%2 = "tf.Cast"(%identity) {Truncate = false} : (tensor<*xi8>) -> tensor<*xf32>
    %3 = "tf.Conv3D"(%input, %2) {
      padding = "VALID", strides = [1, 1, 1, 1, 1],
      attr_map = "strides:0,padding:1,dilations:2"
    } : (tensor<*xf32>, tensor<*xf32>) -> tensor<*xf32>
    func.return %3 : tensor<*xf32>
  }

  func.func private @internal_batch_matmul_fn(

    %2 = "tf.Conv2D"(%1, %0) {attr_map = "0:strides,1:use_cudnn_on_gpu,2:padding,3:explicit_paddings,4:dilations", 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<*xf32>

  %bias = arith.constant dense<3.0> : tensor<16xf32>
  %value = arith.constant dense<4.0> : tensor<16xf32>
  %0 = "tf.Conv2D"(%arg, %filter) {T = "tfdtype$DT_FLOAT", data_format = "NHWC", dilations = [1, 2, 3, 1], padding = "SAME", strides = [1, 4, 5, 1]} : (tensor<256x32x32x3xf32>, tensor<3x3x3x16xf32>) -> tensor<256x8x7x16xf32>

  %rst = "tf.Conv2D"(%arg, %fq) {T = "tfdtype$DT_FLOAT", data_format = "NHWC", dilations = [1, 2, 3, 1], padding = "SAME", strides = [1, 4, 5, 1]} : (tensor<256x32x32x3xf32>, tensor<3x3x3x16xf32>) -> tensor<256x8x7x16xf32>
  func.return %rst : tensor<256x8x7x16xf32>

                          (Arith_ConstantOp:$filter F32ElementsAttr:$filter_value),
                          $strides, $use_cudnn, $padding, $explicit_padding,
                          IsDataFormatNHWC:$data_format, $dilations),
                     (Arith_ConstantOp:$multiplier F32ElementsAttr:$mul_value)),
// TODO(karimnosseir): Add check for $conv is of rank 4.
      (TF_Conv2DOp $input,

  // Compute in NHWC
  %2 = "tf.Conv2D"(%1, %arg1)
    {
      data_format = "NHWC",
      padding = "SAME",
      strides = [1, 1, 1, 1],
      dilations = [1, 1, 1, 1]
    } : (tensor<1x32x32x3xf32>, tensor<1x1x3x8xf32>) -> tensor<1x32x32x8xf32>

  // Convert result back: NHWC -> NCHW
  %3 = "tf.Const"() {value = dense<[0, 3, 1, 2]> : tensor<4xi32>} : () -> tensor<4xi32>

  %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>

  %cst = "tf.Const"() {value = dense<3.000000e+00> : tensor<2x3x512x512xf32>} : () -> tensor<2x3x512x512xf32>
  %0 = "tf.Conv2D"(%arg0, %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<1x3x4x512xf32>, tensor<2x3x512x512xf32>) -> tensor<*xf32>
  func.return %0 : tensor<*xf32>

  %conv = "tf.Conv2D"(%dq_input, %dq_weight) {attr_map = "0:strides,1:use_cudnn_on_gpu,2:padding,3:explicit_paddings,4:dilations", data_format = "NHWC", device = "", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "VALID", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true} : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<*xf32>

                                     Value input, Value filter,
                                     int8_t input_zp_value, ArrayAttr strides,
                                     ArrayAttr dilations,
                                     StringAttr conv_padding,
                                     ArrayAttr explicit_paddings,
                                     Value &padding, int num_dims = 4);