} else {
      return tensorflow::Status(absl::StatusCode::kInvalidArgument,
                                "Unsupported type");
    }
  } else if (auto itype = mlir::dyn_cast<mlir::IntegerType>(element_type)) {
    if (element_type.isSignedInteger()) {
      switch (itype.getWidth()) {
        case 8:
          return DenseElementsAttr::get<int8_t>(shaped_type,

    for (int i = 0; i < 2; ++i) {
      tensor_proto.add_int_val(1);
    }
    NodeDef const_def;
    TF_CHECK_OK(NodeDefBuilder("const", "Const")
                    .Attr("dtype", DT_INT32)
                    .Attr("value", tensor_proto)
                    .Finalize(&const_def));
    Status s;
    Node *const_node = g->AddNode(const_def, &s);
    TF_CHECK_OK(s);

    NodeDef fn_def;

    Location loc = graph_func.getLoc();
    FunctionType ftype = graph_func.getFunctionType();

    func::FuncOp func = rewriter.create<func::FuncOp>(
        graph_func.getLoc(),
        graph_func->getAttrOfType<StringAttr>(SymbolTable::getSymbolAttrName())
            .getValue(),
        ftype);

    func->setAttrs(graph_func->getAttrs());

    llvm::SmallVector<Type> arg_types;
    llvm::SmallVector<Type> res_types;

string AttrView::DefaultValue() const {
  const AttrValue &attr_value = attr_.default_value();
  switch (attr_value.value_case()) {
    case AttrValue::VALUE_NOT_SET:
      return "";
    case AttrValue::kType:
      return DataType_Name(attr_value.type());
    case AttrValue::kS:
      return "\"" + attr_value.s() + "\"";
    case AttrValue::kI:
      return std::to_string(attr_value.i());
    case AttrValue::kF:

                    Attribute value) {
  // Handle the case where the type and value are already tensors.
  if (mlir::isa<TensorType>(type) && mlir::isa<ElementsAttr>(value)) {
    result.addTypes(type);
    result.addAttribute("value", value);
    return;
  }

  // Otherwise, default to the attribute builder.
  ConstOp::build(builder, result, value);

      assert(while_caller.body_function() == func);
      while_callers.push_back(while_caller);
    }
  }
  return while_callers;
}

bool IsResourceType(Type type) {
  return mlir::isa<TF::ResourceType>(getElementTypeOrSelf(type));
}

bool OnlyOperatesOnCompositeDevices(
    TF::TPUExecuteAndUpdateVariablesOp& op,
    const TF::SideEffectAnalysis::Info& side_effect_analysis,

    }
    if (ftype.isF64()) {
      return tflite::TensorType_COMPLEX128;
    }
    return Status(absl::StatusCode::kInvalidArgument, "Unsupported type");
  } else if (auto itype = mlir::dyn_cast<mlir::IntegerType>(type)) {
    switch (itype.getWidth()) {
      case 1:
        return tflite::TensorType_BOOL;
      case 4:
        if (itype.isUnsigned()) {
          return Status(absl::StatusCode::kInvalidArgument,

  else
    return UnrankedTensorType::get(type.getValue());
}
}  // namespace

// Returns whether type can be further refined.
bool CanBeRefined(Type type) {
  auto shape_type = mlir::dyn_cast<ShapedType>(type);
  if (!shape_type) return false;

  // Returns whether type with subtypes can be further refined.

func::FuncOp BuildGELUDecomposition(RankedTensorType type,
                                    PatternRewriter& rewriter,
                                    Block* insertion_point) {
  rewriter.setInsertionPointToStart(insertion_point);

  auto ftype = FunctionType::get(rewriter.getContext(), {type}, {type});
  auto name = rewriter.getStringAttr("gelu_decomp");

LogicalResult ConstantFoldMul(mhlo::MulOp op, PatternRewriter &rewriter) {
  ShapedType type = mlir::dyn_cast<ShapedType>(op.getType());
  Type etype = type.getElementType();
  Attribute result = {};
  if (mlir::isa<FloatType>(etype)) {
    result =
        BinaryFolder<mhlo::MulOp, FloatType, APFloat, std::multiplies<APFloat>>(
            &op);
  } else if (mlir::isa<IntegerType>(etype)) {
    result =