for (size_t i = 0; i < num_spatial_dims; ++i) {
      // In some cases the total padding is odd, so we have 1 leftover, which is
      // why below we check pad_delta > 1.
      int64_t pad_delta = std::abs(padding[2 * i] - padding[2 * i + 1]);
      if (pad_delta > 1) {
        return false;
      }
      int64_t stride = strides[i + 1];
      int64_t input_size = mlir::cast<ShapedType>(conv_op.getLhs().getType())

		// corresponding immediate field is 12 bits wide
		return tgt & 0xfff
	}

	pageDelta := (tgt >> 12) - (pc >> 12)
	if tgt&0xfff >= 0x800 {
		// adjust for sign-extended addition of the low bits
		pageDelta += 1
	}
	// corresponding immediate field is 20 bits wide
	return pageDelta & 0xfffff

    // CHECK: [[SQRT_NEW_V_EPSILON:%.*]] = "tf.AddV2"([[SQRT_NEW_V]], [[EPSILON]])
    // CHECK: [[VAR_DELTA:%.*]] = "tf.Div"([[ALPHA_NEW_M]], [[SQRT_NEW_V_EPSILON]])
    // CHECK: [[OLD_VAR:%.*]] = "tf.ReadVariableOp"([[VAR_HANDLE]]) : (tensor<*x!tf_type.resource<tensor<*xf32>>>) -> tensor<*xf32>
    // CHECK: [[NEW_VAR:%.*]] = "tf.Sub"([[OLD_VAR]], [[VAR_DELTA]])
    // CHECK: "tf.AssignVariableOp"([[VAR_HANDLE]], [[NEW_VAR]])

executed.
  }];

  let arguments = (ins
    Arg<TF_Float32Tensor, [{Value of parameters used in the Adadelta optimization algorithm.}]>:$parameters,
    Arg<TF_Float32Tensor, [{Value of accumulators used in the Adadelta optimization algorithm.}]>:$accumulators,
    Arg<TF_Float32Tensor, [{Value of updates used in the Adadelta optimization algorithm.}]>:$updates,

    DefaultValuedOptionalAttr<I64Attr, "-1">:$table_id,

    *   Added `tf.keras.optimizers.experimental.Optimizer`. The reworked
        optimizer gives more control over different phases of optimizer calls,
        and is easier to customize. We provide Adam, SGD, Adadelta, AdaGrad and
        RMSprop optimizers based on
        `tf.keras.optimizers.experimental.Optimizer`. Generally the new
        optimizers work in the same way as the old ones, but support new