// CHECK-DAG: %[[CST2:.*]] = arith.constant dense<[11, 2, 5]> : tensor<3xi32>
  // CHECK: %[[RESHAPE:.*]] = "tfl.reshape"(%arg0, %[[CST0]]) : (tensor<11x2xi32>, tensor<1xi32>) -> tensor<22xi32>
  // CHECK: %[[TMP:.*]] = "tfl.embedding_lookup"(%[[RESHAPE]], %[[CST1]]) : (tensor<22xi32>, tensor<3x5xf32>) -> tensor<22x5xf32>
  // CHECK: %[[RES:.*]] = "tfl.reshape"(%[[TMP]], %[[CST2]]) : (tensor<22x5xf32>, tensor<3xi32>) -> tensor<11x2x5xf32>

  // Verify filter transformation for grouped convolution.

  // CHECK: %[[RESHAPE:.*]] = mhlo.reshape %arg0 : (tensor<2x2x5x21xf32>) -> tensor<2x2x5x3x7xf32>
  // CHECK: %[[TRANSPOSE:.*]] = "mhlo.transpose"(%[[RESHAPE]])
  // CHECK-SAME: permutation = dense<[0, 1, 3, 2, 4]>
  // CHECK-SAME: (tensor<2x2x5x3x7xf32>) -> tensor<2x2x3x5x7xf32>
  // CHECK: mhlo.reshape %[[TRANSPOSE]] : (tensor<2x2x3x5x7xf32>) -> tensor<2x2x15x7xf32>

  %21 = "tf.Cast"(%20) {Truncate = false, device = ""} : (tensor<1x1xi32>) -> tensor<1x1xi64>
  %22 = "tf.Reshape"(%21, %12) {device = ""} : (tensor<1x1xi64>, tensor<1xi64>) -> tensor<1xi64>
  %23 = "tf.Reshape"(%arg0, %5) {device = ""} : (tensor<1x!tf_type.string>, tensor<1xi32>) -> tensor<1x!tf_type.string>

  func.return %0 : tensor<1x519xf32>
}

// CHECK-LABEL:   func @reshape(
// CHECK-SAME:                  %[[VAL_0:.*]]: tensor<4x6xf32>) -> tensor<2x2x6xf32> {
// CHECK:           %[[VAL_1:.*]] = arith.constant dense<[2, 2, 6]> : tensor<3xi64>
// CHECK:           %[[VAL_2:.*]] = "tf.Reshape"(%[[VAL_0]], %[[VAL_1]]) : (tensor<4x6xf32>, tensor<3xi64>) -> tensor<2x2x6xf32>

  // Converts Tile op to HLO DBroadcastInDim and DReshape ops.
  //   For shape [S1, S2] and multiples [M1, M2],
  //     MS1 = M1 * S1; MS2 = M2 * S2
  //
  //   %out_dim_size = [S1, M1, S2, M2]
  //   %broadcast_dimensions = [1, 3];
  //   %broadcast = mhlo.d_broadcast_in_dim(%input, %out_dim_size, %braodcast_dimensions);
  //   %shape = [MS1, MS2]