// CHECK-SAME: edge_padding_high = dense<2> : tensor<1xi64>,
  // CHECK-SAME: edge_padding_low = dense<1> : tensor<1xi64>,
  // CHECK-SAME: interior_padding = dense<0> : tensor<1xi64>
  %1 = "tf.PadV2"(%arg0, %padding, %arg1) : (tensor<3xf32>, tensor<1x2xi64>, tensor<f32>) -> tensor<6xf32>
  func.return %1 : tensor<6xf32>
}

// -----

// CHECK-LABEL: func @padv2_2D

// CHECK-SAME:                      %[[VAL_1:.*]]: tensor<f32>) -> tensor<11x131xf32> {
// CHECK:           %[[VAL_2:.*]] = arith.constant dense<{{\[\[}}1, 2], [0, 3]]> : tensor<2x2xi64>
// CHECK:           %[[VAL_3:.*]] = "tf.PadV2"(%[[VAL_0]], %[[VAL_2]], %[[VAL_1]]) : (tensor<8x128xf32>, tensor<2x2xi64>, tensor<f32>) -> tensor<11x131xf32>
// CHECK:           return %[[VAL_3]] : tensor<11x131xf32>
// CHECK:         }

  return DenseIntElementsAttr::get(
      tensorflow::GetTypeFromTFTensorShape({shape[0]}, element_type), values);
}

// Returns interior padding to use in HLO Pad op based on the TensorFlow padding
// in TensorFlow PadV2 op.
static DenseIntElementsAttr GetInteriorPadding(ElementsAttr tf_padding) {
  auto length = tf_padding.getShapedType().getShape()[0];
  auto element_type = IntegerType::get(tf_padding.getContext(), 64);