broadcast_shape = tf.shape(y)
  input_value_shape = tf.shape(op.inputs[2])
  _, reduction_axes = tf.raw_ops.BroadcastGradientArgs(
      s0=broadcast_shape, s1=input_value_shape)
  updates_grad_reshaped = tf.reduce_sum(
      grad, axis=reduction_axes, keepdims=True)
  bias_grad = tf.reshape(updates_grad_reshaped, input_value_shape)

  dilations = [1, op.get_attr('dilation_w'), op.get_attr('dilation_h'), 1]

        /*asymmetric_quantize_inputs=*/mlir::BoolAttr());
    rewriter.replaceOp(bmm_op, {fc_op.getResult(0)});

    return success();
  };
};

// Converts batch_matmul operation with a ones tensor to a reduce_sum.
struct ConvertBatchMatMulOpToReduceSum
    : public OpRewritePattern<TFL::BatchMatMulOp> {
  using OpRewritePattern<TFL::BatchMatMulOp>::OpRewritePattern;
  LogicalResult matchAndRewrite(TFL::BatchMatMulOp bmm_op,

                 absl::Span<AbstractTensorHandle*> grad_inputs) override {
    /* Given upstream grad U and a BiasAdd: A + bias, the gradients are:
     *
     *    dA = U
     *    dbias = reduceSum(U, dims = channel_dim)
     */

    AbstractTensorHandle* upstream_grad = grad_outputs[0];
    DCHECK(upstream_grad);

    // Recover data format from forward pass for gradient.
    std::string data_format;