version = 5 : i64
    } : (tensor<4x3x6x5xf32>, tensor<4x3x5x2xf32>) -> tensor<4x3x6x2xf32>
    return %0 : tensor<4x3x6x2xf32>
  }
  // CHECK: func.func private @qdq_for_dot_general_weight_per_channel_default(%[[ARG0:.+]]: tensor<4x3x6x5xf32>)
  // CHECK: %[[CST:.+]] = "tf.Const"() <{value = dense<3.000000e-01> : tensor<4x3x5x2xf32>}> : () -> tensor<4x3x5x2xf32>

// CHECK-SAME:                         %[[VAL_1:.*]]: tensor<4x4x2x2xf32>) -> tensor<1x512x512x2xf32> {
// CHECK:           %[[VAL_3:.*]] = "tf.Const"() <{value = dense<[0, 1]> : tensor<2xi64>}> : () -> tensor<2xi64>
// CHECK:           %[[VAL_4:.*]] = "tf.ReverseV2"(%[[VAL_1]], %[[VAL_3]]) : (tensor<4x4x2x2xf32>, tensor<2xi64>) -> tensor<4x4x2x2xf32>

// CHECK-DAG: %[[BROADCAST:.*]] = stablehlo.broadcast_in_dim %[[CONST_0]], dims = [3] : (tensor<2xf32>) -> tensor<1x3x2x2xf32>
// CHECK-DAG: %[[ADD:.*]] = stablehlo.add %[[CONV]], %[[BROADCAST]] : tensor<1x3x2x2xf32>
// CHECK-DAG: %[[RELU6:.*]] = stablehlo.clamp %[[CONST_2]], %[[ADD]], %[[CONST_1]] : (tensor<f32>, tensor<1x3x2x2xf32>, tensor<f32>) -> tensor<1x3x2x2xf32>

    %2 = "tf.Relu"(%1) {device = ""} : (tensor<1x3x2x2xf32>) -> tensor<1x3x2x2xf32>
    %3 = "tf.FakeQuantWithMinMaxArgs"(%2) {device = "", max = 4.000000e-01 : f32, min = -3.000000e-01 : f32, narrow_range = false, num_bits = 8 : i64} : (tensor<1x3x2x2xf32>) -> tensor<1x3x2x2xf32>

    %6 = "quantfork.qcast"(%5) : (tensor<1x3x2x2xf32>) -> tensor<1x3x2x2x!quant.uniform<i8:f32, 0.0054901962186775953:-19>>
    %7 = "quantfork.dcast"(%6) : (tensor<1x3x2x2x!quant.uniform<i8:f32, 0.0054901962186775953:-19>>) -> tensor<1x3x2x2xf32>
    return %7 : tensor<1x3x2x2xf32>
  }

        : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<1x3x2x2xf32> loc(fused["Conv2D:", "test_opt_out"])
    %2 = "tf.BiasAdd"(%1, %cst_0) : (tensor<1x3x2x2xf32>, tensor<2xf32>) -> tensor<1x3x2x2xf32> loc(fused["BiasAdd:", "model/bias_add"])
    %3 = "tf.IdentityN"(%2) {device = ""} : (tensor<1x3x2x2xf32>) -> tensor<1x3x2x2xf32>
    return %3 : tensor<1x3x2x2xf32>
  }
}

// CHECK-LABEL: func @conv2d_with_bias_disable_quantization

  %2 = "tf.Cast"(%1) {Truncate = false} : (tensor<1x3x2x2xbf16>) -> tensor<1x3x2x2xf32>
  %3 = "tf.IdentityN"(%2) {device = ""} : (tensor<1x3x2x2xf32>) -> tensor<1x3x2x2xf32>
  return %3 : tensor<1x3x2x2xf32>
}

// CHECK: func @cast_bf16_conv_to_fp32

  %2 = "tf.Cast"(%1) {Truncate = false} : (tensor<1x3x2x2xbf16>) -> tensor<1x3x2x2xf32>
  %3 = "tf.IdentityN"(%2) {device = ""} : (tensor<1x3x2x2xf32>) -> tensor<1x3x2x2xf32>
  return %3 : tensor<1x3x2x2xf32>
// CHECK: tf.Conv2D
// CHECK-SAME: loc("Model/conv2d")
}

func.func @conv2d_with_valid_loc(%arg0: tensor<1x3x4x3xf32>) -> (tensor<1x3x2x2xf32>) {

  %1 = "tf.Sub"(%0, %cst_0) {data_format = "NHWC"} : (tensor<1x3x2x2xf32>, tensor<1x1x1x2xf32>) -> tensor<1x3x2x2xf32>
  %2 = "tf.Mul"(%1, %cst_1) : (tensor<1x3x2x2xf32>, tensor<1x1x1x2xf32>) -> tensor<1x3x2x2xf32>
  %3 = "tf.AddV2"(%2, %cst_2) : (tensor<1x3x2x2xf32>, tensor<1x1x1x2xf32>) -> tensor<1x3x2x2xf32>
  func.return %3 : tensor<1x3x2x2xf32>
}

  %3 = "tf.BiasAdd"(%2, %cst_0) {data_format = "NHWC", device = ""} : (tensor<1x3x2x2xf32>, tensor<2xf32>) -> tensor<1x3x2x2xf32>
  %4 = "tf.Relu"(%3) {device = ""} : (tensor<1x3x2x2xf32>) -> tensor<1x3x2x2xf32>
  %5 = "quantfork.qcast"(%4) : (tensor<1x3x2x2xf32>) -> tensor<1x3x2x2x!quant.uniform<i8:f32, 0.0027450981093387976:-19>>