%1 = "tf.MatMul"(%arg0, %0) {T = f32, device = "/device:CPU:0", transpose_a = false, transpose_b = false} : (tensor<3x1xf32>, tensor<1x3xf32>) -> tensor<3x3xf32>
  func.return %1 : tensor<3x3xf32>
}

// CHECK-LABEL: func @gpu_device
func.func @gpu_device(%arg0: tensor<3x1xf32>, %arg1: tensor<!tf_type.resource<tensor<1x3xf32>>>) -> tensor<3x3xf32> {

func.func @main(%arg0: tensor<1x4xf32>) -> tensor<1x3xf32> {
  %0 = stablehlo.constant dense<1.0> : tensor<4x3xf32>
  %1 = stablehlo.dot_general %arg0, %0, contracting_dims = [1] x [0], precision = [DEFAULT, DEFAULT] : (tensor<1x4xf32>, tensor<4x3xf32>) -> tensor<1x3xf32>
  return %1 : tensor<1x3xf32>
}
// CHECK: @main(%[[ARG_0:.+]]: tensor<1x4xf32>) -> tensor<1x3xf32>

      time_major = false} : (
        tensor<1x2x3xf32>,
        tensor<1x1xf32>, tensor<1x1xf32>, tensor<1x1xf32>, tensor<1x1xf32>,
        tensor<1x1xf32>, tensor<1x1xf32>, tensor<1x1xf32>, tensor<1x1xf32>,
        none, none, none,
        tensor<3xf32>, tensor<3xf32>, tensor<3xf32>, tensor<3xf32>,
        none, none,
        tensor<1x3xf32>, tensor<1x3xf32>,
        none, none, none, none) -> tensor<1x2x3xf32>

  %1 = "quantfork.stats"(%0) {bitsNum = 8 : i64, layerStats = dense<[-2.0, 2.0]> : tensor<2xf32>, narrowRange = false} : (tensor<2x3xf32>) -> tensor<2x3xf32>
  %2 = stablehlo.convert %1 : (tensor<2x3xf32>) -> (tensor<2x3xf32>)
  func.return %2 : tensor<2x3xf32>
}

// -----

// CHECK-LABEL: func @dot_redundant_stats

  // CHECK-NEXT: [[r1:%.*]] = tfrt_fallback_async.executeop {{.*}} "tf.BiasAdd"([[r0]], [[result]])
  %3 = "tf.BiasAdd"(%2, %0) {T = f32, _output_shapes = ["tfshape$dim { size: 3 } dim { size: 3 }"], data_format = "NHWC", device = "/device:CPU:0"} : (tensor<3x3xf32>, tensor<3xf32>) -> tensor<3x3xf32>

      time_major = false} : (
        tensor<1x2x3xf32>,
        tensor<1x1xf32>, tensor<1x1xf32>, tensor<1x1xf32>, tensor<1x1xf32>,
        tensor<1x1xf32>, tensor<1x1xf32>, tensor<1x1xf32>, tensor<1x1xf32>,
        none, none, none,
        tensor<3xf32>, tensor<3xf32>, tensor<3xf32>, tensor<3xf32>,
        none, none,
        tensor<1x3xf32>, tensor<1x3xf32>,
        none, none, none, none) -> tensor<1x2x3xf32>

  return %outputs_0 : tensor<1x1xf32>
}

// CHECK-LABEL: @_func
func.func private @_func(%arg0: tensor<1x3xf32>, %arg1: tensor<3x1xf32>) -> (tensor<1x1xf32>) {
  %outputs_0 =  "tf.MatMul"(%arg0, %arg1) {transpose_a = false, transpose_b = false} : (tensor<1x3xf32>, tensor<3x1xf32>) -> tensor<1x1xf32>
  return %outputs_0 : tensor<1x1xf32>
}

# CHECK-NEXT: [[R0:%.*]] = "tf.ReadVariableOp"([[ARG1]]) {{{.*}}} : (tensor<!tf_type.resource<tensor<1x3xf32>>>) -> tensor<1x3xf32>
# CHECK-NEXT: [[R1:%.*]] = "tf.MatMul"([[ARG0]], [[R0]]) <{{{.*}}}> {{{.*}}} : (tensor<3x1xf32>, tensor<1x3xf32>) -> tensor<3x3xf32>
# CHECK-NEXT: return [[R1]] : tensor<3x3xf32>


def Test():

  x = tf.constant([[1.0], [1.0], [1.0]])
  y = tf.compat.v1.get_variable(
      name='y',

    %0 = "tf.Div"(%arg0, %cst_3) {device = ""} : (tensor<2x3xf32>, tensor<f32>) -> tensor<2x3xf32>
    %1 = "tf.AddV2"(%0, %cst_1) {device = ""} : (tensor<2x3xf32>, tensor<f32>) -> tensor<2x3xf32>
    %2 = "tf.Maximum"(%1, %cst_1) {device = ""} : (tensor<2x3xf32>, tensor<f32>) -> tensor<2x3xf32>
    %3 = "tf.Minimum"(%2, %cst_5) {device = ""} : (tensor<2x3xf32>, tensor<f32>) -> tensor<2x3xf32>

      %2 = "tf.XlaCallModule"(%arg0, %1, %0) <{Sout = [#tf_type.shape<?x2>], module = "", version = 9 : i64}> {_original_entry_function = "composite_fn_1"} : (tensor<?x2xf32>, tensor<2x2xf32>, tensor<2xf32>) -> tensor<?x2xf32>
      return %2 : tensor<?x2xf32>
    }