op_state.addOperands(func.getArguments());
  op_state.addTypes(output_tys);
  op_state.addAttributes(attrs);
  mlir::Operation* tf_op = op_builder.create(op_state);
  op_builder.create<mlir::func::ReturnOp>(loc, tf_op->getResults());

  // Run the decompose passes on the module
  TF_RETURN_IF_ERROR(DecomposeGraph(module));

  // Export the result as a FunctionDef.

    // * The 'dilations' attribute is optional, but it must be of the form
    //   [1, X, Y, 1] if exists.

    TFConvOpType tf_op = cast<TFConvOpType>(op);
    if (!TFTypeIsFloat32Tensor(tf_op.getInput()) &&
        !(allow_bf16_and_f16_type_legalization_ &&
          TFTypeIsBFloat16OrHalfTensor(tf_op.getInput())))
      return failure();

    if (!TFDataFormatIsNHWC(op)) return failure();

    IntegerAttr height, width;

//===----------------------------------------------------------------------===//
// TensorFlow op definitions
//===----------------------------------------------------------------------===//

class TF_Op<string mnemonic, list<Trait> traits = []> :
    Op<TensorFlowDialect, mnemonic, traits>;

//===----------------------------------------------------------------------===//
// TensorFlow attribute definitions

        ),
    ],
    tblgen = "@llvm-project//mlir:mlir-tblgen",
    td_file = "ir/tf_ops.td",
    deps = [
        ":tensorflow_ops_td_files",
    ],
)

# We only shard tf_op on name for build performance reasons.
tf_ops_category_list = [
    {
        "name": "ops_a_m",
        "include": "tf.[A-M].*$",
    },
    {
        "name": "ops_n_z",
        "include": "tf.[N-Z].*$",