func.func @main(%arg0: tensor<10xi32>, %arg1: tensor<10xi32>) -> tensor<10xi32>
attributes {tf.entry_function = {inputs = "input0,input1", outputs = "Add"}} {
  %graph = tf_executor.graph {
    %2:2 = tf_executor.island wraps "tf.Add"(%arg0, %arg1) {T = "tfdtype$DT_INT32", device = ""} : (tensor<10xi32>, tensor<10xi32>) -> tensor<10xi32> loc("Add")
    tf_executor.fetch %2 : tensor<10xi32>
  }
  func.return %graph : tensor<10xi32>
}

  %cst1 = arith.constant dense<2.0> : tensor<1024xf32>
  %cst2 = arith.constant dense<2.0> : tensor<1024xf32>

  %0 = "tfl.fully_connected"(%arg0, %cst0, %cst1) {fused_activation_function = "NONE", keep_num_dims = false, weights_format = "DEFAULT"} : (tensor<128x256000xf32>, tensor<1024x256000xf32>, tensor<1024xf32>) -> tensor<128x1024xf32>

  // CHECK: "tfl.pack"(%arg0, %arg1) <{axis = -2 : i32, values_count = 2 : i32}>
  %0 = "tfl.pack"(%arg0, %arg1) {axis = -2 : i32, values_count = 2 : i32} : (tensor<1x4xi32>, tensor<1x4xi32>) -> tensor<1x2x4xi32>
  func.return %0 : tensor<1x2x4xi32>
}

func.func @packNegInputAxis3(%arg0: tensor<1x4xi32>, %arg1: tensor<1x4xi32>) -> tensor<2x1x4xi32> {

  %cst = "tf.Const"() {value = dense<0> : tensor<1x2xi32>} : () -> tensor<1x2xi32>
  %cst_0 = "tf.Const"() {value = dense<1> : tensor<i32>} : () -> tensor<i32>
  %cst_1 = "tf.Const"() {value = dense<2> : tensor<1xi32>} : () -> tensor<1xi32>
  %cst_2 = "tf.Const"() {value = dense<4> : tensor<1x2xi32>} : () -> tensor<1x2xi32>
  %0 = "tf.SpaceToBatchND"(%arg0, %cst_1, %cst_2) : (tensor<1x128x3xf32>, tensor<1xi32>, tensor<1x2xi32>) -> tensor<2x68x3xf32>

  func.return %0 : tensor<*xf32>
}

func.func @xla_func_0(%arg0: tensor<7xf32>, %arg1: tensor<f32>) -> tensor<10xf32> {
  %cst = "tf.Const"() {value = dense<0> : tensor<1x2xi32>} : () -> tensor<1x2xi32>
  %cst_0 = "tf.Const"() {value = dense<1> : tensor<1xi32>} : () -> tensor<1xi32>
  %cst_1 = "tf.Const"() {value = dense<2> : tensor<1xi32>} : () -> tensor<1xi32>

>, tensor<i8>, tensor<f64>, tensor<!tf_type.resource>)

# CHECK-SUBTYPE: func @main(%arg0: tensor<10xi32>, %arg1: tensor<!tf_type.variant<tensor<10xf32>>>, %arg2: tensor<10xi8>, %arg3: tensor<10xf64>, %arg4: tensor<!tf_type.resource<tensor<10xi32>>>) -> (tensor<10xi32>, tensor<!tf_type.variant<tensor<10xf32>>>, tensor<10xi8>, tensor<10xf64>, tensor<!tf_type.resource<tensor<10xi32>>>)...

  // CHECK: return %[[RESULT]] : tensor<1x4xi32>
  %sizes = "tf.Const"() {value = dense<[1, 4]> : tensor<2xi64>} : () -> (tensor<2xi64>)
  %0 = "tf.Slice"(%arg0, %arg1, %sizes) : (tensor<3x4xi32>, tensor<2xi64>, tensor<2xi64>) -> tensor<1x4xi32>
  func.return %0 : tensor<1x4xi32>
}

// CHECK-LABEL: slice_variable_start_negsize

func.func @main(
    %arg0: tensor<10x10xf32>, %scale: tensor<10xf32>, %zp: tensor<10xi32>
  ) -> tensor<10x10xi8> {
  %0 = "tf.UniformQuantize"(%arg0, %scale, %zp) {
    quantization_axis = 1 : i64,
    quantization_min_val = -128 : i64,
    quantization_max_val = 127 : i64
  } : (tensor<10x10xf32>, tensor<10xf32>, tensor<10xi32>) -> tensor<10x10x!tf_type.qint8>

  // CHECK: %[[CUSTOM_AGGREGATOR_2:.*]], {{.*}}, {{.*}}, {{.*}} = "tf.CustomAggregator"(%[[XLA_CALL_MODULE:.*]])
  // CHECK: %[[SUBGRAPH_2:.*]]:2 = "tf.XlaCallModule"(%[[CUSTOM_AGGREGATOR_2]], %[[SUBGRAPH_1]]) <{Sout = [#tf_type.shape<1024x3>, #tf_type.shape<1024x3>], {{.*}}}> {_entry_function = @_stablehlo_main_0

// fixed upstream).
func.func @broadcast_add(%arg0: tensor<1xi32>, %arg1: tensor<1x2xi32>) -> tensor<1x2xi32> {
  // CHECK-NEXT: %[[LHS_BCAST:.+]] = "mhlo.broadcast_in_dim"(%arg0) <{broadcast_dimensions = dense<1> : tensor<1xi64>}>
  // CHECK-NEXT: mhlo.add %[[LHS_BCAST]], %arg1
  %0 = "tf.AddV2"(%arg0, %arg1) : (tensor<1xi32>, tensor<1x2xi32>) -> tensor<1x2xi32>
  func.return %0: tensor<1x2xi32>
}

// CHECK-LABEL: func @broadcast_multi_dim_add