%w = arith.constant dense<127.0> : tensor<512x12xf32>
  %b = arith.constant dense<0.0> : tensor<512xf32>
  %fc = "tfl.fully_connected"(%0, %w, %b) {fused_activation_function = "NONE", keep_num_dims = false, weights_format = "DEFAULT"} : (tensor<1x224x224x3xf32>, tensor<512x12xf32>, tensor<512xf32>) -> tensor<1x112x112x512xf32>

  module {
    func.func @main(%arg0: tensor<2x2x2xf32>, %arg1: tensor<2x2x2xf32>) -> tensor<2x2x2xf32> attributes {_from_xla_call_module} {
      %0 = stablehlo.dot_general %arg0, %arg1,
        batching_dims = [0] x [0],
        contracting_dims = [2] x [1],
        precision = [DEFAULT, DEFAULT]
      : (tensor<2x2x2xf32>, tensor<2x2x2xf32>) -> tensor<2x2x2xf32>
        return %0 : tensor<2x2x2xf32>
    }
  }
)mlir";

  %1 = "tfl.add"(%cst_2, %cst_1) {fused_activation_function = "NONE"} : (tensor<2x2x2xf32>, tensor<  2x2xf32>) -> tensor<2x2x2xf32>
  %2 = "tfl.add"(%cst_0, %cst_2) {fused_activation_function = "NONE"} : (tensor<    2xf32>, tensor<2x2x2xf32>) -> tensor<2x2x2xf32>

  func.return %0, %1, %2 : tensor<2x2xf32>, tensor<2x2x2xf32>, tensor<2x2x2xf32>

    %2 = "tf.XlaCallModule"(%arg0, %1, %0) <{Sout = [#tf_type.shape<?x2>], module = "", version = 9 : i64}> {_entry_function = @composite_dot_general_with_bias_and_relu6_dynamic_fn_2, _original_entry_function = "composite_dot_general_with_bias_and_relu6_dynamic_fn_2", _tfl_quant_trait = "fully_quantizable"} : (tensor<?x2xf32>, tensor<2x2xf32>, tensor<2xf32>) -> tensor<?x2xf32>

}

// CHECK-LABEL: prepareAdd
func.func @prepareAdd(%arg0: tensor<2x2xf32>) -> tensor<2x2xf32> {
  %cst = arith.constant dense<[[0.0, 1.0], [2.0, 255.0]]> : tensor<2x2xf32>
  %add = "tfl.add"(%arg0, %cst) {fused_activation_function="NONE"} : (tensor<2x2xf32>, tensor<2x2xf32>) -> tensor<2x2xf32>
  func.return %add : tensor<2x2xf32>

    %7 = stablehlo.slice %6 [1:3, 2:4] : (tensor<3x4xf32>) -> tensor<2x2xf32>
    %8 = "quantfork.qcast"(%7) {volatile} : (tensor<2x2xf32>) -> tensor<2x2x!quant.uniform<i8:f32, 0.13170163023705575:-1>>
    %9 = "quantfork.dcast"(%8) : (tensor<2x2x!quant.uniform<i8:f32, 0.13170163023705575:-1>>) -> tensor<2x2xf32>
    return %9 : tensor<2x2xf32>
  }

  // CHECK: quantized_dot_general_fn_1

  // CHECK: %[[v0:.*]] = "tf.Transpose"(%arg0, %[[cst]]) : (tensor<2x5x7xf32>, tensor<3xi32>) -> tensor<5x7x2xf32>
  // CHECK: %[[v1:.*]] = "tf.Transpose"(%arg1, %[[cst_0]]) : (tensor<5x3x2xf32>, tensor<3xi32>) -> tensor<5x2x3xf32>
  // CHECK: %[[v2:.*]] = "tf.BatchMatMulV2"(%[[v0]], %[[v1]]) <{adj_x = false, adj_y = false}> : (tensor<5x7x2xf32>, tensor<5x2x3xf32>) -> tensor<5x7x3xf32>

}

// CHECK-LABEL: QuantizeConcat
func.func @QuantizeConcat(tensor<1x2xf32>, tensor<1x2xf32>) -> tensor<2x2x!quant.uniform<u8:f32, 1.000000e-01:128>> {
^bb0(%arg0: tensor<1x2xf32>, %arg1: tensor<1x2xf32>):
  %0 = "tfl.concatenation"(%arg0, %arg1) {axis = 0 : i32, fused_activation_function = "NONE"} : (tensor<1x2xf32>, tensor<1x2xf32>) -> tensor<2x2xf32>

  %w = arith.constant dense<127.0> : tensor<512x12xf32>
  %b = arith.constant dense<0.0> : tensor<512xf32>
  %fc = "tfl.fully_connected"(%arg0, %w, %b) {fused_activation_function = "NONE", keep_num_dims = false, weights_format = "DEFAULT"} : (tensor<1x224x224x3xf32>, tensor<512x12xf32>, tensor<512xf32>) -> tensor<1x112x112x512xf32>
  func.return %fc : tensor<1x112x112x512xf32>

  %0 = "tf_device.cluster_func"(%arg0, %arg1) {func = @_func, use_spmd_for_xla_partitioning = true, use_tpu = true, num_cores_per_replica = 2 : i64} : (tensor<2x4xf32>, tensor<4x2xf32>) -> tensor<2x2xf32>
  %1:2 = "tf.TPUPartitionedOutputV2"(%0) {device = "", partition_dims = [2, 1]} : (tensor<2x2xf32>) -> (tensor<1x2xf32>, tensor<1x2xf32>)