def _ReluGradGrad(op, grad):
  x = op.inputs[1]
  return (gen_nn_ops._relu_grad(grad, x), array_ops.zeros(
      shape=array_ops.shape(x), dtype=x.dtype))

def _guided_grad_relu(op, grad):
    return guided_grad(gen_nn_ops._relu_grad(grad, op.outputs[0]))

def _ReluGrad(op, grad):
  return gen_nn_ops._relu_grad(grad, op.outputs[0])

def _GuidedReluGrad(op, grad):
    return tf.where(0. < grad, gen_nn_ops._relu_grad(grad, op.outputs[0]), tf.zeros_like(grad))

def relu_mask(a, dtype=default_dtype):
  """Produces mask of 1s for positive values and 0s for negative values."""
  from tensorflow.python.ops import gen_nn_ops
  ones = tf.ones(a.get_shape(), dtype=dtype)
  return gen_nn_ops._relu_grad(ones, a)

 def GuidedReluGrad(op, grad):
     return tf.where(0. < grad,
                     gen_nn_ops._relu_grad(grad, op.outputs[0]),
                     tf.zeros(grad.get_shape()))

def _ReluGrad(op, grad):
    t = _VerifyTensor(op.inputs[0], op.name, "ReluGrad input is not finite.")
    return gen_nn_ops._relu_grad(grad, t)

 def _guided_relu_grad(op, grad):
     return tf.select(0. < grad, gen_nn_ops._relu_grad(grad, op.outputs[0]), tf.zeros(grad.get_shape()))