newEI0, newEI1);
 199       }
 200     } else if (InsertElementInst *IE = dyn_cast<InsertElementInst>(I)) {
 201       // Extracting the inserted element?
 202       if (IE->getOperand(2) == EI.getOperand(1))
 203         return ReplaceInstUsesWith(EI, IE->getOperand(1));
 204       // If the inserted and extracted elements are constants, they must not
 205       // be the same value, extract from the pre-inserted value instead.
 206       if (isa<Constant>(IE->getOperand(2)) && isa<Constant>(EI.getOperand(1))) {
 207         Worklist.AddValue(EI.getOperand(0));
 208         EI.setOperand(0, IE->getOperand(0));
 209         return &EI;
 210       }
 211     } else if (ShuffleVectorInst *SVI = dyn_cast<ShuffleVectorInst>(I)) {
 212       // If this is extracting an element from a shufflevector, figure out where
 213       // it came from and extract from the appropriate input element instead.
 214       if (ConstantInt *Elt = dyn_cast<ConstantInt>(EI.getOperand(1))) {
 215         int SrcIdx = getShuffleMask(SVI)[Elt->getZExtValue()];
 216         Value *Src;
 217         unsigned LHSWidth =
 218           cast<VectorType>(SVI->getOperand(0)->getType())->getNumElements();
 219
 220         if (SrcIdx < 0)
 221           return ReplaceInstUsesWith(EI, UndefValue::get(EI.getType()));
 222         if (SrcIdx < (int)LHSWidth)
 223           Src = SVI->getOperand(0);
 224         else {
 225           SrcIdx -= LHSWidth;
 226           Src = SVI->getOperand(1);
 227         }
 228         const Type *Int32Ty = Type::getInt32Ty(EI.getContext());
 229         return ExtractElementInst::Create(Src,
 230                                           ConstantInt::get(Int32Ty,
 231                                                            SrcIdx, false));
 232       }
 233     }
 234     // FIXME: Canonicalize extractelement(bitcast) -> bitcast(extractelement)
 235   }
 236   return 0;
 237 }
 238
 239 /// CollectSingleShuffleElements - If V is a shuffle of values that ONLY returns
 240 /// elements from either LHS or RHS, return the shuffle mask and true.
 241 /// Otherwise, return false.
 242 static bool CollectSingleShuffleElements(Value *V, Value *LHS, Value *RHS,
 243                                          std::vector<Constant*> &Mask) {
 244   assert(V->getType() == LHS->getType() && V->getType() == RHS->getType() &&
 245          "Invalid CollectSingleShuffleElements");
 246   unsigned NumElts = cast<VectorType>(V->getType())->getNumElements();
 247
 248   if (isa<UndefValue>(V)) {
 249     Mask.assign(NumElts, UndefValue::get(Type::getInt32Ty(V->getContext())));
 250     return true;
 251   }
 252
 253   if (V == LHS) {
 254     for (unsigned i = 0; i != NumElts; ++i)
 255       Mask.push_back(ConstantInt::get(Type::getInt32Ty(V->getContext()), i));
 256     return true;
 257   }
 258
 259   if (V == RHS) {
 260     for (unsigned i = 0; i != NumElts; ++i)
 261       Mask.push_back(ConstantInt::get(Type::getInt32Ty(V->getContext()),
 262                                       i+NumElts));
 263     return true;
 264   }
 265
 266   if (InsertElementInst *IEI = dyn_cast<InsertElementInst>(V)) {
 267     // If this is an insert of an extract from some other vector, include it.
 268     Value *VecOp    = IEI->getOperand(0);
 269     Value *ScalarOp = IEI->getOperand(1);
 270     Value *IdxOp    = IEI->getOperand(2);
 271
 272     if (!isa<ConstantInt>(IdxOp))
 273       return false;
 274     unsigned InsertedIdx = cast<ConstantInt>(IdxOp)->getZExtValue();
 275
 276     if (isa<UndefValue>(ScalarOp)) {  // inserting undef into vector.
 277       // Okay, we can handle this if the vector we are insertinting into is
 278       // transitively ok.
 279       if (CollectSingleShuffleElements(VecOp, LHS, RHS, Mask)) {
 280         // If so, update the mask to reflect the inserted undef.
 281         Mask[InsertedIdx] = UndefValue::get(Type::getInt32Ty(V->getContext()));
 282         return true;
 283       }
 284     } else if (ExtractElementInst *EI = dyn_cast<ExtractElementInst>(ScalarOp)){
 285       if (isa<ConstantInt>(EI->getOperand(1)) &&
 286           EI->getOperand(0)->getType() == V->getType()) {
 287         unsigned ExtractedIdx =
 288         cast<ConstantInt>(EI->getOperand(1))->getZExtValue();
 289
 290         // This must be extracting from either LHS or RHS.
