// CHECK:           %[[VAL_2:.*]] = "tf.AddV2"(%[[VAL_0]], %[[VAL_1]]) : (tensor<4x1x1xi32>, tensor<4x4x4x4xi32>) -> tensor<4x4x4x4xi32>
// CHECK:           return %[[VAL_2]] : tensor<4x4x4x4xi32>
// CHECK:         }
func.func @broadcast_multi_dim_add(%arg0: tensor<4x1x1xi32>, %arg1: tensor<4x4x4x4xi32>) -> tensor<4x4x4x4xi32> {

    // CHECK: return %[[RESULT]] : tensor<4x4x4xf32>

    %0 = "tf.Atan2"(%arg0, %arg1) : (tensor<4x1xf32>, tensor<4x1x4xf32>) -> tensor<4x4x4xf32>
    func.return %0: tensor<4x4x4xf32>
  }

  // CHECK-LABEL: func @ternary_op
  func.func @ternary_op(%arg0: tensor<2xi1>, %arg1: tensor<2xi32>, %arg2: tensor<2xi32>) -> tensor<2xi32> {
    // CHECK: mhlo.select %arg0, %arg1, %arg2

// CHECK-NEXT: return %[[back]] : tensor<f32>
}

// CHECK: attribute_cast
func.func @attribute_cast(%arg0: tensor<1x4x4x1xf32>) -> tensor<1x2x2x1xf32> {
  %0 = "tfr.cast"(%arg0) : (tensor<1x4x4x1xf32>) -> !tfr.tensor
  %stride_i32 = arith.constant 2 : i32
  %1 = tfr.call @tf__my_max_pool(%0, %stride_i32, %stride_i32) : (!tfr.tensor, i32, i32) -> !tfr.tensor

    // CHECK-NEXT: %[[CMP:.*]] = mhlo.compare GT, %[[INP]], %[[ZERO]], NOTYPE : (tensor<1x4x4x3xf32>, tensor<1x4x4x3xf32>) -> tensor<1x4x4x3xi1>
    // CHECK-NEXT: %[[RES:.*]] = mhlo.select %[[CMP]], %[[INP]], %[[LEAKY]] : tensor<1x4x4x3xi1>, tensor<1x4x4x3xf32>
    // CHECK-NEXT: return %[[RES]] : tensor<1x4x4x3xf32>
    %0 = "tf.LeakyRelu"(%arg0) {alpha = 2.000000e-01 : f32, device = ""} : (tensor<1x4x4x3xf32>) -> tensor<1x4x4x3xf32>

// -----

func.func @tensor_scatter_update(%tensor: tensor<4x4x4xf32>, %indices: tensor<i32>, %updates: tensor<4x4xf32>) -> tensor<4x4x4xf32> {
  // expected-error @+1 {{op requires indices operand to have at least 1 dimension}}
  %0 = "tf.TensorScatterUpdate"(%tensor, %indices, %updates) : (tensor<4x4x4xf32>, tensor<i32>, tensor<4x4xf32>) -> tensor<4x4x4xf32>
  func.return %0 : tensor<4x4x4xf32>
}

// -----

  func.return %0 : tensor<4x4x4xf32>

  // CHECK-DAG: %[[CST:.*]] = arith.constant dense<[0, 1, 0]> : tensor<3xi32>
  // CHECK-DAG: %[[CST_0:.*]] = arith.constant dense<0> : tensor<3xi32>
  // CHECK-DAG: %[[CST_1:.*]] = arith.constant dense<1> : tensor<3xi32>

  // We cannot fuse this tfl.mul into the preceding conv op because %cst2 is not broadcast-compatible to %cst0.
  %1 = "tfl.mul"(%0, %cst2) {fused_activation_function = "RELU6"} : (tensor<1x4x4x2xf32>, tensor<4x2xf32>) -> tensor<1x4x4x2xf32>

  func.return %1 : tensor<1x4x4x2xf32>

}

func.func @space_to_batch_nd(%arg0: tensor<1x4x4x3xf32>, %arg1: tensor<2xi32>, %arg2: tensor<2x2xi32>) -> tensor<*xf32> {
  %0 = "tf.SpaceToBatchND"(%arg0, %arg1, %arg2) : (tensor<1x4x4x3xf32>, tensor<2xi32>, tensor<2x2xi32>) -> tensor<*xf32>
  func.return %0 : tensor<*xf32>
  // CHECK-LABEL: space_to_batch_nd
  // CHECK: "tfl.space_to_batch_nd"(%arg0, %arg1, %arg2) : (tensor<1x4x4x3xf32>, tensor<2xi32>, tensor<2x2xi32>) -> tensor<*xf32>
}

}

// -----

// CHECK-LABEL: testSpaceToBatchND
func.func @testSpaceToBatchND(%arg0 : tensor<1x4x4x3xf32>, %arg1 : tensor<2xi32>, %arg2 : tensor<2x2xi32>) -> tensor<?xf32> {
  // CHECK: "tfl.space_to_batch_nd"(%arg0, %arg1, %arg2)
  %0 = "tfl.space_to_batch_nd"(%arg0, %arg1, %arg2) : (tensor<1x4x4x3xf32>, tensor<2xi32>, tensor<2x2xi32>) -> tensor<?xf32>
  func.return %0 : tensor<?xf32>
}

// -----