<< "requires crops to be a tensor<2x2>, but got " << crops_type;
  }

  DenseIntElementsAttr crops_attr;
  // Crops are defined as [[crop_top, crop_bottom], [crop_left, crop_right]],
  // and flattened as [crop_top, crop_bottom, crop_left, crop_right]
  llvm::SmallVector<int64_t, 4> crops_values;
  if (matchPattern(op.getCrops(), m_Constant(&crops_attr))) {
    assert(crops_attr.getNumElements() == 4 &&

  // expected-error @+1 {{'tf.BatchToSpace' op requires output height (dimension 1) to be equal to input height (dimension 1) * block_size - crop_top - crop_bottom, but got output height 8, input height 8, crop_top 1, crop_bottom 2, and block_size 2}}
  %0 = "tf.BatchToSpace"(%arg0, %crops) {block_size = 2 : i64} : (tensor<16x8x8x3xf32>, tensor<2x2xi32>) -> tensor<4x8x9x3xf32>
  func.return
}

// -----

dimensions as follows:

    crops = [[crop_top, crop_bottom], [crop_left, crop_right]]}]>:$crops,

    ConfinedAttr<I64Attr, [IntMinValue<2>]>:$block_size
  );

  let results = (outs
    Res<TF_Tensor, [{4-D with shape `[batch, height, width, depth]`, where:

      height = height_pad - crop_top - crop_bottom
      width = width_pad - crop_left - crop_right