DCHECK(while_ctx != nullptr);

  // Record 'summed_grads' as the backprop input associated with 'exit_node'
  std::map<Node*, Output>& backprops = while_backprops_[while_ctx];
  DCHECK(backprops.find(exit_node) == backprops.end());
  backprops[exit_node] = summed_grads;

  // Wait until we have all exit nodes' backprops collected before processing
  // the while loop.

  return result;
}

// The backprop loop counter and main backprop loop run in their own execution
// frame (conceptually, the main forward loop and forward loop counter run
// together in a frame, then the backprop loop counter and backprop loop run
// together in a different frame). This returns the frame name to use for the
// backprop while loops.

  // Build new BackPropFilterOp.
  auto loc = backprop.getLoc();
  auto new_backprop = builder.create<TF::Conv2DBackpropFilterOp>(
      loc, new_result_type, input, new_filter_sizes, backprop.getOutBackprop(),
      strides, backprop.getUseCudnnOnGpu(), backprop.getPadding(),
      backprop.getExplicitPaddings(), backprop.getDataFormat(),
      backprop.getDilations());

    absl::Span<AbstractTensorHandle*> outputs) {
  AbstractTensorHandle* loss;
  AbstractTensorHandle* backprop;
  TF_RETURN_IF_ERROR(ops::SparseSoftmaxCrossEntropyWithLogits(
      ctx, inputs[0], inputs[1], &loss, &backprop,
      "SparseSoftmaxCrossEntropyWithLogits"));
  // `gradient_checker` only works with model that returns only 1 tensor.

// computes loss and backprop of the loss with respect to 'features'.
//
// Softmax cross entropy loss is defined as follows:
//
//  loss = Sum(-labels * Log(Exp(features) / Sum(Exp(features)))
//  loss = Sum(-labels * LogSoftmax(features))
//
// Computing gradient of the loss with respect to features gives us,
//
//  backprop = (Exp(features) / Sum(Exp(features))) - labels

// any gradients to be computed).
//
// Finally, we start a backprop stack with a set of tape entries for which we
// have all gradients available. This set usually is a subset of the set of
// targets (not all since targets which have outputs in the tape will not have
// gradients available initially).
//
// Then we repeatedly pop an entry from the stack, run its backprop, and update

      });
    }
    filter_value = filter.getValue();
    mul_value = multiplier.getValue();
    // In MHLO, Conv filter is in HWIO format, Depthwise conv filter is in HW1O
    // format and backprop input conv filter is in HWOI format.
    // Only fuses multiplier if all dimensions other than the out channel
    // dimension are equal to 1.
    if (!TFL::IsDimensionsDegenerateExceptLastOne(

      Add(params_->body_inputs[0], {one, 0}, params_->body_graph, s_);
  ASSERT_EQ(TF_OK, TF_GetCode(s_)) << TF_Message(s_);
  params_->body_outputs[0] = {add, 0};

  ExpectOK();

  // Create backprop graph
  TF_Output grad_output;
  TF_AddGradients(graph_, outputs_.data(), outputs_.size(), inputs_.data(), 1,
                  nullptr, s_, &grad_output);
  ASSERT_EQ(TF_OK, TF_GetCode(s_)) << TF_Message(s_);

example, suppose y = f(x) and we wish to apply a custom function g for backprop
such that dx = g(dy). In Python,

```python
with tf.get_default_graph().gradient_override_map(
    {'IdentityN': 'OverrideGradientWithG'}):
  y, _ = identity_n([f(x), x])

@tf.RegisterGradient('OverrideGradientWithG')
def ApplyG(op, dy, _):
  return [None, g(dy)]  # Do not backprop to f(x).
```
  }];

  let arguments = (ins

        *   `tf.compat.v1.nn.fused_batch_norm` backprop to `offset` when
            `is_training=False`
        *   `tf.image.adjust_contrast` forward
        *   `tf.nn.depthwise_conv2d` backprop to `filter` when not using cuDNN
            convolution
        *   `tf.image.resize` with `method=ResizeMethod.NEAREST` backprop
        *   `tf.math.bincount` - TODO: confirm exception added