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>

  // CHECK: device = "cpu"
  %2 = "tf.Relu"(%1) {T = f32, _output_shapes = ["tfshape$dim { size: 3 } dim { size: 3 }"], device = "cpu"} : (tensor<3x3xf32>) -> tensor<3x3xf32>
  // CHECK: device = "xpu"
  %3 = "tf.Relu"(%2) {T = f32, _output_shapes = ["tfshape$dim { size: 3 } dim { size: 3 }"]} : (tensor<3x3xf32>) -> tensor<3x3xf32>
  func.return %3 : tensor<3x3xf32>
}

module {
  func.func @main(%arg0: tensor<1x2xi32>) -> tensor<1x2x2xi32> {
  %0 = "tfl.custom"(%arg0) {custom_code = "stablehlo.broadcast_in_dim", custom_option = #tfl<const_bytes : "0x62726F6164636173745F64696D656E73696F6E73000201020119010101072C022401">} : (tensor<1x2xi32>) -> tensor<1x2x2xi32>
  func.return %0 : tensor<1x2x2xi32>
  }
}

// CHECK:      module {
// CHECK-NEXT:  func @main(%arg0: tensor<1x2xi32>) -> tensor<1x2x2xi32> {

    return %2 : tensor<1x3xf32>
  }

  func.func private @composite_dot_general_fn(%arg0: tensor<1x2xf32>, %arg1: tensor<2x3xf32>) -> tensor<1x3xf32> attributes {_from_xla_call_module} {
    %0 = stablehlo.dot_general %arg0, %arg1, contracting_dims = [1] x [0] : (tensor<1x2xf32>, tensor<2x3xf32>) -> tensor<1x3xf32>
    return %0 : tensor<1x3xf32>
  }
}

  %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> {

// CHECK-SAME:    : (tensor<!tf_type.string>, tensor<1x3xf32>) -> tensor<1x1xf32>
// CHECK-NEXT:    return [[RES]] : tensor<1x1xf32>
//
module {
  func.func @serving_default(%arg0: tensor<1x3xf32>) -> tensor<1x1xf32> {
    %0 = "tf.VarHandleOp"() <{container = "", shared_name = "y"}> : () -> tensor<!tf_type.resource<tensor<3x1xf32>>>

# 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',

  %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

  %0 = "tf.Const"() {value = dense<[0, 1, 4, 2, 3]> : tensor<5xi32>} : () -> tensor<5xi32>
  %1 = "tf.Const"() {value = dense<[[0, 0], [0, 0], [0, 0], [1, 1], [1, 1]]> : tensor<5x2xi32>} : () -> tensor<5x2xi32>
  %2 = "tf.Transpose"(%arg0, %0) : (tensor<1x2x4x4x3xf32>, tensor<5xi32>) -> tensor<1x2x3x4x4xf32>
  %3 = "tf.Pad"(%2, %1) : (tensor<1x2x3x4x4xf32>, tensor<5x2xi32>) -> tensor<1x2x3x6x6xf32>

    %cst = arith.constant dense<0.000000e+00> : tensor<5xf32>
    %cst_0 = arith.constant dense<1.0> : tensor<5x3xf32>
    %0 = "tfl.fully_connected"(%arg0, %cst_0, %cst) {fused_activation_function = "NONE", keep_num_dims = false, weights_format = "DEFAULT"} : (tensor<?x3xf32>, tensor<5x3xf32>, tensor<5xf32>) -> tensor<?x5xf32>
    func.return %0, %0 : tensor<?x5xf32>, tensor<?x5xf32>
  }