 291         if (EI->getOperand(0) == LHS || EI->getOperand(0) == RHS) {
 292           // Okay, we can handle this if the vector we are insertinting into is
 293           // transitively ok.
 294           if (CollectSingleShuffleElements(VecOp, LHS, RHS, Mask)) {
 295             // If so, update the mask to reflect the inserted value.
 296             if (EI->getOperand(0) == LHS) {
 297               Mask[InsertedIdx % NumElts] =
 298               ConstantInt::get(Type::getInt32Ty(V->getContext()),
 299                                ExtractedIdx);
 300             } else {
 301               assert(EI->getOperand(0) == RHS);
 302               Mask[InsertedIdx % NumElts] =
 303               ConstantInt::get(Type::getInt32Ty(V->getContext()),
 304                                ExtractedIdx+NumElts);
 305             }
 306             return true;
 307           }
 308         }
 309       }
 310     }
 311   }
 312   // TODO: Handle shufflevector here!
 313
 314   return false;
 315 }
 316
 317 /// CollectShuffleElements - We are building a shuffle of V, using RHS as the
 318 /// RHS of the shuffle instruction, if it is not null.  Return a shuffle mask
 319 /// that computes V and the LHS value of the shuffle.
 320 static Value *CollectShuffleElements(Value *V, std::vector<Constant*> &Mask,
 321                                      Value *&RHS) {
 322   assert(V->getType()->isVectorTy() &&
 323          (RHS == 0 || V->getType() == RHS->getType()) &&
 324          "Invalid shuffle!");
 325   unsigned NumElts = cast<VectorType>(V->getType())->getNumElements();
 326
 327   if (isa<UndefValue>(V)) {
 328     Mask.assign(NumElts, UndefValue::get(Type::getInt32Ty(V->getContext())));
 329     return V;
 330   } else if (isa<ConstantAggregateZero>(V)) {
 331     Mask.assign(NumElts, ConstantInt::get(Type::getInt32Ty(V->getContext()),0));
 332     return V;
 333   } else if (InsertElementInst *IEI = dyn_cast<InsertElementInst>(V)) {
 334     // If this is an insert of an extract from some other vector, include it.
 335     Value *VecOp    = IEI->getOperand(0);
 336     Value *ScalarOp = IEI->getOperand(1);
 337     Value *IdxOp    = IEI->getOperand(2);
 338
 339     if (ExtractElementInst *EI = dyn_cast<ExtractElementInst>(ScalarOp)) {
 340       if (isa<ConstantInt>(EI->getOperand(1)) && isa<ConstantInt>(IdxOp) &&
 341           EI->getOperand(0)->getType() == V->getType()) {
 342         unsigned ExtractedIdx =
 343           cast<ConstantInt>(EI->getOperand(1))->getZExtValue();
 344         unsigned InsertedIdx = cast<ConstantInt>(IdxOp)->getZExtValue();
 345
 346         // Either the extracted from or inserted into vector must be RHSVec,
 347         // otherwise we'd end up with a shuffle of three inputs.
 348         if (EI->getOperand(0) == RHS || RHS == 0) {
 349           RHS = EI->getOperand(0);
 350           Value *V = CollectShuffleElements(VecOp, Mask, RHS);
 351           Mask[InsertedIdx % NumElts] =
 352             ConstantInt::get(Type::getInt32Ty(V->getContext()),
 353                              NumElts+ExtractedIdx);
 354           return V;
 355         }
 356
 357         if (VecOp == RHS) {
 358           Value *V = CollectShuffleElements(EI->getOperand(0), Mask, RHS);
 359           // Everything but the extracted element is replaced with the RHS.
 360           for (unsigned i = 0; i != NumElts; ++i) {
 361             if (i != InsertedIdx)
 362               Mask[i] = ConstantInt::get(Type::getInt32Ty(V->getContext()),
 363                                          NumElts+i);
 364           }
 365           return V;
 366         }
 367
 368         // If this insertelement is a chain that comes from exactly these two
 369         // vectors, return the vector and the effective shuffle.
 370         if (CollectSingleShuffleElements(IEI, EI->getOperand(0), RHS, Mask))
 371           return EI->getOperand(0);
 372       }
 373     }
 374   }
 375   // TODO: Handle shufflevector here!
 376
 377   // Otherwise, can't do anything fancy.  Return an identity vector.
