SmallVector<Value, 2>({one.getResult(), fft_len}), zero->getResult());

    // Insert the rfft_2d.
    auto rfft2d_out_type = tensorflow::GetTypeFromTFTensorShape(
        expanded_output_shape, output_type.getElementType());
    TF::RFFT2DOp rfft2d = rewriter.create<TF::RFFT2DOp>(
        rfft_op.getLoc(), rfft2d_out_type, expanded_input.getResult(),
        expanded_fft_len.getResult());

}

// -----

func.func @rfft2d(%arg0: tensor<10x20x10x30xf32>, %arg1: tensor<2xi32>) -> tensor<10x20x10x30xcomplex<f32>> {
  %0 = "tf.RFFT2D"(%arg0, %arg1) : (tensor<10x20x10x30xf32>, tensor<2xi32>) -> tensor<10x20x10x30xcomplex<f32>>
  func.return %0 : tensor<10x20x10x30xcomplex<f32>>
  // CHECK-LABEL: rfft2d

  SCATTER_ND = 122,
  SELECT_V2 = 123,
  DENSIFY = 124,
  SEGMENT_SUM = 125,
  BATCH_MATMUL = 126,
  PLACEHOLDER_FOR_GREATER_OP_CODES = 127,
  CUMSUM = 128,
  CALL_ONCE = 129,
  BROADCAST_TO = 130,
  RFFT2D = 131,
  CONV_3D = 132,
  IMAG=133,
  REAL=134,
  COMPLEX_ABS=135,
  HASHTABLE = 136,
  HASHTABLE_FIND = 137,
  HASHTABLE_IMPORT = 138,
  HASHTABLE_SIZE = 139,
  REDUCE_ALL = 140,

  let hasCanonicalizer = 1;
}

def TFL_RFFT2dOp : TFL_Op<"rfft2d", [Pure]> {
  let summary = "2D real-valued fast Fourier transform.";

  let description = [{
Computes the 2-dimensional discrete Fourier transform of a real-valued signal
over the inner-most 2 dimensions of `input`.

Since the DFT of a real signal is Hermitian-symmetric, `RFFT2D` only returns the

// CHECK:  %[[CON:.*]] = "tf.ConcatV2"(%[[CST0]], %arg1, %[[CST1]]) : (tensor<1xi32>, tensor<1xi32>, tensor<i32>) -> tensor<2xi32>
// CHECK:  %[[RFF:.*]] = "tf.RFFT2D"(%[[EXP]], %[[CON]]) : (tensor<10x20x1x30xf32>, tensor<2xi32>) -> tensor<10x20x1x30xcomplex<f64>>
// CHECK:  %[[SQE:.*]] = "tf.Squeeze"(%[[RFF]]) <{squeeze_dims = [-2]}> : (tensor<10x20x1x30xcomplex<f64>>) -> tensor<10x20x30xcomplex<f64>>

over the inner-most 2 dimensions of `input`.

Since the DFT of a real signal is Hermitian-symmetric, `RFFT2D` only returns the
`fft_length / 2 + 1` unique components of the FFT for the inner-most dimension
of `output`: the zero-frequency term, followed by the `fft_length / 2`
positive-frequency terms.

Along each axis `RFFT2D` is computed on, if `fft_length` is smaller than the

    * Added `dtensor.get_default_mesh`, for retrieving the current default mesh under the dtensor context.
    * \*fft\* ops now support dtensors with any layout. Fixed bug in 'fft2d/fft3d', 'ifft2d/ifft3d', 'rfft2d/rfft3d', and 'irfft2d/irfft3d' for sharde input. Refer to this [blog post](https://blog.tensorflow.org/2023/08/distributed-fast-fourier-transform-in-tensorflow.html) for details.

*  `tf.experimental.strict_mode`

    "@curl//:__subpackages__",
    "@double_conversion//:__subpackages__",
    "@eigen_archive//:__subpackages__",
    "@farmhash_archive//:__subpackages__",
    "@farmhash_gpu_archive//:__subpackages__",
    "@fft2d//:__subpackages__",
    "@gemmlowp//:__subpackages__",
    "@jsoncpp_git//:__subpackages__",
    "@llvm_openmp//:__subpackages__",
    "@llvm_terminfo//:__subpackages__",
    "@llvm_zlib//:__subpackages__",

    "SCATTER_ND",
    "SELECT_V2",
    "DENSIFY",
    "SEGMENT_SUM",
    "BATCH_MATMUL",
    "PLACEHOLDER_FOR_GREATER_OP_CODES",
    "CUMSUM",
    "CALL_ONCE",
    "BROADCAST_TO",
    "RFFT2D",
    "CONV_3D",
    "IMAG",
    "REAL",
    "COMPLEX_ABS",
    "HASHTABLE",
    "HASHTABLE_FIND",
    "HASHTABLE_IMPORT",
    "HASHTABLE_SIZE",
    "REDUCE_ALL",

                         symbolizeFftType(fft_string).value()),
        rewriter.getI64TensorAttr(fft_length));
    return success();
  }
};

using ConvertRFFTOp = ConvertFFTOp<TF::RFFTOp>;
using ConvertIRFFTOp = ConvertFFTOp<TF::IRFFTOp>;

// The base class to convert TensorFlow FusedBatchNormGrad*Op to HLO
// BatchNormGradOp for training and a sequence of binary ops for inference.