llvm::StringMap<Attribute>& identifier_to_attr) {
  ArrayAttr dilations =
      mlir::dyn_cast<ArrayAttr>(identifier_to_attr["dilations"]);
  const int dilation_h = mlir::cast<IntegerAttr>(dilations[1]).getInt();
  const int dilation_w = mlir::cast<IntegerAttr>(dilations[2]).getInt();
  return rewriter.getI64ArrayAttr({dilation_h, dilation_w});
}

Attribute GetPaddingValue(PatternRewriter& rewriter,

      int dilation_i = mlir::cast<IntegerAttr>(dilations[i]).getInt();
      int out_size = tflite::ComputeOutSize(kTfLitePaddingSame, input_size,
                                            filter_size, stride_i, dilation_i);

      int offset = 0;
      int padding_before = tflite::ComputePaddingWithOffset(
          stride_i, dilation_i, input_size, filter_size, out_size, &offset);

                                       "Conv2D op doesn't have valid padding");
  }
  // Make sure dilations are all ones if set.
  const ArrayAttr& dilations =
      op->template getAttrOfType<ArrayAttr>("dilations");
  if (dilations && !TFIntListIsAllOnes(dilations)) {
    return rewriter.notifyMatchFailure(op, "dilations should be all 1");
  }

  if (!TFL::TFTypeIsFloat32Tensor(op.getInput()) &&

    return errors::InvalidArgument(
        "Conv2DBackpropInput grad requires 1 grad input");
  }

  std::vector<int> dilations, strides, explicit_paddings;
  bool use_cudnn_on_gpu;
  std::string data_format, padding;
  TF_RETURN_IF_ERROR(GetNodeAttr(op.node()->attrs(), "dilations", &dilations));
  TF_RETURN_IF_ERROR(GetNodeAttr(op.node()->attrs(), "strides", &strides));
  TF_RETURN_IF_ERROR(

      (NamedAttr<"explicit_paddings"> $explicit_paddings),
      (NamedAttr<"dilations"> $dilations))),
  [(IsNotInLiftedFunc $res)], [], (addBenefit 1)>;

def LiftDepthwiseConv : Pat<
  (TF_DepthwiseConv2dNativeOp:$res $input, $filter, $strides, $padding,
    $explicit_paddings, IsDataFormatNHWC:$data_format, $dilations),
  (LiftAsTFPartitionedCall<"composite_depthwise_conv2d_fn">
    (ArgumentList $input, $filter),

    (TF_CastOp (TF_IdentityOp $filter), $truncate1),
    $strides, $use_cudnn, $padding, $explicit_padding,
    IsDataFormatNHWC:$data_format, $dilations),
  (CreateXLAConvOpFromTFConv2DOp
    $input, $filter, $input_zp, $conv, $strides,
    $dilations, $padding, $explicit_padding),
  [(IsInt8ElementType $input),
   (IsInt8ElementType $filter),
   (IsConstTensor $input_zp),
   (IsConstTensor $filter),

// CHECK-NEXT: %[[CONV2D_0:.*]] = "tf.Conv2D"(%arg0, %arg1)
// CHECK-SAME: data_format = "NHWC", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "SAME", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true
// CHECK-SAME: attr_map = "0:strides,1:use_cudnn_on_gpu,2:padding,3:explicit_paddings,4:dilations"
// CHECK-NEXT: return %[[CONV2D_0]]

// CHECK-LABEL: private @composite_conv2d_fn_1

// CHECK-NEXT: %[[CONV2D_0:.*]] = "tf.Conv2D"(%arg0, %arg1)
// CHECK-SAME: data_format = "NHWC", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "SAME", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true
// CHECK-SAME: attr_map = "0:strides,1:use_cudnn_on_gpu,2:padding,3:explicit_paddings,4:dilations"
// CHECK-NEXT: %[[BIASADD_0:.*]] = "tf.BiasAdd"(%[[CONV2D_0]], %arg2)
// CHECK-NEXT: %[[RELU6_0:.*]] = "tf.Relu6"(%[[BIASADD_0]])

  %3 = "tf.AddV2"(%2, %1): (tensor<15x28x28x1xf32>, tensor<1xf32>) -> tensor<15x28x28x1xf32>
  func.return %2 : tensor<15x28x28x1xf32>

  %cst_0 = "tf.Const"() {value = dense<0.500000e+00> : tensor<2xf32>} : () -> tensor<2xf32>
  %0 = "tf.Conv2D"(%arg0, %cst) {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<1x3x2x2xf32>