 378   for (unsigned i = 0; i != NumElts; ++i)
 379     Mask.push_back(ConstantInt::get(Type::getInt32Ty(V->getContext()), i));
 380   return V;
 381 }
 382
 383 Instruction *InstCombiner::visitInsertElementInst(InsertElementInst &IE) {
 384   Value *VecOp    = IE.getOperand(0);
 385   Value *ScalarOp = IE.getOperand(1);
 386   Value *IdxOp    = IE.getOperand(2);
 387
 388   // Inserting an undef or into an undefined place, remove this.
 389   if (isa<UndefValue>(ScalarOp) || isa<UndefValue>(IdxOp))
 390     ReplaceInstUsesWith(IE, VecOp);
 391
 392   // If the inserted element was extracted from some other vector, and if the
 393   // indexes are constant, try to turn this into a shufflevector operation.
 394   if (ExtractElementInst *EI = dyn_cast<ExtractElementInst>(ScalarOp)) {
 395     if (isa<ConstantInt>(EI->getOperand(1)) && isa<ConstantInt>(IdxOp) &&
 396         EI->getOperand(0)->getType() == IE.getType()) {
 397       unsigned NumVectorElts = IE.getType()->getNumElements();
 398       unsigned ExtractedIdx =
 399         cast<ConstantInt>(EI->getOperand(1))->getZExtValue();
 400       unsigned InsertedIdx = cast<ConstantInt>(IdxOp)->getZExtValue();
 401
 402       if (ExtractedIdx >= NumVectorElts) // Out of range extract.
 403         return ReplaceInstUsesWith(IE, VecOp);
 404
 405       if (InsertedIdx >= NumVectorElts)  // Out of range insert.
 406         return ReplaceInstUsesWith(IE, UndefValue::get(IE.getType()));
 407
 408       // If we are extracting a value from a vector, then inserting it right
 409       // back into the same place, just use the input vector.
 410       if (EI->getOperand(0) == VecOp && ExtractedIdx == InsertedIdx)
 411         return ReplaceInstUsesWith(IE, VecOp);
 412
 413       // If this insertelement isn't used by some other insertelement, turn it
 414       // (and any insertelements it points to), into one big shuffle.
 415       if (!IE.hasOneUse() || !isa<InsertElementInst>(IE.use_back())) {
 416         std::vector<Constant*> Mask;
 417         Value *RHS = 0;
 418         Value *LHS = CollectShuffleElements(&IE, Mask, RHS);
 419         if (RHS == 0) RHS = UndefValue::get(LHS->getType());
 420         // We now have a shuffle of LHS, RHS, Mask.
 421         return new ShuffleVectorInst(LHS, RHS, ConstantVector::get(Mask));
 422       }
 423     }
 424   }
 425
 426   unsigned VWidth = cast<VectorType>(VecOp->getType())->getNumElements();
 427   APInt UndefElts(VWidth, 0);
 428   APInt AllOnesEltMask(APInt::getAllOnesValue(VWidth));
 429   if (SimplifyDemandedVectorElts(&IE, AllOnesEltMask, UndefElts))
 430     return &IE;
 431
 432   return 0;
 433 }
 434
 435
 436 Instruction *InstCombiner::visitShuffleVectorInst(ShuffleVectorInst &SVI) {
 437   Value *LHS = SVI.getOperand(0);
 438   Value *RHS = SVI.getOperand(1);
 439   std::vector<int> Mask = getShuffleMask(&SVI);
 440
 441   bool MadeChange = false;
 442
 443   // Undefined shuffle mask -> undefined value.
 444   if (isa<UndefValue>(SVI.getOperand(2)))
 445     return ReplaceInstUsesWith(SVI, UndefValue::get(SVI.getType()));
 446
 447   unsigned VWidth = cast<VectorType>(SVI.getType())->getNumElements();
 448
 449   if (VWidth != cast<VectorType>(LHS->getType())->getNumElements())
 450     return 0;
 451
 452   APInt UndefElts(VWidth, 0);
 453   APInt AllOnesEltMask(APInt::getAllOnesValue(VWidth));
 454   if (SimplifyDemandedVectorElts(&SVI, AllOnesEltMask, UndefElts)) {
 455     LHS = SVI.getOperand(0);
 456     RHS = SVI.getOperand(1);
 457     MadeChange = true;
 458   }
 459
 460   // Canonicalize shuffle(x    ,x,mask) -> shuffle(x, undef,mask')
 461   // Canonicalize shuffle(undef,x,mask) -> shuffle(x, undef,mask').
 462   if (LHS == RHS || isa<UndefValue>(LHS)) {
 463     if (isa<UndefValue>(LHS) && LHS == RHS) {
 464       // shuffle(undef,undef,mask) -> undef.
 465       return ReplaceInstUsesWith(SVI, LHS);
 466     }
 467
 468     // Remap any references to RHS to use LHS.
