%1 = "tfl.sum"(%0, %cst) {keep_dims = false} : (tensor<2xf32>, tensor<1xi32>) -> tensor<f32>
  %2 = "tfl.add"(%1, %cst_1) {fused_activation_function = "NONE"} : (tensor<f32>, tensor<1xf32>) -> tensor<1xf32>
  %3 = "tfl.rsqrt"(%2) : (tensor<1xf32>) -> tensor<1xf32>
  %4 = "tfl.mul"(%arg0, %3) {fused_activation_function = "NONE"} : (tensor<2xf32>, tensor<1xf32>) -> tensor<2xf32>
  func.return %4: tensor<2xf32>

  func.return %0#0 : tensor<?x?x?x?xf32>
}

// -----

// CHECK-LABEL: fusedBatchNormV3_training_dynamic_unsupported1

// CHECK:           return %[[VAL_2]], %[[VAL_3]] : tensor<4xf32>, tensor<4xf32>
// CHECK:         }
func.func @broadcast_atan2(%arg0: tensor<1xf32>, %arg1: tensor<4xf32>) -> (tensor<4xf32>, tensor<4xf32>) {
  %0 = "mhlo.broadcast_in_dim"(%arg0) <{broadcast_dimensions = dense<[0]> : tensor<1xi64>}> : (tensor<1xf32>) -> tensor<4xf32>
  %1 = mhlo.atan2 %0, %arg1 : tensor<4xf32>
  %2 = mhlo.atan2 %arg1, %0 : tensor<4xf32>