attrs=['act: {"", "RELU", "RELU6", "TANH"} = ""'],
    derived_attrs=['T: {float, int8}'],
    outputs=['o: T'])
def _composite_fully_connected(input_, filter_, bias, act):
  res = tf.raw_ops.MatMul(
      a=input_, b=filter_, transpose_a=False, transpose_b=True)
  res = tf.raw_ops.Add(x=res, y=bias)
  if act == 'RELU':
    return tf.raw_ops.Relu(features=res)
  elif act == 'RELU6':
    return tf.raw_ops.Relu6(features=res)

// CHECK-NEXT:       inputs: [ 0, 1 ],
// CHECK-NEXT:       outputs: [ 2 ],
// CHECK-NEXT:       builtin_options_type: SubOptions,
// CHECK-NEXT:       builtin_options: {
// CHECK-NEXT:         fused_activation_function: RELU6
// CHECK-NEXT:       }
// CHECK-NEXT:     }, {
// CHECK-NEXT:       opcode_index: 1,
// CHECK-NEXT:       inputs: [ 3, 2 ],
// CHECK-NEXT:       outputs: [ 4 ],
// CHECK-NEXT:       builtin_options_type: AddOptions,

// CHECK-NEXT:       inputs: [ 0, 1 ],
// CHECK-NEXT:       outputs: [ 2 ],
// CHECK-NEXT:       builtin_options_type: SubOptions,
// CHECK-NEXT:       builtin_options: {
// CHECK-NEXT:         fused_activation_function: RELU6
// CHECK-NEXT:       }
// CHECK-NEXT:     }, {
// CHECK-NEXT:       opcode_index: 1,
// CHECK-NEXT:       inputs: [ 3, 2 ],
// CHECK-NEXT:       outputs: [ 4 ],
// CHECK-NEXT:       builtin_options_type: AddOptions,

      # Check activation functions are explicitly present.
      # If present the last op before return should be stablehlo.clamp for relu6
      # and stablehlo.maximum for relu.
      if activation_fn is nn_ops.relu6:
        self.assertRegex(module_str, r'stablehlo.clamp.*\n.*return')
      elif activation_fn is nn_ops.relu:
        self.assertRegex(module_str, r'stablehlo.maximum.*\n.*return')
    else:

  %1 = "tfl.mul"(%0, %arg2) {tac.device = "GPU", fused_activation_function = "RELU6", tac.inference_type = "FLOAT"} : (tensor<1xf32>, tensor<1xf32>) -> tensor<1xf32>

    %1 = "tf.BiasAdd"(%0, %arg2) {data_format = "NHWC", device = ""} : (tensor<*xf32>, tensor<2xf32>) -> tensor<*xf32>
    %2 = "tf.Relu6"(%1) : (tensor<*xf32>) -> tensor<*xf32>
    func.return %2 : tensor<*xf32>
  }

// CHECK-NEXT:       inputs: [ 0, 1 ],
// CHECK-NEXT:       outputs: [ 2 ],
// CHECK-NEXT:       builtin_options_type: SubOptions,
// CHECK-NEXT:       builtin_options: {
// CHECK-NEXT:         fused_activation_function: RELU6
// CHECK-NEXT:       }
// CHECK-NEXT:     }, {
// CHECK-NEXT:       opcode_index: 1,
// CHECK-NEXT:       inputs: [ 3, 2 ],
// CHECK-NEXT:       outputs: [ 4 ],
// CHECK-NEXT:       builtin_options_type: AddOptions,

      {"quantized_ops": ["${main_op}", "BiasAdd", "Relu"], "act_func": "internal_requantize_and_relu_fn", "output_type": "!tf_type.qint8"},
      {"quantized_ops": ["${main_op}", "BiasAdd", "Relu6"], "act_func": "internal_requantize_and_relu6_fn", "output_type": "!tf_type.qint8"},
    ]

def QuantActRangeRelu6Pattern :
  Pattern<
    (TFR_TFRQuantActRangeOp
     (TFR_ConstOp HasStringAttr<"RELU6">:$act),
     (ConstantLikeMatcher F32Attr:$scale),
     (ConstantLikeMatcher I64Attr:$zp)),
    [(TFR_ConstantTensorOp (Arith_ConstantOp (Quantize<"0.0f"> $scale, $zp))),
     (TFR_ConstantTensorOp (Arith_ConstantOp (Quantize<"6.0f"> $scale, $zp)))]>;


def QuantActRangeReluN1To1Pattern :
  Pattern<

  // CHECK: "RELU"
  %1 = tfl.add %arg0, %arg1 {fused_activation_function = "RELU"} : tensor<4xi32>
  // CHECK: "RELU_N1_TO_1"
  %2 = tfl.add %arg0, %arg1 {fused_activation_function = "RELU_N1_TO_1"} : tensor<4xi32>
  // CHECK: "RELU6"
  %3 = tfl.add %arg0, %arg1 {fused_activation_function = "RELU6"} : tensor<4xi32>
  // CHECK: "TANH"