 469     std::vector<Constant*> Elts;
 470     for (unsigned i = 0, e = Mask.size(); i != e; ++i) {
 471       if (Mask[i] < 0)
 472         Elts.push_back(UndefValue::get(Type::getInt32Ty(SVI.getContext())));
 473       else {
 474         if ((Mask[i] >= (int)e && isa<UndefValue>(RHS)) ||
 475             (Mask[i] <  (int)e && isa<UndefValue>(LHS))) {
 476           Mask[i] = -1;     // Turn into undef.
 477           Elts.push_back(UndefValue::get(Type::getInt32Ty(SVI.getContext())));
 478         } else {
 479           Mask[i] = Mask[i] % e;  // Force to LHS.
 480           Elts.push_back(ConstantInt::get(Type::getInt32Ty(SVI.getContext()),
 481                                           Mask[i]));
 482         }
 483       }
 484     }
 485     SVI.setOperand(0, SVI.getOperand(1));
 486     SVI.setOperand(1, UndefValue::get(RHS->getType()));
 487     SVI.setOperand(2, ConstantVector::get(Elts));
 488     LHS = SVI.getOperand(0);
 489     RHS = SVI.getOperand(1);
 490     MadeChange = true;
 491   }
 492
 493   // Analyze the shuffle, are the LHS or RHS and identity shuffles?
 494   bool isLHSID = true, isRHSID = true;
 495
 496   for (unsigned i = 0, e = Mask.size(); i != e; ++i) {
 497     if (Mask[i] < 0) continue;  // Ignore undef values.
 498     // Is this an identity shuffle of the LHS value?
 499     isLHSID &= (Mask[i] == (int)i);
 500
 501     // Is this an identity shuffle of the RHS value?
 502     isRHSID &= (Mask[i]-e == i);
 503   }
 504
 505   // Eliminate identity shuffles.
 506   if (isLHSID) return ReplaceInstUsesWith(SVI, LHS);
 507   if (isRHSID) return ReplaceInstUsesWith(SVI, RHS);
 508
 509   // If the LHS is a shufflevector itself, see if we can combine it with this
 510   // one without producing an unusual shuffle.  Here we are really conservative:
 511   // we are absolutely afraid of producing a shuffle mask not in the input
 512   // program, because the code gen may not be smart enough to turn a merged
 513   // shuffle into two specific shuffles: it may produce worse code.  As such,
 514   // we only merge two shuffles if the result is either a splat or one of the
 515   // two input shuffle masks.  In this case, merging the shuffles just removes
 516   // one instruction, which we know is safe.  This is good for things like
 517   // turning: (splat(splat)) -> splat.
 518   if (ShuffleVectorInst *LHSSVI = dyn_cast<ShuffleVectorInst>(LHS)) {
 519     if (isa<UndefValue>(RHS)) {
 520       std::vector<int> LHSMask = getShuffleMask(LHSSVI);
 521
 522       if (LHSMask.size() == Mask.size()) {
 523         std::vector<int> NewMask;
 524         bool isSplat = true;
 525         int SplatElt = -1; // undef
 526         for (unsigned i = 0, e = Mask.size(); i != e; ++i) {
 527           int MaskElt;
 528           if (Mask[i] < 0 || Mask[i] >= (int)e)
 529             MaskElt = -1; // undef
 530           else
 531             MaskElt = LHSMask[Mask[i]];
 532           // Check if this could still be a splat.
 533           if (MaskElt >= 0) {
 534             if (SplatElt >=0 && SplatElt != MaskElt)
 535               isSplat = false;
 536             SplatElt = MaskElt;
 537           }
 538           NewMask.push_back(MaskElt);
 539         }
 540
 541         // If the result mask is equal to the src shuffle or this
 542         // shuffle mask, do the replacement.
 543         if (isSplat || NewMask == LHSMask || NewMask == Mask) {
 544           std::vector<Constant*> Elts;
 545           const Type *Int32Ty = Type::getInt32Ty(SVI.getContext());
 546           for (unsigned i = 0, e = NewMask.size(); i != e; ++i) {
 547             if (NewMask[i] < 0) {
 548               Elts.push_back(UndefValue::get(Int32Ty));
 549             } else {
 550               Elts.push_back(ConstantInt::get(Int32Ty, NewMask[i]));
 551             }
 552           }
 553           return new ShuffleVectorInst(LHSSVI->getOperand(0),
 554                                        LHSSVI->getOperand(1),
 555                                        ConstantVector::get(Elts));
 556         }
 557       }
 558     }
 559   }
 560
 561   return MadeChange ? &SVI : 0;
 562 }