CodeGenPrepare.cpp

    if (OpInfo.ConstraintType == TargetLowering::C_Memory &&
        OpInfo.isIndirect) {
      Value *OpVal = OpInfo.CallOperandVal;
      MadeChange |= OptimizeMemoryInst(I, OpVal, OpVal->getType(), SunkAddrs);
    }
  }

  return MadeChange;
}

bool CodeGenPrepare::OptimizeExtUses(Instruction *I) {
  BasicBlock *DefBB = I->getParent();

  // If both result of the {s|z}xt and its source are live out, rewrite all
  // other uses of the source with result of extension.
  Value *Src = I->getOperand(0);
  if (Src->hasOneUse())
    return false;

  // Only do this xform if truncating is free.
  if (TLI && !TLI->isTruncateFree(I->getType(), Src->getType()))
    return false;

  // Only safe to perform the optimization if the source is also defined in
  // this block.
  if (!isa<Instruction>(Src) || DefBB != cast<Instruction>(Src)->getParent())
    return false;

  bool DefIsLiveOut = false;
  for (Value::use_iterator UI = I->use_begin(), E = I->use_end();
       UI != E; ++UI) {
    Instruction *User = cast<Instruction>(*UI);

    // Figure out which BB this ext is used in.
    BasicBlock *UserBB = User->getParent();
    if (UserBB == DefBB) continue;
    DefIsLiveOut = true;
    break;
  }
  if (!DefIsLiveOut)
    return false;

  // Make sure non of the uses are PHI nodes.
  for (Value::use_iterator UI = Src->use_begin(), E = Src->use_end();
       UI != E; ++UI) {
    Instruction *User = cast<Instruction>(*UI);
    BasicBlock *UserBB = User->getParent();
    if (UserBB == DefBB) continue;
    // Be conservative. We don't want this xform to end up introducing
    // reloads just before load / store instructions.
    if (isa<PHINode>(User) || isa<LoadInst>(User) || isa<StoreInst>(User))
      return false;
  }

  // InsertedTruncs - Only insert one trunc in each block once.
  DenseMap<BasicBlock*, Instruction*> InsertedTruncs;

  bool MadeChange = false;
  for (Value::use_iterator UI = Src->use_begin(), E = Src->use_end();
       UI != E; ++UI) {
    Use &TheUse = UI.getUse();
    Instruction *User = cast<Instruction>(*UI);

    // Figure out which BB this ext is used in.
    BasicBlock *UserBB = User->getParent();
    if (UserBB == DefBB) continue;

    // Both src and def are live in this block. Rewrite the use.
    Instruction *&InsertedTrunc = InsertedTruncs[UserBB];

    if (!InsertedTrunc) {
      BasicBlock::iterator InsertPt = UserBB->getFirstNonPHI();

      InsertedTrunc = new TruncInst(I, Src->getType(), "", InsertPt);
    }

    // Replace a use of the {s|z}ext source with a use of the result.
    TheUse = InsertedTrunc;

    MadeChange = true;
  }

  return MadeChange;
}

// In this pass we look for GEP and cast instructions that are used
// across basic blocks and rewrite them to improve basic-block-at-a-time
// selection.
bool CodeGenPrepare::OptimizeBlock(BasicBlock &BB) {
  bool MadeChange = false;

  // Split all critical edges where the dest block has a PHI and where the phi
  // has shared immediate operands.
  TerminatorInst *BBTI = BB.getTerminator();
  if (BBTI->getNumSuccessors() > 1) {
    for (unsigned i = 0, e = BBTI->getNumSuccessors(); i != e; ++i)
      if (isa<PHINode>(BBTI->getSuccessor(i)->begin()) &&
          isCriticalEdge(BBTI, i, true))
        SplitEdgeNicely(BBTI, i, this);
  }


  // Keep track of non-local addresses that have been sunk into this block.
  // This allows us to avoid inserting duplicate code for blocks with multiple
  // load/stores of the same address.
  DenseMap<Value*, Value*> SunkAddrs;

  for (BasicBlock::iterator BBI = BB.begin(), E = BB.end(); BBI != E; ) {
    Instruction *I = BBI++;

    if (CastInst *CI = dyn_cast<CastInst>(I)) {
      // If the source of the cast is a constant, then this should have
      // already been constant folded.  The only reason NOT to constant fold
      // it is if something (e.g. LSR) was careful to place the constant
      // evaluation in a block other than then one that uses it (e.g. to hoist
      // the address of globals out of a loop).  If this is the case, we don't
      // want to forward-subst the cast.
      if (isa<Constant>(CI->getOperand(0)))
        continue;

      bool Change = false;
      if (TLI) {
        Change = OptimizeNoopCopyExpression(CI, *TLI);
        MadeChange |= Change;
      }

      if (!Change && (isa<ZExtInst>(I) || isa<SExtInst>(I)))
        MadeChange |= OptimizeExtUses(I);
    } else if (CmpInst *CI = dyn_cast<CmpInst>(I)) {
      MadeChange |= OptimizeCmpExpression(CI);
    } else if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
      if (TLI)
        MadeChange |= OptimizeMemoryInst(I, I->getOperand(0), LI->getType(),
                                         SunkAddrs);
    } else if (StoreInst *SI = dyn_cast<StoreInst>(I)) {
      if (TLI)
        MadeChange |= OptimizeMemoryInst(I, SI->getOperand(1),
                                         SI->getOperand(0)->getType(),
                                         SunkAddrs);
    } else if (GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(I)) {
      if (GEPI->hasAllZeroIndices()) {
        /// The GEP operand must be a pointer, so must its result -> BitCast
        Instruction *NC = new BitCastInst(GEPI->getOperand(0), GEPI->getType(),
                                          GEPI->getName(), GEPI);
        GEPI->replaceAllUsesWith(NC);
        GEPI->eraseFromParent();
        MadeChange = true;
        BBI = NC;
      }
    } else if (CallInst *CI = dyn_cast<CallInst>(I)) {
      // If we found an inline asm expession, and if the target knows how to
      // lower it to normal LLVM code, do so now.
      if (TLI && isa<InlineAsm>(CI->getCalledValue()))
        if (const TargetAsmInfo *TAI =
            TLI->getTargetMachine().getTargetAsmInfo()) {
          if (TAI->ExpandInlineAsm(CI))
            BBI = BB.begin();
          else
            // Sink address computing for memory operands into the block.
            MadeChange |= OptimizeInlineAsmInst(I, &(*CI), SunkAddrs);
        }
    }
  }

  return MadeChange;
}