namespace odml {

// Given a Composite op that wraps a core.aten.avg_pool2d, returns the padding
// configuration required for the `tfl.pad` if the padding part of the op is
// to be done before average pooling.
DenseIntElementsAttr GetPadOpAttr(Builder& builder, mhlo::CompositeOp op);

// Given a Composite op that wraps a core.aten.avg_pool2d, and assuming that
// the padding part is extracted into a tfl.pad op prior to a

    // Refining types may have implications to the attached regions or symbol
    // references so do not update such ops.
    if (!op->getRegions().empty() || HasSymbolRefAttr(op)) return failure();

    IRMapping mapper;
    bool has_type_change = false;
    for (auto [original, updated] : llvm::zip(op->getOperands(), operands)) {
      Type original_ty = original.getType();
      Type updated_ty = updated.getType();

    SmallVector<int32_t>& calibration_methods) {
  for (auto op : region.getOps<TF::CustomAggregatorOp>()) {
    if (aggregator_ops_to_ignore.count(op.getId().str())) continue;

    ids.push_back(op.getId());
    calibration_methods.push_back(op.getCalibrationMethod());
    statistics_outputs.push_back(op.getMin());
    statistics_outputs.push_back(op.getMax());
    statistics_outputs.push_back(op.getHistogram());
  }
}

        BuildOperationRecord op,
        int level,
        Spec<? super BuildOperationRecord> predicate,
        Spec<? super BuildOperationRecord> progressPredicate
    ) {
        if (predicate.isSatisfiedBy(op)) {
            println "${'  ' * level}(${op.displayName}, id: $op.id${op.detailsType ? ", details type: ${simpleClassName(op.detailsType)}" : ''})${op.details ? " $op.details" : ''}"

	}

	// Select mode
	var op covOperation
	cmd := os.Args[1]
	switch cmd {
	case mergeMode:
		op = makeMergeOp()
	case debugDumpMode:
		op = makeDumpOp(debugDumpMode)
	case textfmtMode:
		op = makeDumpOp(textfmtMode)
	case percentMode:
		op = makeDumpOp(percentMode)
	case funcMode:
		op = makeDumpOp(funcMode)
	case pkglistMode:
		op = makeDumpOp(pkglistMode)
	case subtractMode:

  func_op.walk([&should_fold_op_func](TF::IdentityOp op) {
    StringAttr op_device = op->getAttrOfType<StringAttr>(kDeviceAttr);
    if (should_fold_op_func(op.getOperand(), op_device)) {
      op.replaceAllUsesWith(op.getOperand());
      op.erase();
    }
    return WalkResult::advance();
  });

  func_op.walk([&should_fold_op_func](TF::IdentityNOp op) {
    StringAttr op_device = op->getAttrOfType<StringAttr>(kDeviceAttr);

  rewriter.replaceOpWithNewOp<stablehlo::ConstantOp>(adaptor.value().Op(),
                                                     res_attr);
  return success();
}

static LogicalResult FoldDivOp(stablehlo::DivOp op, PatternRewriter& rewriter) {
  auto etype = op.getType().getElementType();
  if (etype.isa<FloatType>()) {
    return FoldDivOpInternal<APFloat>(op, rewriter);
  }
  if (etype.isa<IntegerType>()) {

    return false;

  if (auto arith_const_op = dyn_cast_or_null<arith::ConstantOp>(op)) {
    // arith ConstOp path.
    auto type =
        mlir::cast<ShapedType>(arith_const_op.getType()).getElementType();
    if (!type.isInteger(32) && !type.isInteger(64)) return false;
  } else if (auto const_op = dyn_cast_or_null<TFL::ConstOp>(op)) {
    // ConstOp path.

    // non-quantized int64
    if (allTypesFp || allTypesQuantizedOrInt) return failure();

    Location loc = op->getLoc();
    SmallVector<Value> newOperands;
    newOperands.reserve(op->getNumOperands());
    for (auto operand : op->getOperands()) {
      if (QuantizedType::getQuantizedElementType(operand.getType())) {
        auto newTy = QuantizedType::castToExpressedType(operand.getType());

  // are required to be quantized.
  for (auto arg : func.getBody().begin()->getArguments()) {
    if (UsedAsBias(arg)) {
      AddToWorkListIfUnquantized(arg, &bias_values);
    } else {
      AddToWorkListIfUnquantized(arg, &activation_values);
    }
  }

  func.walk([&](Operation *op) {
    if (!quant::IsOpQuantizable(op) || op->getParentOfType<TFL::CustomTfOp>()) {