%1 = "tf.Conv2D"(%0, %cst) {data_format = "NHWC", device = "", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "SAME", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true} : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<1x3x2x2xf32>

}
// CHECK: func.func @main(%[[ARG:.+]]: tensor<1x3x4x3xf32>) -> tensor<1x3x2x2xf32> {
// CHECK-DAG: %[[CONST_0:.*]] = stablehlo.constant dense<[{{.*}}]> : tensor<2xf32>
// CHECK-DAG: %[[CONST_1:.*]] = stablehlo.constant dense<[{{.*}}]> : tensor<2x3x3x2xf32>
// CHECK-DAG: %[[CONV:.*]] = stablehlo.convolution(%[[ARG]], %[[CONST_1]]) {{.*}} : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<1x3x2x2xf32>

    %0 = "tf.GatherV2"(%arg0, %arg1, %arg2) {attr_map = "0:batch_dims", batch_dims = 0 : i64, device = ""} : (tensor<1024x3x4x3xf32>, tensor<1xi32>, tensor<i32>) -> tensor<1x3x4x3xf32>
    return %0 : tensor<1x3x4x3xf32>
  }

        _stablehlo_module_attrs = {}, _tfl_quant_trait = "fully_quantizable",
        device = ""
      } : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<1x3x4x2xf32>
    return %0 : tensor<1x3x4x2xf32>
  }

  // CHECK: func.func private @qdq_for_conv_weight_per_channel_default(%[[ARG0:.+]]: tensor<1x3x4x3xf32>)
  // CHECK: %[[CST:.+]] = "tf.Const"() <{value = dense<3.000000e-01> : tensor<2x3x3x2xf32>}> : () -> tensor<2x3x3x2xf32>

// -----

func.func @conv_with_non_constant_filter(%arg0: tensor<1x3x4x3xf32>, %arg1: tensor<2x3x3x2xf32>) -> tensor<*xf32> {
  %cst = "tf.Const"() {value = dense<0.000000e+00> : tensor<2xf32>} : () -> tensor<2xf32>
  %0 = "tf.Conv2D"(%arg0, %arg1) {data_format = "NHWC", dilations = [1, 1, 2, 1], explicit_paddings = [], padding = "SAME", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true} : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<*xf32>

    %0 = "quantfork.stats"(%arg0) {layerStats = dense<[1.27501142, 149.824783]> : tensor<2xf32>} : (tensor<1x3x4x3xf32>) -> tensor<1x3x4x3xf32>
    %1 = "tf.PartitionedCall"(%0, %cst_0, %cst) {_tfl_quant_trait = "fully_quantizable", config = "", config_proto = "", device = "", executor_type = "", f = @composite_conv2d_with_bias_and_relu6_fn_10} : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>, tensor<2xf32>) -> tensor<*xf32>

  %0 = "tf.Conv2D"(%output, %cst) <{data_format = "NHWC", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "SAME", strides = [1, 2, 2, 1], use_cudnn_on_gpu = true}> {attr_map = "0:strides,1:use_cudnn_on_gpu,2:padding,3:explicit_paddings,4:dilations", device = ""} : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<1x2x2x2xf32>

  func.func private @quantize_conv_fn(%arg0: tensor<1x3x4x3xf32>) -> tensor<1x3x4x2xf32> attributes {tf._original_func_name = "main_0"} {
    %cst = "tf.Const"() {value = dense<3.00000000e-1> : tensor<2x3x3x2xf32>} : () -> tensor<2x3x3x2xf32>
    %0 = "quantfork.stats"(%arg0) {layerStats = dense<[6.00000000e-6, 9.00000000e-1]> : tensor<2xf32>} : (tensor<1x3x4x3xf32>) -> tensor<1x3x4x3xf32>
    %1 = "tf.XlaCallModule"(%0, %cst) {

func.func private @conv(%input: tensor<1x3x4x3xf32> {tf._user_specified_name = "input_tensor"}) -> tensor<*xf32> attributes {tf._construction_context = "kEagerRuntime", tf._input_shapes = [#tf_type.shape<1x3x4x3>]} {
  %weight = arith.constant dense_resource<__elided__> : tensor<2x3x3x2xf32>
  %bias = arith.constant dense<[7.11401462, 7.05456924]> : tensor<2xf32>

    %2 = "tf.BiasAdd"(%1, %cst_0) {data_format = "NHWC", device = ""} : (tensor<*xf32>, tensor<2xf32>) -> tensor<*xf32>
    func.return %2: tensor<*xf32>
  }
  func.func private @composite_conv2d_fn_1(%arg0: tensor<1x3x4x3xf32>, %arg1: tensor<2x3x3x512xf32>) -> tensor<*xf32> attributes {tf_quant.composite_function} {