# CHECK:  tf.TensorListReserve{{.*}}(tensor<2xi32>, tensor<i32>) -> tensor<!tf_type.variant<tensor<*xf32>>>

# Nested variant type.
# CHECK:  tf.TensorListReserve{{.*}}(tensor<2xi32>, tensor<i32>) -> tensor<!tf_type.variant<tensor<*x!tf_type.variant>>>

# CHECK:  tf.TensorListSetItem{{.*}}(tensor<!tf_type.variant<tensor<*xf32>>>, tensor<i32>, tensor<2x2xf32>) -> tensor<*x!tf_type.variant>

# CHECK-DAG:       %[[VAL_2:.*]] = arith.constant dense<[1, 0]> : tensor<2xi32>
# CHECK-DAG:       %[[VAL_3:.*]] = "tfl.no_value"() <{value}> : () -> none
# CHECK-DAG:       %[[VAL_6:.*]] = arith.constant dense<0> : tensor<i32>

        module = ir.Module.parse(module_bytecode)
        self.assertIn('stablehlo.add %arg0, %cst : tensor<4xf32>', str(module))

  def test_tf_mlir_to_stablehlo(self):
    assembly = """
      module attributes {tf.versions = {bad_consumers = [], min_consumer = 0 : i32, producer = 268 : i32}} {
        func.func @main(%arg0 : tensor<?xf32>) -> tensor<?xf32> {
          %cst = "tf.Const"() {value = dense<1.0> : tensor<f32>} : () -> tensor<f32>

  OwningOpRef<ModuleOp> module_op = ParseModuleOpString(R"mlir(
    module {
      "tf_saved_model.global_tensor"() {type = tensor<2xi32>, value = dense<2> : tensor<2xi32>, sym_name = "__tf_saved_model_x"} : () -> ()
      func.func @main(%arg_0: tensor<!tf_type.resource<tensor<2xi32>>> {tf_saved_model.bound_input = @__tf_saved_model_x}) -> () attributes {tf.entry_function = {inputs = "arg_0:0", outputs = ""}} {
        return
      }

  %perm = arith.constant dense<[1, 0]> : tensor<2xi32>
  %0 = "tfl.transpose"(%arg0, %perm) : (tensor<2x3xi32>, tensor<2xi32>) -> tensor<3x2xi32>
  %cst = arith.constant dense<[[1, 2], [3, 4], [5, 6]]> : tensor<3x2xi32>
  %1 = tfl.add %0, %cst { fused_activation_function = "NONE" } : tensor<3x2xi32>
  func.return %1 : tensor<3x2xi32>
}

// CHECK: %cst = arith.constant dense<[1, 0]> : tensor<2xi32>
// CHECK: [1, 3, 5], [2, 4, 6]

// CHECK: return %[[DEQUANTIZE]] : tensor<8xf32>
}

// CHECK-LABEL: fakeQuantNotFolded
func.func @fakeQuantNotFolded(tensor<8xf32>, tensor<f32>, tensor<f32>) -> (tensor<8xf32>) {
^bb0(%arg0: tensor<8xf32>, %arg3: tensor<f32>, %arg4: tensor<f32>):
  %1 = "tf.FakeQuantWithMinMaxVars"(%arg0, %arg3, %arg4) {num_bits = 5, narrow_range = false} : (tensor<8xf32>, tensor<f32>, tensor<f32>) -> tensor<8xf32>
  func.return %1 : tensor<8xf32>

  %cst0 = mhlo.constant dense<[1.0, 2.0, 3.0, 4.0]> : tensor<4xf32>
  %cst1 = mhlo.constant dense<[[[[1.0, 2.0, 3.0, 4.0], [5.0, 6.0, 7.0, 8.0]]]]> : tensor<1x1x2x4xf32>
  %0 = "mhlo.broadcast_in_dim"(%cst0) <{broadcast_dimensions = dense<3> : tensor<1xi64>}> : (tensor<4xf32>) -> tensor<1x1x2x4xf32>
  %1 = mhlo.multiply %0, %cst1 : tensor<1x1x2x4xf32>

# CHECK-SAME: tf_saved_model.initializer_type = "init_op"
# CHECK-SAME: }
# CHECK-DAG: %[[CST_0:.*]] = "tf.Const"() <{{{.*dense<.*> : tensor<2xi32>.*}}}> {{{.*}}} : () -> tensor<2xi32>
# CHECK: %[[RAND_STD_NORMAL:.*]] = "tf.RandomStandardNormal"(%[[CST_0]])
# CHECK: "tf.AssignVariableOp"(%[[ARG_0]], %[[RAND_STD_NORMAL]]){{.*}}: (tensor<!tf_type.resource<tensor<1x3xf32>>>, tensor<1x3xf32>) -> ()

  // CHECK-NEXT:   s: " 0\n\000\000"
  tf_executor.graph {
    %0:2 = tf_executor.island wraps "tf.Placeholder"() {name = "dummy", dtype = "tfdtype$DT_INT32", value = "\200\n\00\00", listvalue = ["\20\0A"]} : () -> tensor<2xi32>
    tf_executor.fetch
  }
  func.return

  func.func @f(%arg0: tensor<!tf_type.resource<tensor<?xf32>>> {tf_saved_model.bound_input = @v})
  -> (tensor<!tf_type.resource<tensor<?xf32>>> {tf_saved_model.index_path = []})
  attributes {tf_saved_model.exported_names = ["f"]} {
    // CHECK: return
    return %arg0 : tensor<!tf_type.resource<tensor<?xf32>>>
  }