func.func @pack(%arg0: tensor<1xf32>, %arg1: tensor<1xf32>) -> tensor<2x1xf32> {
  %0 = "tfl.pack"(%arg0, %arg1) {axis = 0 : i32, values_count = 2 : i32} : (tensor<1xf32>, tensor<1xf32>) -> tensor<2x1xf32>
  func.return %0 : tensor<2x1xf32>
}

// CHECK:   func @pack(%[[VAL_0:.*]]: tensor<1xf32>, %[[VAL_1:.*]]: tensor<1xf32>) -> tensor<2x1xf32> {
// CHECK-DAG:       %[[VAL_2:.*]] = "tfl.pseudo_const"{{.*}}dense<1> : tensor<4xi32>

    %2 = func.call @func_2_CPU_FLOAT(%0, %1) {tac.interface_name = "func_2"} : (tensor<1xf32>, tensor<1xf32>) -> tensor<2x1xf32>
    func.return %2 : tensor<2x1xf32>
  }

  func.func private @func_2_CPU_FLOAT(%arg0: tensor<1xf32>, %arg1: tensor<1xf32>) -> tensor<2x1xf32> attributes {tac.device = "CPU", tac.inference_type = "FLOAT", tac.interface_name = "func_2"} {

  %3 = "tfl.pack"(%1, %2) {axis = 0 : i32, per_device_costs = {CPU = 2.0 : f32, GPU = -1.0 : f32}, values_count = 2 : i32, tac.device = "CPU"} : (tensor<1xf32>, tensor<1xf32>) -> tensor<2x1xf32>
  func.return %3 : tensor<2x1xf32>
})";
  const std::string kExpectedFB = CreateRuntimeMetadata();
  mlir::DialectRegistry registry;
  registry.insert<mlir::TFL::TensorFlowLiteDialect, mlir::arith::ArithDialect,

  %cst_0 = "tf.Const"() {value = dense<0.400000e+00> : tensor<256xf32>} : () -> tensor<256xf32>
  %cst_1 = "tf.Const"() {value = dense<0.500000e+00> : tensor<256xf32>} : () -> tensor<256xf32>
  %w = "tf.AddV2"(%cst, %cst_1) : (tensor<48x48x3x1xf32>, tensor<256xf32>) -> tensor<48x48x3x256xf32>

}

// -----

// CHECK-LABEL: listPushBack
func.func @listPushBack(%arg0: tensor<!tf_type.variant<tensor<?x1xf32>>>, %arg1: tensor<16x1xf32>) -> tensor<!tf_type.variant<tensor<?x1xf32>>>  {
  %0 = "tf.TensorListPushBack"(%arg0, %arg1) : (tensor<!tf_type.variant<tensor<?x1xf32>>>, tensor<16x1xf32>) -> tensor<!tf_type.variant<tensor<?x1xf32>>>

  %44 = "tf.AddV2"(%43, %37) {device = ""} : (tensor<2x16x1xf32>, tensor<2x16x1xf32>) -> tensor<2x16x1xf32>
  %45 = "tf.Mul"(%42, %35) {device = ""} : (tensor<2x16x1xf32>, tensor<2x16x1xf32>) -> tensor<2x16x1xf32>
  %46 = "tf.AddV2"(%45, %34) {device = ""} : (tensor<2x16x1xf32>, tensor<2x16x1xf32>) -> tensor<2x16x1xf32>

  %2 = "tf.Transpose"(%1, %cst_0): (tensor<1x2xf32>, tensor<2xi32>) -> tensor<2x1xf32>
  func.return %2 : tensor<2x1xf32>

// CHECK: %cst = arith.constant
// CHECK: %[[trans:.*]] = "tf.Transpose"
// CHECK-SAME: -> tensor<2x1xf32>
// CHECK: %[[q:.*]] = "tfl.quantize"(%[[trans]]) <{qtype = tensor<2x1x!quant.uniform<u8:f32, 1.000000e+00>>}>

// execute the production code path.
func.func @main(%arg0: tensor<2x1xf32>, %arg1: tensor<2x3xf32>) -> (tensor<2x4xf32>) {
  %0 = "tfl.quantize"(%arg0) {qtype = tensor<2x1x!quant.uniform<i16:f32, 1.0>>} : (tensor<2x1xf32>) -> tensor<2x1x!quant.uniform<i16:f32, 1.0>>
  %1 = "tfl.dequantize"(%0) : (tensor<2x1x!quant.uniform<i16:f32, 1.0>>) -> (tensor<2x1xf32>)

  func.return %3 : tensor<2x1xf32>
}
}

// CHECK:   func @simpleTest(%[[VAL_0:.*]]: tensor<1xf32>, %[[VAL_1:.*]]: tensor<1xf32>, %[[VAL_2:.*]]: tensor<1xf32>, %[[VAL_3:.*]]: tensor<1xf32>) -> tensor<2x1xf32> {

  %cst_0 = arith.constant dense<[[1.5, -2.5]]> : tensor<1x2xf32>
  %cst_1 = arith.constant dense<[[-3.], [4.]]> : tensor<2x1xf32>

  %0 = "tfl.add"(%cst_0, %cst_1) {fused_activation_function = "NONE"} : (tensor<1x2xf32>, tensor<2x1xf32>) -> tensor<2x2xf32>

  func.return %0 : tensor<2x2xf32>