sym_name = "some_constant",
    type = tensor<1x64xf32>,
    value = dense<42.0> : tensor<1x64xf32>
  } : () -> ()

  // Representation for variables: mutable global tensor.
  // CHECK: tf_saved_model.global_tensor
  "tf_saved_model.global_tensor"() {
    is_mutable,
    tf_saved_model.exported_names = ["some_var", "some.other.name"],
    sym_name = "some_variable",
    type = tensor<?x64xf32>,
    value = dense<42.0> : tensor<1x64xf32>

    %identity_out1, %control_identity1 = tf_executor.island wraps "tf.Identity"(%partitioned_out#0) {device = ""} : (tensor<2x4xf32>) -> tensor<2x4xf32>
    %identity_out2, %control_identity2 = tf_executor.island wraps "tf.Identity"(%partitioned_out#1) {device = ""} : (tensor<2x4xf32>) -> tensor<2x4xf32>

    tf_executor.fetch
  }
  func.return
}

  %5 = "tfl.pseudo_const"() {value = dense<[[-0.435141891, -0.940576493, 1.30446923, -1.02953017], [0.684501767, 0.363370508, -2.29151702, 2.41928673]]> : tensor<2x4xf32>} : () -> tensor<2x4xf32>

  %11 = "mhlo.dot"(%10, %5) : (tensor<32x3920xf32>, tensor<3920x10xf32>) -> tensor<32x10xf32>
  %12 = "mhlo.broadcast_in_dim"(%2) <{broadcast_dimensions = dense<1> : tensor<1xi64>}> : (tensor<10xf32>) -> tensor<32x10xf32>
  %13 = mhlo.add %11, %12 : tensor<32x10xf32>
  %14 = mhlo.maximum %13, %0 : tensor<32x10xf32>
  return %14 : tensor<32x10xf32>
}
// CHECK: ReadVariableOp
// CHECK: mhlo.convolution
// CHECK: mhlo.add

func.func @foo(%arg0: tensor<1x4xf32>, %arg1: tensor<1x4xf32>) -> tensor<1x4xf32> {
  func.return %arg0 : tensor<1x4xf32>
}

func.func @bar(%arg0: tensor<2x4xf32>, %arg1: tensor<2x4xf32>) -> tensor<2x4xf32> {
  func.return %arg0 : tensor<2x4xf32>
}

func.func @foobar(%arg0: tensor<1x4xf32>, %arg1: tensor<1x4xf32>) -> tensor<1x4xf32> {
  func.return %arg0 : tensor<1x4xf32>
}

// CHECK-DAG: func @main

  func.func @batchmatmulv2(%arg0: tensor<1x4x2xf32>, %arg1: tensor<3x2x4xf32>) -> tensor<3x4x4xf32> {
    // CHECK: mhlo.reduce
    // CHECK: mhlo.dot_general
    // CHECK: mhlo.transpose
    %0 = "tf.BatchMatMulV2"(%arg0, %arg1) {T = f32, adj_x = false, adj_y = false, grad_x = false, grad_y = false, device = ""} : (tensor<1x4x2xf32>, tensor<3x2x4xf32>) -> tensor<3x4x4xf32>
    func.return %0 : tensor<3x4x4xf32>
  }

  constexpr absl::string_view
      kModuleXlaCallModuleOpWithDefaultQuantizationMethod = R"mlir(
    func.func @xla_call_module_default_quantization_method(%arg0: tensor<1x1x3xf32>, %arg1: tensor<3x4xf32>) -> tensor<1x1x4xf32> {

// CHECK-LABEL: func @fold_into_mean
func.func @fold_into_mean(%arg0: tensor<1x64x112x112xf32>) -> tensor<1x64xf32> {

  // CHECK: %[[RED_IDX:.*]] = "tf.Const"() <{value = dense<[2, 3]> : tensor<2xi32>}>
  // CHECK: %[[MEAN:[0-9]*]] = "tf.Mean"(%arg0, %[[RED_IDX]])
  // CHECK-SAME: (tensor<1x64x112x112xf32>, tensor<2xi32>) -> tensor<1x64xf32>
  // CHECK: return %[[MEAN]]

    %split:3 = tf_executor.island wraps "tf.Split"(%dim#0, %input#0) : (tensor<i32>, tensor<4x6xf32>) -> (tensor<2x6xf32>, tensor<2x6xf32>)
    tf_executor.fetch
  }
  func.return

// CHECK:           %[[VALUES:.*]], %[[INDICES:.*]] = "tf.TopKV2"(%[[ARG_0]], %[[K]]) <{sorted = true}> : (tensor<3x6xf32>, tensor<i32>) -> (tensor<3x6xf32>, tensor<3x6xi32>)
// CHECK:           return %[[VALUES]], %[[INDICES]] : tensor<3x6xf32>, tensor<3x6xi32>
// CHECK:         }
func.func @convert_sort_to_topk_iota_broadcast(%arg0: tensor<3x6xf32>) -> (tensor<3x6xf32>, tensor<3x6xi32>) {
  %0 = "mhlo.iota"() <{ iota_dimension = 0 : i64 }> : () -> tensor<6xi32>