// CHECK-LABEL: init_all_tables
  func.func @init_all_tables()
  attributes {tf_saved_model.exported_names = ["__tf_saved_model_session_initializer"]} {
    %cst = arith.constant dense<[1, 2, 3, 4]> : tensor<4xi64>
    %cst_0 = arith.constant dense<["a", "b", "c", "d"]> : tensor<4x!tf_type.string>

// CHECK-LABEL: broadcast_to_to_reshape_i64
func.func @broadcast_to_to_reshape_i64(%arg0: tensor<4x4x4xf32>, %arg1 : tensor<4xi64>) -> tensor<1x4x4x4xf32> {
  %0 = "tfl.broadcast_to"(%arg0, %arg1) : (tensor<4x4x4xf32>, tensor<4xi64>) -> tensor<1x4x4x4xf32>
  // CHECK: "tfl.cast"
  // CHECK-SAME: (tensor<4xi64>) -> tensor<4xi32>
  // CHECK-NEXT: "tfl.reshape"
  // CHECK-SAME: (tensor<4x4x4xf32>, tensor<4xi32>) -> tensor<1x4x4x4xf32>

  %1 = mhlo.constant dense<[[[[0, 1, 2, 3]], [[0, 1, 2, 3]]]]> : tensor<1x2x1x4xi32>
  %2 = "mhlo.broadcast_in_dim"(%0) <{broadcast_dimensions = dense<[0, 1, 2, 3]> : tensor<4xi64>}> : (tensor<1x1x1x4xi32>) -> tensor<1x2x1x4xi32>
  %3 = mhlo.multiply %1, %2 : tensor<1x2x1x4xi32>
  // CHECK:      return %[[RES]] : tensor<1x2x1x4xi32>
  return %3 : tensor<1x2x1x4xi32>
}

  func.return %2 : tensor<?x?x?x?xf32>

// CHECK-DAG: %[[CST:.*]] = arith.constant dense<0> : tensor<4xi64>
// CHECK-DAG: %[[CST0:.*]] = arith.constant dense<[1, 9, 23, 768]> : tensor<4xi64>
// CHECK-DAG: %[[CST1:.*]] = arith.constant dense<1> : tensor<4xi64>

}

func.func @real(%arg0: tensor<1 x complex<f64>>) -> tensor<1xf64> {
  %0 = "tf.Real"(%arg0) : (tensor<1 x complex<f64>>) -> tensor<1xf64>
  func.return %0: tensor<1xf64>

// CHECK-LABEL: real
// CHECK:  "tfl.real"(%arg0) : (tensor<1xcomplex<f64>>) -> tensor<1xf64>
// CHECK:  return
}

func.func @imag(%arg0: tensor<1 x complex<f64>>) -> tensor<1xf64> {

func.return %0 : tensor<2x?x2x4xf32>

// CHECK-LABEL:     convert_dot_general_dynamic_batch_dim
// CHECK:           %0 = "tfl.pseudo_const"() <{value = dense<[0, 1, 3, 2]> : tensor<4xi64>}> : () -> tensor<4xi64>
// CHECK-NEXT:      %1 = "tfl.cast"(%0) : (tensor<4xi64>) -> tensor<4xi32>
// CHECK-NEXT:      %2 = "tfl.transpose"(%arg1, %1) : (tensor<2x?x4x3xf32>, tensor<4xi32>) -> tensor<2x?x3x4xf32>

  // CHECK-NEXT:   mhlo.return %[[CMP]]
  // CHECK-NEXT: }) : (tensor<8xf64>) -> tensor<8xf64>
  // CHECK-NEXT: return %[[SORT]]
  %output = "tf.XlaSort"(%input) : (tensor<8xf64>) -> (tensor<8xf64>)
  func.return %output : tensor<8xf64>
}

// -----

// CHECK-LABEL: @xlasort_const
func.func @xlasort_const() -> (tensor<2x3xi64>) {

//CHECK-NEXT: return %0 : tensor<1x161x1xf32>
//CHECK-NEXT:}

func.func @convert(%arg0: tensor<2xf64>) -> tensor<2xf32> {
  %0 = "vhlo.convert_v1" (%arg0) : (tensor<2xf64>) -> tensor<2xf32>
  return %0 : tensor<2xf32>
}

//CHECK:func.func private @convert(%arg0: tensor<2xf64>) -> tensor<2xf32> {
//CHECK-NEXT: %0 = "vhlo.convert_v1"(%arg0) : (tensor<2xf64>) -> tensor<2xf32>
//CHECK-NEXT: return %0 : tensor<2xf32>
//CHECK-NEXT:}

    // CHECK-NEXT: %15 = mhlo.constant dense<1.000000e+00> : tensor<f32>
    // CHECK-NEXT: %16 = mhlo.constant dense<[32, 12, 12, 64]> : tensor<4xi64>
    // CHECK-NEXT: %17 = "mhlo.rng"(%14, %15, %16) <{rng_distribution = #mhlo.rng_distribution<UNIFORM>}> : (tensor<f32>, tensor<f32>, tensor<4xi64>) -> tensor<32x12x12x64xf32>
  %cst = "tf.Const"() {value = dense<[32, 12, 12, 64]> : tensor<4xi32>} : () -> tensor<4xi32>

func.func @testMaximumOfZeroToReluInt64OnGpu(%arg0: tensor<4xi64>) -> tensor<4xi64> {
  // CHECK: %[[RESULT:.*]] = "tf.Relu"(%arg0) {device = "/job:localhost/replica:0/task:0/device:GPU:0"} : (tensor<4xi64>) -> tensor<4xi64>
  // CHECK: return %[[RESULT]]
  %cst_0 = arith.constant dense<0> : tensor<i64>
  %0 = "tf.Maximum"(%arg0, %cst_0) {device = "/job:localhost/replica:0/task:0/device:GPU:0"} : (tensor<4xi64>, tensor<i64>) -> tensor<4xi64>