Failed to convert mhlo dialect to Linalg dialect

[wwwwcu]

I want to convert the conv2d operator from mhlo to linalg dialect. It seems like that it can't process the situation when the layout of conv2d is NCHW.
The mhlo dialect of conv2d.mlir is shown below:

module  {
  func @main(%arg0: tensor<64x3x7x7xf32>, %arg1: tensor<1x3x224x224xf32>) -> tuple<tensor<1x64x112x112xf32>> {
    %0 = "mhlo.convolution"(%arg1, %arg0) {batch_group_count = 1 : i64, dimension_numbers = {input_batch_dimension = 0 : i64, input_feature_dimension = 1 : i64, input_spatial_dimensions = dense<[2, 3]> : tensor<2xi64>, kernel_input_feature_dimension = 1 : i64, kernel_output_feature_dimension = 0 : i64, kernel_spatial_dimensions = dense<[2, 3]> : tensor<2xi64>, output_batch_dimension = 0 : i64, output_feature_dimension = 1 : i64, output_spatial_dimensions = dense<[2, 3]> : tensor<2xi64>}, feature_group_count = 1 : i64, lhs_dilation = dense<1> : tensor<2xi64>, padding = dense<3> : tensor<2x2xi64>, precision_config = ["DEFAULT", "DEFAULT"], rhs_dilation = dense<1> : tensor<2xi64>, window_strides = dense<2> : tensor<2xi64>} : (tensor<1x3x224x224xf32>, tensor<64x3x7x7xf32>) -> tensor<1x64x112x112xf32>
    %1 = "mhlo.tuple"(%0) : (tensor<1x64x112x112xf32>) -> tuple<tensor<1x64x112x112xf32>>
    return %1 : tuple<tensor<1x64x112x112xf32>>
  }
}

I use "./mlir-hlo-opt ./conv2d.mlir -hlo-legalize-to-linalg -o conv2d_linalg.mlir" to do the transformation.
I traced the workflow and found out when it checks "HasCanonicalDimensionNumbers()", it returns false.
legalize_to_linalg.cc

  if (dimension_numbers.input_batch_dimension().getInt() != 0 ||
      dimension_numbers.input_feature_dimension().getInt() !=
          (input_spatial_rank + 1)) {
    return false;
  }