def preprocess_device_grads()

in scripts/tf_cnn_benchmarks/variable_mgr.py [0:0]

• 21 lines of code
• 13 McCabe index (conditional complexity)

  def preprocess_device_grads(self, device_grads):
    compact_grads = (self.benchmark_cnn.params.use_fp16 and
                     self.benchmark_cnn.params.compact_gradient_transfer)
    defer_grads = (self.benchmark_cnn.params.variable_consistency == 'relaxed')

    grads_to_reduce = [[g for g, _ in grad_vars] for grad_vars in device_grads]
    algorithm = batch_allreduce.algorithm_from_params(self.benchmark_cnn.params)
    reduced_grads, self._warmup_ops = algorithm.batch_all_reduce(
        grads_to_reduce, self.benchmark_cnn.params.gradient_repacking,
        compact_grads, defer_grads, self.benchmark_cnn.params.xla_compile)
    if self.benchmark_cnn.enable_auto_loss_scale:
      # Check for infs or nans
      is_finite_list = []
      with tf.name_scope('check_for_inf_and_nan'):
        for tower_grads in reduced_grads:
          with tf.colocate_with(tower_grads[0]):
            # TODO(tanmingxing): Create fused op that takes in a list of tensors
            # as input and returns scalar boolean True if there are any
            # infs/nans.
            is_finite_list.append(tf.reduce_all(
                [tf.reduce_all(tf.is_finite(g)) for g in tower_grads]))
        self.grad_has_inf_nan = tf.logical_not(tf.reduce_all(is_finite_list))
    reduced_device_grads = [[
        (g, v) for g, (_, v) in zip(grads, grad_vars)
    ] for grads, grad_vars in zip(reduced_grads, device_grads)]
    return self.benchmark_cnn.devices, reduced_device_grads